Human management and hybridization shape treegourd fruits in the Brazilian Amazon Basin

Abstract Local people's perceptions of cultivated and wild agrobiodiversity, as well as their management of hybridization are still understudied in Amazonia. Here we analyze domesticated treegourd (Crescentia cujete), whose versatile fruits have technological, symbolic, and medicinal uses. A wild relative (C. amazonica) of the cultivated species grows spontaneously in Amazonian flooded forests. We demonstrated, using whole chloroplast sequences and nuclear microsatellites, that the two species are strongly differentiated. Nonetheless, they hybridize readily throughout Amazonia and the proportions of admixture correlate with fruit size variation of cultivated trees. New morphotypes arise from hybridization, which are recognized by people and named as local varieties. Small hybrid fruits are used to make the important symbolic rattle (maracá), suggesting that management of hybrid trees is an ancient human practice in Amazonia. Effective conservation of Amazonian agrobiodiversity needs to incorporate this interaction between wild and cultivated populations that is managed by smallholder families. Beyond treegourd, our study clearly shows that hybridization plays an important role in tree crop phenotypic diversification and that the integration of molecular analyses and farmers’ perceptions of diversity help disentangle crop domestication history.


| INTRODUCTION
Amazonia is an important center of plant domestication (Clement, 1999;Meyer, Duval, & Jensen, 2012). Its great biological and cultural diversity (Balée, 2013) make it an especially interesting area to study the role of human societies in plant domestication and diversification (Balée, 2013;Clement, de Cristo-Araújo, D'Eeckenbrugge, Pereira, & Rodrigues, 2010). The distinction between wild and cultivated is one of the basic questions of plant domestication (Lévi-Strauss, 1950;Pickersgill, 2013;Terrell et al., 2003). The distinction, however, often goes unnoticed, given the lack of understanding of how local people perceive biological diversity in traditional societies (Caillon & Degeorges, 2007). People's perceptions of cultivated and wild diversity, as well as their management practices that deal with plant hybridization, are still understudied in Amazonia, especially for tree species (Moreira, Lins, Dequigiovanni, Veasey, & Clement, 2015;Rollo et al., 2016;Smith & Fausto, 2016). Hybridization between related cultivated and wild plants may be favored or discouraged by local farmers (Jarvis & Hodgkin, 1999). It can promote domestication and diversification (Gompert & Buerkle, 2016;Miller & Gross, 2011), because hybrids often present interesting traits that can be selected and maintained (Goldschmidt, 2013;Miller & Gross, 2011;Zohary & Spiegel-Roy, 1975), but may not be adaptive in natural environments (Ellstrand, 2003). This is especially true of introgressive hybrids, as backcrossing to one parent maintains its useful characteristics with minor influence of the other parent (Ellstrand, 2003;Harrison & Larson, 2014). It follows that hybridization and introgression between cultivated and wild plants, as well as the human practices that maintain diversity, are important for effective agrobiodiversity conservation. The distinction between wild and cultivated and its linkages with natural ecosystems are essential for a broader understanding of agriculture (Aumeeruddy- Thomas, Hmimsa, Ater, & Khadari, 2014). More efforts are necessary for its recognition and implementation by public agricultural and forestry policies (Michon, Nasi, & Balent, 2013).
The relationship between the two species is largely speculative.
Crescentia amazonica was hypothesized to be the wild progenitor from which treegourd was domesticated (Ducke, 1946). Alternatively, treegourd was domesticated in Mesoamerica and later distributed to Amazonia (Gentry, 1980). In this case, it was hypothesized that C. amazonica was feralized C. cujete (Gentry, 1980). Their chloroplast genetic diversity does not support the possibility of C. amazonica being derived from C. cujete (Moreira et al., 2016). Nonetheless, a close relationship between C. cujete and C. amazonica is recognized by local human populations. In Guyana, people "called spirits" when C. amazonica was found, because they recognized it is a sort of "shadow" of the domesticated C. cujete (van Andel, 2000). As the two species co-occur in Amazonia and Crescentia species are hypothesized to be largely interfertile (Gentry, 1980), gene flow might be abundant, but has not yet been shown at the molecular level.
In this study, we asked whether (i) hybridization plays a significant role in shaping genetic and morphologic diversity in Crescentia species, and whether (ii) hybrid and introgressed individuals are managed by Amazonian smallholders. To address these questions, we combined a genetic study based on chloroplast (single nucleotide polymorphisms-SNP) and nuclear (simple sequence repeats-SSR) markers with local farmer interviews. Using these datasets, we analyzed (i) the genetic differences between C. amazonica and C. cujete in the Amazon Basin, and gene flow between them; and (ii) the relationship between genetic and morphological diversity, and how people use and perceive this diversity.

| Field sampling and interviews
We visited rural and peri-urban villages in 36 municipalities distributed along the major rivers of the Brazilian Amazon Basin (Figure 1). The broad geographical sampling followed two criteria for village selection: dependence on river resources and treegourd use in daily life.
Data were collected after an informed consent invitation that was read collectively in each village and signed by a local representative.  proc. no. 408.611, 2013). We collected leaves for genetic analyses of each treegourd (N = 469) found in domestic areas in the villages (Table S1). We considered domestic areas to include homegardens, swiddens, ports near the river, football fields, trails, and old homestead sites. Clones propagated from the same individual were avoided in order to better assess available diversity. Fruits were measured and photographed and their shape was categorized according to Arango-Ulloa et al. (2009). We performed semi-structured interviews about use, management, and history of most trees collected, and in nine municipalities we practiced participant observation of daily activities and urban farmers' markets. To map wild treegourd distribution, we surveyed herbarium records of C. amazonica, and photographs of its fruit were presented to farmers in all villages visited to stimulate their memory of its presence in local flooded forests. We collected

| Genetic analysis
DNA was extracted from dried leaves (N = 234) using the CTAB 5% protocol (Doyle & Doyle, 1990) with minor modifications. For nuclear SSR analysis, we genotyped all 234 samples, among which 184 were also analyzed for chloroplast SNPs (Table S1). Collected leaves without enough DNA or with low quality were excluded. To develop the nuclear SSRs, a barcoded library of C. cujete genomic DNA was sequenced (Moreira et al., 2016) using an Illumina MiSeqv.3 (San Diego, California, USA). QDD software 3.1.2 (Meglécz et al., 2009) was used to identify nuclear SSR motifs and design primers from 113,865 merged reads. The parameters used to select the nSSR primers were as follows: one primer pair for each read to avoid repeated regions of the genome; avoidance of mononucleotide microsatellites; preference for perfect microsatellites with ≥8 repeats; and avoidance of primers that are very close (≤20 bp) to the target SSR. A total of 1,436 SSR were identified, of which 1,068 were perfect SSR motifs with 819 di-, 191 tri-, 47 tetra-, 10 penta-, and 1 hexa-repeat motifs, and 368 compound motifs. The primers were designed for the perfect SSRs (Table S2). We performed preliminary tests of amplification of 15 SSRs using C. cujete samples (N = 3). Five of the primer pairs failed to amplify, even using different temperatures and DNA concentrations, and were discarded.
The remaining 10 SSR primers were labeled with fluorescence (FAM, NED, HEX; Applied Biosystems, Foster City, California, USA) and genotyped in an ABI 3130xL Genetic Analyzer (Applied Biosystems) using GS-500 LIZ as the size standard (Applied Biosystems). Although all 10 SSR were polymorphic, at least for cultivated samples (N = 221), two loci were excluded (SSR2 and SSR9) because they failed to amplify in 60% of the samples. We kept the remaining eight SSRs (Table S3) for hybridization analysis, as they also amplified successfully for C. amazonica. Locus amplifications were made in simplex and multiplexed for fragment analysis, using the PCR kit (Qiagen, n.206143, Hilden, Germany) with the following program: 95°C for 15 min; 38 cycles, each of 94°C for 30 s, T a °C for 1.30 min, and 72°C for 1 min; and a final step of 60°C for 30 min. Fragment size and allele identification were determined using GeneMapper (Applied Biosystems). We also obtained SNPs observed in the whole chloroplast using previously described approaches (Moreira et al., 2016;Scarcelli et al., 2016).

| Diversity analyses
The eight nuclear SSRs were used to genotype 234 treegourds, among which 221 were from domestic areas and 13 from flooded forests (Table 1). Relationships among individuals were assessed with Structure 2.3, using the admixture model (Pritchard, Stephens, & Donnelly, 2000).
We varied the number of genetic clusters (K) from K = 1 to 20, with 100.000 burn-in, 100.000 iterations, and five different runs for each K F I G U R E 1 Geographical distribution of Crescentia cujete and Crescentia amazonica treegourds collected for this study along major rivers of Brazilian Amazonia (N = 234). Proportions of admixture identified by Structure at K = 2 are indicated. Proportions above 0.9 were considered pure and below 0.9 were classified as admixed. Crescentia amazonica records in northern South America are from the Global Biodiversity Information Facility value. The ad hoc ΔK (Evanno, Regnaut, & Goudet, 2005) was used to identify the most likely number of clusters in the matrix. We considered an individual to belong to a given cluster if its proportion of admixture was <0.10, that is, with more than 0.90 of the individual's SSR profile attributable to the given cluster. In this study, we used the term "admixed" for ancestry between 0.90 and 0.60, and the term "hybrid" for ancestry from 0.60 to 0.40 (Table S1). We also calculated a hybridization index using Introgress 1.2.3 (Gompert & Buerkle, 2010) and compared it with the Structure admixture proportions. The whole chloroplast sequences of 174 domestic treegourds and 10 from flooded forests were analyzed (Table S1). We built a haplotype network based on 250 SNPs using the median joining algorithm (Bandelt, Forster, & Röhl, 1999). The network was visualized using the samples with <4% of missing data, according to software requirements in POPART 1.7 (Leigh & Bryant, 2015). The nuclear and chloroplast comparison defined paternal and maternal introgression, respectively, and determined the final botanical identification (170 C. cujete and 14 C. amazonica). To assess the impact of the uneven number of samples of the two species, we performed a complementary Structure analysis using the same sample size for both species (14 C. amazonica, 14 C. cujete; Fig. S1), with the C. cujete samples chosen at random. Finally, we assessed the potential impact of null alleles on admixture inferences (Fig. S2), by coding any missing data as a homozygote recessive allele (Falush, Stephens, & Pritchard, 2007).
Nuclear genetic diversity of C. cujete and C. amazonica was explored with principal components analysis (PCA) executed with stats R package (R Core Team, 2015). Nuclear genetic diversity and species differentiation were estimated using hierfstat (Goudet, 2005).
Pairwise F ST were calculated and statistically assessed using 1,000 bootstraps (Nei, 1987). The significance of F IS was measured as deviation from Hardy-Weinberg equilibrium using pegas R (Paradis, 2010).

| Genetic structure revealed with nuclear SSRs
The Structure analyses identified two clusters as the most likely structure (K = 2, Figure 2a), based on the ad hoc ∆K approach (Evanno et al., 2005; Fig. S1). The two clusters correspond to domesticated C. cujete and wild C. amazonica, the two accepted botanical species already described in the Amazon Basin. We also found a significant amount of admixture between them ( Figure 2a, Table 1), especially along the Solimões-Amazonas River (Figures 1 and 2b).
The frequency of individuals with admixture proportions suggesting hybridization and introgression was similar in both groups. Among the 234 samples, 200 were assigned primarily to C. cujete, of which 24 (12%) were admixed to some degree, and 23 were assigned to C. amazonica, of which five (21%) were admixed (Table 1). Eleven plants presented admixture proportions between 40% and 60% and were classified as hybrids. These proportions were similar when the C. cujete sample size was reduced at random to be equal to the C. amazonica sample size (r 2 = .99, p < 10 −15 , Fig. S1). At the second possible grouping (K = 3, Fig. S3), similar admixture proportions were also found (r 2 = .98, p < 10 −15 ), while the original C. cujete group was subdivided without relation to geography or fruit morphology (Fig. S3).

| Comparison between chloroplast SNPs and nuclear SSR diversity
The chloroplast analyses also clearly identified the two botanical species ( Figure 3), with 63 mutational differences between the C. cujete and C. amazonica chloroplast sequences. In the haplotype network, T A B L E 1 Summary of nuclear (SSR) admixture proportions (N = 234) and chloroplast (SNPs) haplotypes (N = 184) of the Crescentia cujete and Crescentia amazonica collections analyzed in this study and the habitats they were collected in. A. Admixture proportions for C. cujete and C. amazonica in columns and chloroplast haplotypes in lines. Not confirmed means the chloroplast was not analyzed in plants that were genotyped with nSSR. B. Habitats in which pure and admixed treegourds were collected Flooded forest (N = 13) 0 0 0 0 13 b a Crescentia amazonica with "pure amazonica" assignment in Structure at K = 2 is predominant in flooded forests (N = 10), but one was found cultivated (N = 1). Admixed C. amazonica were only found in domestic areas. b All of these are likely to be C. amazonica, but three were not analyzed for their chloroplast haplotypes.
samples with C. cujete nuclear assignments showed exclusively C. cujete chloroplast haplotypes ( Figure 3, Table 1 the Negro, Solimões, and Amazonas rivers are ordered west to east; the Branco River is ordered north to south; the Madeira River is ordered south to north. (c) Samples are ordered by seven fruit shapes (see Figure 4) and fruit size, with size increasing from left to right samples showed mean H s values of 0.58 and 0.33, respectively (Table   S4). C. amazonica showed extremely low diversity, with fixed alleles at six loci. These low diversity values for C. amazonica are probably due to marker development from C. cujete with consequent poor transferal due to the significant divergence between species. Although such bias might lead to imprecise estimation of C. amazonica diversity, it does not have an impact on the identification of hybrids, as admixture is based on allele frequency differences and not diversity per se (Pritchard et al., 2000). haplotypes. Nucleotide diversity (π) was 3 × 10 −3 and 1.2 × 10 −3 , for C. amazonica and C. cujete, respectively. Chloroplast differentiation between species was high (F ST = 0.89, p < .05).
F I G U R E 3 Chloroplast haplotype median joining network of Crescentia cujete and Crescentia amazonica from Brazilian Amazonia based on 250 chloroplast SNPs. Nuclear ancestry of each sample (N = 184) was evaluated using Structure (Table 1). The size of the circle reflects the number of individuals presenting the same haplotype. Numbers of mutations are indicated as hatch marks and numbers between haplotypes. Colors represent percentage of nuclear admixture, where C. amazonica >0.9 (dark blue); admixed C. amazonica <0.9 and >0.6 (light blue); hybrids <0.6 and >0.4 (violet); admixed C. cujete <0.9 and >0.6 (orange); and C. cujete >0.9 (red). A clear chloroplast difference is observed between C. amazonica and C. cujete, although 17% (N = 4) of the C. amazonica ancestry samples (N = 23, Table 1) have C. cujete haplotypes

| Admixture proportions correlated with fruit size
There is ample morphological diversity of treegourd fruits in the Brazilian Amazon Basin (Figure 4). We recorded seven types of C. cujete fruit shapes in domestic areas (N = 167) and one fruit type of C. amazonica in flooded forests (N = 10). In domestic areas, two of them account for 86.4% of the plants: 63% flattened (type 1) and 23.4% oblong (type 2). The other five types were rare: 0.6% was cuneate (type 3), 3.6% elongated (type 4), 1.8% globular (type 5), 4.8% rounded-drop (type 6), and 1.2% oblong-drop (type 7). The two C. amazonica samples found in domestic areas showed type 2 fruits. While the fruits of C. amazonica in the flooded forest have small diameters ranging from 2.5 to 5.4 cm (median: 4.85 cm), in domestic areas diameters were slightly larger, ranging from 4 to 8.4 cm (median: 6.2 cm) (Figure 4). Among C. cujete, there is great size variation, especially within types 1, 2, and 5, with variation from 5.2 to 29 cm. Smaller fruits of C. cujete were correlated with C. amazonica admixture (r 2 = .34, p = 1.26 × 10 −6 , Figure 5), especially the oblong (type 2) fruit (r 2 = .62, p = 4.81 × 10 −6 ). This relation remains even if we exclude the most extreme admixed samples (wild ancestry higher than 0.8, r 2 = .23, p = 1.3 × 10 −4 ). The smaller C. cujete fruits, between 3 and 10 cm in diameter, are those with the highest admixture proportions ( Figure 5). Some C. amazonica ancestry could be observed in all shapes found in domestic areas, except in the rare type 7 (Figure 2c).

| Use, management and perception of fruit diversity
We recorded 11 current domestic uses of C. cujete fruits that are related to fruit morphology and were categorized in levels of specialization. Five uses are related with fruit size, without restriction of shape.
The small gourds that are mainly used for food consumption (xibé, a meal of water and manioc flour), açaí (juice from Euterpe oleracea or E. precatoria), water, or some particular medicine were reported in 31% of the municipalities. In the upper Negro River, these are called cuiupi. Large gourds are preferred for use as baskets to store seeds, seedlings, and cleaning products or to serve food (fish, tapioca), and were reported in 66% of the municipalities. Large gourds are also used as a unit of measurement of manioc during the preparation of flour and were reported in 47% of the municipalities, for the daily bath in the river (39%), to bail water from inside the canoe (39%), and pulp of broken fallen fruits as fodder for livestock (16%). Diameter (cm) C Four uses require greater specialization because they need a combination of size and fruit shape. Larger (20-24 cm) oblong fruits (type 2) are used to make a kind of bag with fiber handles (called coió, comboró, mocó), recorded in 30% of the municipalities, and used to transport things to the swiddens or to fish. Flattened fruits (type 1) with relatively smaller sizes (13-15 cm) are used in different contexts. They are involved in tacacá commerce, a typical Amazonian soup served in treegourd bowls (called paranã), whose manufacture was recorded in 11% of the municipalities. They are also used to make bowls to prepare blessings (14 cm) during religious events, recorded in 8% of the municipalities, and more rarely as parts of clothing sewn for regional festivals (2.8% of the municipalities).
Finally, there is an opportunistic use of fruit diversity, recorded in 16% of the municipalities, which incorporates different fruit shapes and sizes into the repertoire of manufactured objects sold as handicrafts. In this repertoire, even the wild species, C. amazonica, with its small fruits, is included, as recorded in four municipalities along the Solimões-Amazonas River. Its fruits are designed as small cups, painted and sold to cosmetic and food companies as "ecological cups," and are worth U$ 6 per 100 cups for handicrafters (price in 2014).
Other wild fruit type handicraft designs include the reproduction of ancient musical instruments called maracá, and also as toys for children (carrapeta). The use of these artifacts was recorded in domestic contexts in 8% of the municipalities, all along the Solimões-Amazonas River, where admixture is relatively frequent and with higher levels ( Figure 2b).
While large fruits have a general name (cuia and coité) with variations that include shape information (e.g., long gourd), the smaller fruits are distinguished with different local names that are related to flooded environments: cuia do igapó, cuiupi, paranã, and maracá. "Cuia do igapó" is the wild tree that grows spontaneously in the igapó, a local word for a forest environment that is periodically flooded by black or clear water (Irion, Junk, & Mello, 1997), and has smaller size fruit ("it is the same as cuia, but is small, is from the forest, is not planted").
The other smaller gourds mentioned depend on human intervention to survive and are found exclusively in domestic contexts. Cuiupi was cited along the Madeira River as similar to the wild type ("there is cuiupi in the lakes, near the river, but I prefer the big one"), while along the upper Negro River it is the general term for small gourds. Paranã is a local word to describe secondary river channels, often between the edge of the forest and the white-water floodplain system rich in nutrients and sediments (Irion et al., 1997), the habitat of C. amazonica. Maracá develops increased fruit size in domestic contexts ("maracá is the cuia do igapó, but it is bigger") and its fruit production is restricted to flooding season ("the fruit fails, does not give all the time"). Based on genetic analyses and people's designation of varieties, cuia do igapó refers to C. amazonica, while cuiupi and paranã are C. cujete with moderate admixture proportions. Maracá is a special case, as it is applied not only to admixed C. cujete, but also to C. amazonica.
Cultivation by seeds, but also by stems, maintains highly admixed samples (Figure 5b), as many desirable gourds have smaller sizes, which is partly due to admixture effects ( Figure 2c). Most treegourds are propagated by stem cuttings (61% of the sampled trees), and less frequently by seeds (37%). Stem cuttings have a purpose: They ensure a fast way to produce fruits ("by seeds takes more time") and the maintenance of fruit morphology, avoiding those fruits that break easily ("by stems it is better for the cuiupi not to become soft and break"). This is also the preferred method among traders who cultivate large clonal areas in upland environments for tacacá bowls, the flattened smaller gourd (type 1) with moderate levels of admixture called paranã. Seed propagation can be a spontaneous event when flood water brings seeds (28%) or spontaneously propagated through discarded fruit pulp near the house (71%). Seedlings, despite the lack of guarantee in fruit morphology, can survive floods better than cuttings in floodplain landscapes. People stated that: "it is not correct to plant seeds, but stems do not take here," "when it grows by seeds, from the pulp, it grows smaller." In the floodplain, stem cuttings require greater effort ("to make big cuia takes much work"), and success is not guaranteed ("it is not all the stems that work, here I've tried hard"). Thus, seed propagation is a way to deal with the high annual flooding events in the floodplain. Another motivation is that seedlings are more suitable to produce smaller fruit sizes that are also useful, especially for handicraft purposes, as observed in the social movement of craftswomen along the middle Amazonas River.

| Identification of geographically widespread admixture
The tenuous nature of reproductive barriers among Crescentia species (Gentry, 1980)  proportions (Barton & Hewitt, 1985;Gompert & Buerkle, 2016). In this case, one question is whether these admixture proportions reveal hybridization after secondary contact or a long-term divergence.
The high number of substitutions between C. cujete and C. amazonica chloroplast sequences suggests ancient divergence between the two species. As C. cujete is a species with cultivated populations, the origin of this domestication is unlikely to be older than other domestications in the Americas, which started by 11,000 years (Piperno, 2011). The chloroplast sequence divergence that we found suggests that the two species diverged earlier. Therefore, C. cujete was likely introduced by humans into South America (Gentry, 1980), and the admixture observed is secondary contact.
The hybrids are concentrated along the east-west axis of the Amazon Basin, the Solimões and Amazonas rivers ( Figure 1). Although proximity to floodplains is an important parameter for the occurrence of hybrids, admixed individuals are also found beyond C. amazonica's known distribution. One example is the occurrence of admixture along the Negro River, where no C. amazonica has been collected to date (Figures 1 and 2b). This pattern might result from social networks and propagule exchange of admixed plants by humans between rivers, such as between the Orinoco and Negro basins (Hornborg, 2005;Lathrap, 2010).

| Flooded forests are a source for cultivated treegourd phenotypic diversity
Fruit size variation of domesticated C. cujete in Amazonia is partly shaped by admixture between wild and cultivated plants. Note here that we do not have a common garden experiment to evaluate the fruit phenotype. Establishing such a common garden will be difficult because treegourd is of minor economic importance outside local communities, and consequently, there is no conservation or breeding in a research institution to allow working in already available common gardens. However, most of the cultivated plants measured shared a cultivated environment in homegardens. If size was simply associated with variability of the environment, we would not detect a significant association with introgression. The absence of a common environment certainly adds more variability in size, but consequently also renders significant correlations with admixture more difficult to detect.
Our results showed that pollen gene flow occurs in both directions between these Crescentia species. Bat pollination observed in both species (Fleming, Geiselman, & Kress, 2009) certainly favors this pollen flow between villages and flooded forests. However, the hybrid plants were restricted to human-managed areas. Hybrid and introgressed C. cujete/C. amazonica are certainly selected against in the flooded forest and favored in human areas, as adaptation to natural environments is likely to be reduced by hybridization with domesticated populations (Ellstrand, 2003).
Both of these objects made from smaller treegourds are considered by oral histories as the first things to be in the world in different cultures of Mesoamerica, the Antilles, and South America (Heiser, 1993), such as the Taíno from the Dominican Republic (Martin, 1999), the Guarani from southern Brazil (Montardo, 2002), and the Tukano from Amazonian Brazil and Colombia (Hugh-Jones, 2009). These smaller fruits come from admixed individuals (Figures 4 and 5), which are maintained and dispersed by people. In the Brazilian Amazon Basin, they are recognized as local varieties (cuiupi, paranã, maracá, and cuia do igapó), whose names and associated traditional ecological knowledge are explicitly related to C. amazonica habitat, or wild-cultivated hybridization and its morphological consequences. The association with flooded environments is also present in the Tupi origin of the name cuiupi (from kuy' y) that refers to gourds of the water (Ferreira, 2004). The name maracá (mbara'ka) refers to the small rattles played in order to talk to lakes and heal sick people (Andía, 2015). Maracá was also used by Ducke (1946) when he described C. amazonica collected in flooded forests of the Solimões River in 1937. All of this confirms that the small fruited varieties used to manufacture important objects in the Amazon Basin, such as maracás, are the result of human selection of hybrid and admixed trees, and highlight that treegourd diversity is partly dependent on hybridization between homegarden and flooded forest genepools.
Although there is an historical use of these admixed treegourds, there is no evidence of ancient cultivation of C. amazonica. Our documentation of cultivation of C. amazonica along the Solimões-Amazonas River, especially in Santarém handicrafter villages, appears to be a recent practice. Treegourd handicrafts have been famous since before the Colonial period (Medina, 1934;Patiño, 1967;Rodrigues Ferreira, 1933) and gained prominence recently as a Brazilian Cultural Heritage (IPHAN, 2015). Handicrafts are motivated not only by social and economic demands, as highlighted by Santos (1982) and Carvalho (2011), but also ecological pressures, as severe flooding in these areas influenced people to use seedlings as an alternative way to produce treegourds for sale. As a result, we observed a high frequency of admixture in these handicrafter villages located in the middle Amazonas River (Figure 2b). The most common cultivation practice of C. cujete, however, is vegetative propagation, not only in Amazonia, but also in Mexico (Aguirre-Dugua et al., 2012). This is the traditional way to maintain useful fruit phenotypes, a practice that allows management of hybridization also (Miller & Gross, 2011). The admixed treegourds have been dispersed by humans along Amazonian rivers and potentially over larger geographical areas, which might create a complex pattern of geographical admixture, as observed for several other Neotropical fruit trees and their wild populations [Spondias (Miller, 2008); Inga (Dawson et al., 2008); Chrysophyllum (Petersen, Parker, & Potter, 2014)], as well as in the Old World genus Prunus (Delplancke et al., 2012). Therefore, human activity not only maintains, but promotes congener interaction, as expected with other crops (Anderson, 2005;Riesenberg & Wendel, 1993). Hybrids are perceived and propagated in a dynamic way, so that hybridization is managed according to people's needs. This is in agreement with local farmers' practices and experimentation observed worldwide (Brush, 2000;García-Marin et al., 1986;Hughes et al., 2007;Jarvis & Hodgkin, 1999).

| CONCLUSIONS AND PERSPECTIVES
We provided evidence that variation in fruit size of the two Crescentia species found in the Brazilian Amazon Basin is related to their admixture proportions. New morphotypes that arise from hybridization are clearly recognized by people and named as local varieties (maracá, cuiupi, paranã), whose symbolism is emblematic for Amazonian cultures. Beyond treegourd, our study clearly shows that hybridization plays an important role in crop phenotypic diversification. We also showed that the integration of molecular analyses and farmers' perceptions of diversity can help disentangle crop domestication history. The specific traditional uses suggest that admixture management is an ancient human practice, also used in current traditional communities. We found that treegourd phenotype diversity depends partially on gene flow between homegardens and flooded forests. These results highlight the linkages between agriculture and forest ecosystems necessary for effective conservation of Amazonian agrobiodiversity. This is especially important as traditional ecological knowledge and floodplain conservation are neglected by development models for Amazonia (Castello et al., 2013;Posey & Balick, 2006).