Author for correspondence: Daniel A. HermsTel: +1 330 2023506Fax: +1 330 2633686Email: firstname.lastname@example.org
• Here, the growth–differentiation balance hypothesis (GDBH) was tested by quantifying temporal variation in the relative growth rate (RGR), net assimilation rate (NAR), and phenylpropanoid concentrations of two willow species (Salix sericea and Salix eriocephala) across five fertility levels.
• Initially, RGR increased and total phenylpropanoids declined (although every individual phenolic did not) as fertility increased, but NAR was unaffected. Subsequently, NAR and phenylpropanoids declined in the low fertility treatment, generating a quadratic response of secondary metabolism across the nutrient gradient. As above- and below-ground growth rates equilibrated, NAR and phenylpropanoids increased in the low fertility treatment, re-establishing a negative linear effect of fertility on secondary metabolism.
• A transient quadratic response of secondary metabolism is predicted when GDBH is integrated with models of optimal phenotypic plasticity, occurring when low NAR imposes carbon constraints on secondary metabolism in low nutrient environments. Once plants acclimate to nutrient limitation, the equilibrium allocation state is predicted to be a negative correlation between growth and secondary metabolism.
• Although both willow species generally responded according to GDBH, the complexity observed suggests that prediction of the effects of nutrient availability on secondary metabolism (and other plastic responses) in specific cases requires a priori knowledge of the physiological status of the plant and soil nutrient availability.
GDBH states that there is a physiological trade-off between growth and secondary metabolism imposed by developmental constraints in growing cells, and competition between primary and secondary metabolic pathways in mature cells. GDBH integrates this trade-off with responses of net assimilation rate (NAR) and relative growth rate (RGR) to resource availability to predict that nutrient (or water) availability will have a parabolic effect on secondary metabolite concentration (Fig. 1; Herms & Mattson, 1992). NAR reflects the balance between carbon gain (via photosynthesis) and losses (via respiration, exudation, volatilization, and leaching) per unit leaf area per unit time, and thus integrates environmental effects on net carbon acquisition at the whole-plant level over the specified growing period. Mathematically, RGR is the product of NAR and leaf area ratio (LAR), which is the ratio of total leaf area to total plant mass (McDonald, 1990; Lambers & Poorter, 1992).
The quadratic response of secondary metabolism predicted by GDBH to occur across a fertility gradient is based on three main physiological assumptions (Herms & Mattson, 1992): differential investment of photosynthate into new leaf area is the major determinant of phenotypic variation in RGR (Potter & Jones, 1977; Körner, 1991); NAR (and photosynthesis) is less sensitive to nutrient availability than is RGR, being reduced only when nutrient deficiency is severe (McDonald, 1990; Luxmoore, 1991); and secondary metabolism diverts resources from production of new tissue, and vice versa (Chapin, 1989).
Growing meristems, which are strong photosynthetic sinks, are provisioned by carbon sources that include neighboring, mature leaves (Marcelis, 1996). If increased nutrient availability has little effect on the net carbon gain of source leaves (e.g. Luxmoore, 1991; Ericsson, 1995), it will not increase their carbon budget, and production of new leaves can be supported only if source leaves export a greater proportion of photosynthate to growing meristems, retaining less to support other processes, including secondary metabolism (Körner, 1991). Thus, rapidly growing plants are predicted by GDBH to have low secondary metabolite concentrations (Fig. 1). Moderate nutrient deficiency imposes limitations on the growth of sink tissues with little effect on NAR (sink limitation, sensu Patrick, 1988), causing carbohydrates that otherwise would be exported to growing meristems to accumulate in source leaves (Geiger et al., 1996), where they may support increased secondary metabolism (Fig. 1; Waterman & Mole, 1989). Resource limitation severe enough to limit NAR is predicted to decrease both growth and secondary metabolism (Herms & Mattson, 1992) because all functions are carbon limited (source limitation, sensu Patrick, 1988) (Fig. 1).
These responses can have a temporal component. For example, as plants acclimate to nutrient limitation, initial constraints on photosynthesis can relax, which can increase NAR (Ingestad, 1982; Ingestad & Ågren, 1992), and possibly secondary metabolism, which may generate a negative rather than a quadratic relationship between nutrient availability and secondary metabolite production.
Although many studies have used GDBH as a framework for examining environmental effects on secondary metabolism, Stamp (2003, 2004) concluded that most experimental protocols were not aligned closely enough with assumptions and predictions of GDBH to provide informative tests. Few studies employed the three or more fertility levels necessary to document a parabolic response, and none of these measured both RGR and NAR (or their surrogates or components, such as plant size and photosynthesis rate), the interrelationships of which are assumed by GDBH to be key determinants of secondary metabolite concentrations.
Our objective was to test GDBH by quantifying the RGR, NAR, and secondary metabolite production of the North American shrub willows Salix eriocephala and Salix sericea across a gradient of five fertility levels ranging from very low to optimal. Phenylpropanoids are the major pool of secondary compounds in both species, with foliar concentrations exceeding 20% of dry leaf mass (Orians et al., 2000): Salix eriocephala produces condensed tannins but no phenolic glycosides, while S. sericea is rich in phenolic glycosides with lower concentrations of condensed tannins (Nichols-Orians et al., 1993; Orians & Fritz, 1995). Variation in the phenylpropanoid concentration has been shown to mediate interactions between willows and herbivores (Rowell-Rahier, 1984; Tahvanainen et al., 1985; Fritz et al., 2001).
Methods and Materials
Experimental plant material
On 4 May 2001, approx. 200 stem cuttings were taken from eight individuals of each species (one Salix sericea Marshall and one Salix eriocephala Michx. full-sibling family). Stems were cut into 10-cm sections, dipped in 1% indole-3-butyric acid (IBA) and 0.5% 1-napthalene acetic acid (NAA; Dip ’N Grow Inducing Concentrate®, Clackamas, OR, USA), and placed in a mist bed for 20 d. On 24 May, rooted cuttings were transferred to small (11.5 cm top diameter × 12 cm deep) plastic pots containing commercial potting medium (Pro-Mix ‘GSX’®, Premier Horticulture, Inc., Quakertown, PA, USA), and maintained in a glasshouse for c. 3 wk. On 18 June, plants were transplanted to larger plastic pots (27 cm top diameter × 24 cm deep) containing the same growing medium, and placed outdoors under shade cloth (25%) to acclimate for 7 d. Plants were not fertilized before the initiation of the experiment.
Experimental design and treatments
The experiment was designed as a randomized complete block, with five fertility levels. For each species, each fertility level (assigned randomly) was replicated three times within each block. On 24 June, the 152 plants of each species with the most uniform growth were divided into eight blocks (19 plants per species per block) based on stem diameter, with the largest plants assigned to one block, the next largest to another, and so on. From each block, four plants of each species were then randomly harvested to determine initial leaf area, and above- and below-ground biomass (as described in ‘Destructive plant harvests’). The remaining 120 plants of each species were relocated to a nursery, where they were arranged in eight blocks (15 plants per species per block) on a gravel bed exposed to ambient weather conditions. Each block consisted of five rows of six plants, with 1.0 m between pots in a row and 1.5 m between rows, which was sufficient to prevent plants from shading each other.
The nutrient treatment was initiated on 25 June, when five fertility levels were applied via a computer-controlled fertigation (combined fertilization and irrigation) system: 0, 25, 50, 150 and 200 ppm nitrogen (N), with N:P2O:K2O supplied in a ratio of 3 : 1 : 2. Sources of nutrients were calcium nitrate, potassium nitrate, and mono-potassium phosphate. The irrigation water contained sulfur, magnesium, and sodium at 30, 17, and 47 ppm, respectively. Mineralization of the composted softwood bark in the potting medium (30–35% by volume) also provided essential macro- and micronutrients, and, with the irrigation water, was the only source of nutrients for plants in the 0 ppm N treatment. No plants exhibited foliar necrosis, chlorosis, discoloration, or other pathological symptoms of nutrient deficiency.
Precise fertility levels were maintained automatically by applying nutrients with each irrigation event (1000 ml per event) that were scheduled to maintain optimal moisture levels (between 75 and 100% container holding capacity) (White & Mastalerz, 1966). Fertigation events were triggered when container moisture levels reached 75% of capacity as estimated by an evapotranspiration model calculated from weather data collected on site (q-com gem3 software system, Q-COM Corperation, Irvine, CA, USA) (Lee et al., 2000). Targeted moisture levels were maintained as plants grew, and pots therefore dried more quickly, by increasing the leaf area term of the model. The leaf area term was increased more quickly in the higher fertility treatments to increase the frequency of irrigation events in direct proportion to plant growth rate, thereby maintaining nearly constant levels of nutrient availability in each treatment. To confirm that targeted water levels were maintained through the experiment, a computer-monitored tensiometer was inserted to a depth of 8 cm in one pot per treatment (selected randomly) in each block. Nutrient solutions were delivered from tanks via five output lines connected to the fertigator, and dispensed via an emitter in each pot. Soluble salt concentrations in container leachate were monitored throughout the study, and did not reach excessive values.
Destructive plant harvests
To quantify plant biomass and leaf area at the beginning of the experiment, four plants of each species from each block were harvested on the day before initiation of fertility treatments. To document ontogenetic variation in plant response to the nutrient gradient, plants were destructively harvested 23, 40, and 85 d after initiation of the fertility treatments (on 17 July, 3 August, and 17 September, respectively). One of the three replicate plants per block (selected randomly) was harvested on each sampling date, for a total of eight plants per fertility treatment per harvest for each species (80 plants total for each harvest). Thus, a total of 304 plants were destructively harvested (240 treatment plants, and 64 pretreatment plants).
RGR, NAR, and biomass allocation
Immediately before each destructive harvest, leaf mass per unit area (LMA; g m−2) was estimated from a representative sample of 12 fully expanded leaves from each plant. The area of individual leaves was measured using a computerized image analyzer (Ci-400 Computer Image Analysis System and software; CID Inc., Vancouver, WA, USA), after which leaves were dried (60°C for 48 h) and weighed (to the nearest 0.01 mg). LMA was then calculated as the quotient of the mass and area of the leaf sample. At harvest, plants were partitioned into leaf, stem, and root fractions. After the shoots had been severed, roots were extracted by submerging the pot and carefully washing away the container medium. Each plant fraction was dried at 60°C for 48 h, and then weighed (nearest 0.1 mg). The mass of the 12 representative leaves was added to the harvested leaf mass to quantify the total leaf dry mass. Indices of plant growth and allocation were then calculated from dry mass and leaf area measurements according to the following equations:
Total plant mass (g) = total leaf mass + total stem mass + total root mass
RGR (g g−1 d−1) =[ln(final total mass) – ln(initial total mass)]/time
Total leaf area (m2) = total leaf mass/LMA
LAR (cm2 g−1) = total leaf area/total plant mass
NAR (g m−2 d−1) = RGR/LAR
SWR (g g−1) = stem mass/total plant mass
RWR (g g−1) = root mass/total plant mass
(SWR, stem weight ratio; RWR, root weight ratio.)
The initial mass of each plant used to calculate growth, RGR, and NAR over each harvest interval was estimated as the treatment mean for each block at the beginning of each harvest interval.
Just before a plant was harvested, foliage of two age cohorts (referred to as immature and mature foliage, respectively) was sampled for analysis of secondary metabolites (condensed tannins in both species, and phenolic glycosides in S. sericea). Immature leaves were defined as the youngest, fully expanded leaves (i.e. still lighter in color than older leaves). Their plasticron position varied from 5 to 7 (with the youngest apical leaf at least 2 cm long designated as leaf 1) depending on the fertility treatment. Mature leaves were sampled from positions 14 to 16.
On each harvest date, a representative sample of 12 leaves of each age cohort was collected from several shoots throughout the canopy. Leaves were removed from the stem with the petiole intact, placed in paper envelopes and immediately flash-frozen in liquid nitrogen, placed under dry ice in a cooler, and transported to a freezer within 30 min of sampling. Samples were stored at –80°C until they were dried without thawing at –4°C (Orians, 1995) in a tray lyophylizer, after which they were weighed (to 0.01 mg). The mass of these samples was then added to the total leaf mass for each plant. Dried leaf samples were stored in airtight desiccators at –18°C before and after they were milled (40 mesh) until they were analyzed for phenylpropanoid concentration.
Condensed tannins were analyzed using standard techniques (Hagerman & Butler, 1989; Orians, 1995; see Albrectsen et al., 2004 for details). Concentrations of foliar condensed tannins (mg g−1 dry weight) were calculated from purified willow tannin standards (0.2–2.0 mg ml−1) prepared in a similar manner. Foliar phenolic glycoside concentrations of S. sericea were also analyzed using standard high performance thin layer chromatography (HPTLC) techniques (Lindroth & Koss, 1996; see Albrectsen et al., 2007 for details). Total foliar phenylpropanoid concentrations of S. sericea are defined as the sum of the concentrations of the two individual phenolic glycosides and condensed tannins.
The effects of nutrient level on plant responses were assessed for each harvest date by analysis of variance (proc glm, Type III sums of squares; SAS Institute, 1999), with data reported as least square means ± 1 standard error (SE). Because the fertility treatment is a quantitative factor, relationships among means were examined using orthogonal contrasts to test for linear, quadratic (means of intermediate fertility levels are greater or less than means of lowest and highest levels), or asymptotic (mean of highest or lowest fertility level is greater or less than means for the other four treatments, which do not differ) treatment effects (Chew, 1976). Coefficients for the unequally spaced treatment levels were calculated according to Robson (1959). Tests for normality of residuals and homogeneity of variances revealed that data transformations were not required.
RGR, NAR, and biomass allocation
Nutrient availability had a positive linear effect on the total plant mass and total leaf area of both species on all three harvest dates (Table 1, Fig. 2). The RGR of both species also increased with increasing nutrient availability during the first two harvest intervals, with the most dramatic effect observed between days 23 and 40 (Fig. 3). However, nutrient availability had no effect on the RGR of either species between days 40 and 85, although total plant mass and leaf area continued to increase in all treatments (Fig. 2).
Table 1. F-values for orthogonal contrasts from analysis of variance (ANOVA) for tests of linear, quadratic, and asymptotic effects of five nutrient levels on growth and biomass allocation of Salix eriocephala and Salix sericea at 23, 40, and 85 d following initiation of treatments
The effects of nutrient availability on NAR were also dynamic, with the two species responding in a similar manner over time (Fig. 3). Nutrient availability had no effect on the NAR of either species through day 23, but did have significant asymptotic effects on the NAR of both species during the second harvest interval. Between days 23 and 40, NAR was significantly lower in the 0 ppm N treatment relative to the four higher fertility treatments, which did not differ. This pattern changed between days 40 and 85, when the NAR of both species increased substantially in the 0 ppm N treatment. Over this time period, the NAR of S. eriocephala was actually greater in the 0 ppm N treatment than in the four other treatments, which did not differ, while in S. sericea nutrient availability had a negative linear effect on NAR, which declined gradually as nutrient availability increased.
The two species displayed striking correspondence in their allocation responses to nutrient availability over the 85-d growing period (Table 1, Fig. 4). Nutrient availability had a positive linear effect on the LAR of both species on all harvest dates, with treatment effects being strongest in the early stages of the experiment. Relative to their initial state, the LAR of both species was much lower in the two lowest fertility treatments at day 23, while the LAR of S. eriocephala remained constant in the three highest fertility treatments. LAR in S. sericea also remained constant at the intermediate nutrient level (50 ppm N), but increased in the two highest treatments (150 and 200 ppm N) (Fig. 4). The LAR of both species peaked at day 40 and was lowest at day 85 in all treatments (as was variation among the fertility levels). LMA was not affected by nutrient availability in either willow species (Table 1).
Effects of nutrient availability on below-ground growth were generally opposite to those observed on LAR (Table 1, Fig. 4). Linear contrasts for effects of nutrient availability were highly significant for all three harvests, with RWR decreasing as nutrient availability increased. Temporal patterns of variation in RWR in response to nutrient availability were complex, but strikingly similar in the two species. Relative to their initial state, the RWR of both species increased in the two lowest fertility treatments at day 23, and then decreased sharply at day 40. Conversely, the RWR of plants receiving the three highest nutrient regimes steadily decreased until day 40. The RWR of both species stabilized between days 40 and 85, remaining nearly constant for the duration of the experiment, although at a higher level in the low than in the high fertility treatment. On the final harvest date, plants in the lowest fertility treatment had allocated 40% of their biomass below ground, compared with 25% for plants in the high fertility treatments.
The effects of nutrient availability on foliar secondary metabolite concentrations were also similar in the two species (Table 2, Fig. 5). On day 23, condensed tannin concentrations of S. eriocephala declined in both immature and mature foliage as nutrient availability increased. The same pattern was observed for total phenylpropanoid concentrations in mature and immature foliage of S. sericea. This pattern changed on day 40, as a positive quadratic response of condensed tannin concentrations to nutrient availability was observed in both immature and mature foliage of S. eriocephala, and a positive quadratic response of total phenylpropanoid concentrations was also observed in immature foliage of S. sericea. Nutrient availability continued to have a negative linear effect on total phenylpropanoid concentrations in mature foliage of S. sericea. The quadratic effects observed on day 40 reverted on day 85 to negative linear effects of nutrient availability on condensed tannin concentrations in mature foliage of S. eriocephala, as well as total phenylpropanoid concentrations in immature and mature foliage of S. sericea. There remained a marginally significant positive quadratic effect of nutrient availability (P = 0.082) on tannin concentration in immature foliage of S. eriocephala on day 85.
Table 2. Effects of five nutrient levels on concentrations of individual phenylpropanoid compounds in immature and mature foliage of Salix sericea harvested 23, 40, and 85 d after initiation of treatments, with probability of significance of orthogonal contrasts from analysis of variance (anova) for linear, quadratic, and asymptotic treatment effects
Foliage age class
Concentration (mg g−1)
Fertility treatment (ppm N)
Data are expressed as least square means ± 1 standard error.
196.0 ± 2.8
191.2 ± 3.4
179.0 ± 3.6
165.6 ± 3.4
162.1 ± 10.3
155.6 ± 5.6
152.6 ± 6.8
159.4 ± 5.6
144.1 ± 5.8
151.3 ± 5.4
37.7 ± 1.2
37.9 ± 1.5
34.1 ± 2.5
30.8 ± 2.0
28.0 ± 2.1
23.9 ± 1.4
21.5 ± 2.0
21.5 ± 1.9
16.5 ± 1.4
15.5 ± 1.4
86.9 ± 4.8
79.8 ± 4.3
63.3 ± 6.4
51.3 ± 2.8
50.3 ± 3.0
87.6 ± 5.1
80.6 ± 7.4
58.2 ± 4.0
48.7 ± 1.4
66.0 ± 14.8
171.2 ± 5.1
191.9 ± 11.0
189.0 ± 11.7
176.7 ± 10.3
169.0 ± 10.4
143.0 ± 17.1
178.4 ± 7.5
165.4 ± 7.6
159.3 ± 12.7
162.8 ± 4.6
33.2 ± 1.9
32.1 ± 4.0
34.8 ± 3.2
35.1 ± 2.3
37.7 ± 1.1
22.4 ± 3.1
24.8 ± 2.6
22.9 ± 2.2
18.8 ± 1.8
18.5 ± 1.9
49.3 ± 4.8
59.1 ± 8.2
64.8 ± 5.9
44.9 ± 5.5
44.3 ± 3.3
94.5 ± 4.8
67.2 ± 5.3
59.7 ± 4.8
39.5 ± 9.0
42.6 ± 4.1
167.9 ± 15.6
181.5 ± 20.5
141.2 ± 16.5
136.2 ± 20.2
136.1 ± 21.5
155.6 ± 7.3
156.3 ± 4.0
136.0 ± 10.8
138.0 ± 7.6
123.5 ± 10.3
36.6 ± 2.8
38.5 ± 1.5
36.1 ± 2.0
36.3 ± 1.6
35.8 ± 2.3
30.3 ± 2.4
30.0 ± 1.6
31.0 ± 1.5
25.0 ± 2.6
24.5 ± 1.7
99.7 ± 13.9
83.7 ± 6.7
73.2 ± 7.0
78.5 ± 6.2
72.3 ± 9.0
83.3 ± 8.1
81.2 ± 2.9
77.0 ± 8.6
68.1 ± 5.2
61.8 ± 7.1
Effects of nutrient availability on individual phenolic compounds of S. sericea did not always correspond with fertility effects on total phenylpropanoid concentrations (Table 2). On days 23 and 40, nutrient availability had no effect on salicortin concentration in mature foliage. Rather, the negative linear effects on total phenylpropanoid concentrations observed on these dates were a result of effects on 2′-cinnamoylsalicortin and condensed tannins. However, nutrient availability did have a negative linear effect on salicortin concentration in immature foliage on day 23, which did correspond with effects on total phenylpropanoid concentration. The quadratic effect of nutrient availability on total phenylpropanoid concentration in immature foliage observed on day 40 was a result of the response of salicortin and condensed tannins, which counteracted a positive effect of nutrient availability on 2′-cinnamoylsalicortin. On day 85, the effect of nutrient availability was not significant for any of the three individual compounds in immature foliage, despite the overall significant negative linear effect on total phenylpropanoid concentrations. However, in mature foliage, nutrient availability had significant negative linear effects on all three compounds.
Effects of nutrient availability on total phenylpropanoid levels of both willows largely conformed to predictions of GDBH, which states that secondary metabolite concentrations are contingent on the relationship between NAR and RGR. In our study, this relationship varied across fertility treatments, as well as over time within a treatment, as did above- and below-ground allocation patterns (LAR and RWR). Although complex, these patterns were highly consistent across the two species, which suggests programmed responses to nutrient availability rather than random variation.
The negative linear response of total phenylpropanoid concentrations to the fertility gradient observed on day 23 in immature and mature foliage of both species is predicted by GDBH when RGR increases with no effect on NAR (Fig. 1; sink-limited range of abscissa). It is possible that the high NAR observed in the low nutrient treatments was facilitated by nutrient reserves acquired before the start of the experiment (Ingestad, 1982). By day 40, however, the NAR of both species had declined substantially in the 0 ppm N nutrient treatment, possibly as a result of low nutrient uptake coupled with dilution of previously acquired nutrients (Ingestad, 1982). Total phenylpropanoid concentrations were also lowest in this treatment. Consequently, the linear effects of nutrient availability on secondary metabolism observed on day 23 transitioned to a positive quadratic response in immature and mature foliage of S. eriocephala, and in immature foliage of S. sericea, on day 40. These quadratic responses conform to GDBH when both RGR and NAR are source-limited at the low end of the fertility gradient and increase as nutrient availability increases, as we observed (Fig. 1; full range of abscissa).
Between days 40 and 85, the NAR of both species increased substantially in the lowest nutrient treatment. In response to increased NAR, GDBH predicts that secondary metabolite concentrations should also increase as carbon limitations relax. Accordingly, total phenylpropanoids increased in the 0 ppm N treatment between days 40 and 85, and the quadratic response of secondary metabolite concentration to nutrient availability reverted to a negative linear response in immature foliage of S. sericea and mature foliage of S. eriocephala (Fig. 1; sink-limited range of abscissa). Tannin concentrations of immature foliage of S. eriocephala also increased, but not sufficiently to generate a negative linear effect, and we continued to observe a marginally significant quadratic effect of nutrient availability. Although a quadratic response in mature foliage of S. sericea was never observed, it is possible that a transient parabolic response occurred either before or after day 40.
Integrating GDBH with models of optimal phenotypic plasticity
A parabolic response of secondary metabolism to resource availability is a cornerstone prediction of GDBH that has been tested only rarely (Stamp, 2004). Nutrient availability also had quadratic effects on foliar terpene concentrations in camphorweed (Heterotheca subaxillaris) (Mihaliak & Lincoln, 1985) and on phenolics in tomato (Lycopersicon esculentum; Wilkens et al., 1996), while water availability had a quadratic effect on constitutive terpene production in grand fir (Abies grandis) stems (Lewinsohn et al., 1993). However, the fleeting nature of the parabolic responses was not predicted by Herms & Mattson (1992). It is proposed that GDBH can be integrated with models of optimal phenotypic plasticity to predict that the parabolic response represents a temporary state imposed by carbon stress in extremely low nutrient environments.
Models of optimal phenotypic plasticity predict that plants will respond to altered nutrient regimes to increase acquisition of limiting resources, for example by increasing the root:shoot ratio in response to nutrient limitation (Bloom et al., 1985; Hirose, 1987; Ingestad & Ågren, 1991, 1992; Shipley & Meziane, 2002). Once a plant has acclimated to its respective nutrient treatment, these models also predict that plants will achieve an equilibrium allocation state characterized by equivalent above- and below-ground growth rates (balanced growth), resulting in a stable root:shoot ratio. As plants acclimate to low nutrient conditions and NAR and RGR increase, GDBH predicts that secondary metabolism should also increase as carbon limitations relax, and the parabolic response of secondary metabolism to nutrient availability should transition to an equilibrium state characterized by a negative correlation between growth rate and secondary metabolism. This acclimation process would be represented in Fig. 1 by a temporal shift along the abscissa to the right as the plant transitions from a state of source limitation to one of sink-limited growth (sensu Patrick, 1988).
The temporal variation observed in dry matter allocation (such as LAR and RWR), NAR, RGR, and secondary metabolism was remarkably similar in the two willows and highly consistent with these predictions. As nutrient availability increased, LAR increased and RWR decreased, as predicted by optimality models. Constant RWR between days 40 and 85 suggests that plants had achieved an equilibrium state of balanced above- and below-ground growth (Ingestad, 1982; Anten et al., 1995). As allocation patterns equilibrated, NAR increased sharply in the low fertility treatments (possibly as a result of acclimation responses that buffered effects of nutrient limitation on photosynthesis). Total phenylpropanoid concentrations also increased in the lowest nutrient treatment, and the quadratic response transitioned to a negative linear effect.
Effects of nutrient availability on RGR diminished over time as plastic responses increased RGR in low nutrient environments, while allometric increases in stem relative to foliage biomass (stem weight ratio; Table 1) in high fertility treatments may have increased respiratory losses (e.g. Ågren, 1985; Poorter & Garnier, 1999). Despite having no effect on RGR over the last 45 d, nutrient availability continued to have a positive linear effect on absolute growth rate, causing treatment effects on total leaf area and biomass to widen over time.
An underlying premise of GDBH as extended by Herms & Mattson (1992) is that phenotypic plasticity in secondary metabolism has been shaped by natural selection. Once plants are acclimated to low fertility, high concentrations of secondary metabolites are predicted to enhance resistance to biotic and abiotic stressors when compensatory growth is physiologically constrained (e.g. Chapin, 1991; Chapin et al., 1993). Low secondary metabolism in resource-rich environments is considered a constraint imposed by high resource demands of rapid growth selected for by competitors (Herms & Mattson, 1992). There is strong evidence that phenotypic plasticity in secondary metabolism is under genetic control (Han & Lincoln, 1997), including that for some of the phenylpropanoids examined in these willows (Orians et al., 2003). Moreover, variation in secondary metabolism in response to nutrient availability has been shown to be regulated by gene expression (Bongue-Bartelsman & Phillips, 1995; Scheible et al., 2004), which also suggests a response to selection.
It is important to note that not all individual phenylpropanoids conformed to predictions of GDBH. In S. sericea, it was found that condensed tannin concentrations were generally more responsive to nutrient availability than were phenolic glycosides (although salicortin and 2′-cinnamoylsalicortin did respond in concert with tannins on some sampling dates), reinforcing the contention of Herms & Mattson (1992) that predictions of GDBH are more relevant to large pools of compounds (such as total concentrations of the phenylpropanoids measured here) than to individual compounds. Divergent responses of individual compounds have also been observed in other studies, and can result from internal metabolic trade-offs within a pathway and/or differing selection pressures on individual compounds (Herms & Mattson, 1992; Berenbaum, 1995).
Experimental and ecological implications of complex temporal responses
The spatial and temporal variation that was observed in both willows has important experimental and ecological implications. Our results reinforce Stamp's (2004) analysis that conclusions regarding effects of nutrient availability on secondary metabolism will depend on treatment levels employed, when plants are sampled, and the rate at which, and degree to which, plants acclimate to changes in fertility. For example, nutrient availability declines steadily following a single fertilizer application, which can result in resource allocation patterns that shift continuously as plants attempt to acclimate to a moving target (Ingestad & Ågren, 1991, 1992). Any single measure of secondary metabolism may represent a snapshot of a transient state that never equilibrates (Stamp, 2004).
An important implication of a quadratic response is that secondary metabolite concentrations can either increase or decrease in response to increased resource availability, depending on the initial physiological status of the plant and the magnitude of the treatment (Herms & Mattson, 1992). Results of a recent meta-analysis are consistent with the pattern proposed in Fig. 1: correlations between growth and secondary metabolism tended to be positive in low nutrient environments and negative in fertile environments (Koricheva, 2002).
We do not concur with Stamp's (2004) argument that good tests of GDBH must employ steady-state nutrient addition rates. Nonequilibrium conditions are ecologically relevant because nutrient availability can be highly dynamic under natural conditions (Attiwill & Adams, 1993). We do, however, agree with her assessment that tests will not be highly informative if they do not quantify variation in RGR and NAR (or highly correlated indices such as growth and photosynthesis), as well as secondary metabolism (Stamp, 2004).
The dynamic responses that were observed contribute to an increasing body of evidence that temporal responses of secondary metabolism to environmental variation may be ecologically important (e.g. Crone & Jones, 1999; Wallin & Raffa, 2001). Although our results are generally consistent with GDBH, the complexity of the responses suggests that predicting effects of nutrient availability on secondary metabolism (and other plastic responses) in specific cases requires detailed a priori knowledge of the physiological state of the plant and the nutrient status of the environment.
We thank Brian Brannigan, Bryant Chambers, Bob Fritz, Cathy Love, and Ben Marthey for technical assistance, and Jessica Prenger for operation of the fertigation system. Bob Fritz provided stock plants used to propagate the willows. Critical reviews by Dr Richard Norby and two anonymous referees substantially improved the manuscript. This project was supported in part by the USDA National Research Initiative (Award 00-35316-9250 to DAH), National Science Foundation (NSF) (DEB 9981568 to CMO), the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning, and by state and federal funds appropriated to the Ohio Agricultural Research and Development Center and The Ohio State University.