We performed measurements in an upland forest (locally called terra firme) permanent plot at the Amazon Tall Tower Observatory (ATTO) site in central Amazonia. The ATTO site is located about 150 km northeast of Manaus in the Uatumã Sustainable Development Reserve (02° 08.9’ S, 59° 00.2’ W, 130 m a.s.l.). The site is situated in a humid tropical climate zone, with a mean annual temperature of 26.7°C and precipitation of 2376 mm and characterized by a pronounced wet season from December to May and a dry season from July to October, with a transitory moderately wet period in between the seasons (Botía et al., 2022). Vegetation in the terra firme plot is dense (leaf area index of 5.3 m² m^-2), mature, and non-flooded, with a mean canopy height of 35 m (Gomes Alves et al., 2023). The soil is a highly weathered and well-drained ferralsol (Chauvel et al., 1987). More details on the experimental site are provided by Andreae et al. (2015).

Located inside the terra firme plot is an 80 m high tower (INSTANT, 02°08.7520′ S, 58°59.9920′ W) with a StarDot RGB camera (model NetCam XL 3MP) installed on top of it at 81 m height facing west. For more details on the camera setup, radiometric calibration, and detection of phenological stages see Lopes et al. (2016). The camera (PhenoCam) monitored upper-crown surfaces of 194 liana-free trees from July 2013 to November 2018 generating an image-derived leaf longevity dataset (i.e., PhenoCam dataset) that allowed the classification of trees into three categories of leaf phenological type which were defined as follows: i) brevideciduous (BD) - trees that lost all of their foliage/part of their foliage and flushed new leaves concentrated in the drier months of the year; ii) evergreen (EV) - trees that showed detectable flushing events and massively flushed new leaves, predominantly in the drier months of the year; and iii) no flushing detected (NF) - evergreen trees that possibly added and lost leaves throughout the years and did not show detectable flushing crown events during the monitoring period (Botía et al., 2022). Leaf phenological type classifications agree with satellite vegetation indices retrieved from MODIS-MAIAC (Multi-Angle Implementation of Atmospheric Correction) for this region (Gonçalves et al., 2020) and branch-level monitoring of leaf age distributions for trees from this plot (Gomes Alves et al., 2023). At the PhenoCam view, the BD group contained 49 trees from 45 species, the EV group 83 trees from 60 species, and the NF 62 trees from 53 species. Only 36 species in the dataset had replicate trees available, and trees of the same species showed different leaf phenological types. Such intra-specific variability in leaf phenological type has been observed in another tropical forest (Park et al., 2019). Moreover, leaf phenological types are subject to phenotypic plasticity, and studies have observed that “random” events such as herbivore attacks, pathogens, and environmental changes caused by extreme events can alter leaf-out phenology patterns in some trees (Borchert, 1999; Cleland et al., 2007; Gonçalves et al., 2020).

Of the 194 trees in the PhenoCam dataset, we were able to sample branches from 175 trees from 124 species of angiosperms. All of the trees occupied the upper canopy layer of the plot and were the most representative in terms of canopy dominance. For all trees, we measured diameter at breast height (DBH, diameter at 1.3 m height), leaf-level isoprene E _c, net photosynthesis rate (A _n), leaf morphological traits (leaf dry matter content, LDMC; leaf mass per area, LMA; leaf thickness, LT; leaf toughness or Force to Punch, FtP) and collected leaves for leaf stable C isotope and elemental analyses, terpene (mono- and sesquiterpene) storage analysis and total phenolics content analysis. We sampled the trees and performed measurements between October 15 - November 9, 2022. This period corresponds to the transition between dry and wet seasons, when tree canopies are mostly composed of mature leaves (Alves et al., 2018; Gonçalves et al., 2020), and variation in leaf age is expected to be low.

Given the logistical challenges of studying tall tropical trees, often exceeding 20 meters in height, leaf measurements were obtained from cut branches immediately placed in water. This method provides a practical solution for conducting gas exchange and isoprene emission measurements, enabling the capture of key ecological processes without compromising leaf viability (Llusia et al., 2014; Albert et al., 2018; Jardine et al., 2020; Taylor et al., 2021; Gomes Alves et al., 2022). Branches with diameters of at least 2 cm were collected from sun-exposed areas of the canopy to avoid shade-adapted leaves. Senescent, immature, or visibly damaged leaves were excluded, ensuring that only physiologically active leaves were analyzed. After collection, branches were immediately re-cut underwater to prevent embolism formation in open vessels, stored in water bottles for transport, and re-cut once more under water at the field camp to restore xylem flow before isoprene E _c and gas exchange measurements (section 2.5).

We selected one visibly mature and healthy leaf of the branch to measure leaf-level isoprene E _c and A _n, then removed the branch from the water, wrapped the lower end of the stem in moist absorbent paper, and placed it in a closed plastic bag for further leaf morphological trait measurements. We selected between 10-20 leaves (fewer larger leaves and more smaller leaves were collected) that were immediately frozen in liquid nitrogen and further taken to Manaus for terpene storage analysis and selected another set of 10-20 leaves that were dried in an oven at 60°C for 72 h, ground and weighed for leaf stable C isotope and elemental analyses and total phenolics content analysis at the Max Planck Institutes for Biogeochemistry (MPI-BGC) and Chemical Ecology (MPI-CE). Finally, we selected four leaves (including the one used to measure isoprene E _c) to measure LDMC, LMA, and LT, and another four leaves to measure FtP. For compound leaves, we considered a leaflet as an equivalent of a simple leaf for all leaf measurements described below. Detailed descriptions of leaf morphological trait measurements, terpene storage analysis, stable C isotope and elemental analyses, and total phenolics content analysis are presented in the Supplementary Material ( Supplementary Methods S1–S4 ).

We measured leaf-level isoprene E _c using a combined LI-6800 portable gas exchange (LiCor Inc., USA) and proton-transfer-reaction quadrupole mass spectrometer (PTR-QMS, IONICON Analytik, Innsbruck, Austria) system, which allows real-time measurements of isoprene emissions under defined environmental conditions of the LI-6800 leaf chamber. We installed a hydrocarbon filter (Restek Pure Chromatography, Restek Corporations, USA) at the air inlet of the LI-6800 to remove isoprene from incoming ambient air. All tubing in contact with the sampling air was PTFE and does not exchange isoprene. At the beginning of each day and before each measurement, we obtained a chamber blank sample from the empty leaf chamber. We separately enclosed the leaf (for compound leaves we considered a leaflet as the equivalent of a simple leaf lamina) in the leaf chamber under standard conditions: photosynthetic photon flux density (PPFD) of 1000 μmol m^-2 s^-1, leaf temperature of 30°C, flow rate of air going into the leaf chamber of 400 μmol s^-1, CO₂ and H₂O concentrations of 420 μmol mol^-1 and 21 mmol mol^-1 and relative humidity of ~60%. The stability criterion for measurements was defined as one standard deviation of the mean A _n, and we visually monitored A _n until the value reached a plateau, beginning measurements when the instrument had reached the defined stability criterion. Leaves displaying no signs of photosynthetic activity were excluded from analyses. A _n was transformed to photosynthesis per leaf dry mass (A _mass) and expressed in units of μg C g^-1 h^-1.

The air exiting the LI-6800 leaf chamber was redirected to the PTR-QMS, which operated in standard conditions with a drift tube voltage of 600 V, drift tube pressure of 2.2 mbar, and E/N 120 Td. Measurements were performed for 10 minutes, and during each PTR-QMS measurement cycle the following mass-to-charge ratios (m/z) were monitored: 21 (H₃ ¹⁸O₊), 32 (O₂ ⁺), and 37 (H₂O-H₃O⁺) with a dwell time of 500 ms each; 41 (isoprene fragment), 69 (isoprene) with a dwell time of 1 s each. Humidity-dependent calibrations (using water-bubbled nitrogen to dilute standard gas, simulating ambient relative humidity) were performed with a certified standard gas provided by Apel-Riemer Environmental, Inc. ( Supplementary Table S1 ), at the beginning and end of the measurement campaign. The mixing ratios of isoprene were calculated from the calibration curves (R² = 0.99). The detection limit of the PTR-QMS was calculated as three times the standard deviation of isoprene (ppb) detected in the water-bubbled nitrogen background of the calibration curves and was equal to 0.93 ppb. Cross-validation for isoprene data obtained by in situ PTR-QMS measurements and by adsorbent cartridges analyzed via GC-FID was performed in a previous study showing a coefficient of determination (r ²) of 0.88 (Yáñez-Serrano et al., 2015). Once mixing ratios of isoprene (ppb) from the samples were obtained, isoprene emission capacity per area E _c,A was determined using the equation (E _c,A = Rppb × Q/S), where E _c,A (nmol m^-2 s^-1) is the leaf flux of isoprene emission; Rppb (nmol mol^-1) is isoprene concentration of the outgoing air; Q is the flow rate of air into the leaf chamber (400 x 10^-6 mol s^-1); S is the area of leaf within the chamber (0.0002 m² or 0.0006 m²). Values of isoprene E _c,A were transformed to units of isoprene emission capacity per dry mass (E _c,M, µg C g^-1 h^-1).

Because the number of replicates per species available in our sampling plot prevented characterizing species-level variation, we focused on individual-level analyses and controlled potential species-level effects by performing mixed-effects models with species as random factor. To evaluate if the presence of isoprene emissions or terpene storage changed between leaf phenological types, we performed chi-squared (χ²) analysis to compare observed and expected proportions of detected isoprene E _c,A, and mono-/sesquiterpene storage in the full dataset and to compare observed and expected proportions of detected isoprene E _c,A and mono-/sesquiterpene storage in each leaf phenological type.

To evaluate if the magnitudes of isoprene emissions and terpene storage changed between leaf phenological types, we performed mixed-effects pair-wise comparisons of the magnitude of isoprene E _c,A and relative abundances of stored mono-/sesquiterpenes between leaf phenological types. To evaluate whether the interactions between leaf phenological types and functional traits influenced the presence of isoprene emissions or the variation in isoprene emission rates, we performed univariate mixed effects linear regression models (UMELMs) of detected isoprene E _c,M/magnitude of isoprene E _c,M ~ functional trait * leaf phenological type + (1| Species). Given that some traits had missing data (NA), UMELMs were performed with a reduced sample size of n = 154 for detected isoprene E _c,M and n = 81 for the magnitude of isoprene E _c,M (only trees with detected isoprene E _c,M). We performed univariate models instead of a single multiple model containing all functional traits measured because our number of observations did not allow for the inclusion of all these variables and their interactions in a single multiple model (Harrell, 2001; Burnham and Anderson, 2002; Babyak, 2004).

UMELMs were performed using the lmer function of the LME4 R package (Bates et al., 2015). The p-values of mixed effects pairwise comparisons and UMELMs were obtained with the EMMEANS package (Lenth, 2024). Distributions of detected isoprene E _c,A, and mono-/sesquiterpene storage between leaf phenological types and results of mixed effects models are presented as plots from the GGPLOT2 package (Wickham, 2016). All statistical analyses were performed using R version 4.3.2 through the platform RStudio 2023.9.1.494 (R core team, 2023).

We have detected isoprene emissions in 88 trees from 69 species, which corresponds to c. 50% of all trees and 55.6% of all species. Most studies so far have reported only between 20-38% of tropical tree species as isoprene emitters (Harley et al., 2004; Loreto and Fineschi, 2015), but recent studies have shown that this number can be even larger, up to 76% (Jardine et al., 2020; Mu et al., 2022). Results of the distribution of isoprene emitters between leaf phenological types showed slight differences compared to what has been observed by Gomes Alves et al. (2023) through emission probability modeling, suggesting a prevalence of non-emitters of isoprene within the group of massively flushing evergreen (EV) trees. Because evergreen trees are more dominant in the central Amazon forest compared to brevideciduous trees (Condit et al., 2000; Aleixo et al., 2019), a higher fraction of non-emitters in this group may imply lower regional fluxes in this region. However, contrary to our hypothesis, we did not see significant differences in isoprene emission rates comparing leaf phenological types. The fact that brevideciduous trees in this forest may not lose all their foliage and only remain leafless/partially leafless for shorter periods, combined with the fact that this forest is mostly composed of dry-season flushing evergreen trees (Condit et al., 2000; Aleixo et al., 2019; Mesquita Pinho, 2021), could indicate that the observed seasonality of higher isoprene emissions at the end of the dry season/early wet season is probably being driven by a higher fraction of mature leaves in the canopy, regardless of leaf phenological type (Alves et al., 2018; Gomes Alves et al., 2023).

Meanwhile, we have detected sesquiterpene storage in a significantly large number of trees, with no differences in relative abundances of monoterpene and sesquiterpene storage comparing leaf phenological types, contradicting our initial hypothesis. These results show that terpene storage is widespread regardless of leaf phenological types and emphasize how tropical tree species are more complex and do not hold to general assumptions or plant trait coordinations found in temperate tree species (Dani et al., 2014). Sesquiterpenes require more carbon for production (15C) compared to isoprene (5C), with a yield rate of secondary organic aerosol (SOA) formation that can reach up to 70% (Griffin et al., 1999b, 1999a), while for isoprene it has been reported as <6% (Kroll et al., 2005; Xu et al., 2014). Studies have observed significant temperature-induced emissions of these heavier terpenes at temperatures above 35°C (Nagalingam et al., 2023; Robin et al., in preparation). Considering that tropical forest canopies can frequently experience such high temperatures (Jardine et al., 2017; Manzi et al., 2024), the large number of sesquiterpene-storing species we observed may indicate a strong potential for higher fluxes of these compounds than previously estimated by emission models (e.g., Guenther et al., 2012), and may incur higher carbon losses to the atmosphere if the forest is under more frequent stress (e.g.; heatwaves, insect outbreaks), but more research is needed to test this.

Even though isoprene emission rates did not vary between leaf phenological types, interactions between functional traits and leaf phenological types were significantly related to variations in the presence and magnitude of isoprene emissions. Much had been discussed on the roles of isoprene in increased thermotolerance (Singsaas et al., 1997; Pollastri et al., 2014, 2019) and oxidative stress protection (Vickers et al., 2009; Morfopoulos et al., 2013, 2014; Rodrigues et al., 2020), and recent research has demonstrated that the presence of isoprene emissions is related to multiple up and down regulations of gene expression, transcription factors, and protein abundance (Behnke et al., 2010; Harvey and Sharkey, 2016; Lantz et al., 2019; Zuo et al., 2019; Frank et al., 2021; Monson et al., 2021; Dani et al., 2022; Weraduwage et al., 2023; Srikanth et al., 2024). The current view is that isoprene occupies a unique metabolic position, mediating processes that govern the supply of photosynthetic substrates and the requirements for secondary metabolite products, enabling plants to allocate resources to defense while minimizing the impact on growth (Monson et al., 2021).

Isoprene-emitting trees from the NF group were less likely to store monoterpenes and had tougher and less photosynthetically active leaves (i.e., higher mechanical defense). Meanwhile, in brevideciduous trees, isoprene emission rates were higher with a higher diversity of stored sesquiterpenes and higher total phenolics content (i.e., higher chemical defense). There are only a few multi-omic studies evaluating how the presence of isoprene emissions regulates chemical and mechanical defenses (Harvey and Sharkey, 2016; Zuo et al., 2019; Monson et al., 2020, 2021). In isoprene-emitting (IE) poplar leaves, for example, the presence of isoprene emissions was associated with increased expression of genes involved in the accumulation of lignin, supporting our observation of isoprene emitters from the NF group with higher FtP and CN ratios (Monson et al., 2020, 2021). On the other hand, opposite relationships between isoprene emission and terpene accumulation have been observed: in IE poplar and tobacco leaves, the presence of isoprene emissions was related to reductions in the expression of genes and proteins involved in terpene biosynthesis (Zuo et al., 2019; Monson et al., 2020, 2021); however, Harvey and Sharkey (2016) observed increases in transcript abundances of terpene synthesis-related genes under fumigation with isoprene for Arabidopsis plants. As for phenolics, the observed relationship between higher isoprene emission rates and higher phenolics content in BD trees corroborates with Behnke et al. (2010) and Monson et al. (2020), which showed that the presence of isoprene emissions was associated with an upregulation in the expression of genes in the phenylpropanoid pathway with consequent increases in production of phenolic compounds.

Trees from EV and BD groups presented regular flushing events in the early and mid-dry seasons, respectively, which means that their canopies have a more homogeneous leaf age composition (Lopes et al., 2016; Gonçalves et al., 2020). In contrast, since NF trees did not show detectable flushing events and probably flushed/lost new leaves continuously throughout the years, canopies from these trees likely have greater leaf age heterogeneity. Although we did not measure visually old leaves, and this group did not show significantly lower A _mass or LMA compared to the other groups ( Supplementary Figure S2 ), it is possible that we measured slightly older leaves for these trees, and isoprene emissions decrease together with photosynthesis as leaves get older (Schnitzler et al., 1997; Gomes Alves et al., 2023). On the other hand, part of the NF group could be composed of trees that did not have detectable flushing events because they flush dark green leaves that are not detected as young by the PhenoCam, hence more research at the branch or leaf levels on these trees is needed to understand the mechanisms driving their leaf renewal.

One hypothesis to explain the evolutionary drivers of leaf-out phenology in upland central Amazon forests suggests that trees flush new leaves in the dry season as a way to avoid increased herbivory pressure in the wet season, since young leaves have fewer structural defenses (i.e., softer, thinner), and are more palatable to herbivores, which are more abundant in the rainy season (Wright and van Schaik, 1994; Coley and Barone, 1996; Lopes et al., 2016). Perhaps, in brevideciduous trees, the presence of isoprene emissions could be influencing metabolic regulation towards a more functionally diverse chemical-based defense, that protects the large fractions of synchronized newly flushed and vulnerable young leaves (Coley and Barone, 1996; Lopes et al., 2016). Meanwhile, in evergreen trees that continuously flushed/lost leaves (NF), isoprene may be associated with an upregulation of defense towards a more long-term, lignin-based structural defense that supports longer leaf longevities (Monson et al., 2021). This suggests that isoprene emission in this forest could be involved in the co-regulation of a chemical-mechanical defense trade-off (van der Meijden et al., 1988; Pedersen et al., 1995) between brevideciduous (BD) and evergreen trees with continuous flushing (NF), which is reinforced by the observation that isoprene emitters in NF are also less likely to store monoterpenes.

Recent studies have demonstrated how isoprene is intrinsically interconnected with broad patterns of gene expression, and that leaf phenological types are under strong genetic control, being less of an observable trait and more of a dynamic response that results from gene-environment interactions (Satake et al., 2024). For example, studies on the seasonal expression of BVOC synthesis-related genes in two tree species of Fagaceae (Quercus glauca and Lithocarpus edulis) showed that genes downstream of the MVA pathway, involved in sesquiterpene production, had increased expression during the period that matches leaf flushing for these trees (Satake et al., 2023, 2024). Even though we did not measure visually young leaves, brevideciduous trees possibly produce new leaves earlier than evergreen trees, so it is possible that their canopies were overall composed of slightly younger leaves in comparison to other leaf phenological types, and thus had increased expression of sesquiterpene synthase genes. Considering this underlying molecular component of plasticity in leaf phenological types and associations with pathways of isoprene and sesquiterpene synthesis, perhaps our results suggest that, in this resource-abundant, species-rich, ecologically-complex upland terra-firme central Amazon forest, the direction of isoprene’s regulation over growth and defense is possibly being influenced by leaf phenological type, although more research is needed to test this hypothesis.

Lastly, isoprene emission rates were significantly higher with characteristic resource-acquisition traits (Wright et al., 2004), like lower mechanical resistance (low FtP), and higher nitrogen content (lower CN) and A _mass. It is reasonable that, independent of leaf phenological type, emission rates from isoprene emitters would be higher with such traits that enable faster leaf metabolism, hence providing sufficient carbon uptake to support stronger emission rates (Loreto and Sharkey, 1990; Delwiche and Sharkey, 1993; Loreto et al., 1996; Magel et al., 2006; Sharkey and Monson, 2017). Still, isoprene emitters from evergreen trees were generally constrained towards more resource-conservative strategies (higher FtP and lower A _mass, and CN) which, given the predominance of evergreen trees in central Amazon forests (Condit et al., 2000; Aleixo et al., 2019), emphasizes the importance of incorporating leaf phenological type when estimating regional and global fluxes.

Although our results showed that isoprene emissions and terpene storage did not significantly vary between evergreen and brevideciduous trees, they revealed that interactions between traits and leaf phenological types drive variations in the presence of isoprene emissions and isoprene emission rates. Isoprene-emitting trees with no detectable flushing were less likely to store monoterpenes and had tougher and less photosynthetically active leaves, while brevideciduous trees showed higher isoprene emission rates with a higher diversity of stored sesquiterpenes and total phenolics content. Recent studies have revealed that isoprene is an integrative compound that co-regulates both growth and defense responses by promoting changes in gene expression patterns and protein abundances. Our results perhaps suggest that the direction of this co-regulation is influenced by leaf phenological types, and that isoprene emissions participate in co-regulating a chemical-mechanical defense trade-off between brevideciduous and evergreen trees with continuous flushing in central Amazonia. Moreover, we detected isoprene emissions and sesquiterpene storage in a greater number of trees than expected, which indicates a greater potential for emissions of these compounds than previously thought (Harley et al., 2004; Guenther et al., 2012; Loreto and Fineschi, 2015).

Isoprene and sesquiterpene emissions, directly and indirectly, influence atmospheric processes and cloud formation, with sesquiterpenes having a yield rate of particle formation almost 10 times that of isoprene (Griffin et al., 1999b, 1999a; Kroll et al., 2005; Xu et al., 2014). Warmer climates might favor the predominance of thermotolerant isoprene-emitting trees (Singsaas et al., 1997; Pollastri et al., 2014, 2019; Taylor et al., 2018), and increased heat stress and herbivore outbreaks can induce stronger sesquiterpene emissions (Nagalingam et al., 2023; Robin et al., in preparation), but the exact effects of current global climate changes and multiple stressors (e.g. extreme heat events, more frequent and intense droughts and flooding, elevated CO₂ and O₃) on forest-atmosphere emission feedbacks are uncertain (Yáñez-Serrano et al., 2020; Satake et al., 2024). The Amazon forest is the greatest source of volatile isoprenoid emissions to the atmosphere (Jardine et al., 2020; Mu et al., 2022), and a better understanding of the dynamics between emissions, leaf phenological types and functional traits in this forest is essential to provide a more mechanistic understanding of emissions and improve their representation in models.