Introduction

Tropical forests are the global epicentres of insect diversity and abundance, yet tropical insects remain understudied (Stork 2018). Insects are ideal organisms for studying the impacts of climate change as their small size and ectotherm physiology make them particularly sensitive to environmental changes (Rezende and Bozinovic 2019). In addition, tropical insects are sensitive to climate change as they live closer to their upper thermal envelopes than temperate species (Deutsch et al. 2008). It is therefore important to understand the processes and drivers shaping tropical insect communities in order to predict how they are shifting under environmental change.

One framework that has been used extensively to understand how species could respond to further climate change is the ‘bottom-up’ and ‘top-down’ system of control in ecosystems (Preszler and Boecklen 1996; Hunter 2001; Vidal and Murphy 2018; Wenda et al. 2022). Insect herbivores experience ‘bottom-up’ and ‘top-down’ controls. Here, we refer to bottom-up control as selection pressures driven by plant traits such as leaf nitrogen content, leaf area and defensive phytochemicals. We refer to top-down control as selection pressures driven by predators and other natural enemies. The relative importance of bottom-up and top-down controls of insect herbivory have important ecological consequences on processes such as plant biomass accumulation, decomposition, nutrient cycling, plant community, and carbon sequestration via removal of herbivores, altering herbivore metabolic rates (Hawlena and Schmitz 2010a, b) and feeding patterns (Schmitz 2003, 2006), and carbon turnover and allocation in plants (Strickland et al. 2013). Bottom-up control of herbivory occurs via impacts to herbivore survival and fitness due to plant nutrients, and plant defences, which include herbivore-induced plant volatiles that attract natural enemies (Sarfraz et al. 2009; Zhou et al. 2015; Kergunteuil et al. 2019). Bottom-up effects also impact higher trophic levels via plant species mediation of prey quality (Razeng and Watson 2015; Ugine et al. 2021). Top-down control of herbivory occurs via direct removal of herbivores through predation or parasitization, or indirectly by driving behavioural change of herbivores (Long and Finke 2015) and may be more effective in controlling herbivores than bottom-up controls (Vidal and Murphy 2018). However, predator taxa vary in their capacities to facilitate arthropod herbivore top-down control. For instance, ants, through removal of herbivores, may be more important drivers of top-down control of herbivory than vertebrate insectivores or other predatory arthropods in warmer regions such as lowland tropical rainforests (Sam et al. 2022) and birds may be more effective at driving top-down control at higher elevations (Sam et al. 2015).

It has been hypothesised that higher temperatures could result in decreases in foliar nitrogen with associated changes in defensive compounds leading to increased herbivory (Coley 1998). In addition, as insect predators are more sensitive to higher temperatures (Agosta et al. 2018), warming temperatures could lead to a reduction in top-down control of insect herbivores. Just how the relative strengths of bottom-up and top-down controls may shift under changing temperatures remains poorly understood, particularly in the tropics.

Elevational gradients are ideal study systems to understand how climate shapes ecological interactions (Sundqvist et al. 2013; Tito et al. 2020). Temperature, humidity, and UV-B radiation all shift with elevation. These abiotic factors can limit the survival and fitness of species and determine population dynamics (Carbonell et al. 2017) and plant productivity (Lata and Prasad 2011; Bellard et al. 2012; Pereira 2016). Climatic changes along elevation can shape leaf traits (Drollinger et al. 2017; Kfoury et al. 2019; Martin et al. 2020), herbivory (Galmán et al. 2019; Sohn et al. 2019; Sam et al. 2020) and predation (Van Atta et al. 2015; Roslin et al. 2017). These changes to environmental conditions, across relatively small geographic areas, make elevational gradients ideal study systems to understand how climate shapes species interactions, and to make predictions on how climate change will alter these relationships.

Bottom-up and top-down controls of insect herbivores can vary with elevation, leading to changes in herbivory (van de Weg et al. 2014; Fyllas et al. 2017; Roslin et al. 2017; Moreira et al. 2018). Herbivorous insects can be influenced by bottom-up controls, such as plant nutrition, across elevations (Coley 1998; Callis-Duehl et al. 2017; Kergunteuil et al. 2018; Njovu et al. 2019; Zhang et al. 2019). At higher elevations, herbivores may be limited by leaf nitrogen, as nitrogen-fixation decreases with lowering temperatures and the ratio of carbon-compounds in leaves increases (Martin et al. 2020). Conversely, the same can be true at lower elevations, where higher temperatures may cause nutrient dilution in plants, which could drive higher herbivory rates (Welti et al. 2020). Insect herbivores carefully moderate their nutrient intake by source and quantity of food eaten to optimise their fitness (Behmer 2009; Kraus et al. 2019). Nutrient dilution can therefore exacerbate nutritional imbalances in herbivores, impacting their fitness and herbivory rate (Filipiak and Filipiak 2022). Herbivorous insects experience more top-down control via higher predation pressure at lower elevations but are often released from some predation pressure at higher elevations (Godschalx et al. 2019). Higher predation pressure at lower elevations is recognised as a global pattern in nature (Table 1). Therefore, both bottom-up and top-down controls may shape patterns of herbivory across elevation, with a range of ecological consequences for carbon cycling, plant community dynamics and vectoring of plant diseases (Prather et al. 2013).

Table 1 Studies addressing macroecological patterns of predation across elevation
Full size table

By studying both leaf nutrition and predation pressure on herbivory across elevation, we aim to better understand how climate shapes biotic interactions and invertebrate-mediated ecosystem processes. We used permanent study plots in tropical rainforest in Mengla, Xishuangbanna, in Yunnan Provence, China (Kitching et al. 2013), to assess how insect herbivory and predation patterns respond to elevation. We quantified these ecosystem processes along this elevational gradient to test the following hypotheses: (1) The ratio of carbon to nitrogen increases, whereas the predation rate decreases with increasing elevation (Fig. 1). (2) As a result of decreased bottom-up and top-down controls at higher elevations, herbivory increases with elevation. Through investigating the relationships between plants, herbivores, and predators across a climatic gradient, we can better understand the drivers of ecosystem function and how species interactions could shift with further climate change.

Fig. 1
figure 1

Conceptual framework. We expect parallel decreases in temperature, leaf nutrition, folivory rate and predation rate as elevation increases

Full size image

Methods

Study site

Yunnan Province is located in South-West China. It shares borders with Laos, Myanmar, Vietnam and Tibet. A permanent elevational transect was established in Mengla, Yunnan province (101° E, 21° N), as part of the Queensland-Chinese Academy of Sciences (QCAS) Biodiversity Project in 2011 (Fig. 2). Several studies of biodiversity have taken place in these plots (e.g., Ashton et al. 2016; Song et al. 2016; Fontanilla et al. 2019). The climate is monsoonal tropical with some humid sub-tropical areas. Mean annual precipitation is 1493 mm, of which > 80% falls during the wet season (Cao et al. 2006). The wet season is May to October, with peak rainfall in June to August, which also corresponds to the warmest period. The dry season is November to April, the peak being December to February. The mean annual temperature is 21.4 °C (Cao et al. 2006). Temperatures average 17.9 °C in December, the coolest month, and 26.4 °C in June, the warmest month. Along the transect in Mengla twenty 20 × 20 m plots, 300 m apart have been established. The plots are grouped into sets of five at four elevations separated by intervals of approximately 200 m (Fig. 2). The four elevational bands are 800, 1000, 1200 and 1400 m above sea level (m.a.s.l). The lowest elevational plots are located at 800 m.a.s.l. as forest below this has been cleared for rubber tree plantations and other land uses. In the present study data were collected during the wet season between June and August 2019.

Leaf nutrition

Both ubiquitous plant species (i.e. those found across the entire elevational gradient but not necessarily in each plot) and dominant plant species (i.e. those that were most abundant at each elevation, although not necessarily present throughout the elevational gradient, Table 2) were analysed for carbon to nitrogen ratio (C:N) on this elevational gradient. The ubiquitous species were Castanopsis calathiformis (Skan), Croton kongense (Gagnep.) and Tabernaemontana corymbosa (Roxb. Ex Wall.). All species represented in this analysis are native to the study site. Due to the scarcity of C. kongense, leaves were taken from individuals outside of (but in the vicinity of) the 20 × 20 m plots. At each elevation between five and eight leaves (depending on leaf size) were picked from five individual plants of each species. These were dried in a climate control cabinet at 70 °C for 3 days, ground up using a stainless-steel automatic grinder and dried for a further 24 h at 65 °C before weighing for stable isotope analysis (SIA). SIA was carried out at the Institutional Centre for Shared Technologies and Facilities of Xishuangbanna Tropical Botanical Garden, Chinese Academy of Sciences. Samples were combusted in an elemental analyser (Vario Isotope Cube) at 920 °C in the combustion tube and at 600 °C in the reduction tube under air pressure at 1000–1250 bars. Resulting gases were interpreted by IsoPrime100 isotope ratio mass spectrometer (IRMS). Stable isotope ratios are reported using the delta (δ) notation which for δ15N, and δ13C is: [(Rsample/Rstandard) − 1], where R is the ratio of the heavy to light isotope (e.g., 13C/12C), and measured values are expressed in per mil (‰). After IRMS analysis, samples were mass corrected to determine the original sample delta values. Reported δ13C and δ15N values are relative to the international standards V-PBD and atmospheric nitrogen, respectively. Carbon to nitrogen ratio (C:N) was calculated from these data. This information can inform how C:N changes across elevation, within species, which may affect folivory.

Table 2 The plant species studied at each altitudinal band and their average carbon to nitrogen ratio values
Full size table

Predation

Model caterpillars were employed to compare predation across elevation. Model caterpillars were 3 cm long and 0.5 cm in diameter, following the methods of Roslin et al. (2017). Model caterpillars were made of an equal blend of Newplast™ plasticine in colours light green and green which is malleable, oil-based, and non-toxic. We modelled artificial caterpillars by pressing the clay through a metal extruder to ensure that each caterpillar had a smooth surface (Sam et al. 2015). Caterpillars were bent into the locomotive pose of geometrid caterpillars, which are common in the study area.

Over the course of 3 months from June to August in 2019, 2008 plasticine caterpillars were deployed across the 20 plots. A total of ~ 20 plasticine caterpillars were deployed randomly in each plot. Caterpillars were attached to an artificial standardised leaf (hereafter leaves or leaf) affixed with green wire to vegetation at a height of ~ 1.5 m and at least 1 m away from each other. Caterpillars were removed after 1 day and analysed for attack marks. The model caterpillars were examined by eye using a magnifying glass and microscope when necessary to assess attack marks. Using bite mark guides developed from previous studies (Low et al. 2014) any marks observed were documented broadly as invertebrate, lizard, bird or rodent. Marks left by non-predatory arthropods (e.g. grasshoppers) were also documented. There were no controls made for the confounding effect of alternative prey items on observed rates of predatory foraging activity on plasticine caterpillars.

Folivory

To determine folivory, five 1 × 1 m quadrats were sampled within each 20 × 20 m plot across the elevational gradient. Within each quadrat each leaf on up to thirty plants (representing most of the plants within the quadrat) that were 2 m height or less were examined for signs of folivory. Evidence of folivory was assigned a weighting based on the proportion of leaf loss observed: viz: 0 for undamaged leaves, 0.2 (1–20% area loss), 0.4 (21–40% area loss), 0.6 (41–60% area loss), 0.8 (61–80% area loss), 1 (81–100% area loss). Damage was scored by rapid visual assessment, which has been shown to be an accurate form of assessment (Robertson and Duke 1987). Accuracy was assessed by comparing a sub-set of 30 by-eye measurements with the area loss recorded using a leaf area meter app BioLeaf—Foliar Analysis™ (2016). Rapid visual assessments were found to be accurate to within the 20% interval. This method only detects feeding by leaf miners and chewing insects, which were analysed together to elucidated overall changes in folivory along the elevational gradient. Damage from sap-sucking insects requires more detailed observation of the leaf surface that was not practical in this field study. Average density of plants within the plots was not recorded.

Microhabitat temperature and humidity

Our microhabitat data comes from published (Song et al. 2016; Fontanilla et al. 2019) and unpublished microhabitat temperature and humidity measurements recorded from 11th July to 8th September 2015 in each of the twenty plots used in this study. Hourly temperature was recorded from each plot using thermologgers (DS1923 Hygrochron® iButton®, Maxim, CA, USA) placed 1.3 m above the ground (Song et al. 2016).

Statistical analysis

All statistical analysis was undertaken using R (version 4.2.3) and RStudio™ (2015). All models accounted for nesting of survey plots within elevational bands. To test for cofounding effects between elevation and mean microhabitat variables we ran linear regression models. Microhabitat temperature and elevation were found to be highly correlated (Adjusted R-squared: 0.9359), so microhabitat temperature was removed from all models. Goodness of fit for models was assessed using AIC values.

Leaf nutrition

C:N data for both dominant and ubiquitous species exhibited a normal distribution and were analysed using linear mixed effects regression with a Gaussian family and identity link function. Two models were made: one for dominant species at each elevational band and one for species that were ubiquitous across the elevational gradient. The maximal models for both included elevation, and mean microhabitat humidity as fixed effects and plant species as a random effect. Plot number was not used in these models as some leaves were collected from outside plots (but within the elevational band) due to scarcity. Removing mean microhabitat humidity did not improve the fit of the model for dominant plants but did for ubiquitous plants. Interaction terms between elevation and microhabitat humidity improved the fit for the dominant plant model but were not important in the ubiquitous plant model. Our final model for dominant plants contained only elevation as a fixed effect and plant species as a random effect, it also contained interaction terms. Our final model for ubiquitous plants contained only elevation as a fixed effect and plant species as a random effect. Variance components were estimated with restricted maximum likelihood. We used the anova() function in the stats package (R Studio Team 2020) to assess significance of fixed effects in these models. We used the ranova() function in the lmerTest package (Kuznetsova et al. 2017) to compute an ANOVA-like table with tests of random-effect terms in the model to test the random effect of plant species on leaf C:N. Additionally, Tukey multiple comparisons of means with 95% family-wise confidence level was carried out to explore significant differences in C:N across elevational bands and plant species for both models.

Predation

All attacks (arthropod and vertebrate) were analysed together due to the extremely low number of caterpillars attacked by vertebrates (see “Results”). Attack data were analysed as a proportion of those attacked versus not attacked using generalized linear mixed effect regression with a binomial family and logit link function. The maximal model included elevation and microhabitat humidity as fixed effects, and the day of sampling and the plot number as random effects. This model also included interaction terms. Removal of interaction terms and microhabitat humidity improved the fit. The final model therefore included only elevation as a fixed effect, and day of sampling and plot number as random effects. We used the Anova() function in the car package (Fox and Weisberg 2019) to assess the significance of fixed effects on predation activity. We also used the compare_means() function in the ggpubr package (Kassambara 2023) to perform a comparison of means using t-tests with all other groups as a reference to assess differences in predation rate across elevation. We compared anova output of models with stepwise deletions of random effects to test the significance of day of sampling and plot effects on predation activity.

Some model caterpillars fell from their leaves and were therefore not counted. To ensure that the number of model caterpillars which fell at each elevation had no significant influence on our attack rate results we performed the same analysis as above. The number of caterpillars that fell were analysed as a proportion of those that did not fall, with elevation as a fixed effect and day of sampling and plot number as random effects.

Folivory

Leaf area loss data were averaged at the plant level. Folivory was then analysed using negative binomial generalized mixed effects regression with log link function. The maximal model included elevation and microhabitat humidity as fixed effects and quadrat and plot number as a random effect. We added individual plant ID number to the model to test for any plant-level effect on folivory but found it had no significant impact. We found that removal of humidity improved the fit of the model. The final model contained only elevation as a fixed effect and quadrat and plot numbers as random effects. We used the anova() function to assess significance of elevation on folivory. We compared anova output of models with stepwise deletions of random effects to test the significance of plant number and quadrat on folivory.

Fig. 2
figure 2

Map of study sites with individual plots and elevational bands highlighted. Recreated (with permission) from Song et al. (2016). Inlay shows the location of Mengla within Yunnan province

Full size image

Results

Leaf nutrition

From the 139 leaf samples analysed for C:N we found that, contrary to our hypothesis, the C:N in leaves did not change in a uniform way along the elevational gradient. This was true for both the ubiquitous species (F(56) = 3.32, P = 0.07, Fig. 3) and the dominant species (F(80) = 3.48, P = 0.07, Fig. 4) sampled. Due to the stark increase in leaf C:N at 1200 m.a.s.l. for dominant plant species (Fig. 4) we ran a pairwise post-hoc test to assess changes in C:N across elevational bands. This showed that leaf C:N was significantly different between the elevational band at 1200 m.a.s.l. and all other elevational bands (P < 0.0001 for all comparisons). There were no differences in leaf C:N across any other elevational band pairings, so there was no overall effect of elevation on leaf C:N. The RANOVA or ANOVA-like tests of random-effect terms for ubiquitous and dominant species showed that species had a significant effect on C:N in leaves (P < 0.001 for ubiquitous and dominant species). Pairwise comparisons for ubiquitous species showed that C. calathiformis was highly significantly different from C. kongense and T. corymbosa (P < 0.0001), and C. kongense and T. corymbosa were moderately significantly different (P < 0.05). Means and SD for leaf C:N in the ubiquitous species were 24.30 ± 2.73, 16.86 ± 1.65, 15.07 ± 2.04 for C. calathiformis, C. kongense and T. corymbosa, respectively.

Fig. 3
figure 3

Elevation effect on foliar carbon to nitrogen ratio within three species that are ubiquitous across elevation from 800 to 1400 m.a.s.l. These three species were present at all elevations used in this study but were not present in all plots

Full size image
Fig. 4
figure 4

Elevation effect on foliar carbon to nitrogen ratio within species that are dominant at each elevation from 800 to 1400 m.a.s.l

Full size image

Predation

Of the 2008 plasticine caterpillars deployed, 121 were not recovered or had fallen from their leaves, 1655 were not marked, 177 were attacked (some were attacked by multiple predatory groups) and 55 were marked by non-predatory arthropods. Attack marks were left by lizards (2), a rodent (1), and predatory arthropods (174). Most marks left by predatory arthropods were attributed to ants (101) and then beetles (29) but differentiating between marks made by large ants and beetles is difficult, and not all marks left by predatory arthropods could be identified to a finer level, so they are grouped. Marks left by non-predatory arthropods suggest foraging predominately by grasshoppers but also by cockroaches (see Table 3).

Table 3 Bite mark contributions on plasticine caterpillars along the elevational gradient.
Full size table

Predation showed a negative relationship with elevation (Chi(3) = 25.86, P < 0.001), as hypothesised (Fig. 5). Comparison of means showed that attack rates at 1400 m were almost 10% lower than that at 800 m. Comparison of anova output showed no differences between models indicating that random effects did not influence predation activity.

Fig. 5
figure 5

Attack rate, determined by the number of attack-marks left on plasticine caterpillars by predator species, across elevation

Full size image
Fig. 6
figure 6

Folivory rate, determined by the proportion of leaf area loss, across elevation

Full size image

To ensure results were not influenced by the number of model caterpillars that fell these were removed from analysis and we compared rates of loss between plots. Overall, recovery rate of caterpillars was 85.9%. There were significant differences in the numbers that fell, these differences were related to both elevation and the day of sampling. Number of fallen caterpillars decreased with increase in elevation.

Folivory

In total, 11,597 leaves were assessed for leaf area loss due to insect herbivores. Despite differences in folivory among elevations, there was no overall trend in folivory across the elevational gradient (F = 2.39, Chi(3) = 0.11, P = 0.8) (Fig. 6). Average leaf area loss was 18%, 22%, 17% and 21% for 800, 1000, 1200 and 1400 m.a.s.l., respectively. Comparison of anova output showed no differences between models indicating that random effects did not influence folivory.

Discussion

Predation, but not herbivory, declined with elevation in this study. Our findings suggest that top-down control of herbivores decreased with elevation across this gradient, although this did not impact folivory. This result supports the well-recognised global pattern that predation decreases with elevation (Purcell and Avilés 2008; Tvardikova and Novotny 2012; Sam et al. 2015; Hoffman and Avilés 2017; Roslin et al. 2017; Camacho and Avilés 2019). This is thought to be driven largely by ant predation (Roslin et al. 2017). In our study, beetles contributed almost as many marks as ants at lower elevations but attack marks by beetles dropped off dramatically above 1200 m.a.s.l., contributions from other predators (and parasitoids) in our study were negligible along the elevation. The opposite result was found in the tropical Andes, where non-ant predators were responsible for most of the predation events at the higher elevations as predatory ants communities diminished (Camacho and Avilés 2019), while a study in Papua New Guinea found that attacks by non-ant arthropod predators were not correlated with elevation (Sam et al. 2015). Other studies have found negligible predation by non-ant arthropods in comparison to ant predation, but do not offer a break-down of relative predation rates across elevation (Hoffman and Avilés 2017; Roslin et al. 2017). Declines in ant foraging activity with elevation probably drove the overall trends in predation along our gradient as ants were the key predator. Ants are active predators and thermophiles, which contributes to their dominance as predators at low elevations where temperature is higher (Carroll and Janzen 1973; Hölldobler and Wilson 1990; Sam et al. 2015; Molleman et al. 2016; Libra et al. 2019).

Bottom-up control of herbivores by plant nutrients across this gradient was not observed as changes in C:N did not correlate with elevation. Herbivory is influenced by leaf C:N (Castagneyrol et al. 2017; Ruiz-Guerra et al. 2021). For example, one study from Tanzania found C:N was lowest at the mid-elevation, and showed that herbivory was highest at this point and decreased as C:N increased towards the mountain top and lowlands (Njovu et al. 2019). Leaf nutrition is a key mediator of herbivory as nitrogen is needed for insect development: nitrogen limitation can curb growth, reproduction and performance, leading to reduced herbivory (Li et al. 2016; Hansen et al. 2020). We found no evidence of nitrogen limitation in this forest. If bottom-up controls were more important than top-down controls in our study site, this could explain the lack of pattern in folivory along this elevational gradient. Other bottom-up factors such as plant type, secondary plant metabolites, or physical defences may be influencing herbivory across this gradient as has been found elsewhere (Descombes et al. 2017; Galmán et al. 2018; Moreira et al. 2018), but these were not included in our study.

We expected to see an increase in C:N and a decrease in predation along this elevational gradient, due to the effect of lower temperatures on foliar nitrogen and insect activity, respectively (Bruijnzeel and Veneklaas 1998; Rezende and Bozinovic 2019). For C:N no such pattern was observed. This could be because our elevational gradient was both short (800–1200 m.a.s.l.) and tropical (17.9–26.4 °C). Previous studies of change in foliar nitrogen across elevation have had contrasting results (Midolo et al. 2019) but those that found increasing C:N with elevation employed longer gradients, often in temperate locations, incorporating higher altitudes which are subject to cold temperature suppression of biogeochemical processes (Reich and Oleksyn 2004; Fisher et al. 2013; Martin et al. 2020; Ohdo and Takahashi 2020). Our result suggests that, at local scale, moderate climate change may not alter C:N ratio, and there will not be knock-on effects of leaf C:N on folivory. However, changes to leaf nitrogen availability are likely to drive higher folivory as the climate continues to warm (Crous et al. 2018; Dusenge et al. 2019; Johnson et al. 2020; Prieto and Querejeta 2020; Mason et al. 2022). Although environmental change along this elevational gradient was insufficient to drive changes in C:N it may have been sufficient to suppress predator activity as higher trophic levels have lower thermal tolerances than lower trophic levels (Agosta et al. 2018; Wenda et al. 2023).

In this study we used plasticine caterpillars as a proxy for predator activity. Smooth, green model caterpillars are readily attacked as they are associated with palatable, defenseless prey (Heinrich 1979) and the predators responsible for attack marks can often be identified to a fine level (e.g. ant, wasp, large or small bird etc.) (Low et al. 2014). Model caterpillars may under-estimate predation pressure (Nimalrathna et al. 2023) as they do not illicit plant-defense cues, frass or prey chemical cues (Howe et al. 2009; Sam et al. 2015), used by some predators and parasitoids to locate their prey or host (Ananthakrishnan 1990). However, unpublished data by Coley (cited in Richards and Coley 2007) showed that attack rate does not differ between models and live caterpillars. Additionally, both treatments may result in similar trends across an environmental gradient (Libra et al. 2019). In this study we found that the number of plasticine caterpillars that fell (and were therefore not included in analysis) decreased with increase in elevation. Even with the exclusion of potentially predated caterpillars due to falling, and therefore relatively fewer caterpillars contributing to data at low elevation, we saw a clear trend in higher predatory foraging activity at low elevations along this gradient. While model caterpillars may not be ideal for assessing absolute predation pressure, they are useful for assessing changes in predation pressure along an ecological gradient as we can assume bias is conserved. There is a need for more understanding of insect species interactions, such as predation, in the tropics and model caterpillars are a simple, inexpensive, and effective way to collect these data.

While our hypothesis for C:N and folivory were not supported by our data these results do advance our appreciation of the complexity of the driving forces behind changes in insect food web interactions along elevational gradients and promote the need for more research to understand these forces, especially in understudied areas where insects may be more vulnerable to climate change (Deutsch et al. 2008). Topographical features found along the gradient should be accounted for in the analysis. Increases in climate extremes, particularly temperature, will also mediate changes in insect trophic interactions and ecological functioning (Harvey et al. 2020). There is a need for long-term data collection to account for these acute but important climate events.

Our results suggest that microhabitats may initially buffer some aspects of insect food webs against climate change at the local scale. Microhabitats heat up slower than macrohabitats, decreasing the impact of ambient environmental changes on communities (Scheffers et al. 2014). Insect herbivores and their predators are responding in different ways to climate change (Voigt et al. 2003). These asymmetric effects across trophic levels will drive shifts in community structure and assemblage, impacting bottom-up and top-down control of insect herbivores.