Yield response of field-grown soybean exposed to heat waves under current and elevated [CO2]
Abstract
Elevated atmospheric CO2 concentration ([CO2]) generally enhances C3 plant productivity, whereas acute heat stress, which occurs during heat waves, generally elicits the opposite response. However, little is known about the interaction of these two variables, especially during key reproductive phases in important temperate food crops, such as soybean (Glycine max). Here, we grew soybean under elevated [CO2] and imposed high- (+9°C) and low- (+5°C) intensity heat waves during key temperature-sensitive reproductive stages (R1, flowering; R5, pod-filling) to determine how elevated [CO2] will interact with heat waves to influence soybean yield. High-intensity heat waves, which resulted in canopy temperatures that exceeded optimal growth temperatures for soybean, reduced yield compared to ambient conditions even under elevated [CO2]. This was largely due to heat stress on reproductive processes, especially during R5. Low-intensity heat waves did not affect yields when applied during R1 but increased yields when applied during R5 likely due to relatively lower canopy temperatures and higher soil moisture, which uncoupled the negative effects of heating on cellular- and leaf-level processes from plant-level carbon assimilation. Modeling soybean yields based on carbon assimilation alone underestimated yield loss with high-intensity heat waves and overestimated yield loss with low-intensity heat waves, thus supporting the influence of direct heat stress on reproductive processes in determining yield. These results have implications for rain-fed cropping systems and point toward a climatic tipping point for soybean yield when future heat waves exceed optimum temperature.
1 INTRODUCTION
As the global population is expected to exceed 9.7 billion by 2050 (UN DESA, 2017), improving yield under increased climate variability remains a persistent challenge across cropping systems (Lobell & Gourdji, 2012). Average atmospheric CO2 concentration ([CO2]) has surpassed 400 µmol/mol and on the current trajectory could reach in excess of 550 µmol/mol with a concomitant increase in global mean temperature approaching 2°C by 2080 with more extreme heating on land surfaces (IPCC, 2014, 2018). Recent climate models also project more frequent extreme climate events such as heat waves (i.e., extremely hot weather outside the historical means; Karl et al., 2008). These abrupt periods of excessive temperature may have a greater effect than gradual changes in mean temperature, as heat waves that surpass the temperature optimum (Topt) of physiological and developmental processes lead to more variable yields or even to crop failure (Gornall et al., 2010; Hatfield et al., 2011; Herring et al., 2016). Such an impact on cultivated regions will significantly affect global food production and security (Lesk, Rowhani, & Ramankutty, 2016).
Soybean (Glycine max), a C3 plant, is among the most important industrial and food crops of the world. Together, the United States, Brazil, and Argentina produce 88% of the world's soybean that is utilized for livestock feed, vegetable oil, biodiesel, and direct human consumption (FAO, 2015). In the Midwestern United States, the sensitivity to changes in [CO2] varies across soybean cultivars, but field studies at elevated [CO2] consistently show photosynthesis (AN), water-use efficiency (WUE), and yield all increase (Ainsworth & Long, 2005; Bernacchi, Kimball, Quarles, Long, & Ort, 2007; Bernacchi et al., 2006; Bishop, Betzelberger, Long, & Ainsworth, 2015; Leakey et al., 2009; Sanz-Sàez et al., 2017). These benefits, referred to as the “CO2 fertilization effect,” are threatened by increasing frequency and intensity of heat waves driven by the “CO2 greenhouse effect” (Ainsworth & Ort, 2010; Pryor et al., 2014). Yet, field studies investigating the impact of heat waves on crops including soybean are few (Siebert & Ewert, 2014), and modeled projected yields of soybean exposed to heat waves under elevated [CO2] are inconsistent. One model suggests the higher Topt of soybean at anthesis could safeguard this crop from future heat waves (Gourdji, Sibley, & Lobell, 2013), while another model predicts up to a 26% overall reduction in yield by 2080 when heat waves occur together with increasing mean temperature with soybean yields in the Midwestern United States among the most susceptible to heat stress (Deryng, Conway, Ramankutty, Pricee, & Warren, 2014; Teixeira, Fischer, van Velthuizen, Walter, & Ewert, 2013). Currently, most yield projections for C3 crops incorporate the “CO2 fertilization effect,” but whether this yield stimulation is maintained when field-grown cultivars are exposed to intermittent heat waves is uncertain (but see Jin et al., 2017; Wing, Monier, Stern, & Mundra, 2015).
Separately, heat stress and elevated [CO2] are opposing environmental conditions that affect primary production through cellular-, leaf-, and plant-level processes in C3 species. At the cellular level, heat stress impacts AN mainly through ribulose-1,5-bisphosphate (RuBP) carboxylase-oxygenase (Rubisco)-limited (Vc,max; maximum carboxylation rate) or RuBP regeneration-limited (Jmax; maximum electron transport rate) processes (Bernacchi, Pimentel, & Long, 2003; Bernacchi, Singsaas, Pimentel, Portis, & Long, 2001; Farquhar, von Caemmerer, & Berry, 1980). At ambient [CO2] and saturating light, Vc,max primarily limits AN as temperature increases because photorespiration is enhanced due to decreasing specificity of Rubisco for CO2 compared to O2 and the lower ratio of CO2:O2 in solution (Jordan & Ogren, 1984; Sage, Way, & Kubien, 2008). Alternatively, as Topt is exceeded, electron transport capacity can decrease, and Jmax may limit AN (Cen & Sage, 2005). At the leaf level, heat stress reduces AN when stomatal conductance (gs) decreases under increasing vapor pressure deficit (VPD; Berry & Björkman, 1980) or with reductions in soil moisture, especially when heat waves coincide with drought (Gray et al., 2016; Perdomo, Capó-Bauçà, Carmo-Silva, & Galmés, 2017). Finally, at the plant level, heat stress adversely affects growth and yield by accelerating crop development and shortening the grain filling period and exceeding the Topt of reproductive processes (Hatfield et al., 2011; Sage et al., 2015). In the latter case, even if crops were to maintain high rates of AN, heat stress during key reproductive stages that impairs flower and fruit development could result in crop failure (Sage et al., 2015; Wheeler, Craufurd, Ellis, Porter, & Vara Prasad, 2000). Within the Topt of temperate C3 plants (~20–35°C; Schrader, Wise, Wacholtz, Ort, & Sharkey, 2004), the Topt of soybean varies between ~30°C during the vegetative phase and ~23°C post-anthesis (Baker, Allen, Boote, Jones, & Jones, 1989; Hatfield et al., 2011; Jumrani, Bhatia, & Pandey, 2018). Two critical stages in soybean development are flowering (R1) and pod-filling (R5). Temperatures ≥32°C decrease soybean flower initiation and pollen viability at R1 and decrease pod set and yield at R5 (Egli & Wardlaw, 1980; Koti, Reddy, Kakani, Zhao, & Reddy, 2004; Puteh, Thuzar, Mondal, Abdullah, & Halim, 2013; Salem, Kakani, Koti, & Reddy, 2007).
In contrast to heat stress, elevated [CO2] increases primary production in C3 crops, although this response is influenced by the duration of exposure (Long, Ainsworth, Rogers, & Ort, 2004). In the short term, elevated [CO2] stimulates AN as carboxylation is enhanced with fewer resources lost to photorespiration (Drake, Gonzàlez-Meler, & Long, 1997; Kromdijk & Long, 2016) and WUE improves. This heightened efficiency of AN corresponds with improved growth and yield in soybean (Ainsworth et al., 2002). Under long-term exposure to elevated [CO2], higher rates of AN may be offset by declining photosynthetic capacity (Long et al., 2004). This loss of capacity may be induced by an accumulation of nonstructural carbohydrates that occurs when plants are “sink limited” or unable to utilize the excess photosynthate generated at higher [CO2]. In soybean, sink limitation typically occurs in determinate and non-nodulating cultivars (Ainsworth et al., 2002; Ainsworth, Rogers, Nelson, & Long, 2004; Rogers et al., 2004).
While the main effects of temperature and elevated [CO2] have been broadly examined in soybean and most C3 species, it remains difficult to make any generalization about the interaction, even though their co-occurrence characterizes the present and future climate system. A classic review of the topic predicts that elevated [CO2] would have the greatest effect on AN at higher temperature because photorespiration is inhibited as CO2 increases (Long, 1991). Most field studies that include elevated [CO2] and temperature interaction focus on changes in mean temperature (i.e., gradual changes) and rarely exceed 4°C above ambient (Chaturvedi, Bahuguna, Shah, Pal, & Krishna Jagadish, 2017; Fitzgerald et al., 2016; Rosenthal et al., 2014; Ruiz-Vera et al., 2013). Recently, Siebers et al. (2015) measured a 10% reduction in yield when field-grown soybean was exposed to heat waves that increased canopy temperature 6°C above ambient. While Siebers et al. (2015) did not include an elevated [CO2] and heat wave interaction, the significant reduction in yield following brief (3-day) heat waves reveals an urgent need for field studies to extend beyond shifts in mean temperature and to investigate interactions with elevated [CO2] (IPCC, 2014; Kreyling, Jentsch, & Beier, 2014; Thompson, Beardall, Beringer, Grace, & Sardina, 2013). Moreover, the intensity of future heat waves will be influenced by the degree of change in mean temperature from low-emission (1.5°C RCP2.6) to high-emission (4.5°C RCP.8.5) scenarios (IPCC, 2014).
Here, we utilized Free Air CO2 Enrichment (FACE; as described by Miglietta et al., 2001) to elucidate how predicted changes in atmospheric [CO2] interact with heat wave intensity to influence soybean yield through cellular-, leaf-, and plant-level processes. Moreover, we coupled in situ measurements with a dynamic global vegetation model (DGVM) to determine whether heat waves alter soybean yields primarily through direct changes in carbon assimilation or alternative processes, such as heat stress on reproductive tissues. We focused on 5-day heat waves that increased canopy temperature 5 and 9°C above ambient, because 5°C heating events are historically representative and heat waves >7°C, while rare for this region (Siebers et al., 2015; Westcott, 2011), will become more likely as global climate change advances. We asked the following questions: (a) how is CO2 fertilization during R1 and R5 reproductive stages affected across the cellular, leaf, and plant level by low- (+5°C) intensity and high- (+9°C) intensity heat waves; and (b) how do the measured versus modeled yields compare when soybean, grown under present [CO2] and elevated [CO2], are exposed to heat waves during key reproductive (R1 and R5) stages? In general, we predicted that the “CO2 fertilization effect” would be maintained from the cell to plant level as long as heat waves did not exceed Topt.
2 METHODS
2.1 Field site
This study was located at the SoyFACE research facility, Savoy, IL, USA (40°02′N, 88°14′W, 228 m above sea level). Briefly, the field site supports half soybean (G. max) and half maize (Zea mays) rotated annually. Soils across the 32 ha field site are deep, organically rich, and of the Flanagan/Drummer series (Ort et al., 2006). The regional climate is characterized as temperate with cold winters and warm summers. Mean annual temperature is 10.9°C with an average low temperature of −8.5°C in January and a high temperature of 29.4°C in July. On average, annual precipitation is 1,050 mm/year. Additional information about the SoyFACE facility can be found at: http://www.soyface.uiuc.edu.
2.2 Experimental design
For both years of the study, soybean (cv. Pioneer P35T58R, Hi Bred International Des Moines) was sown in 0.38 m row spacing with a planting density of 29 plants/m2 in June (Table 1). Experimental plots (Figure S1) were arranged in a randomized complete block design (n = 4) as previously described by Ruiz-Vera et al. (2013) and assigned one of two [CO2]: (a) ambient (400 µmol/mol; n = 4); and (b) elevated (600 µmol/mol; n = 4) targeting CMIP5 under the RCP8.5 emission scenario for 2080 (Collins et al., 2013). Within each experimental plot, three 7 m2 subplots were established comprising either an ambient [CO2] control (A) and two heat wave (H) subplots (A + H) or an elevated [CO2] control (E) and two H subplots (E + H; Figure S1). One subplot of A + H and E + H was used for heating during R1 (flowering; Board & Kahlon, 2011) and one was used for heating during R5 (pod-filling; Board & Kahlon, 2011). On average, 5 day heat waves increased canopy temperature above the control by 9 and 5°C during the 2015 and 2016 growing seasons, respectively (Table 2). Within each heated subplot, six overhead infrared (IR) heating units (Mor Electric Heating Association Inc.) were installed in a circular array 1.2 m above the soybean canopy for a total heated area of 7 m2 as previously described by Ruiz-Vera et al. (2013) and Siebers et al. (2015).
| Year | Date (DOY) | |||
|---|---|---|---|---|
| Seed sown | R1 heat wave | R5 heat wave | Final harvest | |
| 2015 | 5 June (156) | 18–22 July (199–203) | 8–12 August (220–224) | 13 October (286) |
| 2016 | 6 June (158) | 23–27 July (205–209) | 13–17 August (226–230) | 12 October (286) |
- Abbreviations: DOY, day of year; R1, flowering; R5, pod-filling.
| Treatment period | A | A + H | E | E + H | Significant effects (p < .1) |
|---|---|---|---|---|---|
| Tcan (°C) | |||||
| 2015 R1 | 23.3 ± 0.1 | 32.5 ± 0.8 | 23.4 ± 0.2 | 32.5 ± 1.4 | n/a |
| 2015 R5 | 22.1 ± 0.3 | 31.6 ± 0.8 | 22.2 ± 0.1 | 31.2 ± 0.5 | n/a |
| 2016 R1 | 24.4 ± 0.1 | 29.7 ± 0.3 | 24.7 ± 0.2 | 30.0 ± 0.1 | n/a |
| 2016 R5 | 22.4 ± 0.03 | 27.4 ± 0.2 | 22.6 ± 0.1 | 27.8 ± 0.1 | n/a |
| θ v | |||||
| 2015 R1 | 33.8 ± 1.4 | 28.1 ± 1.9 | 35.8 ± 1.6 | 30.4 ± 2.2 | [CO2], H |
| 2015 R5 | 27.1 ± 1.1 | 24.9 ± 1.1 | 29.2 ± 1.3 | 25.6 ± 1.3 | H |
| 2016 R1 | 36.3 ± 2.0 | 32.4 ± 2.7 | 36.2 ± 1.6 | 30.8 ± 2.5 | H |
| 2016 R5 | 38.9 ± 1.9 | 37.5 ± 2.1 | 37.7 ± 1.6 | 35.7 ± 1.8 | ns |
Note
- All data are mean ± 1 SE.
- Abbreviations: A, ambient [CO2]; E, elevated [CO2]; H, heat; n/a, data that were not statistically analyzed; ns, nonsignificant differences; R1, flowering; R5, pod-filling.
2.3 Sensor measurements
Soybean canopy temperature was measured continuously in each ambient and heated plot using an IR radiometer (IRR; SI-121; Apogee Instruments, Inc.) maintained at 1.2 m above the canopy throughout the growing season as previously described by Kimball (2005) and Siebers et al. (2015). Soil volumetric water content (θv) was measured in each subplot by vertically inserting a handheld soil moisture probe (Hydrosense II; Campbell Scientific) under the canopy. The measurements of θv, which occurred during gas exchange measurements (see below), are reported as mean θv for days 1, 3, 5 of each heat wave and day 6 (post-heat wave). All values of θv were integrated over the top 20 cm because aboveground heating treatments at this site have been shown to impact soil moisture at this depth (Rosenthal et al., 2014).
2.4 Gas exchange and biochemistry measurements
Leaf-level photosynthesis (AN), stomatal conductance (gs), intercellular CO2 concentration (Ci), leaf VPD (VPDL), and transpiration (ET) were measured using LI-6400 portable photosynthesis systems with leaf chamber fluorometers (LI-COR). Instantaneous WUE (WUEINST) was calculated as AN/ET. For all measurements, irradiance (LED light source; LI-COR) was set to ambient conditions. Block temperature was also set to ambient temperature and at 5 or 9°C above ambient temperature in the heated subplots. Reference CO2 was set to 400 µmol/mol in ambient plots and 600 µmol/mol in elevated [CO2] plots. Midday measurements of gas exchange occurred between 12:00 and 14:00 hr 24 hr prior to heating, on days 1, 3, and 5 during each heat wave, and 24 hr following a heat wave to measure recovery. Data on day 3 of the 2016 R5 treatment were not collected due to rain. In addition to leaf gas exchange, midday leaf water potential (Ѱmd) was measured at 12:00–14:00 hr. Leaves were collected and placed immediately into plastic bags, stored in a cooler, and returned to the laboratory to be measured with a Scholander-type pressure chamber (PMS, Model 1000, PMS Instrument Co.).
AN versus Ci concentration (AN–Ci) curves were measured on day 4 of each heat wave. For all leaves sampled, the petioles were cut in the field predawn, immediately placed in water, and returned to the laboratory where the petioles of all leaves were recut under water and kept in a dark growth chamber (Bernacchi, Morgan, Ort, & Long, 2005). Prior to measurements, leaves were acclimated at 1,200 µmol m−2 s−1 irradiance for 30 min. For all measurements, light levels were set at saturating irradiance (1,700 µmol m−2 s−1), the block temperature was set to 25°C, and the reference [CO2] was set at the respective growth [CO2]. At steady state, measurements were initially recorded at the growth [CO2], after which [CO2] was decreased incrementally to 50 µmol/mol and then increased incrementally to 1,600 µmol/mol. The measurements of AN–Ci were used to solve for Vc,max and Jmax using equations from Farquhar et al. (1980). All values were corrected for temperature following Bernacchi et al. (2001, 2003).
To determine the concentration of soluble sugars, 7.1 mm2 leaf discs were collected from the uppermost leaves of two separate plants at midday (12:00–13:00 hr) on the last day of each heat wave. Each leaf disc was placed in a 2 ml vial and immediately placed in liquid nitrogen and stored at −80°C. Total nonstructural carbohydrate (TNC: glucose, fructose, sucrose [hereafter referred to as GFS], and starch) concentration was determined according to methods previously described in Jones, Outlaw, and Lowry (1977). Leaf TNC concentration was calculated as the sum of GFS and starch concentration per leaf disc.
2.5 Mid-season and final harvests
Within 3 days following a heat wave, mid-season harvest was conducted in one undisturbed plot (0.16–0.30 m2) per subplot. For each harvested plot, the total number of plants, flowers (R1), and pods (R5) were recorded. Leaf area (LA) per plant at R1 and R5 was measured directly using a leaf area meter (LI-3100; LI-COR). Aboveground biomass (AGB) was oven dried at 68°C for at least 72 hr and weighed to the nearest 0.01 g. At the end of each growing season (Table 1), final harvest occurred in a second undisturbed plot (0.16 m2) per subplot. The total number of plants and pods per plant was recorded. All harvested biomass was oven dried and weighed as described above. Total seed count and mass to determine seasonal yield were completed following oven drying and threshing. For all harvested parameters, the mean per plant (calculated as the mean harvested parameter divided by the actual number of plants harvested) was multiplied by the planting density in a 1 m2 area to standardize values and avoid bias toward any one treatment based on differences in the number of plants harvested per plot.
2.6 Field data analysis
For each measurement variable, data obtained from a single subplot (A, A + H, E, E + H) within a block were averaged such that each subplot represented an experimental unit (n = 4). Data analyses were completed using regression analysis (PROC Reg; SAS v 9.4) and mixed-model ANOVA (PROC Mixed) with [CO2], heat (H), day of year (DOY), and all interactions as fixed effects and block as a random effect. Mixed-model ANOVA analysis was followed by LSMEANS for multiple comparison when the model was significant. Data that did not meet the assumptions of ANOVA were log transformed. Because of the small sample size (n = 4), α was set a priori to <0.10 to avoid potential type II errors.
2.7 Model description and analysis
Simulations were run using the DGVM Agro-IBIS (Agricultural version of the Integrated Biosphere Simulator; Foley et al., 1996; Kucharik, 2003; Kucharik & Brye, 2003; Kucharik et al., 2000; Kucharik & Twine, 2007; Twine & Kucharik, 2008) to determine if changes in measured yield could be attributed solely to changes in carbon accumulation (i.e., AN) during each heat wave or to other processes, such as heat stress on reproductive tissues. Agro-IBIS simulates canopy phenology, soil hydrology, and biogeochemical properties of an ecosystem. In addition, the model simulates leaf-level physiology, including AN and gs, which is scaled to the canopy level. Leaf-level processes are influenced by leaf temperature, moisture stress, nitrogen stress, and [CO2], the last of which was evaluated by Twine et al. (2013) using data from elevated [CO2] field studies.
Agro-IBIS incorporates a semi-mechanistic model to calculate leaf AN (Farquhar et al., 1980), and gs is determined with an empirical model (Ball, Woodrow, & Berry, 1987). As in Twine et al. (2013), the default temperature-dependent functions for Vc,max and electron transport were replaced with equations from Bernacchi et al. (2001, 2003). In addition, the equation for triose phosphate utilization limitation to AN was adjusted to account for temperature dependency (Harley & Sharkey, 1991).
Prior to simulations for 2015 and 2016, a spin-up period was used to generate stable soil conditions at a 0.5° × 0.5° location encompassing the SoyFACE facility using historic climatological datasets (Twine et al., 2013). The model was then run in 2015 and 2016 on hourly time steps using meteorological data from the Surface Radiation Network (SURFRAD) site (Bondville, IL) located <10 km from the field site. Data included temperature, solar radiation, wind speed, and relative humidity. Gaps in the SURFRAD data were either filled using the average of the corresponding conditions for the same hour on the day preceding and following the day containing the gap or using data from the Energy Farm (University of Illinois at Urbana-Champaign). Precipitation data for this study were obtained from Willard Airport located <3 km from the field site. Meteorological conditions used to drive the model were the same for all simulations with the exception of the increase in hourly temperature during the heated treatments (A + H, E + H) according to heating intensity (Table 2).
Crop growth was simulated using actual planting dates (Table 1) and accumulated growing degree-days needed to reach flowering, pod-filling, and harvest. Atmospheric [CO2] was set to growth concentration in the respective A and E experimental plots. Field measurements of Vc,max and Jmax obtained from A subplots were used to calculate AN for the entire growing season for both A and A + H subplots excluding heat wave periods where Vc,max and Jmax were directly measured in A + H subplots (see AN–Ci methods). The same was true for E and E + H subplots. Additional non-AN–related parameters within the model were initiated at the same values used in Twine et al. (2013) but allowed to vary to enable simulated biomass measurements (leaf + stem following R1 heating and leaf + stem and pod following R5 measurements) to correspond more closely with measured values. However, these non-AN–related parameters remained the same from A to A + H treatments and E to E + H treatments to ensure changes in simulated yields between non-heated and heated plots were due to changes in AN-related parameters only.
3 RESULTS
3.1 Climate and canopy temperature
Although the total growing season precipitation was 8% higher in 2015 (58.4 cm) than in 2016 (53.6 cm), most rainfall occurred during June prior to the R1 and R5 stage heat wave treatments. In 2016, 5% more rain fell in July compared to 2015 (July), which included the R1 heat wave treatment, and 30% more rain fell in August of 2016 compared to 2015 (August), which included the R5 heat wave treatment (Figure S2a,b).
Overall, daily mean maximum air temperature (Tmax) and minimum air temperature (Tmin) were similar between growing seasons (Figure S2a,b). Canopy temperature (Tcan) ranged from ~22 to 25°C across all non-heated plots in both years (Table 2). In 2015, heated Tcan was 9°C higher than non-heated canopies, which corresponded with significantly lower θv (10%–17%) at 20 cm depth for heated plots in R1 and R5 (Table 2) compared to the 2015 unheated plots. Moreover, in both years, mean θv was significantly higher in elevated [CO2] plots irrespective of heating treatment during R1. In 2016, heating increased Tcan by 5°C compared to non-heated canopies (Table 2), which corresponded with significantly lower mean θv (10%–15%) in heated plots during R1; but not when canopies were closed during R5.
3.2 Cellular-level response
Vc,max was not affected by either H or [CO2] treatments during R1 in either year (Figure 1a,e). In contrast, during R1 2015, Jmax significantly decreased by 13% in A + H and by 9% in E + H plots (Figure 1b). Moreover, there was a significant [CO2] × H interaction for Jmax during R1 in 2016, suggesting that elevated [CO2] was mitigating heating effects on Jmax under the lower intensity (+5°C) heat wave (Figure 1f). While there was no treatment effect on Vc,max during R5 2015, Jmax decreased by 25% in response to heat waves under ambient [CO2] and 27% under elevated [CO2] (Figure 1c,d). In 2016, there were significant [CO2] and H main effects on Vc,max during R5 and a significant [CO2] × H interaction effect on Jmax (Figure 1g,h).
Under higher intensity heat waves (+9°C) in 2015, there was a significant H effect on mean TNC during R1 and significant [CO2] and H main effects when the canopies closed during R5 (Table 3). In 2016, there were significant [CO2] and H effects on mean TNC during R1 and a significant [CO2] effect irrespective of heating during R5 (Table 3). Overall, mean TNC was highest in E plots, and heating reduced TNC with greater reductions in elevated [CO2] as compared to ambient [CO2] (Table 3).
| Treatment period | A | A + H | E | E + H | Significant effects (p < .1) |
|---|---|---|---|---|---|
| 2015 R1 | 29.5 ± 2.7 | 15.9 ± 2.2 | 51.2 ± 12.8 | 18.5 ± 3.1 | H |
| 2015 R5 | 45.2 ± 2.8 | 25.1 ± 3.3 | 75.0 ± 16.8 | 32.2 ± 5.4 | [CO2], H |
| 2016 R1 | 24.6 ± 2.9 | 21.4 ± 2.7 | 47.7 ± 5.8 | 27.0 ± 3.1 | [CO2], H |
| 2016 R5 | 28.0 ± 5.6 | 22.6 ± 2.2 | 32.3 ± 3.1 | 30.5 ± 0.7 | [CO2] |
Note
- All data are mean TNC (mmol/m2) ± 1 SE.
- Abbreviations: A, ambient [CO2]; E, elevated [CO2]; H, heat; R1, flowering; R5, pod-filling.
3.3 Leaf-level response
For both years of the study, the “CO2 fertilization effect” was mostly prominent at the leaf level prior to each heat wave. Elevated [CO2] typically increased AN but not during heating when AN was similar in A + H and E + H plots (Figures 2a and 3e) or in A and E + H plots (Figures 2e and 3a). That AN was significantly affected by [CO2] and H was controlled by various factors. For example, there was a significant H × DOY interaction for VPDL across reproductive stages as VPDL consistently increased in response to heating during 2015 and 2016 (Figures 2b,f and 3b,f). This corresponded to a significant H × DOY interaction for gs and Ψmd during R1 and R5 2015, and significant [CO2] and H effects during R1 2016 along with a significant [CO2] × H × DOY interaction during R5 2016 (Figures 2 and 3c,g; Table S1). As a result, gs in soybean consistently decreased in heated plots (Figures 2c,g and 3c,g), likely in response to significantly lower mean θv (Table 2) and changes in Ψmd in heated plots (Table S1). Ci often responded to both H and [CO2] with the highest Ci in E plots and slightly lower Ci in E + H plots (Figures 2d,h and 3h). However, H did not significantly affect Ci in R1 2016 (Figure 3d). Post-heat wave recovery occurred for all reproductive stages in 2015 except for in elevated [CO2] following R1 heating (Figure 2a,e; Table S2). In 2016, recovery only occurred in ambient [CO2] following R5 heating (Figure 3a,e; Table S2). In this study, “recovery” was determined when there was no statistical difference between heated and non-heated means within [CO2] treatments.
3.4 Plant-level response: AGB and yield components
Following R1 of 2015, mid-season AGB was significantly lower in A + H and E + H plots under the high-intensity (+9°C) heat wave (Figure 4d). These results were attributable to significantly lower flower, leaf, and stem biomass when heated plots were compared to their non-heated counterparts (Figure 4a–c). Following R5 of 2015, there was a significant [CO2] × H interaction on leaf, stem, and AGB with heating lowering biomass in ambient conditions but not in elevated [CO2] (Figure 4f–h). Pod biomass remained significantly lower in heated plots (Figure 4e) irrespective of [CO2]. These results corresponded with significantly lower number of pods harvested from soybean grown in heated plots compared to non-heated counterparts (Table S3).
In contrast to 2015, [CO2] effects were most prominent in 2016 under the low- (+5°C) intensity heat wave. For both reproductive stages, mid-season AGB was significantly higher in E and E + H plots due to increased stem and leaf mass (Figure 5b–d,f–h). Interestingly, while there was no statistical treatment effect on pod mass (R5), the trend toward greatest pod mass occurred in A + H and E + H plots (Figure 5e), which also corresponded with the highest pod count (Table S3).
At final harvest, soybean exposed to the +9°C heat wave in 2015 had significantly lower seed yield (SY) compared to the non-heated plants within a [CO2] treatment irrespective of heat wave treatment period (Table 4). R1 (2015) final yield was 15% and 17% lower in A + H and E + H plots compared to the respective non-heated plots within [CO2] treatment and was likely due to lower seed count in heated plots (Table S4), while R5 (2015) final yield was 26% and 17% lower in A + H and E + H plots, which was attributable to significantly lower seed and pod count in heated plots (Table S4). In the 2016 field season, soybean grown under +5°C heat waves maintained significantly higher yield in elevated [CO2] plots for both reproductive stages (Table 4). During R5 (2016), final yields were also 9% and 20% higher in A + H and E + H plots (Table 4), which corresponded with slightly higher pod and seed count (A + H) and the greatest 100-seed weight (E + H; Table S4). Moreover, R5 (2016) soybean in heated plots had significantly higher AGB (stem, pod wall, seed mass) compared to A and E plots (Table S5).
| Treatment period | A | A + H | E | E + H | Significant effects (p < .1) |
|---|---|---|---|---|---|
| 2015 R1 | 511.1 ± 17.2 | 431.6 ± 29.8 | 548.8 ± 29.7 | 456.0 ± 29.1 | H |
| 2015 R5 | 511.1 ± 17.2 | 379.7 ± 53.0 | 548.8 ± 29.7 | 452.8 ± 26.0 | H |
| 2016 R1 | 282.0 ± 40.5 | 283.5 ± 32.0 | 338.6 ± 63.4 | 361.0 ± 21.8 | [CO2] |
| 2016 R5 | 282.0 ± 40.5 | 307.4 ± 34.0 | 338.6 ± 63.4 | 408.2 ± 65.8 | [CO2], H |
Note
- All data are means ± 1 SE (g/m2).
- Abbreviations: A, ambient [CO2]; E, elevated [CO2]; H, heat; R1, flowering; R5, pod-filling.
3.5 Modeled response
Using measured Vc,max and Jmax values and meteorological data to drive the model, Agro-IBIS accurately simulated AGB (i.e., leaf + stem) within 2% of measured yields for both reproductive stages and years (Table 5). In 2015, simulated heat waves reduced end-of-season yields to a lesser extent than the actual heat waves, suggesting that heat treatments in the field likely affected reproductive physiology (not included in the model) in addition to carbon assimilation (Table 5). While measured yields were similar or increased with the lower intensity heat waves of 2016, the model still predicted yield losses (Table 5). In 2016, R1 measured final yields increased in response to heating in both ambient (+0.35%) and elevated (+6.49%) [CO2], whereas simulated yields declined with heating in both ambient (−5.7%) and elevated (−4.4%) [CO2] (Table 5). Heating during the R5 period of 2016 also increased measured yields in both ambient (+8.9%) and elevated (+20%) [CO2] conditions, but simulated R5 yields again declined with heating in both the ambient (−15%) and elevated (−8.1%) [CO2] conditions, albeit to a lesser extent than when compared to R5 of 2015 (Table 5).
| Treatment | R1 | R5 | ||||||
|---|---|---|---|---|---|---|---|---|
| Leaf + stem (g/m2) | Yield (g/m2) | Leaf + stem +pod (g/m2) | Yield (g/m2) | |||||
| M | S | M | S | M | S | M | S | |
| 2015 | ||||||||
| A | 294 | 295 | 511 | 507 | 741 | 745 | 511 | 508 |
| A + H | 187 | 189 | 432 | 458 | 549 | 544 | 380 | 386 |
| E | 305 | 300 | 549 | 550 | 864 | 870 | 549 | 555 |
| E + H | 250 | 255 | 456 | 524 | 868 | 861 | 453 | 462 |
| Percent change | ||||||||
| A versus A + H | — | — | −15.5 | −9.66 | — | — | −25.6 | −24.0 |
| E versus E + H | — | — | −16.9 | −4.73 | — | — | −17.5 | −16.8 |
| 2016 | ||||||||
| A | 313 | 313 | 282 | 281 | 538 | 525 | 282 | 297 |
| A + H | 289 | 294 | 283 | 265 | 566 | 569 | 307 | 253 |
| E | 399 | 402 | 339 | 340 | 705 | 705 | 339 | 357 |
| E + H | 370 | 371 | 361 | 325 | 744 | 742 | 408 | 328 |
| Percent change | ||||||||
| A versus AH | — | — | 0.35 | −5.69 | — | — | 8.87 | −14.8 |
| E versus EH | — | — | 6.49 | −4.41 | — | — | 20.4 | −8.12 |
- Abbreviations: A, ambient [CO2]; Agro-IBIS, Agricultural version of the Integrated Biosphere Simulator; H, heat; E, elevated [CO2]; R1, flowering; R5, pod-filling.
4 DISCUSSION
Although it is widely known that elevated [CO2] generally enhances and heat stress generally inhibits C3 plant productivity, less is known about the interaction of these two variables, especially during key reproductive phases in temperate crops such as soybean. In our study, soybean yields improved under elevated [CO2], but high intensity (+9°C) heat waves in 2015 resulted in significant yield loss despite elevated [CO2]. Negative effects on yield due to heating during R1 (2015) were associated with significant reductions in carbon assimilation-related processes at the cellular, leaf, and plant level; however, yield loss after R5 (2015) heating was more likely driven by heat stress on reproductive processes (i.e., lower pod biomass following the heating period), especially in E + H plots where leaf and stem biomass were unaffected by heating. Alternatively, low intensity (+5°C) heat waves during R1 and R5 in 2016 maintained or improved yields, and the “CO2 fertilization effect” appeared to increase in these conditions. The lower heat intensity and increased growing season rainfall in 2016 likely uncoupled negative H effects at the cellular and leaf level from the plant level and reproductive processes. Modeling based on carbon assimilation alone underestimated yield loss due to heating in 2015 and overestimated yield reductions due to heating in 2016, indicating that H effects on reproductive processes also contributed to changes in yield.
4.1 Soybean yields increased under elevated [CO2], but high intensity (+9°C) heat waves in 2015 resulted in significant yield loss irrespective of [CO2]
Soybean yields were higher in elevated [CO2] plots irrespective of heat treatment (Table 4). However, mean Tcan during the R1 (2015) heat wave exceeded the critical temperature (Tcrit; 32°C) for soybean, leading to a significant yield loss in A + H compared to A (−15.5%) and E + H compared to E (−16.9%) plots. During R5 heating (2015), mean Tcan surpassed Topt (23–30°C) for soybean but remained slightly below Tcrit, and yields also decreased in A + H (−25.6%) and E + H (−17.5%) plots compared to the non-heated plots within [CO2] treatment. Previous studies with soybean have shown that R3-R5 (i.e., seed development) phases are most susceptible to heat stress (Boote et al., 2005; Egli & Wardlaw, 1980). Indeed, our 2015 results corroborate this with greater yield reductions when heating occurred during R5 compared to R1, albeit elevated [CO2] alleviated some H effects in E + H plots. H effects on yield loss during R5 2015 were also likely exacerbated by lower mean θv that occurred across all treatments (Table 2).
4.2 Reduced carbon assimilation at the cellular, leaf, and plant level drove yield reductions during R1 2015 heating, whereas heat stress on reproductive processes likely drove yield reductions during R5
At the cellular level, it is widely known that Vc,max and Jmax are sensitive to [CO2] and temperature (Bernacchi et al., 2001, 2003; Farquhar et al., 1980), but neither [CO2] nor heating changed Vc,max during R1 or R5 2015 (Figure 1a,c). Earlier field studies that exposed soybean to short-term heat stress (+6°C above ambient) or maintained higher growing season temperature (+4°C above ambient) under elevated [CO2] report higher and lower rates of Vc,max under heating depending on developmental phase or acclimation (i.e., downregulation) of Vc,max under elevated [CO2] irrespective of temperature (Rosenthal et al., 2014; Siebers et al., 2015). Most nodulated soybean with indeterminate growth, such as the one used in this study, are not sink limited or may be intermittently sink limited (Irigoyen et al., 2014; Rogers et al., 2004), which likely explains why Vc,max did not downregulate under elevated [CO2]. This is supported by mean foliar TNC, which did not reach levels previously found to cause downregulation of photosynthesis in soybean at this study site (Rogers et al., 2004), except in E plots during R5 2015 (Table 3). Lack of a significant H effect on Vc,max in 2015 (Figure 1a,c) suggests that Rubisco activity was either lower in leaves sampled at predawn or was not significantly constrained by temperature (Rosenthal et al., 2014; Vu, Allen, Boote, & Bowes, 1997). In contrast to Vc,max, H significantly reduced Jmax irrespective of [CO2] during R1 and R5 of 2015 (Figure 1b,d). Siebers et al. (2015) also reported significantly lower Jmax in soybean when Tcan reached 31.5°C, which corresponds to the temperature range during R1 and R5 in this study (Table 2). The consistent reduction in Jmax in response to heating at both reproductive stages suggests that RuBP regeneration capacity limited AN at the cellular level in heated plots in 2015 (Cen & Sage, 2005).
Heating also negatively affected carbon assimilation at the leaf level. Heat waves during R1 and R5 of 2015 significantly decreased mean θv (Table 2) and increased VPDL compared to non-heated controls (Figure 2). As a result, soybean in heated plots were significantly more water stressed (i.e., more negative Ψmd) during the R1 and R5 heat waves (Table S1). This corresponded with significantly lower gs, Ci, and AN in A + H and E + H plots for both reproductive stages (Figure 2). These results were not unexpected as stomatal limitation of AN under heating and lower rates of gs in response to [CO2], θv, and VPDL have been documented in soybean (Ainsworth & Long, 2005; Leakey, Bernacchi, Ort, & Long, 2006; Medina & Gilbert, 2015; Rosenthal et al., 2014).
While carbon assimilation at the cellular and leaf level was lower in heated plots during both reproductive stages, these effects paralleled changes at the plant level only after R1 heating (Figure 4a–d). Following the 2015 R1 heat wave, mid-season AGB was significantly lower in heated plots due to decreased flower, stem, and leaf mass (Figure 4a–d), the latter of which corresponded with significantly lower LA per plant (Table S3). Heating also decreased the number of flowers, and, although not statistically significant, the total number of seeds (final harvest). Previous studies of soybean and other legumes indicate that short-term and seasonal heat stress at anthesis (i.e., during flowering) can decrease the number, size, and variability of flowers, leading to lower pod set irrespective of [CO2] (Gross & Kigel, 1994; Jumrani et al., 2018; Koti, Reddy, Reddy, Kakani, & Zhao, 2005). Therefore, lower yields following 2015 R1 heating are attributed to a combination of increased water stress, lower carbon assimilation, and heat stress on reproductive processes, which is consistent with other soybean and legume studies (Boote et al., 2005; Ferris, Wheeler, Hadley, & Ellis, 1998; Jumrani et al., 2018). By contrast, mid-season R5 AGB was higher in elevated [CO2] plots irrespective of heating (Figure 4f–h). However, mid-season pod mass and pod number remained lower in A + H and E + H plots as compared to non-heated controls (Figure 4e; Table S3). Therefore, cellular- and leaf-level responses in soybean may be uncoupled from plant-level and reproductive processes under projected climate change, such as the interaction between elevated [CO2] and heat stress (Morrison & Lawlor, 1999; Ziska & Bunce, 1994, 1995). These findings also indicate that enhanced photosynthesis and growth under elevated [CO2] can be counteracted by the negative effects of heat stress on reproductive growth during R5 development in soybean (Baker et al., 1989; Bannayan, Soler, Garcia, Guerra, & Hoogenboom, 2009; Prasad, Boote, Allen, & Thomas, 2002). Thus, yield loss in 2015 was driven by heat stress on carbon assimilation and reproductive processes with the extent of each varying with reproductive stage.
4.3 Low-intensity (+5°C) heat waves of 2016 maintained or enhanced yields while also enhancing the “CO2 fertilization effect”
In 2016, the low-intensity +5°C heat wave treatment during R1 increased Tcan from 24.4 to 30°C, which was within the Topt range (23–30°C) for soybean. Overall, mean θv was lower in H plots, but there were interannual differences in θv. During R1, mean θv increased by 13% and 1% from 2015 to 2016 in A + H and E + H plots, respectively (Table 2). This was attributable to higher rainfall (+8%) during R1 2016 compared to R1 2015. These conditions were associated with no change in R1 2016 yields in A + H plots and significantly higher yield in elevated [CO2] plots irrespective of heating as compared to A plots (Table 4). During R5, heating increased Tcan from 22.5 to 27.5°C, which remained within the Topt range for soybean. Favorable Tcan, along with an increase in precipitation (+30%) during R5 2016 compared to R5 2015, corresponded with no treatment effect on mean θv. Under these environmental conditions, heating enhanced R5 2016 yields in A + H (+9%) plots and in E + H (+20%) plots as compared to non-heated controls. In addition, the “CO2 fertilization effect” on yields increased from 20% in non-heated plots to 33% in heated plots during R5 2016.
4.4 Low-intensity heat waves of 2016 largely uncoupled cellular- and leaf-level processes from plant-level and reproductive processes during R1 and R5
Similar to results from 2015, there was no treatment effect on Vc,max during R1 (Figure 1e). As previously noted, these results are supported by lower mean foliar TNC concentrations than are required to cause downregulation (Rogers et al., 2004) and by mean Tcan remaining below Tcrit for soybean. However, [CO2] and H affected Vc,max during R5 (Figure 1g), which corroborates previous studies showing elevated [CO2] stimulates Vc,max under increasing temperature in C3 crops (Alonso, Pérez, & Martínez-Carrasco, 2009; Stirling, Davey, Williams, & Long, 1997). However, it is unclear why heating effected Vc,max in 2016 but not under high intensity heating in 2015. Under lower intensity heat waves (+5°C), a significant [CO2] × H interaction occurred with Jmax during R1 and R5, with a greater difference between heated and non-heated plots within elevated [CO2] as compared to ambient [CO2] (Figure 1f,h) when Tcan was within optimum for soybean (Alonso et al., 2009).
During R1 2016, leaf-level AN decreased in heated plots compared to non-heated plots in both [CO2] treatments (Figure 3a), even though E + H plots had similar Vcmax, Jmax, and Ci compared to E plots (Figures 1e,f and 3d). In addition, AN was similar between E + H and A plots (Figure 3a) despite higher Ci in E + H plots (Figure 3d) and no difference in cellular-level parameters (Figure 1e,f). These results suggest heat negatively affected AN through processes other than those related to Ci and biochemical limitations. One such limitation could be mesophyll conductance, which has been shown to increasingly limit AN in C3 crops as temperatures increase (Bernacchi, Portis, Nakano, von Caemmerer, & Long, 2002). During R5 2016, most leaf parameters displayed a significant H × DOY interaction, supporting H effects were sustained.
Although significant H effects occurred at the cellular and leaf level during R1 and R5, the plant-level response (i.e., growth and yield) was primarily moderated by [CO2] under +5°C heat waves in 2016. Mid-season AGB for 2016 R1 and R5 increased in elevated [CO2] plots compared to ambient [CO2] plots irrespective of heating (Figure 5d,h). While in contrast to AGB in the 2015 high-intensity (+9°C) heat wave (Figure 4d,h), these data suggest AGB in soybean can increase in response to rising temperature under elevated [CO2], provided that warmer temperatures remain optimal for growth, such as occurred in 2016 (Baker et al., 1989). Similar results occurred in a long-term heating experiment in field-grown soybean where AGB in elevated [CO2] and elevated temperature (+4°C above ambient) conditions was highest in a relatively cool year but decreased to levels lower than ambient conditions in a hotter year (Ruiz-Vera et al., 2013). Mid-season yield components also responded to treatment differences, although not significantly (Figure 5). Flower mass (R1) declined by 30% in A + H compared to A and increased by 40% in E + H compared to E (Figure 5a), whereas pod mass (R5) increased by 31% and 18% in A + H and E + H plots compared to controls, respectively (Figure 5e). In 2016, end-of-season yields following either R1 or R5 heating were highest in E + H plots (Table 4), which was likely attributable to significantly more pods and seeds produced following R1 heating and heavier seeds (i.e., greater 100-seed weight) at R5 (Tables S3 and S4). That yields increased in response to heating following the R5 (2016) heat wave was surprising, but soybean has been shown to maintain consistent rates of pod set and seed growth from 22 to 30°C (Egli & Wardlaw, 1980), which encompassed the mean Tcan of this study. Regardless of the mechanism, lower heat wave intensity combined with higher θv, elevated [CO2], and no significant negative effects on reproductive processes likely surpassed any negative cellular and leaf level heating effects on 2016 yield (Siebers et al., 2015).
4.5 Modeling based on carbon assimilation suggests effects of heat on reproductive processes in both years also played a role in determining yields
Coupling in situ measurements with the AgroIBIS model, we investigated whether this predominately carbon assimilation-based model could simulate yield when soybean, grown under ambient and elevated [CO2], was exposed to short-term heat waves of varied intensity. Although simulated yields were within 1% of measured yields in A and E plots for both reproductive phases and years, there were substantial deviations between modeled and measured yields in heated plots (Table 5).
As previously noted, our Tcan in 2015 exceeded 30°C, and negative effects on reproductive physiology in warm and cool season species can occur between 30 and 40°C (Sage et al., 2015). Thus, the underestimated yield in 2015 heated plots suggests that heat stress likely had direct effects on reproductive physiology that were not captured by the model. This is supported by significantly lower midseason flower and pod numbers produced in A + H and E + H plots following R1 and R5 heat waves in 2015.
Modeled results from 2016 projected yield loss in heated plots compared to non-heated plots, albeit to a lesser extent than in 2015. Yet, the R1 heat wave did not offset yields, and the R5 heat wave increased yields in heated plots (Table 5). For many crop models, the uncertainty around yield loss in response to temperature change is high, and for soybean, it has been especially large on a global scale (Deryng et al., 2014; Siebert & Ewert, 2014). In this study, there are several factors that likely contributed to the overestimation of modeled yield loss. For both reproductive phases in 2016, negative cellular- and leaf-level heating effects were uncoupled from the plant level. Following each heat wave, plant biomass was significantly higher in E and E + H plots, whereas the number of flowers/m2 was highest in E + H plots and the number of pods/plant was higher in A + H and E + H plots (Table S3). Furthermore, a significant negative relationship occurred during R5 between pods/plant and soil moisture (Table S6), the latter of which was associated with θv near field capacity (ca. 41% 0–20 cm depth; Rosenthal et al., 2014), and observable lodging, which is not accounted for in the model. This may have also caused some of the difference in yield between 2015 and 2016. Although air temperature and precipitation were similar between years, the timing and 5% and 30% increase in rainfall during R1 (July) and R5 (August) in 2016 likely contributed to lower yield compared to 2015 when most rainfall occurred in June prior to R1 and R5 stages (Figure S2a,b). Together, these results suggest that carbon-based crop models will need to integrate stress responses of processes other than carbon assimilation, such as direct heat stress on reproductive processes, to accurately simulate yield in future climate conditions.
4.6 Overall, the benefits of [CO2] fertilization decreased when Topt was exceeded, and the inability of the crop model to account for heat on reproductive processes led to differences between measured and modeled yields
Agricultural regions cover ~40% of the terrestrial biosphere and support an increasing global population through agriculture, livestock, and biofuel production (Foley et al., 2005; Godfray et al., 2010). Developing methods for crop adaptation to climate variability and model parameterization is challenging, as crop response to extreme climate events is species specific and often contingent on the reproductive stage (Reichstein et al., 2013). Most yield projections for C3 crops depend on [CO2] fertilization, but it is uncertain whether this effect is maintained in crops exposed to heat waves (Wing et al., 2015). Our results show that [CO2] fertilization was sustained under high intensity (+9°C) and low intensity (+5°C) heat waves for both reproductive stages; however, the benefits decreased under high intensity heat waves leading to significant yield loss. Irrespective of heat intensity, heating effects at the cellular and leaf level were often uncoupled from the plant level for both reproductive stages, indicating that reproductive processes in soybean were also sensitive to heat stress and should be included in future crop adaptation and modeling analyses (Sage et al., 2015). We also found a trade-off between heat stress and [CO2] fertilization. Under high intensity (+9°C) heat waves (2015), heat stress was the prominent driver of yield loss as Topt was exceeded, whereas, in agreement with Long (1991), soybean yields peaked in heated plots under elevated [CO2] provided Tcan did not exceed optimum under low intensity (+5°C) heat waves (2016). These results have implications for rain-fed cropping systems that are vulnerable to environmental change and point toward a climatic tipping point in soybean yield when future heat waves exceed the temperature optimum.
ACKNOWLEDGEMENTS
This research was supported by the United States Department of Agriculture—Agricultural Research Service. We thank Bradley Dalsing (SoyFACE), Chris Montes, Kristen Bishop, Craig Yendrek, Pauline Lemonnier, Andy VanLoocke, Kelsie Ferin, Patrick Edmonds, Matt Siebers, Phil Lopez, and countless undergraduate students. MLT dedicates this paper in memory of Michael T. Friggens—friend, colleague, and brilliant engineer of field science.
CONFLICT OF INTEREST
The authors have no conflict of interest to declare.
