Megaherbivores modify forest structure and increase carbon stocks through multiple pathways

We investigated the mechanisms that influence food preferences of forest elephants by combining data from a global database of plant nutritional values (23) with forest elephant feeding data from seven different sites across tropical Africa: West (N = 4), Central (N = 2), and East (N = 1 site in two separate studies) (24). The nutritional data covered 145 plant species and 1,343 measurements of essential biomolecules ( proteins, minerals, fat, and carbohydrates), gross energy, and structural and defensive compounds (fibers, tannins, and phenols), which reduce food palatability and digestibility (% of assimilated food). Results from an ordinal logistic regression suggest that elephants prefer to feed on leaves low in fibers, tannins, and phenols (Fig. 1A). We found moderate to strong evidence that less-preferred species are more likely to contain higher percentages of these defensive compounds compared with highly preferred species (P = 0.004 to 0.063). As the percentage of defensive compounds of leaves increases from 0 to 10% to 30 to 70%, the probability of a plant species being least preferred by elephants increases from ~37% to ~65 to 90%, depending on the defensive trait (Fig. 1A). High protein and mineral content of leaves increases the probability of a plant species being in the high-preference category compared with medium preference (P ≤ 0.001), whereas nonstructural carbohydrates and fat decrease the probability of preference (P ≤ 0.05). We found no consistent pattern across preference groups in the other nutritional properties of leaves (Fig. 1A and SI Appendix, Fig. S1). It appears that the main determinants of elephant preferences are defensive structural properties rather than essential biomolecules. These choices are not correlated with the relative abundance of plant species. Both at Ndoki and Kibale National Park (Uganda), there is very strong evidence that preference is inversely correlated to availability, with very similar slopes and fit in the linear model of both sites (SI Appendix, Fig. S2, R² = 0.51 to 0.62, P < 0.001). The analysis of fruit properties consumed and non-consumed by elephants also provides very strong evidence that elephant-consumed fruits are larger (P < 0.0001) and contain more fiber and minerals (P < 0.01) compared with non-consumed fruits (Fig. 1B). We found moderate evidence for differences between the two groups in total tannins, sugars, and fat (Fig. 1B, P < 0.05), the latter being the only property that was higher in non-consumed fruit. No statistical differences were found in crude tannins and proteins across the two groups. At Ndoki, the analysis of the relationship between total diameter at breast height (DBH), a possible proxy for total fruit production, and percentage of detection in dung did not reveal any significant relationship (SI Appendix, Fig. S3A). However, the correlation between presence of seeds in dung and fruit volume revealed a positive correlation (SI Appendix, Fig. S3B, R² = 0.16, P = 0.003). This may suggest that larger fruits are selected for their size and not for their abundance. However, detectability of seeds during dung sampling might be biased by other factors (see Discussion). Overall, the analysis of nutritive properties suggests that elephants seek more palatable and less fibrous leaves and large fruits, which are also high in sugars and minerals but have the drawback of being more fibrous (i.e., less digestible) and less fatty.

The comparison between nutritional properties of fruits and leaves consumed by elephants revealed very strong evidence that, compared with fruits, leaves contain roughly twice as much protein (P < 0.0001) and minerals (P < 0.0001), but six times less sugars and nonstructural carbohydrates (P < 0.0001) (Fig. 1C and SI Appendix, Fig. S4). Moderate evidence suggests that fruits contain less phenols and more fat compared with leaves. There were no significant differences in gross energy content between the two organs (Fig. 1C). Thus, fruits provide short-term usable energy, although gross energy content is similar to leaves, which instead contain biomolecules useful for longer term physiological processes. Forest elephants are making deliberate choices that are mostly independent of leaf availability or fruit production and more dependent on nutritional or morphological (i.e., fruit) properties.

We then investigated whether WD is a good predictor of leaf and fruit nutritional properties. Results from linear regression models revealed strong to very strong evidence that in leaves, WD is positively correlated with fibers (R² = 0.11, P < 0.001) and phenolics (R² = 0.12, P = 0.007) (Fig. 2A). There was weak-to-moderate evidence of a negative correlation between WD and fat (R² = 0.14, P = 0.025) and a positive correlation with gross energy (R² = 0.07, P = 0.057) and tannins (R² = 0.04, P = 0.084). We found no evidence of a correlation between WD and protein or minerals, and little-to-weak evidence of a negative correlation with total nonstructural carbohydrates (Fig. 2A and SI Appendix, Fig. S5). Whereas leaves from higher WD plants had higher content of structural and defensive properties, we found strong evidence that fruits from higher WD plants were lower in essential biomolecules compared with fruits produced by lower WD species (Fig. 2). In particular, we observed a negative correlation between WD and minerals, proteins, and fat (R² = 0.07 to 0.15, P < 0.001) and a positive correlation with sugars (R² = 0.16, P = 0.014) (Fig. 2B). Note however that total nonstructural carbohydrates (sugars plus starches) did not show any significant correlation with WD (SI Appendix, Fig. S5B). The data revealed weak-to-moderate evidence of WD being positively correlated with fruit phenols and fiber content (R² = 0.03 to 0.10, P = 0.043 to 0.085) (Fig. 2B). No other statistically significant relationships were found for the other nutritional properties (SI Appendix, Fig. S5B).

The elephant feeding data and browsing preferences included 197,557 feeding records from 730 plant species for which WD could be determined (eight total studies, SI Appendix, Table S1). The actual number of feeding records is higher because three studies did not report their total sample size (SI Appendix, Table S1). At all sites, except Bia National Park (Ghana) and Santchou Wildlife Reserve (Cameroon), feeding preference metrics were reported in terms of relative preference for single species in relation to all consumed plants. These metrics could be assimilated into three groups indicating high, intermediate, and low preference (Methods). Data from Ndoki included number of feeding events and relative quantity consumed, and, along the two sites mentioned above, was not included in the global analysis. At Ndoki 74% of feeding events involved leaves (or leaves and another plant part) and 21% bark. At the other sites, leaf feeding represented 88% of all events followed by bark at 8%. The remaining feeding signs recorded included roots, piths, branches, and stems. Results from an ordinal logistic regression provided strong evidence that globally the probability of a species of being highly preferred increases as WD decreases across all groups and between low and medium groups (Fig. 3, P = 0.003 to 0.027). In the model with aggregated data from five studies, the average WD was systematically lower in higher preference groups but statistically different only between the low and high groups (Fig. 3, t test – P < 0.01). At Ndoki, the only site where both feeding frequency and quantity consumed were available, very strong evidence suggested that higher probability of preference was associated with decreasing WD (Fig. 3, P < 0.0001). At Ndoki, we also found very significant differences between medium and high (ordinal regression, P < 0.0001) and low and high (t test, P < 0.0001) preference groups. At Tai (Ivory Coast) and Kibale (Uganda, 1996 study), similar patterns of higher WD in low preference groups were found with strong to weak evidence, respectively (SI Appendix, Fig. S6). Data from Bia provided moderate evidence that the probability for high preference increases with WD (P = 0.026). However, this study did not report site-relative preferences and many species were only recorded as browsed once, resulting in many species in the low preference group (see Methods). At Santchou, there was no correlation between preference rank and WD based on 16 species (SI Appendix, Fig. S6). In five out of seven studies, there was moderate to very strong evidence that the high-preference group had the lowest WD compared with the intermediate and/or low-preference groups (Fig. 3 and SI Appendix, Fig. S6). Only at Bossematie (Ivory Coast) there was no significant trend. Analyses from the aggregated sites, Ndoki, and Tai, provided moderate to very strong evidence that the WD of trees dispersed by elephants was higher compared with the WD of species browsed by them (Fig. 3). Overall, these results are compatible with the hypothesis that elephants increase forest community WD through dispersal of higher WD species and browsing of lower WD species.

Our data from Ndoki (understory and overstory) and LuiKotale (overstory) showed moderate to strong evidence that high WD species are slightly more abundant than low WD ones with abundance measured as a percentage of total stems or total DBH (R² = 0.05 to 0.06, P = 0.003 to 0.019); however, the linear model explains only a small fraction of the variability (SI Appendix, Fig. S7). As shown previously for nutritive properties of leaves, elephants seem to make specific choices regardless of the abundance of plant species. For example, in Ndoki understory vegetation plots, Rinorea welwitschii and Diospyros bipindensis were recorded 559 and 468 times respectively (from a dataset of 6,548 tree stems from at least 151 species, see Methods). Yet, of 5,458 feeding events, only two involved D. bipindensis and R. welwitschii was never browsed.

Following previous classifications (6, 25), we identified five dispersal modes in Ndoki and LuiKotale: gravity/dehiscence (gravity), wind, elephants and other animals (non-obligate), elephants (obligate), and other animals (total of 307 species, complete list in Dataset S1). The analysis of the variation of WD as a function of dispersal mode revealed that obligate species had the highest average WD in both sites, but their average was statistically different only compared with certain groups (Fig. 4A). At LuiKotale, only obligate and gravity had statistically different WD than non-obligate species (P ≤ 0.05). At Ndoki, data showed moderate to very strong evidence that obligate species had statistically higher WD compared with non-obligate, other animals, and wind-dispersed but not gravity species. The total number of species identified at LuiKotale (n = 103) was almost half that of Ndoki (n = 204), we thus urge caution when interpreting the results from LuiKotale.

There were differences in the distribution of stem size classes by dispersal mode at the two sites (Fig. 4B). Obligate and wind-dispersed tree communities were characterized by few smaller trees, a higher number of larger trees, and were overrepresented in the 125 to 250-cm DBH range compared with trees with other dispersal modes (Fig. 4B). In LuiKotale, obligate trees represented the largest proportion of stems with DBH > 150 cm (35 to 75%), whereas in Ndoki, these formed the second (18 to 38%) or third largest group of stems with DBH 125 to 225 cm and DBH > 225 cm, respectively. The proportion of wind-dispersed trees increased with size class at both sites similarly. Non-obligate and other-animals’ trees were most abundant in the lower size classes between 40 and 125 cm, but in Ndoki their abundance increased in trees with DBH > 150 cm (Fig. 4B). Gravity trees were particularly abundant in Ndoki in the intermediate size class because of the presence of many very large stands of monodominant Gilbertiodendron dewevrei.

The distribution of AGC in trees (DBH ≥ 40 cm) grouped by dispersal mode reveals diverse patterns in Ndoki and LuiKotale (Fig. 5). In Ndoki, AGC is more evenly distributed among dispersal modes. Abiotically-dispersed trees account for ~50% of AGC and obligate for ~15%. In LuiKotale, trees dispersed by other animals store 52% of AGC and ~19% is stored in obligate trees (second largest biomass pool). In Ndoki, our sampling of vegetation included the monodominant G. dewevrei forest, which occupies a proportion of Ndoki (19) following watercourses. If G. dewevrei forest was removed from the analysis, and only mixed species terra firma forest considered, the contribution of obligate species would increase by ~2 to 3% in DBH classes >40 cm and >70 cm, respectively. When considering only larger trees (DBH ≥ 70 cm), the percentage of AGC stored increases in obligate (23% LuiKotale and 19% Ndoki) and abiotically-dispersed trees (57% Ndoki) (Fig. 5). Notably, at both sites the few obligate species have the highest relative contribution to AGC, despite their low stem count (i.e., highest AGC to stem ratio represented by bar widths in Fig. 5). This is explained by their high WD, highest AGC per cm of diameter, and high relative abundance in the large size classes (diameter > 125 cm) (Fig. 4B). The loss of forest elephants likely greatly diminishes or prevents the recruitment of obligate trees in addition to negatively affecting non-obligate species (6, 25). We quantified the loss of AGC by simulating a replacement of obligate trees with trees with other dispersal modes proportionally to their relative total DBH (Methods). The loss of AGC was estimated to be 9.2% (SD ± 0.07) at LuiKotale and 5.8% (SD ± 0.02) at Ndoki. Thus, the “other” trees cannot completely compensate for the contribution of obligate trees to AGC (but see Methods and Discussion for limitations of this simulation). The important role of large trees in AGC (26, 27) and the widespread decline of forest elephants make the plight of obligate species critical for the future of AGC in African tropical forests.

Both savanna and forest elephants are the largest megavertebrates in their respective ecosystems, and there are similarities in their ecological roles in the physical and trophic structuring of ecosystems. The effects of savanna elephants on their environment have been heavily studied (1), yet few studies have quantified the impacts of forest elephants. We synthesized the literature and selected studies that provided quantifiable measures of the mechanisms of ecosystem engineering by elephants expressed in terms of rates, equations, or data. Of all the possible ecological processes influenced by elephants (28), only a few have been quantified and most of them only once or twice. Many other studies exist on seed dispersal or browsing preferences, but their consequences on ecosystem properties could not be quantified or generalized with equations. Both savanna and forest elephants topple small trees to access foliage, scar, and debark trunks but the impacts of these behaviors on tree mortality in forests are poorly quantified (Table 1). Data on debarking, scarring, and forest properties (forest openness, stem density, AGC, and WD) come from single studies (Table 1). Only one study quantified forest properties as a function of elephant trails (29) and another reported a 1.4% annual mortality rate of large trees (DBH >10 cm) due to elephants. The latter might be significant when compared with the background mortality of African tropical forests (21), but the single-site results route cannot be generalized. However, three general conclusions can be drawn on other processes based on our synthesis. First, the mortality of seedling and saplings is several times higher than larger trees. Second, distance from trails is a key parameter when assessing the effect of elephants on forest properties. Third, there is a clear relationship between canopy openness, reduced regeneration, and elephant feeding preference; however, this has not been estimated in more quantitative terms such as visitation frequency or biomass consumption. Less robust conclusions can be drawn on forest elephant effects on the density of small trees (elephants likely reduce sapling density) and the mortality rate of large trees due to debarking (trees debarked by elephants likely have higher mortality than trees not debarked by elephants).

The Ndoki Forest (“Ndoki” 1.5 to 3°N, 16 to 17°E) lies in the northern Republic of Congo. The climate is transitional between the Congo-Equatorial and subequatorial zones with a mean annual rainfall of ca. 1,400 mm (Ndoki Forest records) (6, 19). Topography varies from terra firma uplands and flat plateaus to the northwest to the extensive floodplain of the Likouala aux Herbes River to the southeast. Soils are of three main types: arenosols to the north and west, ferrasols to the southeast in the Likouala aux Herbes basin on terra firma, and gleysols in the swamps further southeast. Ndoki is embedded in wet Guineo-Congolian lowland rainforest, transitioning to the north into dry Guineo-Congolian lowland rainforest, and into swamp forests to the south. Terra firma is dominated by Sterculiaceae-Ulmaceae semideciduous forest, with large patches of monodominant G. dewevrei forest along watercourses and upland plateaus, and swamp forests (19). The Ndoki fauna includes several large charismatic species such as forest elephants, western lowland gorillas (Gorilla gorilla), common chimpanzees (Pan troglodytes troglodytes), forest buffalo (Syncerus caffer nanus), bongo (Tragelaphus eurycerus), and leopards (Panthera pardus). The human population density is low (<1 inhabitant/ km²) and the immediate study area contains no permanent human settlement.

The LuiKotale research site is located within the equatorial rainforest (2°470S, 20°210E), at the south-western fringe of Salonga National Park in the Democratic Republic of the Congo (25). The study site covers >60 km² of primary evergreen lowland tropical forest. The climate is equatorial with abundant rainfall (>2,000 mm/y) and two dry seasons, a short one around February and a longer one between May and August. Mean temperature at LuiKotale ranges between 21 °C and 28 °C with a minimum of 17 °C and a maximum of 38 °C (2007 to 2010). Two major habitat types can be distinguished. The dry (terra firma) forest and the wet temporarily and permanently inundated forest. The dry habitat dominates with heterogeneous species composition covering 73% and patches of monodominant Gilbertiodendron spp. covering 6% of the site. The wet habitat consists of heterogeneous forest temporarily (17%) and permanently (4%) inundated (25). The LuiKotale fauna includes several large species such as elephants (almost extinct within the last 30 years), bonobos (Pan paniscus), forest buffalo, bongo (Tragelaphus eurycerus), and leopards (Panthera pardus). Similarly to Ndoki, the human population density is low (<1 inhabitant/ km²), and the immediate study area contains no permanent human settlement.

Fresh elephant trails were followed opportunistically over the course of 2-y in Ndoki across a range of habitat types including permanent swamps, seasonally inundated forests, and terra firma open and closed canopy forest. In the case of woody species, a single-feeding event was defined as all fresh feeding signs on an individual plant, regardless of plant parts consumed, though all parts consumed were also recorded. At each feeding site, data were collected on location (using a handheld GPS) estimated age (fresh [<24 h] or recent [24 to 48 h]), plant species, plant part consumed (leaf, stem, bark, wood, roots, etc.), estimated amount consumed on a 1 to 4 scale (rare, few, moderate, and abundant). Five thousand six hundred and forty-eight feeding events were recorded. Quantifying diet selection based on secondary evidence is open to several sources of bias—for example, one cannot detect a feeding event of a sapling that was completely consumed. In an attempt to reduce and standardize observation bias, we quantified feeding events based on identifiable remains (e.g., a terminal branch stripped of its leaves) in close association with fresh elephant prints, and feel confident we captured gross trends on diet selection.

Over a 3-y period throughout the Ndoki Forest, the seed content of 855 piles of fresh intact elephant dung was quantified. Dung piles were broken apart with sticks, and fibers were thoroughly teased apart. In each dung pile, all seeds were identified to species, and the percentage of presence of each species was calculated based on all sampled dung piles.

Our data of forest elephant feeding preferences at Ndoki were combined with data from the MegaFeed database, which contains feeding preferences of all elephant species including the forest elephant (24). We only retained data from studies that quantified feeding preferences per plant species through ordinal ranking, count of browsing events, selection ratio, or browsing frequency. We excluded studies providing only a list of consumed species. Crops were removed. We combined Ndoki and MegaFeed data covering a total of eight studies at seven different sites. Five out of eight studies classified feeding preference in three categories: rare, medium, and high. The Ndoki data contained four categories that were recategorized in three by combining the rare and medium categories into “low.” The remaining two studies had different data types compared with the other five. The data from Bia (36) reported the number of browsing events per tree species. We assigned species to three categories (low, medium, and high browsing preferences) based on the frequency distribution of browsing events. Species with less than three browsing events were assigned to the “low” category, species with more than six were assigned to the “high” category, and the species in between to the “medium” category. Feeding preferences at Santchou (37) were reported with an ordinal scale and thus are presented without using categories. The feeding preference index at Kibale also accounted for relative abundance of elephant-preferred species in relation to the availability of all trees. Dispersal mode of trees was determined following (6, 11) and complemented with data collected at LuiKotale (25). The species classified as obligate elephant included: Ochna gilletiana, Omphalocarpum lecomteanum, Omphalocarpum procerum, Autranella congolensis, Balanites wilsoniana, Detarium macrocarpum, Drypetes gossweileri, Irvingia excelsa, Irvingia gabonensis, Irvingia grandifolia, Irvingia robur, Klainedoxa gabonensis, Mammea Africana, Maranthes sp., Omphalocarpum elatum, Panda oleosa, Tridesmostemon omphalocarpoides, and Picralima nitida. The complete species list is provided in Dataset S1. Note that not all obligate elephant species indicated by ref. 25 were classified as such, as we found evidence in literature that some of those species can be dispersed also by other animals.

Tree inventory data were collected in Ndoki (along and perpendicularly from nine large elephant trails, 5,674 trees DBH > 40 cm) and LuiKotale (16 1-ha plots, 6,579 trees DBH >10 cm). In Ndoki, 1,664 understory circular plots were enumerated, in which 6,479 trees and shrubs were measured and identified. Tree species data (browsing preference plus forest inventories) from other sites spanned over several decades, and species names were homogenized and updated following the taxonomy provided by World Flora Online through their associated R package.

We used the R package “BIOMASS” to assign WD to each feeding record starting at the species level, to the genus, and finally to the site family average. If none of these were available, we assigned the plot-average WD for the inventory data. Feeding data records without WD were removed because the plot-average WD was not available. AGC was calculated using the “BIOMASS” package with the following equation:

where E is a measure of environmental stress estimated from site coordinates (47). We simulated the loss of AGC due to the lack of recruitment of elephant-obligate trees by adapting a methodology used to study the consequences of changes in tree species composition on AGC (48). We replaced the DBH of obligate trees with DBH of trees with other dispersal modes proportional to each dispersal mode contribution to total DBH. To do so, we first created a normal distribution of average AGC/DBH, which represents the variability in WD and size among trees in each dispersal mode. We then sampled randomly from the distribution until the new total DBH was reached for every dispersal mode. The total DBH after the replacement process is similar to total DBH before the removal of obligate trees. This process was repeated 10,000 times for each of the two sites and the difference between pre- and postreplacement calculated for each iteration. The mean and standard deviation of the 10,000 iterations were used to estimate the loss of obligate trees on AGC. These simulations do not account for species succession, seed dispersal dynamics, or the time scale at which these processes occur. The absence of elephant trampling and consumption, which could affect AGC and species composition (5), is also not considered. Additionally, some of obligate trees might still be dispersed by other biotic or abiotic means, and adults might live hundreds of years in the absence of logging or extreme mortality events.

We gathered nutritional values of plant species consumed by elephants from PNuts, a global database of plant nutritional properties (49). PNuts contains nutritional values of leaves, bark, roots, fruits, and stems. We retained only data for leaves and fruits because they were the most comprehensive and included several nutritional properties. We selected the nutritional properties for which more data were available, these properties were: crude protein (in the main text and figures referred as “Protein” or CP), acid detergent fiber (“Fibers” or ADF), crude tannins (“Cr. Tannins” or CT), total tannins (“Tot. Tannins” or TT), total phenols (“Tot. Phenols” or TP), ash (“Minerals”), water structural carbohydrates (“Sugars” or WSC), total nonstructural carbohydrates (starch + sugars,"Nonstr. Carb." TNC), and gross energy (GE). The retained data covered 1,343 records (fruits and leaves) and 145 plant species included in the forest elephant diet and 45 species and 346 records of fruit not consumed by elephants. Note that the total species for which we have found nutritional properties are only a subset of the wide range of species consumed by elephants, which in our study was ~800 across all sites. Fruit volume was calculated by multiplying fruit length and width found in the African Plant Database (50). The same database was used to retrieve seed length.

We researched the literature using Google Scholar and Web of Science to find studies investigating the physical effect of forest elephants on the ecosystem. The following keywords were used: “forest elephant,” “L. cyclotis,” “ecosystem engineering,” “ecosystem engineer,” “regeneration,” “mortality,” “tree density,” “stem density,” “debarking,” and “nutrients.” We also examined any relevant publication within the references cited by the articles found during the systematic literature search.

Linear regressions were performed with the R “stats” package function “lm.” For each regression, we checked for normality of the data with Q–Q plots. The homogeneity of variances was checked by examining the graphs of residuals vs. fitted values for each model. In the cases where a trend in the residuals was detected, the data were log-transformed and the homogeneity of the variances reexamined. The log-transformed variables are indicated in their respective figures’ descriptions. We used ordinal logistic regressions to analyze the association between WD or nutritional properties and browsing preferences (51). In our case, the ordinal logistic regression allows to calculate the probability of a species being in a certain feeding group. The results estimate the slope of the regression as well as the odds-ratio of being in successive categories (low, medium, and high preferences). For each estimate, a t-value along with a P-value was estimated. P-values were calculated by comparing t-value against the standard normal distribution. From these models, we calculated the specific probability of an observation being in each level of the ordinal category in our fitted model by simply calculating the difference between the fitted values from each pair of adjacent stratified binomial models (51). The model slope is based on the low preference group as the “focal group” being compared with the other two preference groups. These analyses were performed with functions “polr” and “ggpredict” from R package “MASS” and “ggeffect,” respectively. The aggregated analysis of WD across preference groups on the four sites (five studies) was carried out by including a random effect to account for site effect. For this ordinal logistic regression with random effect, we used the function “clmm” from R package “ordinal.” In this analysis, the Ndoki, Santchou, and Bia sites were removed because the methodology used to quantify elephant food preferences was different from the other sites. The five studies’ aggregate and the Ndoki analysis were presented as the main results. The aggregate result allows to discern a general pattern beyond site-specific trends, and the Ndoki data are ideal because they report not only frequency of feeding but also quantity, which is critical when assessing the potential effects of biomass consumption. The single-site analysis of the relationships between WD and browsing preferences was conducted using an ordinal logistic regression model and are presented in the SI Appendix. Analysis of leaf nutritional properties across preference groups was performed following the same procedure. Given that ordinal logistic regression models do not provide any information on statistical differences between the low and high categories, we compared the means of these two groups using additional t tests. The normality of the distributions was verified, and if this was not verified (only in data from Ndoki) a nonparametric Wilcoxon test was used. The same procedure was followed to compare the means of other categories (fruit eaten or not eaten by elephants, fruit vs. leaf properties). Complete test results are included in the Dataset S1. We report our scientific findings by following recently suggested methodology that avoids interpreting P-values with arbitrary cutoff point but instead uses evidence language associated with ranges of P-values (52). Evidence language includes: very strong, strong, moderate, weak, and little or no evidence according to P-value ranges (52).

F. Berzaghi, F. Bretagnolle, and S.B. designed research; F. Berzaghi performed research; F. Berzaghi, F. Bretagnolle, C.D., and S.B. contributed new reagents/analytic tools; F. Berzaghi analyzed data; F. Berzaghi prepared figures; and F. Berzaghi, F. Bretagnolle, C.D., and S.B. wrote the paper.

The authors declare no competing interest.