Study Area

The study was located in the private Inti-Llacta Nature Reserve, on the western flanks of the Ecuadorian Andes (00°02′N, 78°43′W). The area lies at 1860 m a.s.l. and regular precipitation averages 3200 mm annually (Anthelme et al. 2011). The Reserve is representative of fragmented Andean cloud forest, with as much as 41% of its forests having been clear-cut between 1966 and 1990. Forested patches are dominated by secondary forest elements such as the trees Cecropia spp. (Cecropiaceae) and Alnus acuminata (Betulaceae), the understorey palm Chamadorea pinnatifrons and the mid-storey palm Prestoea acuminata (Arecaceae). A small number of deforested areas in the Reserve remain grazed by large, domestic herbivores (cows and horses). This is because the objective to protect the natural environment must be counterbalanced by the needs of the local people to maintain agricultural activities for subsistence. In terms of their vegetation, they are dominated by spotted patterns of the exotic tussock grass S. sphacelata (Poaceae), surrounded by short herbaceous vegetation; shrubs and juvenile trees are almost absent, probably because they are inhibited by the presence of the tussock grasses [F. Anthelme & R. Montufar, pers. obs.; see Sarmiento (1997) for patterns at the regional scale].

Interacting Species

The genus Ceroxylon comprises 12 species, all generally restricted to the Andean TMCFs (Mulligan 2010; Sanín & Galeano 2011). Ceroxylon echinulatum is endemic to the cloud forests of Ecuador and northwest Peru, and its distribution is patchy and highly fragmented. It is a slow-growing, dioecious palm tree that frequently reaches or grows through the canopy, can form dense populations, and shows little resilience to deforestation (Montúfar et al. 2011; Sanín et al. 2013). Similar to other species that have been selected for restoration purposes (e.g. Psidium guajaba; Sarmiento 1997), C. echinulatum is a keystone species of the cloud forest, in particular because of its large production of fleshy fruits consumed by efficient seed dispersers (Sanín & Galeano 2011). Thus, the species is an interesting phytometer to determine to what extent nurse plants can facilitate the recovery of TMCF (Anthelme et al. 2011). Recent observations in deforested areas of the Andean cloud forest showed that, despite the presence of some spared adult palms with fruits and young seedlings (only beneath female adult palms), populations of Ceroxylon spp. are unable to regenerate in the long term (Anthelme et al. 2011; Sanín et al. 2013).

Setaria sphacelata is a tall C₄ tussock grass that was introduced in Ecuador around 1980 and has become the dominant grass in pastures created following the deforestation of TMCF (Brenner et al. 2012). The species has a relatively low nutrition value for large herbivores because of the low nutrient content of its leaves, particularly the older ones located at the tussock base (Hacker 1987). Spatial patterns observed in the field show common positive spatial associations between seedlings of C. echinulatum and mature tussocks of S. sphacelata, but only at the foot of female trees of C. echinulatum (F. Anthelme & R. Montufar, pers. obs.). Therefore, S. sphacelata was hypothesized to be a biotic refuge for other plants, especially the lower, older segments of tussocks [shared defense mechanism sensu Callaway (2007)].

In Situ Manipulation Experiment

An in situ experiment was implemented in a deforested area of c. 1 ha, dominated by an assemblage of short native/exotic grasses and S. sphacelata (c. 25% cover at the soil surface), and in the adjacent forest. The deforested area was regularly grazed by five or six cows and two horses. In March 2009, 600 natural seedlings of C. echinulatum were collected under 20 mother trees in the forest and grown together in a greenhouse at the site under study. After 3 months, 210 of these seedlings were transplanted to the deforested area and in the adjacent forest for the experiment. We selected ‘seedling’ (individuals with two or three undivided leaves; Anthelme, Montúfar Galárraga & Pintaud 2010) as our target life stage because of the expectation of greater sensitivity to interactions with neighbouring plants than older life stages (e.g. Venn, Morgan & Green 2009).

We used a randomized block design (30 blocks) that manipulated the biotic stress (herbivory) and abiotic stress applied to transplanted seedlings in the deforested area (Fig. 1). Seedlings were transplanted in each block within a 3 × 2 factorial design, with three levels of neighbours (open: microsites occupied by short, herbaceous vegetation; nurse: presence of S. sphacelata; shade: presence of artificial shade retaining c. 50% of light intensity; Fig. 1) and two levels of herbivory (presence/absence, achieved using metal exclusion fences). To this design, we added an additional treatment in the adjacent secondary forest, where domestic herbivores were absent (forest; 30 transplanted seedlings; Fig. 1). One seedling was transplanted for each replication of each treatment (total: 210). Comparing forest and open treatments allowed the effects on seedlings of abiotic stress as a result deforestation to be measured. In the deforested area, possible amelioration of the abiotic microenvironment by artificial shade or the presence of nurse plants was assessed by comparing the performance of seedlings in the open vs. the nurse and shade treatments, respectively. The level of light interception by artificial shade was equivalent to the shading effects provided by S. sphacelata on palm seedlings. To test for the effects of herbivory in the three neighbouring treatments in the deforested area, we replicated them outside metal fences in each of the 30 blocks (treatments open_h, nurse_h and shade_h, respectively; Fig. 1).

The performance of transplanted seedlings was monitored every 2 months between March 2009 and March 2010, as well as in November 2010. For each seedling of C. echinulatum, we measured two performance variables to obtain a clear picture of the effects of S. sphacelata: survival, and the green area of leaves (cm²). The green area (GA) of each surviving individual reflected the photosynthetic potential of each individual. Leaves were extrapolated to be diamond-shaped and showing little morphological variation (Sanín & Galeano 2011; pers. obs.). Accordingly, their GA was calculated as:

GA(one leaf)=L×0·5l×X×Y,

where L is the length of the leaf, l is its width, X is the portion of leaf material removed by herbivores (ratio of leaves grazed, estimated visually in each leaf and taken separately as the index of leaf predation), and Y is the portion of the leaf that was not green. Variables X and Y were estimated visually and systematically for every single leaf of each individual. The GAs of all leaves of each individual were summed to provide the GA at the individual level.

In order to explore the mechanisms underlying the patterns of interactions found, we measured atmospheric water stress, light intensity and soil temperature for each treatment without herbivores, every 30 min over 12 days in a relatively dry period (March 2009). Water stress was estimated at 15 cm above the soil surface (approximately the height of seedling leaves) via the vapour pressure deficit (VPD) (kPa), which was calculated with atmospheric temperature and relative humidity (Anthelme, Michalet & Saadou 2007). Atmospheric temperature and relative humidity were measured with HOBO-Pro RH/Temp data loggers (Onset, Bourne, MA, USA), with at least four simultaneous measurements assigned randomly in each treatment. Light intensity (lux) was measured using UA-002 data loggers (Onset), with eight simultaneous measurements for each treatment. Data loggers in the nurse treatment were placed directly under the peripheral cover of S. sphacelata stems. Temperature in the soil at a depth of 5 cm was measured with U23-01 data loggers (Onset), as the average value of three simultaneous measurements in each treatment.

Data Analyses

In the deforested area, the effects of herbivory (yes/no) and neighbourhood (open/shade/nurse) on seedling survival were analysed using the mixed-effects Cox model (Therneau & Grambsch 2000), which is a modification (by allowing the inclusion of random covariates, i.e. the ‘blocks’ in this study) of the commonly used Cox's Proportional Hazards model. To test the significance of each fixed variable, we performed a likelihood ratio test (LRT) by comparing the deviances of a pair of nested models: a null model including only the random variable, and an alternative model including the random variable plus each fixed factor. The interaction between the fixed factors was included only if the LRT of the model indicated significant difference with both the simpler models. The effects of herbivory and neighbourhood on GA were analysed using a randomized-block linear model (fixed effects) with comparisons between pairs of treatments within blocks provided by simultaneous post hoc Tukey tests. This latter analysis was conducted using data from July 2009, after which the mortality of seedlings did not permit the performance of further statistical tests.

The block structure of our design in the deforested area allowed us to make paired comparisons among treatments (see Fig. 1). We calculated the intensity of the facilitative effects on Ceroxylon seedlings generated by (i) S. sphacelata and (ii) artificial shade, by comparing seedling performance in these treatments with that in the open_h and open treatments. Additionally, we compared each above treatment with forest in order to determine to what extent the performance of seedlings under conditions of shade or the presence of nurse plants can challenge the performance of seedlings in their natural habitat (with random assignment of forest replicates to each of the 30 blocks). These paired comparisons were made with the Relative Interaction Index (RII; Armas, Ordiales & Pugnaire 2004). RII was calculated with the survival and GA performance variables as:

RII=(Bw−Bo)/(Bw+Bo),

where Bw and Bo represent the performance of potential beneficiary species within and outside the nurse/artificial shade areas, respectively. RII values <0 reflect competitive interactions, while RII values >0 reflect facilitative interactions. As seedlings in the forest were not allocated to a specific block, we assigned them randomly to each block and calculated additional RII values with the forest treatment taken as a control (absence of deforestation and herbivory).

Variation in the ratio of leaves grazed throughout the experiment was analysed with two-sample t-tests (and one-sample t-tests when all values of one treatment reached the maximum value of 100%). Variation of each of the abiotic parameters among the open, nurse and forest treatments were analysed with anova and two-sample t-tests for average and maximal values. Correlations between abiotic parameters were tested with linear regressions. All analyses were made using minitab 15 and r 2.14.1 software (R Development Core Team 2011).

The Effects of Deforestation and Herbivores on Seedlings

Throughout the experiment, the seedlings in forest displayed higher survival than the seedlings in open_h (two-sample t-tests: P < 0·001 at each date; Fig. 2a). They were almost unaffected by mortality (survival ratio after 20 months: 0·86 ± 0·06 SE), whereas in open_h most of the seedlings died at the very beginning of the experiment (survival after 2 months: 0·07 ± 0·05) and all individuals died after 10 months (Fig. 2a; for other treatments see Fig. S1, Supporting information). Cumulative variation in the area of leaves was also significantly different between the two treatments (two-sample t-tests: P < 0·001 at each date; Fig. 2b). Whereas GA regularly increased in forest, (+70·90 cm² ± 0·13 after 20 months), it reached its minimum value after 4 months in open_h and maintained it up until the end of the experiment (−51·92 cm²).

Effects of Shade and Nurse Plants With and Without Herbivory

The presence of a neighbourhood had a significant positive effect on the survival, as well as the GA, of seedlings; whereas, the presence of herbivores impacted negatively upon the performance of seedlings (P_survival < 0·001 and P_GA < 0·01 in both cases; Table 1). The interaction between the neighbourhood and herbivory variables was significant only for GA (P_survival = 0·099 and P_GA = 0·032, respectively). The net effects of nurse vs. open (and nurse_h vs. open_h) on the survival and the GA of seedlings were significantly positive throughout the experiment (e.g. survival after 20 months of experiment: RII_NURSE = 0·60; RII_NURSEH = 0·20; GA after 4 months: RII_NURSE = 0·40; one-sample t-tests ‘not superior than zero’: P < 0·01; Fig. 3a,c,e). Meanwhile, the effect of shade vs. open was significantly positive throughout the experiment when considering survival, reaching a plateau from 10 months (RII_SHADE: 0·37 ± 0·11; Fig. 3b), but was significantly lower than the effect of nurse vs. open, except after 12 months (simultaneous post hoc Tukey tests: P < 0·05). The shade_h treatment did not influence survival (Fig. 3d). The effect of shade on the GA of seedlings was increasingly positive up until between 2 and 10 months of the experiment, after which the low number of surviving seedlings yielded insufficient repetitions for interpretation (RII after 10 months: 0·46 ± 0·14; Fig. 3f).

Table 1. Summary of the models analysing the effects of herbivory (presence/absence), neighbourhood (open/shade/nurse), and their interactions on the performance of Ceroxylon echinulatum seedlings in the deforested area. Survival curves were analysed using the mixed-effects Cox model, while GA in July 2009 was analysed using a randomized-block linear model (fixed effects) and post hoc Tukey tests (P < 0·05). d.f., degrees of freedom. Different letters mean significant variations among treatments

Variables	χ survival 2	P Survival	FGA	P GA	d.f.	Comparisons among treatments (P > 0·05)
Herbivory (yes/no)	51·15	0·000	8·35	0·006	1	Survival: no > yes GA: no > yes
Neighbourhood (open/shade/nurse)	30·58	0·000	7·84	0·001	2	Survival: nurse > shade > open GA: nurse > shade = open
Herbivory × Neighbourhood	4·61	0·099	3·68	0·032	2

The nurse treatment took the survival of seedlings to a similar level as forest (after 20 months: RII_{NURSE vs. FOREST} = 0·20, one-sample t-test: P > 0·05; Fig. 3a); however, the GA was increasingly lower in the former throughout the experiment (after 20 months: RII_{NURSE vs. FOREST} = −0·47; Fig. 3e). In contrast, in the presence of herbivory, the effect of S. sphacelata (nurse_h) was highly negative when compared to forest, from the very beginning of the experiment (RII_2 months: −0·93 ± 0·05; Fig. 3c). RII_GA was not calculable for treatments with herbivory because of too few data being available.

Abiotic Characterization of Experimental Treatments

Light intensity reached an average value of 1331 ± 90 lux in the forest treatment, which was 9 times lower than in open and 3·7 times lower than in nurse (Fig. 4). The VPD in open was significantly higher than that in forest during all periods of the day (Fig. 4a–c; two-sample t-tests at each time: P < 0·05). These differences peaked at 12:00 when average VPD in the open treatment reached 0·59 ± 0·11 kPa, compared to 0·01 ± 0·00 kPa in forest. The nurse treatment reduced the average VPD by more than half (two-sample t-test: P < 0·001), and buffered the maximum VPD to 0·29 ± 0·07 kPa at 12:00. In open, the light intensity was a major driver of VPD variation, with or without the presence of S. sphacelata (R² = 0·22 and 0·30, respectively; see Fig. S2, Supporting information). In forest, variation in VPD was barely influenced by light intensity (R² = 0·01). The average soil temperature was higher in open than in forest, while nurse provided intermediate values (anova: P < 0·001).

Herbivore Damage to Palm Seedlings

In the deforested area with the presence of herbivores, the leaf area removed by herbivores reached 100% within only 6 months after transplantation, regardless of the presence/absence of artificial shade (Fig. 5; open_h vs. shade_h; P > 0·05 on all dates). In contrast, the impact of herbivores was significantly attenuated and delayed by the presence of S. sphacelata (nurse_h: 78 ± 8% of leaf area grazed after 20 months; nurse_h vs. open_h: P < 0·01 on all dates), but it remained significantly higher than in forest on all dates (forest: 17 ± 6% of leaf area grazed after 20 months; P < 0·001).

Palm Seedlings Beyond Their Physiological Limits under Deforestation and Herbivory

As is the case for most plants in TMCF, C. echinulatum is a shade-tolerant species during its earlier life stages, with seedlings particularly intolerant to drought and full light intensity (Holmgren et al. 2012). These traits explain the high mortality of palm seedlings observed in deforested, non-grazed areas after 20 months of the experiment (93%; see Fig. S1, Supporting information). Their mortality is likely the result of higher maximum levels of atmospheric water stress, light intensity and soil temperature (our data), as well as lower minimum levels of soil humidity (Anthelme et al. 2011). The additional presence of herbivores in deforested areas eliminated all remaining palm seedlings, thus corroborating our first hypothesis that the combination of deforestation and herbivory is responsible for the absence of C. echinulatum regeneration in deforested areas. This interpretation is supported by a significant loss in the GA of leaves observed in open, grazed areas with time following the transplantation of seedlings. In this situation, the loss of leaf area and exacerbated atmospheric water stress took palm seedlings away from their physiological limits, making their survival and growth dependent on habitat amelioration, especially by the presence of nurse plants (Gross et al. 2010; Holmgren et al. 2012). Similar to the case for seedlings of C. echinulatum, many other tree seedlings present in these cloud forests may also not be able to withstand the stresses generated by deforestation and grazing. Therefore, facilitation among plants must be considered a consistent option for the restoration of TMCF.

Facilitation Through Reduction of Abiotic Stress and Herbivory

The tussock grass S. sphacelata was shown to be an efficient nurse plant for palm seedlings in deforested TMCF without herbivory. The mechanisms involved are a reduction of atmospheric water stress (by half; our data), and the provision for soils of a similar humidity as that observed in the adjacent forest during the driest period of the day (Anthelme et al. 2011). This direct habitat amelioration effect explains the increasing positive effect of S. sphacelata observed on Ceroxylon seedlings with time, which is expected to persist in the long term. Owing to the fact that the nurse-plant effects were significantly more positive than those of artificial shade, they were likely not only caused by the shading effects of S. sphacelata, but also (i) by the dense aggregation of tussock aerial stems, which maintains high atmospheric humidity in its core (Anthelme, Michalet & Saadou 2007) and (ii) by the more humid and cooler soil provided by tussock grasses, via specific mechanisms such as hydraulic lift (Prieto, Armas & Pugnaire 2012). Therefore, we provide for the first time experimental evidence for strong facilitative interactions among plants through (abiotic) habitat amelioration in a tropical, productive environment. Our results support the conceptual model of Holmgren & Scheffer (2010). They are also in line with the stress-gradient hypothesis (more facilitation with more abiotic stress) provided that the relative abiotic stress endured by plants – not only the absolute harshness defining an ecosystem – is taken into account in the stress-interaction relationship (see also Smit, Rietkerk & Wassen 2009; Bagousse-Pinguet et al. 2014).

In the presence of herbivores, the effects of S. sphacelata on Ceroxylon seedlings remained positive throughout the 20-month experiment. The facilitative mechanisms involved are therefore partly indirect, that is, encompassing a third biotic component (herbivores). First, they are likely related to ‘shared defense’ because the oldest leaves at the base of Setaria tussocks are barely eaten by large herbivores (Hacker 1987). A second indirect mechanism of facilitation likely to occur is ‘external avoidance’ (sensu Milchunas & Noy-Meir 2002) because seedlings are partially hidden from herbivores by tussocks. These interpretations are supported by the fact that artificial shade, which generated no hiding or protective effects on seedlings, did not increase the performance of seedlings in the presence of herbivores (see Fig. 1; on the contrary, we observed that horses were rather attracted to our ‘shade’ design, in such a way that reduced facilitation might in part be indebted to the methodology used). Consequently, facilitation by nurse tussocks can be explained by both a reduction in relative abiotic stress [model of Holmgren & Scheffer (2010)] and a reduction in the effects of herbivory, up to a certain threshold [model of Smit et al. (2007)]. This corroborated our second hypothesis.

However, combining the effects of abiotic stress and herbivory resulted in a reduction in the facilitative effects exerted by the nurse tussocks, at least when considering the GA of leaves (see the significant ‘Herbivory × Neighbourhood’ effects in Table 1). This result is in line with various studies showing that intense herbivory in an already (abiotically) stressed habitat reduces facilitation among plants (Graff & Aguiar 2011; Soliveres et al. 2011). Furthermore, it lends support to the conceptual and mathematical models that predict the combined influence of stress and disturbance on facilitation among plants to not simply be additive (Smit et al. 2007; Malkinson & Tielbörger 2010; Bagousse-Pinguet et al. 2014). More experimental data are required to fully understand this complex relationship. We suggest that the level of protection against herbivory provided by the nurse tussocks may decline at higher intensities of herbivory because S. sphacelata, which is usually a low palatable species, would switch to become a viable option as a source of nutrition under such conditions.

The facilitative effects generated by S. sphacelata on C. echinulatum under herbivory gradually faded with time during the experiment, in such a way that we can reasonably expect there to be no nurse-plant effects in the long term. This trend might reduce the relevance of using facilitative interactions for long-term restoration projects (e.g. Prévosto et al. 2012). However, while this uncertainty surrounding facilitative effects may generally represent a strong limitation when considering their use for restoration, recent studies on the sister species Ceroxylon quindiuense demonstrated that juvenile individuals of more than 2 m in height allocate the majority of their resources to developing subterranean meristems. This strategy permits them to be highly resilient to damage of their aerial meristems (Sanín et al. 2013). Furthermore, leaves of tropical forest plants, when ageing, develop more chemical and physical defences to protect themselves from herbivory (Coley & Barone 1996), and older palm leaves may be more resilient to predation (Mendoza, Piñero & Sarukhan 1987). Therefore, we assume that the facilitative effects provided by S. sphacelata on C. echinulatum should ensure the long-term survival of these individuals by protecting them until they reach a life stage at which they are more resistant to herbivory (at least at a height of 1 m, which is the height almost all Setaria tussocks reach).

From Theory to Application: Is Restoration of Degraded Cloud Forest Achievable With Tussock Grasses?

Recent reviews have evidenced the important role that facilitative interactions can play in the restoration of degraded ecosystems, including in tropical (dry) regions (Padilla & Pugnaire 2006; Gómez-Aparicio 2009). A meta-analysis that took into account various degraded ecosystems worldwide demonstrated that trees may be predominantly facilitated by nurse shrubs or nurse trees whereas herbs – in particular perennial grasses – may have mostly a negative effect on the survival and growth of trees (Gómez-Aparicio 2009). However, the combined effects of deforestation and herbivory in TMCF generate patterns of impoverished plant communities without trees and shrubs and dominated by invasive tussocks at the regional scale (Aide, Ruiz-Jaen & Grau 2010; Brenner et al. 2012). Therefore, two options for forest nurse-based restoration are possible: planting nurse trees/shrubs, or using existing tussock grasses as potential nurse plants.

The first option should yield the most efficient results in terms of restoration given the higher nurse potential of these life-forms. One experiment using the native nitrogen-fixing tree A. acuminata (Betulaceae) for this purpose in the Colombian cloud forest yielded positive results, but did not consider the effects of herbivores, which were absent from the study site (Murcia 1997). In comparison, empirical data taking into account the effects of herbivores on the regeneration of forest tree species show that shrubs are better nurse plants than trees, especially because of their multistemmed architecture and their crown being located closer to the soil surface (Gómez-Aparicio, Zamora & Gómez 2005). From this viewpoint, and at similar cost, planting shrubs rather than trees may yield better results in grazed deforested areas. However, the expense of planting trees or shrubs reduces the feasibility of each of these restoration techniques (Aide, Ruiz-Jaen & Grau 2010).

An alternative nurse-based restoration option highlighted by our data is to consider existing tussock grasses as potential nurse beds for forest restoration. Even though tussock grasses have in general been found to be competitive with other plants, by developing high densities of roots and aerial stems (Maestre, Bautista & Cortina 2003; Hooper, Legendre & Condit 2005), observed patterns in the deforested cloud forest show positive spatial associations between S. sphacelata and young individuals of C. echinulatum, but only at the base of mother palms where palm seeds and seedlings were found aggregated, probably because of a deficit in the number of animal dispersers (Anthelme et al. 2011; pers. obs.). This demonstrates that the success of restoration using nurse tussocks requires transplanting palm seedlings under the cover of tussocks, thus compensating the limited dispersal generally observed in degraded cloud forest (e.g. Sarmiento 1997). The level of investment needed is minor, with the possibility of collecting seedlings directly in adjacent forests, or at the foot of mother trees in deforested areas. The fact that exotic tussock grasses may be detrimental to the biodiversity of open areas in the tropics (e.g. Hooper, Legendre & Condit 2005) is not relevant here as the tussocks are already established in the deforested areas at high levels of abundance. On the contrary, using pre-established S. sphacelata as nurse plants for C. echinulatum may lead to a reduction of its abundance along with succession.