This study was conducted in eight Ticuna indigenous communities, along approximately 400 km of the upper Solimões River in Western Brazilian Amazonia (Figure 1 and Table 1). The Ticuna people are the largest indigenous group in Brazil and are distributed in three countries: Brazil, Colombia, and Peru. In Brazil, their communities are located in the state of Amazonas and are distributed along both margins of the Solimões River and its tributaries, where our sampling was performed. The upper Solimões River was chosen as the study area because it has a high concentration of cultivated P. cecropiifolia populations and is considered to be the center of domestication (Clement, 1989; Clement et al., 2010). In this region, cultivated populations occur in terra-firme areas and wild populations occur in adjacent floodplain forests.

Pourouma cecropiifolia is a fruit tree species that occurs in the Amazon rainforest, from Western to Central Amazonia. The species is found in wild conditions in Bolivia, Colombia, Ecuador, Venezuela, Peru, and Brazil. In Brazil, it occurs in the state of Acre and in the state of Amazonas. In areas of primary forest, it occurs mainly in terra-firme forests, but abundance in this phyto-physiognomy varies within the distribution area of the species. In the region of upper Solimões River (near the city of Tabatinga, Amazonas, Brazil), for example, it occurs mainly in floodplains, being scarce in terra-firme forests. These floodplains are relatively high floodplains, which are flooded for a period of 2–3 months, where the maximum level of flooding is around 1.5 m (personal observation in the year 2015). The occurrence of P. cecropiifolia was not recorded in low floodplains or in chavascais (almost permanently flooded areas), where there is a great abundance of pioneers, such as species of the genus Cecropia (also of the family Urticaceae) and grasses.

The fruiting of the species occurs annually between September and December (Lopes et al., 1999). The main pollinator agents of P. cecropiifolia are insects of the family Apidae, Oxytrigona obscura, Trigona dellatarreana, and Trigona sp. (Falcão and Lleras, 1980). The seed dispersing agents are mainly small-sized primates, bats, and humans. It is considered a fruit of easy propagation, fast growth, precocity and good productivity. Villachica (1996) reports that, in plantations, the trees begin to bear fruit at 2 years, reaching an optimum of production between the fifth and sixth year, with subsequent progressive decrease.

For the Ticuna people, P. cecropiifolia is an important traditional fruit and a symbolic component of their culture, and is widely consumed and cultivated in Ticuna fields and agroforests. Moreover, it is reported in Ticuna myths as a plant associated with fauna and mythical entities of the forest.

In each of the eight communities, we sampled 10 adult plants in a terra-firme area under cultivation (domesticated population) and 10 adult plants in a nearby forested area (wild population); 3 km was the average distance between paired populations. The paired sampling model follows Dawson et al. (2008), who studied domestication of Inga edulis in Western Amazonia. For each individual plant we measured the following morphological characteristics: (i) number of fruits per bunch (mean of five bunches), (ii) fruit length (cm), (iii) fruit diameter (cm), (iv) fruit mass (g), (v) seed mass (g), (vi) peel mass (g), (vii) pulp mass (g) – by difference (iv – v – vi), (viii) pulp:fruit mass ratio – ratio vii:iv, (ix) seed:fruit mass ratio – ratio v:iv, (x) diameter at breast height (DBH) (cm), (xi) plant height (m), to determinate the (xii) plant height:DBH ratio (m/cm) and (xiii) branch wood density (g/cm³) (correlation among characteristics are showed in Supplementary Figure S2). For fruit and seed metrics, we used two fruits from each of five bunches. DBH was measured at 1.30 m above ground level. Plant height was estimated using a hypsometer (Vertex Laser VL400 Ultrasonic-Laser Hypsometer III, Haglöf of Sweden). Wood density was determined by the ratio between the dry weight and wet volume of a lateral terminal branch section, with approximately two centimeters in diameter. The wood samples had the wet volume measured by water displacement and were dried for 72 h at 105°C.

We collected 300-g soil samples in the 0–30 cm layer close to each tree sampled. The 10 individual samples from each population were dried, homogenized and mixed in the laboratory to make a composite sample that represented the soil of each population. The composite sample was analyzed to evaluate phosphorus (P), potassium (K), calcium (Ca), and magnesium (Mg) concentrations (EMBRAPA, 1997). The clay, sand and silt content were determined by granulometric analysis to characterize soil texture (EMBRAPA, 1997).

Light availability was estimated for each tree and averaged over the population. We used the Crown Illumination Index, which describes the environment luminosity inside the forest, on a scale ranging from 1, where there is diffuse incident light, up to 4, where there is direct light on the canopy (Keeling and Phillips, 2007).

To evaluate the morphological differences between domesticated and wild individuals, we compared the means and amplitudes of variation of 10 characteristics using Kernel density graphs and performed an ANOVA between the two groups (wild and domesticated) for each characteristic, using R software (R Core Team, 2016). A principal component analysis (PCA) of the 10 morphological characteristics was also used to evaluate the differentiation between wild and domesticated individuals, and to evaluate which characteristics are most strongly correlated with domestication, also using R. To test the multivariate differences between wild and domesticated individuals, we performed an ANOVA on the two principal axis of the PCA. To identify and classify groups of wild and domesticated populations, we performed a cluster analysis based on Normal Mixture Modeling, performed with the mclust package (Fraley and Raftery, 2002; Fraley et al., 2012) in R. We also identified which group has the greatest phenotypic variability by comparing the variances between the wild and domesticated populations for each characteristic.

To evaluate whether domestication altered allometric patterns of fruit components, we used data from the literature of 74 non-domesticated species with drupe fruits like those of P. cecropiifolia (Ibarra-Manriquez and Oyama, 1992; Chen et al., 2010; Shiels and Drake, 2011; Bentos et al., 2012). We adjusted Niklas (1994) potential regression model (SM = a FMˆ^b) of the relationship between fruit mass (FM) and seed mass (SM) for these non-domesticated species (including non-domesticated P. cecropiifolia populations), and used this model for domesticated P. cecropiifolia individuals (n = 80), performed in the qpcR package (Spiess, 2014) in R. We then compared their shape factors (a, exponent of the potential relation between variables) and scaling factors (b, intercept of the potential relation between variables). The differences between the two equations were evaluated by the overlap of the confidence intervals of the shape and scaling values. We performed a covariance analysis to test the differences in the relations between seed mass and fruit mass considering the three groups – wild individuals of P. cecropiifolia, domesticated individuals of P. cecropiifolia and the other species.

A PCA of the environmental conditions [described above, plus the sum of bases (K + Ca + Mg), which represents a fertility index] was used to evaluate the differences between the environmental conditions of the forest and cultivated sites. To compare the multivariate differences between forest and cultivated sites, we performed an ANOVA on the first two principal axis of the PCA. We evaluate the effect of environmental conditions on the 10 morphological characteristics through simple linear regressions using all populations together to encompass all the environmental and morphological variation observed in the study. We used the PCA axis that best represented the environmental conditions to evaluate their relationships with morphological characteristics. To evaluate the individual effect of environmental variables, we performed simple linear regressions between each environmental variable and each morphological characteristic. All analyses were run in R.

Finally, we constructed a conceptual model to present an overview of the combined direct and indirect effects of domestication and environmental conditions on the plant phenotype.

Domestication increased the length, diameter and mass of fruits, seed mass, pulp mass and pulp:fruit mass ratio. In contrast, domestication reduced the number of fruits per bunch, seed:fruit mass ratio (seed:fruit allometry), plant height:DBH ratio and wood density (Figure 2, Table 2, and Supplementary Table S1). Domesticated fruits had 20× greater mass than wild fruits. About 64% of the domesticated fruit is composed of edible pulp, compared to only 34% in wild fruits (Table 2 and Supplementary Table S1). On the other hand, the average values of fruits per bunch, seed:fruit mass ratio, plant height:DBH ratio and wood density were 24.8%, 45.5%, 42.1%, and 21.7% higher in wild populations, respectively (Table 2 and Supplementary Table S1). We found significant differences (p < 0.01) between wild and domesticated populations for all 10 characteristics evaluated (Supplementary Table S2).

The first axis of the PCA with the 10 morphological characteristics explained 73.7% of the data variation highlighting the multivariate differences between wild and domesticated plants (F = 1794, p < 0.001). The second axis explained 7.9% of the data variation and did not differentiate wild from domesticated plants (F = 0.394, p = 0.531) (Figure 3). The characteristics most associated with the domestication syndrome, mass, proportion and size of fruits and their components (seed and pulp), were highly and positively correlated (±90%) with axis 1. Hence, PC1 is the axis that best reflects the domestication syndrome.

Reinforcing the pattern found in the PCA, the clustering and classification analysis (Normal Mixture Modeling) distinguished among groups of domesticated and wild populations for seven morphological characteristics (Figure 4). Fruit length, fruit diameter, fruit mass, seed mass, pulp:fruit mass ratio and seed:fruit mass ratio discriminated two groups, the domesticated populations and the wild populations. Pulp mass, however, allowed discrimination of three groups, dividing the domesticated populations into two groups, including four populations in and close to Tabatinga, with higher values of pulp mass than the other four populations further east in the study area. Using the clustering and classification analyses, the number of fruits per bunch, plant height:DBH ratio and wood density did not differentiate wild populations from domesticated populations.

Among the fruit characteristics, fruit length, fruit diameter, fruit mass, seed mass, pulp mass, and pulp:fruit mass ratio presented higher variances in domesticated populations, indicating greater phenotypic variability in these characteristics in plants under human selection (Table 2). Fruit mass and pulp mass presented much greater variances within the domesticated group. The greater amplitude of variation in these characteristics is also apparent in the density curves (Figure 2). Number of fruits per bunch, seed:fruit mass ratio and plant height:DBH ratio presented higher variances in wild populations, while wood density presented similar variances in wild and domesticated populations.

The general allometric model SM = 0.63 FM^0.89 (where SM is seed mass and FM is fruit mass) based on 74 species that have fleshy fruits with only one seed (including P. cecropiifolia wild populations) presented a higher value of the shape factor than the model adjusted to the characteristics of domesticated individuals (SM = 0.44 FM^0.60). There was no difference in the confidence intervals of the scaling factor between the equations. In the equation for domesticated individuals the confidence interval of the scaling factor ranged from 0.35 to 0.55 and in the general equation for 74 wild species it ranged from 0.52 to 0.74. The confidence intervals of the shape factor values of the equations did not overlap. In the equation of domesticated individuals, the confidence interval of the shape factor ranged from 0.50 to 0.69 and in the general equation for 74 wild species it ranged from 0.83 to 0.96. This shows that the two equations are different and that the observed values of seed mass in the P. cecropiifolia domesticated plants are lower than the values predicted by the general allometric equation. The seed:fruit mass ratio changed from approximately 0.9:1 in wild plants to 0.6:1 in domesticated plants (Figure 5A). In comparison, the correlation between the observed seed mass and the predicted seed mass by the model is higher in wild P. cecropiifolia plants (r² = 0.87) than in domesticated plants (r² = 0.66, Figure 5B). In the ANCOVA, we found significant differences in the intercept and in the slopes between the groups of other species, the wild individuals of P. cecropiifolia and the domesticated individuals of P. cecropiifolia [F(2,215) = 267.14, p < 0.001]. The interaction between fruit mass and groups was also significant [F(2,215) = 228.33, p < 0.001], showing that the allometric relation between seed mass and fruit mass change as a function of the groups.

The environmental conditions where the wild and the domesticated groups occur in the study area were also very different from each other (Supplementary Figure S1 and Supplementary Table S3). The first principle component explained 91.9% of the data variation, and differentiated floodplain forests from cultivated sites [F(1,14) = 18.77, p < 0.001]. The second axis explained 7.3% and did not differentiate floodplain forests from cultivated sites (F = 3.128, p = 0.098). Domestication is strongly correlated with variations in light availability (CII), sum of bases, and calcium, magnesium, phosphorus, silt and sand (Supplementary Figure S3). Cultivated sites (terra-firme) have 28% higher light availability, poorer soils (16x lower sum of bases), and 63% sandier soils than floodplain forests (right side of Supplementary Figure S1); the floodplain forest sites have lower light availability and more fertile silty soils (Supplementary Table S3). Only the clay and potassium contents were slightly altered in cultivated areas and are less correlated with domestication (Supplementary Figure S3).

The environmental conditions had significant effects on all the morphological characteristics (Table 3). The mass and dimensions of fruits (Figures 6B–D), seeds (Figure 6E), pulp (Figure 6F), and pulp:fruit mass ratio (Figure 6G) increase in environments with higher available light and poorer sandier soils (Supplementary Figure S4). The number of fruits per bunch (Figure 6A), seed:fruit mass ratio (Figure 6H), plant height:DBH ratio (Figure 6I) and wood density (Figure 6J) increase in environments with less available light and more fertile silty soils (Supplementary Figure S4). Analyzing only those characteristics less associated with domestication, we found a significant increase in wood density due to the increase in potassium content (Supplementary Figure S4).

The increase in dimensions and mass of fruits and seeds are the characteristics most modified during the domestication process of fruit species (Meyer et al., 2012). Domestication in fruit trees selects extreme phenotypes for size and mass of fruits, and eliminates phenotypes that differ from the preferred phenotype, reducing their frequencies in domesticated populations (Zohary, 2004). In P. cecropiifolia, humans selected for fruits with larger sizes and larger masses than those found in wild populations, in which large fruits are not favored by natural selection (McCouch, 2004). Dispersers of wild P. cecropiifolia in the forest are usually small-sized primates, such as Saguinus mystax, S. fuscicollis (Knogge and Heymann, 2003), Callimico goeldii, Saguinus labiatus (Porter, 2001) and Cebus apella (Gómez-Posada, 2012). Smaller fruits may provide an advantage over large fruits for dispersal by small-sized primates, because they can be more easily removed, transported and dispersed (Tanksley, 2004). On the other hand, in domesticated plants, the larger fruit size does not have an adverse effect, because humans guarantee the seeds’ dispersal and the seedlings’ establishment.

The number of fruits per bunch was smaller in domesticated populations than in wild populations. This negative correlation between number and size is common in fruit trees (Browning, 1985), due to the reallocation of photoassimilates to fruit size, which demands a decrease in the number of fruits per bunch for biomechanical reasons.

Within the domesticated populations we found geographic variation for fruit size, where a group of populations on the western side of the sample area had larger fruit than a group of populations to the east. This finding supports Clement’s (1989) proposal of a larger-fruited landrace close to the triple frontier (Brazil, Colombia, and Peru). Whether this extends as far as Iquitos, Peru, where Ducke (1946) commented on the abundance and popularity of P. cecropiifolia, remains to be investigated.

The lower mean values found in plant height:DBH ratio and in wood density can be explained by changes in edaphic and light conditions, which will be better detailed later in the specific section where the environmental effects on the morphological characteristics will be discussed. However, it is also possible that domestication generated a reduction in the height, diameter and wood density of the trees, due to the reallocation of photoassimilates to the harvestable product, changing the ratio between the biomass of the harvestable product (the fruits and seeds in the case of P. cecropiifolia) and the total plant biomass (Li et al., 2012). This ratio is called the ‘harvest index,’ and a negative correlation with plant height is common in many annual crops, such as rice (Li et al., 2012) and sorghum (Can and Yoshida, 1999) due to greater translocation of photoassimilates from the vegetative tissues to grains (Zou et al., 2003). For tree crops, this negative correlation is also expected (Cannell, 1985).

In addition to the morphological differences between wild and domesticated populations that characterize the domestication syndrome, phenotypic variability is also expected to be greater in useful parts (Pickersgill, 2007). This expectation was observed in domesticated populations of banana (Li et al., 2013), peach palm (Clement, 1988) and tomato (Tanksley, 2004). Although genetic variability generally decreases in domesticated populations, phenotypic variability of selected parts may increase with domestication (McCouch, 2004; Purugganan and Fuller, 2009) due to dispersal and diversification after initial domestication (Meyer and Purugganan, 2013). During the dispersal process, the genetic material under selection is shared and disseminated among different human groups with cultural peculiarities, which may have, for example, different food preferences. This may also promote diversity in domestication syndromes (Milla et al., 2015). In the case of domesticated P. cecropiifolia populations, the Ticuna report fruits with more fibrous pulp and others whose pulp has higher water content, thus generating large variation in pulp and fruit mass. However, the Ticuna also report that they seldom select for the juicier pulp, as the fruits “explode” when they fall on the ground, a common occurrence during harvesting, and are unfit for transport or sale.

Humans selected P. cecropiifolia plants to have larger fruits. In response to selection to increase fruit mass, a faster increase in pulp mass than in seed mass is expected, resulting in fruit with a higher relative proportion of pulp (Martinez et al., 2007; Chen et al., 2010). The increase in pulp mass in P. cecropiifolia is mainly due to the increase in carbohydrates (fibers, cell walls, and starch) (Lopes et al., 1999). Due to the correlation among fruit components, the increase in fruit mass also leads to an increase in seed mass. However, the seeds of domesticated P. cecropiifolia do not follow the same tendency of increase observed in wild plants (MS ∼ MF^0.89; Niklas, 1994). Seeds contain proteins and oils, which are energetically more “expensive” than carbohydrates (Lopes et al., 1999). Therefore, it is likely that seed mass does not increase in the same proportion as fruit mass in domesticated plants, because the highest selective pressure is on the fruit, which responds with more carbohydrates, and not on the seed, which needs to retain a balance of proteins, carbohydrates, and oils to guarantee good germination.

Contrary to expectations for the variation in dimensions and mass of fruits and their components in response to edaphic conditions (Janick and Paull, 2008), we found fruits with larger masses and dimensions, and larger masses of seed and pulp in the domesticated populations, which occur in soils 16x poorer in nutrients and 63% sandier than floodplain soils, where wild populations occur. Having larger fruits with larger seeds and pulp mass in poorer and sandier soils is an indication that the domesticated phenotype is mainly a result of alterations in the genotype resulting from domestication, considering that for P. cecropiifolia the fruit characteristics are the most important for humans and are under direct human selection. The increase of fruit dimensions may have been due to the preferential selection of trees with large fruits and the elimination of trees with small fruits (Zohary, 2004). Considering that the size of the plant affects biomass allocation (Milla and Matesanz, 2017) and considering that the high light availability of the cultivated environment where the domesticated populations of P. cecropiifolia grow generates a decrease in the total size of the plant, it is possible that there is a trade-off that may have led to a lower allocation to vegetative parts and a higher allocation to reproductive organs, as is typical of changes in harvest index (Li et al., 2012).

The subtle variations in vegetative characteristics suggests that they are not under direct selection. These changes are possibly results of changes in the environmental conditions caused by human management, an indirect effect of domestication (Harlan, 1992; Zohary, 2004) (Figure 7). Due to high light availability in cultivated landscapes, the plants of domesticated populations invested less in growth in height, because the competition with other plants for light is smaller when compared to individuals in forest landscapes that have smaller canopy openings (Cannell, 1985). With the greater light availability in the cultivated environment, domesticated plants possibly grow faster than wild plants, and consequently have lower wood density (Poorter, 1999; Poorter et al., 2008). The plant height:DBH in domesticated plants is also affected by the sandier soils of the cultivated areas. Sandy soils maintain less water and nutrients due to their lower surface areas (Tarboton, 2003), which affects water supply in dry seasons.

Under natural conditions, wild P. cecropiifolia plants produce proportionately larger seeds due to the difficult conditions of establishment in the shaded understory, as observed in other forest species (Salisbury, 1974; Westoby et al., 1996). In cultivated landscapes where the P. cecropiifolia domesticated plants grow, the low fertility of cultivated soils (Westoby et al., 1990; Hammond and Brown, 1995) and the high light availability promote reduction of seed mass (Eriksson et al., 2000; Moles et al., 2005). In contrast, human selection for larger fruits and the positive correlation between fruit mass and seed mass promote an increase in seed mass in the domesticated plants. Therefore, the combined but contrasting environmental effects and human selection lead domesticated populations to have a proportionately smaller increase in seed mass than wild populations.

By evaluating individually the variables not correlated with landscape domestication, it was possible to observe which morphological changes are effectively responding to environmental variations. The positive correlation between wood density and potassium content in the soil, and the existence of one domesticated population with wood density similar to the wood density of the wild populations, in a site with a high potassium content, shows the environmental effect on the phenotypic plasticity of the P. cecropiifolia individuals. This suggests that, although the effect of environmental conditions in cultivated landscapes can be superimposed on the domestication effect, we cannot ignore the plastic capacity of individuals in explaining the morphological variations of plants under human selection.

Due to the difficulty in dissociating the effect of environmental conditions from the effect of human selection, we suggest that reciprocal transplant experiments of domesticated plants to uncultivated landscapes and wild plants to cultivated landscapes will be needed to effectively differentiate domestication effects from environmental effects. In addition, we consider that in future studies it will be necessary to evaluate experimentally the effect of luminosity on fruit mass, seed mass and in seed:fruit allometry considering the high light availability in cultivated landscapes and because it is expected that there is a positive correlation between light intensity and fruit mass (Moles et al., 2005; Janick and Paull, 2008), but a negative correlation with seed mass (Eriksson et al., 2000; Moles et al., 2005).

In this study, the domesticated phenotype is a result of a combination of human selection and environmental conditions in the sites where the plants are cultivated. We observed strong environmental modification created by humans in cultivated landscapes that is exacerbated by the fact that wild populations occur in flooded areas, while domesticated populations occur in upland areas. These changes in environmental conditions between natural and cultivated sites, in addition to genetic selection by humans, promoted the phenotypic changes in domesticated populations. However, the forces of genetic selection through human management and of natural selection through environmental conditions are intrinsically mixed and discriminating the magnitude of each component, and the environment by genotype interaction, requires a well-designed common garden experiment (Falconer and Mackay, 1996).