Assembly Processes under Severe Abiotic Filtering: Adaptation Mechanisms of Weed Vegetation to the Gradient of Soil Constraints

Questions Effects of soil on vegetation patterns are commonly obscured by other environmental factors; clear and general relationships are difficult to find. How would community assembly processes be affected by a substantial change in soil characteristics when all other relevant factors are held constant? In particular, can we identify some functional adaptations which would underpin such soil-induced vegetation response? Location Eastern Serbia: fields partially damaged by long-term and large-scale fluvial deposition of sulphidic waste from a Cu mine; subcontinental/submediterranean climate. Methods We analysed the multivariate response of cereal weed assemblages (including biomass and foliar analyses) to a strong man-made soil gradient (from highly calcareous to highly acidic, nutrient-poor soils) over short distances (field scale). Results The soil gradient favoured a substitution of calcicoles by calcifuges, and an increase in abundance of pseudometallophytes, with preferences for Atlantic climate, broad geographical distribution, hemicryptophytic life form, adapted to low-nutrient and acidic soils, with lower concentrations of Ca, and very narrow range of Cu concentrations in leaves. The trends of abundance of the different ecological groups of indicator species along the soil gradient were systematically reflected in the maintenance of leaf P concentrations, and strong homeostasis in biomass N:P ratio. Conclusion Using annual weed vegetation at the field scale as a fairly simple model, we demonstrated links between gradients in soil properties (pH, nutrient availability) and floristic composition that are normally encountered over large geographic distances. We showed that leaf nutrient status, in particular the maintenance of leaf P concentrations and strong homeostasis of biomass N:P ratio, underpinned a clear functional response of vegetation to mineral stress. These findings can help to understand assembly processes leading to unusual, novel combinations of species which are typically observed as a consequence of strong environmental filtering, as for instance on sites affected by industrial activities.


Introduction
, little is known about adaptation mechanisms that underpin this phenomenon.
We present the results of a survey undertaken on an exceptional model locality in Eastern Serbia, where large-scale and long-term fluvial deposition of pyrite-rich tailings from a copper (Cu) mine has caused a remarkable change of alluvial soils from the unaffected calcareous, towards the acidic, nutrient-poor, Cu-enriched soils over very short distances (field scale; [16], [17]). The major research question was: how would a model system of cereal weed assemblages (characterized by species abundances, life forms, Ellenberg indicator values, chorotype and leaf element concentrations) respond to a substantial change in soil characteristics if all other factors (altitude, climate, season, year, agricultural practices, surrounding vegetation, and landscape context) are held constant. This study aimed to explore the role of several ecophysiological adaptations in explaining a response of spontaneous vegetation to a complex gradient of soil properties.

Research locality
Since the 1940 s, the alluvial, strongly calcareous soils in the lower course of the Timok river (Serbia) have been drastically altered by the long-term release of mining waste from the Bor copper mines (highly sulphidic copper tailings) directly into the local river system (for details see [17]). More than 10 000 ha of arable land is permanently lost for agriculture, and large additional areas have been severely degraded, most prominently at meander positions where the Timok flow slows down and creates a braided pattern of lateral river channels. The fluvial deposition of mining waste over the arable fields (via regular floods) has brought about clear and consistent gradients in soil properties and cereal crop responses along the transects perpendicular to the water channel; these gradients were strongly indicated by the concentrations of total sulphur (S tot ) in soil [16], [17]. This pollution has created multiple constraints for plant growth: nutrient deficiency (primarily P and micronutrients), excessive concentrations of available Cu, and, as a result of the oxidative weathering of deposited pyrite, very low pH (about 4) and excessive concentrations of extractable Al. The research area is under both sub-continental and sub-Mediterranean climatic influences ( Figure  S1). Outside of the fluvial influence, the thermpohilous varieties of the climazonal Quercion frainetto-cerridis Rudski 1940Rudski (1949 forests are often replaced by more xerothermic extrazonal vegetation of the Quercion pubescentis-petraeae Br.-Bl. 1931 alliance [18] because of carstic geology, dissected terrain and shallow calcareous soils on steep slopes. A major association of winter cereal weeds in the lowlands of the Timok valley has been described as Consolida regalis-Polygonum aviculare ass. var. lathyrethosum aphacae Kojić 1973; this therophytic community belongs to the Caucalidion R.Tx.1950 alliance and is characterized by the dominance of sub-Mediterranean and Eurasian floristic elements [19].

Field trials
The survey included 26 privately owned fields of winter cereals (18 wheat and 8 fodder barley fields), partially damaged by waste deposition, which were located within the 5 km radius from the N 44˚04340, E 22˚319100 (about 70 km downstream from the pollution source). No endangered or protected species were sampled. No permission was required for this field study according to the Serbian legislation. The consent to take plant and soil samples in the cereal fields was obtained from local farmers. The fields were located in a rather narrow band between the unpolluted soils on the distal part, and severely polluted soils (where cropping had been abandoned) on the proximal part of a transect perpendicular to the river channel. A satellite image of the research area (fields belonging to the villages Rajac, Braćevac and Tamnič) is available at: http://maps.google.com/ maps?ll544.082508,22.519602andz513andt5handhl5en); the size of the fields commonly ranged between 0.5 and 1 ha. The length of the studied transects through the partially damaged fields was on average less than 15 m. Miniscule differences in elevation along these transects were impossible to detect with the standard GPS devices (presumably the microlelevation gradient was not more than 0.5 m [16]). Herbicides are sparingly applied in this low-input system: systemics (commonly Glyphosate) are applied in the autumn before sowing, and occasionally some auxinic herbicides during the stem-elongation phase.

Weed survey
While the selection of the fields to be surveyed was rather opportunistic, vegetation and soil sampling within each field was carried out according to the flexible systematic model, a form of stratified sampling where samples were taken on the basis of the visual symptoms in the cereal crop (Table 1). In each field, it was possible to clearly distinguish two to four (depending on the site-specific topography) physiognomically uniform areas (zones) along the transect perpendicular to the river channel, as previously described [16]. Of a total of 100 samples (relevées), 26 samples each were taken in Zones 1, 3 and 4, while Zone 2 was represented by 22 samples (plots). Because of the narrow and elongated shape of the smallholder fields surveyed (see the link to the map), the area of each relevée was selected to be a rectangle of 30 m 2 (commonly 5 m66 m). The internet associated W3 TROPICOS nomenclatural database of the Missouri Botanical Garden, and the associated authority files (http://www.tropicos.org/ Home.aspx), were used as the reference.
To reduce the noise and increase the chance of capturing the response of weed vegetation to soil, we have excluded vernal ephemerals and late-summer (i.e. postharvest, stubble) species, which are normally less dependent on the soil properties, and more on the season [20]. The vegetation of the selected 26 cereal fields was continuously monitored (checked every 10 days) during the period from 15 th May to 14 th July 2009. Cover-abundance of weed species was estimated on the extended nine-grade Braun-Blanquet scale [21] at every check; the final scores used for each species were the highest recorded values during the monitoring period.

Plant sampling and analyses
Weed shoot biomass assessments and leaf sampling were carried out at the onset of milky development of the cereal crop, immediately after anthesis (Zadoks growth stage Z71-75), when the typical weed association (Consolido-Polygonetum avicularis) is fully developed. In each sample, one 1 m61 m quadrat was laid for destructive sampling; biomass was clipped, sorted by species, air dried and weighed. For each weed species encountered, ten leaves were taken for a composite sample for tissue elemental analyses, as well as 20 flag (youngest fully emerged) leaves from a cereal crop. Crop plants were sampled for chemical analysis only if they were alive and able to set at least four seeds in the spike.
Leaf samples were thoroughly washed with deionized water, dried at 70˚C and digested with a mixture of conc. HNO 3 and H 2 O 2 (3:2) in a microwave oven. Upon digestion, the concentrations of elements shown in this paper (P, S, Ca, Fe, Cu, and Al) were determined by inductively coupled plasma optical emission spectrometry (ICP-OES; SpectroGenesis EOP II, Spectro Analytical Instruments GmbH, Kleve, Germany). The concentrations of N in leaf samples were determined by a CHNOS elemental analyzer (Vario ELIII, Elementar Analysensysteme GmbH, Hanau, Germany). The certified reference material (GBW10015 Spinach; Institute for Geophysical and Geochemical Exploration, Langfang, China) was used to assess the accuracy and precision of plant analyses.

Soil sampling and analyses
A composite soil sample was obtained by sampling a soil core (30 cm height, 6.5 cm diameter) at three locations in each weed relevée. Soil samples were analyzed in fine earth fraction (,2 mm), after drying and sieving through a 2-mm mesh screen. The pH was measured in water (soil:water51:2.5), and the concentrations of total C, N and S by the CHNOS elemental analyzer. Organic carbon (C org ) was calculated from total C and CaCO 3 . Different extraction procedures were applied to determine plant-available concentrations of elements

Statistical analyses
For all the subsequent analyses, the records on the nine-grade Braun-Blanquet scale were transformed to a quasi-metric 1-9 scale of ordinal transform values (OTV, [21]). Species not achieving at least 20% frequency in at least one sampling zone, and species with frequency less than 10% along the transect, were excluded from further analyses; as well as outliers flagged by more than 2.3 standard deviations from the average distance between the species in a sample space. The final set comprised 84 weed species recorded in 100 relevées (Table S4). All multivariate analyses were done by PC-ORD version 6.15 software (MjM Software Design, Gleneden Beach, USA). Sørensen distance was used in all analyses. Only one data transformation was applied: relativization of OTV by sample unit totals. To test whether the visually distinguishable zones of crop growth disorders (as defined in Table 1), used as a basis for sampling, really reflect the significant difference in weed vegetation along the gradient, the nonparametric Multi-Response Permutation Procedure [24] was used (Table S1). Indicator Species Analysis (ISA), using relativized OTV, was done after Dufrêne & Legendre [25]. Weed species with very high indication (Indicator Value IV.30%, P value of the Monte Carlo test ,0.01) for a certain visual zone of cereal crop disorder were further classified by agglomerative cluster analysis with the space-conserving flexible beta method (b520.25).
The unconstrained ordination of weed vegetation data was done by Nonmetric Multidimensional Scaling (NMS), using OTV data and Sørensen distance. A final 2-d solutions with instability ,10 25 and final stress (Monte Carlo test for stress in relation to dimensionality) of 17.55 for unrelativized and 17.76 for relativized OTV data were selected after 98 and 68 iterations for the relativized and unrelativized data, respectively. The solutions were varimax rotated. To measure the strength of association between the species, 262 contingency tables of presence-absence were used. Plexus values (w coefficient) were calculated as a standardized x 2 statistic for species association; Yates' correction was applied.
Univariate analysis (ANOVA) was done by STATISTICA 6 (StatSoft Inc., Tulsa, USA). Variables with approximately lognormal distribution were log-transformed prior to analyses. A posteriori comparison of means was done by Tukey's test, with a50.05. Ellenberg indicator values [26] for each plot were calculated as averages weighted by the species cover-abundance (OTV). Elemental concentrations in weed vegetation (per 1 m 2 ) were calculated as an average of leaf mineral concentrations in each species, weighted by the relative contribution of species to the total plot biomass (community-weighted means). The regulatory coefficient H was calculated from the equation: log (leaf N:P)51/H6[log (soil N:P)]+log (b); this coefficient is a continuously variable parameter which quantifies the degree of stoichiometric homeostasis of an organism across a range of environmental supply [27].

Floristic changes
The fluvial deposition of mining waste over arable fields caused a decrease in the total plant biomass (crop + weeds) along the transects matching the spatial gradient perpendicular to the water channel ( Figure 1). While the increasing pollution load (indicated by the increasing soil S tot concentrations) caused a severe growth reduction of cereal crop, the weed biomass remained at higher levels than on the relatively unaffected (S tot ,2 g kg 21 ) soils along the gradient ( Figure 1). The floristic composition of weed assemblages (recorded over the 2months survey period) significantly varied among all the a priori delineated zones of crop growth disorders (Table S1). The weed assemblages on the most-severely degraded soils (Zone 4) were clearly set apart and had the highest floristic homogeneity (Table 2); the analysis of species association further showed that, along the whole transect, the w coefficient higher than +0.5 was only found in the most severely degraded soils, namely among the following species: Rumex acetosella, Agrostis capillaris, and Persicaria lapathifolia (see Cluster C1, Figure 2). The differentiation of weed vegetation along the soil gradient was chiefly achieved by a remarkable shift in dominance relations (species abundances), and not by a prominent presence/absence turnover. Of the total of 84 weed species recorded in the survey, 63% were present on both relatively undamaged calcareous soils of the Zone 1 and on the most severely altered acidic soils where substantial reduction of crop growth occurs in Zone 4 (Table S2).
Though not a single weed species was found to be confined to one visual zone of crop growth disorders, 44 species very clearly indicate the turnover of vegetation along the soil gradient ( Figure 2). Three major groups of weeds with similar response to the pollution-induced change of soil properties along the transects (Clusters A, B, and C; Figure 2) accounted for about 80% of variation in distances between weed species on the polluted fields. The succession of these three groups of weeds along the soil gradient was spatially explicit and visually very apparent over very short distances (Figure 3), commonly of less than 15 m. Accordingly, free ordination of the weed relevées showed a clear separation of vegetation along the NMS Axis 1, which corresponded to increasing pollution indicated by the increasing soil S tot concentrations ( Figure 4). Fifteen soil chemical parameters, which considerably varied along the transects (plantavailable concentrations of Ca, Mg, K, P, B, Fe, Zn, Mn, Al, Cu; C org , N tot , S tot , pH, CEC) were overlaid on the ordination of weed samples; nine of them were correlated by more than 10% with the ordination scores ( Figure 4, Table 3; the remaining soil properties are shown in Table S3). Little systematic variations in weed vegetation coincided with a wide gradient of soil Cu concentrations (Table 3, Figure 4). There was a high degree of multicolinearity among the soil parameters that might influence plant growth in the concentration ranges measured (Table 3): Kendall's correlation coefficients (tau) with the NMS Axis 1 for pH, Ca, P and Al were 20.72, 20.62, 20.61 and 0.63, respectively (for the unrelativized abundances, Figure 4a; and the values were even higher for the relativized data, Figure 4b).
Distinct pattern of a turnover in dominance of major weed species along the soil gradient is shown in Figure 5. Relative abundances show the trends when different competition pressure from the cereal crop is accounted for. Weeds  Table 1.  typical for the unpolluted calcareous soils (including some rather rare, specialized calcicoles like Nigella arvensis and Myagrum perfoliatum; Figure 5a, d) were the first to disappear with increasing soil acidification by the deposited mining waste. The middle parts of the soil gradient were dominated by the common, widespread species known to have a rather broad environmental adaptations (Figure 5b, e), whereas the dominant weeds of the severely altered soils were species which did not occur on the unpolluted nearby soils of the surrounding (Figure 5c, f).  The soil concentrations of mineral elements ( Table 3) that were strongly correlated with the main gradients in weed vegetation ( Figure 4) were differentially reflected in the foliage of weed species (Figures 7 and 8). At the weed assemblage level, the leaf concentrations of the pollutants deposited with the mining waste (S, Fe, Cu and Al) roughly reflected the increase of pollution load along the spatial transects (Figure 7), but these trends did not indicate any relation with the spatial succession of the three major groups of indicator species (see Figures 2, 3 and 5). All three ecological groups of weeds successfully maintained leaf Cu concentrations in a relatively narrow range along the soil gradient (Figure 7d). Only one species (Chenopodium botrys) consistently had a markedly elevated leaf Cu concentration (over 100 mg kg 21 DW), but it never reached an OTV.4. Moreover, the two major species adapted to the most acidified soils apparently had contrasting adaptation strategies to high extractable Al (Figure 7f). As a consequence, differences in leaf Al concentrations at the vegetation relevée level between the common weeds of a broad ecological valence (Cluster B) and weeds that strongly indicate the most severely altered soils (Cluster C) could not be detected (Figure 7e).
Contrary to the pollutants, trends in leaf concentrations of the elements whose soil concentration decreased due to waste deposition (Ca and P, Figure 8), differed among the three ecological groups of species (see Figure 5). Plantavailable soil Ca decreased by 58% along the gradient (Table 3). Wheat and barley maintained a very narrow range of leaf Ca concentrations (not more than 20%decrease in the most severely polluted soils compared to non-polluted ones, Figure 8a), whereas their growth was significantly reduced (see Figure 1). In contrast, leaf Ca concentration in weed vegetation decreased 2.5-fold along the gradient (Figure 8a). Thus, continuous weed vegetation growth along the soil gradient ( Figure 1) is dependent on succession of species with differential Ca concentrations (Figure 8b). The highest leaf Ca concentrations, observed in species typical for unpolluted calcareous soils (Cluster A, see Figure 2), decreased when the complex soil pollution induced the overall disappearance of this group along the gradient (see Figure 5a, d). While the other two successive groups did not show such a severe change of leaf Ca along the soil gradient, the group of species adapted to most severely degraded soils (Cluster C) consistently maintained the lowest Ca levels (Figure 8b).
The trends in leaf P (Figure 8c) in the three species groups (defined by indicator species analysis, Figure 2) can be related to the change in abundances of these species groups along the soil pollution gradient ( Figure 5). A sharp decrease of leaf P levels (below 1 g kg 21 ) in the species adapted to unpolluted calcareous soils (Cluster A, Figure 8c) was accompanied by a loss of their vitality and competitive ability (reflected in their decreased abundances; Figure 5a, d). Common species that dominated the middle portions of the pollution gradient (Cluster B) were Table 3. Soil chemical properties along the pollution gradient correlated by .10% with the NMS ordination scores of weed relevé es.

Parameter
Visual zones of crop growth disorders apparently better adapted to the multiple stresses caused by soil pollution, and concomitantly they maintained leaf P concentrations ( Figure 8c) and abundance (Figure 5b, e) on relatively strongly polluted soils, where Cluster A members already diminished (i.e. until Zone 4). Finally, complex adaptations of a ''novel'' group of species (a combination of species that did not occur as such on unpolluted nearby soils, Cluster C) to the multiple constraints on the most Finally, the sustenance of weed growth (Figure 1), accompanied by the remarkable change in floristic composition of weed vegetation (Figures 2-5) was underpinned by the maintenance of the N:P ratio in the biomass of weeds along the soil pollution gradient, while this ratio had not been sustained in the crop biomass (Figure 9a). Moreover, the same pattern of change of the biomass N:P ratio between the weeds and the cereal crops was observed along the gradient of plant-available P in the polluted soils (Figure 9b). Figure 9c shows that the weed biomass N:P ratio was only very weakly influenced by the external supply of N and P (in the range of soil N:P values measured in this study). This was also indicated by a high value of the regulatory coefficient H of 10.8 (Figure 9c). The value of the

Discussion
Causality in environment-vegetation relationship in studies based on observational analysis like the one reported here is difficult to prove, and it is widely denigrated as mere correlation [28]. The present study has shown that exceptionally strong gradients in weed vegetation (accounting for over 70% of the observed variation) were caused by a complex pollution-induced soil gradient ( Figure 4); all the other plausible environmental factors were kept constant over short transects investigated (on average less than 15 m), and differential competition pressures from the cereal crop were accounted for by relativization  (Figures 2, 4b and 5d-f). Broadly speaking, change in soil properties from calcareous (pH.7.5) to strongly acidic, leached, nutrient-poor soils (average pH 4.6, Table 3) was accompanied by a change of species adaptations which indicate different habitat conditions ( Figure 6) as well as by a shift from the dominance of ''calcicole'' (Cluster A, Figure 2) species to ''calcifuges'' (Cluster C, see also Figures 4 and 5). These findings are in accordance with a general substitution of weed species belonging to Caucalidion lappulae Tx. 1950 alliance of warmer, drier, calcareous soils by species of the Aperetalia spicaeventi Tx. and Tx. 1960 order in moister, cooler climates and acidic soils which however takes place over a large geographic gradient [29]. The species composition of the analyzed weed vegetation (Table S4) was significantly affected by (a) a rather low level of field management, and (b) surrounding vegetation matrix of predominantly fallow land in different successional stages. The former had enabled the maintenance of some species that have become rare in moreintensively managed agricultural landscapes, while the latter contributed to a rather high proportion of non-segetal species. As a result, a high proportion of species known as pseudometallophytes (as defined by [30]; Clusters B and C, Figure 2, see also Table S4) constituted an important response of weed vegetation to soil pollution gradient.
While species co-occurrence is a result of two simultaneous processes with contrasting effects: ''habitat filtering'' and ''limiting similarity'' [31], in severe environments like post-mining sites communities are assembled primarily by environmental filtering [11]. The assemblages on the most severely affected soils (Zone 4) were characterized by: (a) dominance of few typical acidophytic species (Apera spica-venti, Vicia hirsuta, Rumex acetosella, and Agrostis capillaris, in Cluster C, Figure 2), which are not sustained by the regional climate (semiarid conditions, Figure S1), carstic geology and calcareous soils and xerothermic natural vegetation; (b) the lowest variability of weed relevées along the soil gradient (Table 2, Figure 4), and the strongest association among the dominant species (Cluster C, Figure 2); and c) an unlikely combination of species that normally do not occur together (for example, Agrostis capillaris is not a cereal weed; Chenopodium botrys is also not a segetal species, and is not found on moist, acid soils in cool areas; Vulpia myuros and Xanthium italicum vs. Equisetum palustre and Persicaria lapathifolia have contrasting preferences for humidity). These findings clearly demonstrate the key effect of severe abiotic filtering on weed assembling. Firstly, small-scale soil factors (Zone 4, see Table 3) overrode the hierarchically higher filters of regional flora and climate which normally have a primary effect on plant communities [32]. Likewise, the increasing differences between weed communities along the soil pollution gradient (Figure 4) are in accordance with the reported phenomenon that increasing intensity of abiotic constraint leads to increasing differences of assembled communities from the regional species pool [33]. Secondly, intensification of habitat constraints has been shown to increase aggregation of dominant species [33], [34], and in particular it can increase the degree of association of specialist weed species at the field scale [6]. The dominant species Agrostis capillaris, Rumex acetosella and, less apparently Persicaria lapathifolia (see Figure 5c, f) are frequently found pioneer colonizers on mining-affected, acidic and nutrient-poor substrates, which are often also characterised by high concentrations of plant-available heavy metals [35], [36]; likewise, they possess certain tolerance to flooding, what is altogether a wide range of adaptations not usually encountered in nature. Shared ecological tolerance (i.e. trait convergence) is a common feature of communities assembled primarily by environmental filtering [37]. Thus, specific adaptations required to survive under complex, anthropogenically induced soil constraints are apparently the main biological reason underlying the consistent co-occurrence of a small number of dominant species in Zone 4. Thirdly, non-historic species combinations due to strong environmental change are the key feature of the novel ecosystem concept [13], [14]. The main functions of novel ecosystems are often preserved although biotic composition is strongly altered [15]; the underlying biological mechanisms leading to a particular novel species composition are however widely unknown.
While it was possible to demonstrate the response of weed vegetation (  Figures 2-5) to a complex soil gradient (Table 3; see also Table S3), the causal hierarchy of specific soil factors has remained in domain of black box. For wheat and barley on the same soils, growth reduction occurred due to a myriad of possible interactions among the major soil constraints (P deficiency, low pH, excessive concentrations of Al and Cu), as discussed previously [16], [17]. In the present study, the measured response parameters of the vegetation (biomass, floristic composition and leaf mineral composition) could not show the type of adaptations to all the significant pollution-induced soil constraints (shown in Figure 4 and Table 3). For instance, while soil pH is a well-known major factor that underpins vegetation response [29], our methodology could indicate only complex adaptations to acidic soils along the pollution gradient (Ellenberg indicator value for soil reaction; Figure 6a). Likewise, we did not detect any trend in leaf concentrations of elements deposited by soil pollution (Figure 7) which could be related to the induced gradient in weed vegetation. Moreover, the relatively high concentrations of available soil Cu (Table 3; for example, concentrations of only about 30 mg kg 21 were toxic to cereals; [38]) were not found to be significantly correlated with any observed pattern in weed vegetation ( Figure 4). Trends in vegetation response to the elevated soil Cu concentrations were, inter alia, obscured by the importance of pseudometallophytes, species that commonly possess certain tolerance to metals, mostly based on exclusion [30], what was reflected in the narrow range of leaf Cu concentrations (Figure 7d). Even though adaptations to metal-contaminated soil have been extensively investigated at the level of individual species, the effects on vegetation have seldom been examined [39]. An overriding effect of metals on the vegetation patterns in metalcontaminated soils is usually assumed; nevertheless, this conclusion has commonly been reached without concomitant consideration of the plant availability of the major nutrients in these soils [40]. On the other hand, several earlier reports have pointed out nutrient deficiency and in particular P deficiency as a key factor that affects the structure of vegetation on soils polluted by metals [35], [41]- [43].
Contrary to the elements whose soil concentrations were anthropogenically increased, we did show clear connection between soil-induced species turnover and the underlying functional adaptations to decreasing availability of Ca and P, as well as a strong capacity of vegetation to maintain biomass N:P ratio along the gradient. The pollution-induced decrease of soil Ca concentrations (Table 3), highly correlated with the changes in weed vegetation (Figure 4), was accompanied by a clear spatial succession of species with differential leaf Ca concentrations (Figure 8b). In their natural environments, plants adapted to calcareous soils (''calcicoles'') commonly have higher shoot Ca than plants adapted to acidic soils (''calcifuges''); calcifuges likely have lower Ca requirements [8], [44]. These two functional vegetation types possess widely differing key physiological adaptations to other particular soil constraints beside Ca levels (low pH, P deficiency, Al-, Mn-and Fe-toxicity on acidic, and P-and micronutrient deficiency in calcareous soils [45]), and their occurrence is further complexly affected by climatic gradient and some other evolutionary and historic processes [46]. Next, soil-induced species turnover along the gradient (Figures 3 and 4) can be related to a spatial sequence of ecological groups which successively achieve both dominance (see Figure 5) and the highest leaf P concentration (Figure 8c). A large scale-survey in UK [8] showed that ''normal'' P leaf concentrations in the species encountered also in our study (in Clusters B and C, Figure 2) ranged between 2 and 4 g kg 21 , what was comparable to the leaf P status at the ''peak abundance'' of each ecological group in our study (see Figures 5 and 8c). The floristic changes induced by a complex soil pollution gradient conveyed the plasticity of the vegetation response to the external P supply; the available soil P decreased more than three times (Table 3), while weed vegetation maintained a rather low variation in leaf P (decrease by 40%, calculated from Figure 8c). Finally, the N:P ratio in weed biomass varied in a rather narrow range (Figure 9). At the level of weed vegetation in the damaged fields, the value of the regulatory coefficient H was extremely high (Figure 9c). So far, H reported for individual species was below 5 [27], [47].
Leaf concentrations of mineral elements have very different potential to be used as true species traits. The strength of homeostatic regulation of concentration of mineral elements in plants in response to substrate widely differs among elements as well as among plant species and functional groups, so that meaningful results might be uncertain to obtain [8], [9]. Moreover, for most weed species critical deficiency and toxicity concentrations of mineral elements in leaves have not been established [48]. For instance, leaf Ca concentrations (under optimal conditions) vary considerably across plant species, more than P concentrations and much more than N concentrations [48]. Three physiotypes for Ca nutrition with intrinsically different leaf Ca levels have been described, and Ca deficiency is rarely encountered in nature [44]. Therefore, though a spatial succession of calicole by calcifuge species along the gradient of decreasing soil Ca levels was demonstrated (Figures 5 and 8b; Table 3), it remained unclear whether very different leaf Ca levels (2.5-fold, Figure 8b) along the induced vegetation gradient were a trait required per se by the imposed environmental filters. Furthermore, the mass N:P ratio as an indicator of nutrient limitation and nutrient cycling in ecosystems has extensively been used to study vegetation change [47], though controversies on its reliability exist (as discussed by [49]). Interspecific variation of N:P ratio is primarily caused by varying leaf P concentrations [47]. In our study soil N levels were rather low along the whole gradient (Table 3) and we did not detect significant change of leaf N along the gradient (not shown). The reason might be the internal N-use efficiency which enables plants to maintain the critical N tissue concentrations by regulating growth rate, e.g. [50]; hence, leaf N concentrations below the critical level are not likely to be recorded in viable spontaneously growing plants. Finally, P has previously been indicated as the only mineral element whose leaf concentrations (closely correlated with leaf N) could be considered a legitimate species trait [8]. Plants have different strategies to cope with P deficiency which occurs on both calcareous and acidic soils [51]. Thus, nutrient acquisition and utilization strategies might have changed with species composition along the soil gradient (as demonstrated by [52]). The present study showed that the final effect of a complex of physiological adaptations to changing soil constraints, conveyed by the induced species turnover, was homeostasis of leaf P and N:P ratio in weed vegetation. Environmental filtering (primarily by severe soil constraints), as the key assembly process, determined weed species cooccurrence by filtering out species with the same adaptations to the dominant constraint along the gradient (as shown by [37]); the sum effect of these adaptations was indicated by leaf nutrient concentrations (Figures 8 and 9).
A very high variability in species composition induced by strong soil gradient was accompanied by low variability of both leaf P concentrations ( Figure 8c) and biomass N:P ratio ( Figure 9). Thus, under severe environmental conditions and using leaf nutrient concentrations as species traits, this study supports the core findings of a manipulative experiment of [53] that environmental filtering can cause the composition of traits to converge even while species composition diverges. This divergence of species composition led to novel associations with unusual species combinations but only weakly affected biomass production (Figure 1), supporting a wider notion on functionality of novel ecosystems [15]. Finally, novel ecosystems, recognized to emerge due to anthropogenic disturbance, have been studied mostly from the perspective of management and restoration [13]- [15], [54]; this study indicated one of the underlying physiological drivers of ''novelty'', confirming the intriguing potential of accidental experiment sites for purely ecological research.

Conclusions
Using cereal weeds as a model system on an ''accidental experiment'' location with a distinct gradient of polluted soil over short distances (less than 15 m), we demonstrated links between gradients in soil properties and floristic composition of spontaneous vegetation. The response of vegetation to a gradient from highly calcareous to highly acidic, nutrient-poor and Cu-enriched soils included a substitution of calcicoles by calcifuges, and an increase in abundance of species known as pseudometallophytes, with preferences for Atlantic climate, broad geographical distribution, hemicryptophytic life form, adapted to low-nutrient and acidic soils, with lower concentrations of Ca, and very narrow range of Cu concentrations in leaves. The trends of abundance of the successive ecological groups of indicator species along the soil gradient underpinned a clear functional response to mineral stress, systematically reflected in the maintenance of leaf P concentrations and strong homeostasis in biomass N:P ratio. These findings help to understand the driving forces of assembly processes leading to unusual, novel combinations of species which are commonly observed as a consequence of strong environmental filtering, as for instance on sites affected by industrial activities.