This study was carried out in the Amazon rainforest in two of the main producing states of E. precatoria. The plant material (leaves) of E. precatoria was collected in 19 municipalities (Figure 1), 16 in the state of Amazonas and three in the state of Rondônia. The municipalities are located on the margins of different hydrographic basins, which, for the most part, serve as the only route of river transport of people, food, and genetic material to other areas of the Amazon, mainly the capitals of these states. These municipalities are all within the geographic distribution of the species (Lorenzi, 2010) and present a substantial record of açaí in the aggregated data of the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatística – IBGE, 2019). Each sampled collection site was called a population. Samples ranging from 17 to 20 plants were collected in each of the 19 populations, totaling 377 samples (Supplementary Table 1). The collection was carried out within the rules of SisGen (National System for the Management of Genetic Heritage and Associated Traditional Knowledge, Decree no 8.772, of May 11, 2016, which regulates Law no 13.123, of May 20, 2015, registration AA9537B).

DNA extraction was performed using the CTAB protocol, as described by Doyle and Doyle (1990), and quantified with GelRed dye. Subsequently, the genomic DNA was standardized at 10 ng/μl for use in amplification. The 377 DNA samples were amplified by polymerase chain reaction (PCR), using 12 microsatellite loci developed for E. precatoria (Epr01, Epr02, Epr05, Epr13, Epr14, Epr15, Epr18, Epr19, Epr21, Epr31, Epr32, and Epr36) (Ramos et al., 2016). The amplification products were quantified with GelRed dye. The amplified PCR products were subjected to an automatic DNA analyzer by ABI 3130XL capillary electrophoresis in the Genetic Analyzer (Applied Biosystems). GeneScan^TM -500 ROX^® standard size (Life Technologies do Brasil Ltda.) was used to determine the size of the alleles. The amplified fragments were observed and analyzed with GeneMapper v4.0 software (Applied Biosystems).

The genetic diversity, compared among the populations collected in the states of Amazonas and Rondônia, was obtained using the following genetic parameters: total number of alleles (A_T), average number of alleles/locus (A), observed heterozygosity (H_O) and expected heterozygosity (H_E), fixation index (f), and Hardy-Weinberg equilibrium (HWE). These parameters were calculated using the divBasic function of the diveRsity package (Keenan et al., 2013) on the R platform (R Core Team, 2019).

The linkage disequilibrium (LD) and the number of private alleles (Ap) were calculated using the softwares GDA (Lewis and Zaykin, 2002), and the tests LD and Nm_private function of the genepop package (Rousset, 2008) on the R platform (R Core Team, 2019). HWE and LD were performed by Fisher’s exact test with 20,000 permutations. The significance level (P ≥ 0.05) of HWE and LD was adjusted with Bonferroni’s correction (Rice, 1989). The number of null alleles was calculated using the null function of the genepop package (Rousset, 2008).

For analysis of the genetic structure, Wright’s F statistics were used (Wright, 1951), including F_IT (total inbreeding levels in individuals of all populations), F_IS (fixation index within subpopulations owing to the reproductive system), and F_ST (genetic differentiation owing to subdivision), and calculations were performed using the algorithms of Weir and Cockerham (1984). An F_ST matrix was estimated from the general fixation index and in pairs of populations (Weir and Cockerham, 1984). The Wright F-statistics (1951) and the FST matrix were calculated and evaluated with significance (P ≤ 0.05) from 20,000 bootstrapping, using the diffCalc function of the diveRsity package (Keenan et al., 2013).

To verify the degree of genetic variation according to hierarchical levels among and within populations and among all individuals sampled, molecular variance analysis (AMOVA) was performed, followed by estimating the extent of genetic differentiation with phi-statistics, both using the poppr.amova function in the poppr package (Kamvar et al., 2014). The significance of the components of variance through the Monte-Carlo test, ΦCP, differentiation among populations, ΦCAP, differentiation among samples within populations, ΦCA, and differentiation among samples were evaluated using a permutation test implemented in the randtest function in the ape4 package (Chessel et al., 2004; Dray and Dufour, 2007; Dray et al., 2007; Bougeard and Dray, 2018) with 9,999 permutations.

In order to determine the correlation coefficient among the F_ST geographic and genetic distance matrices (Weir and Cockerham, 1984), the Mantel test (Mantel, 1967; Smouse et al., 1986) was performed with 999 replicates, using the mantel. rtest function of the ade4 package (Chessel et al., 2004; Dray and Dufour, 2007; Dray et al., 2007; Bougeard and Dray, 2018). The geographic distance matrix was calculated using the DIVA-GIS v7.5 program (DIVA-GIS, 2012).

Within the analysis of genetic structure, two Bayesian analysis techniques were performed to determine the number of clusters within the set of evaluated samples. In the first analysis, Structure software (Pritchard et al., 2000) from the Admixture model was used. This is normally applied to real or natural populations. The number of clusters (K) was defined from 1 to 21, and for each K, 10 iterations were performed with a Burn-in of 5,00,000 followed by 10,00,000 Markov Chain Monte Carlo (MCMC) iterations.

The number of clusters was estimated using the Evanno ΔK method, indicating that the most likely K is where the change is greatest in the second order rate of change in Pr (X| K) among successive K values (Evanno et al., 2005). With the K value selected, a consensus was reached on the iterations carried out in this cluster through CLUMPP v.1.1.2 (Cluster Matching program and Permutation Program) (Jakobsson and Rosenberg, 2007) and with Distruct v. 1.1 (Rosenberg, 2004), a graphical visualization of the population structure was performed. In the second analysis of the population genetic structure, Discriminant Principal Component Analysis (DACP) was used (Jombart et al., 2010), considering the clusters as unknown (Roullier et al., 2013). First, the probability of adherence of each sample to the different groups was determined, using k-means (Jombart et al., 2010). Then, the possible ideal number of clusters was identified, using k-means, comparing and discriminating possible groupings using Bayesian Information Criterion (BIC) (Jombart et al., 2010), using 10,00,000 permutations within the find.structure and dapc functions of the adegenet package 2.1.0 (Jombart and Ahmed, 2011) on the R platform (R Core Team, 2019).

From the 12 loci analyzed, 11 were in Hardy-Weinberg equilibrium (P > 0.00417) (Supplementary Table 2), and the analysis of loci pairs showed 15.7% with linkage disequilibrium. The 12 highly polymorphic loci had a high content of genetic information, with a total of 211 alleles, ranging from 8 (Epr15) to 27 (Epr22) alleles per locus, with an average of 17.6 alleles per locus. The expected heterozygosity (H_E) varied between 0.53 for Epr32 and 0.77 for Epr01 loci. The observed heterozygosities (H_O) showed higher values than H_E, with an exception for Epr02 (0.73) and Epr14 (0.66) loci. H_O ranged between 0.56 for Epr32 locus and 0.86 for Epr13 and Eper05 loci. The average of all loci for H_O was 0.72, higher than the average for H_E (0.66). The fixation index (f) showed values below zero, which shows an excess of heterozygotes, although the Epr02 (f = 0.084) and Epr14 (f = 0.100) loci showed a small degree of inbreeding for the evaluated populations. However, the negative value found for the average (f = –0.108) shows the absence of inbreeding and excess heterozygotes for all populations and loci.

The 19 natural populations of E. precatoria showed high levels of genetic diversity. An average of 5.97 alleles per locus, ranging from 4.08 in Parintins to 7.92 for Guajará Mirim and Porto Velho populations (Table 1). High values were observed for H_O and H_E for all populations, with H_O showing values equal to, or slightly higher than H_E in 18 populations, with a single exception for the population of Atalaia do Norte, AM. The mean H_O of all populations (0.72) was also higher than the H_E (0.66). The average of f values per locus was below zero (f = –0.100), showing an excess of heterozygotes (Table 1). The allelic richness was 5.13.

The analyzes also showed a total of 37 private alleles observed in 16 populations, for all evaluated loci, except for the Epr36 locus (Supplementary Table 2), indicating a mean frequency of private alleles of 0.0986. Most of the studied loci showed linkage equilibrium, with percentages of LD that varied between zero percentage for the Borba population to 51.5% for the Atalaia do Norte population (Supplementary Table 2). Possible null alleles were detected for all microsatellite loci used. However, the frequency of these possible null alleles was found for some populations with frequencies less than 0.20. Null alleles with a frequency of less than 0.20 do not compromise population assessments (Dakin and Avise, 2004). Loci Epr22, Epr13, Epr14, and Epr19 presented higher frequencies in a single population. These alleles were maintained in the current study, for keeping their frequencies below 0.20.

Estimates of Wright’s F (Wright, 1951) in the 19 populations of E. precatoria sampled suggest that diversity is more concentrated within populations than among populations; however, high levels of genetic differentiation are indicated by an F_ST of 0.1820 (Wright, 1978). The estimates of F_IT (0.1215) and F_IS (-0.0740) were lower compared to F_ST (Table 2). The F_ST and F_IT fixation indexes were outside the upper and lower limits of the bootstrappping, which indicates that the estimates are significantly different from zero. However, the F_IS was not significant because it is lower than zero, indicating that the frequency of homozygotes is less than expected under EHW, showing an absence of inbreeding crosses (Table 2). In the F_ST estimates pair by pair (Table 3), most of the sampled populations showed significant differences. The largest statistical difference was observed among the sampled populations of Parintins and Itacoariara (39.57%) and the smallest among the populations of Anorí and Codajás (1.68%), the latter showing no statistical difference (p < 0.05) (Table 3).

The AMOVA results confirm the existence of genetic structure in the set of populations evaluated. Significant differences were observed among populations (18%; p = 0.004); however, most of the genetic variation occurred among the individuals evaluated (87.8%; p = 0.004) (Table 4). The significance of the components of variance of differentiation among populations (ΦCP = 0.1796), among samples within populations (ΦCAP = -0.0697) and among samples (ΦCA = 0.1224) is compatible with the results identified in statistical estimates of Wright’s F(1, 951), such as F_ST, F_IS, and F_IT, respectively. Under this scenario, the Mantel test, using the genetic matrices of F_ST and geographical distances of the populations, showed positive and significant correlations (r = 0.4044, p = 0.001), proving the existence of isolation by distance among the populations evaluated.

The estimate of the number of genetically homogeneous populations (K) among all individuals sampled in the 19 populations, as performed by the Structure software, showed the occurrence of three clusters (K = 3) (Figure 1). The classification of populations into three groups (K = 3) shows consistency in relation to geographical distance (Figure 3). The groups formed, considering K = 3, were as follows: group I covers the populations of the geographical basins of the Solimões, Negro, Uatumã and Amazonas Rivers (Ucurará, São Sebastião do Uatumã, Itapiranga, Anorí, Codajas, Coari, Presidente Figueiredo, Manacapuru, Itacoatiara, Novo Airão); group II is composed of the populations of the Solimões, Madeira and Amazonas Rivers (Parintins, Borba, Manicoré, Tabatinga, Benjamin Constant, Atalaia do Norte), and group III is found in populations of the Mamoré and Madeira Rivers (Nova Mamoré, Guajará Mirim and Porto Velho) (Figures 1, 2).

Using DACP analysis, population structure is formed by eight groups (Figure 3 and Table 5). The seven eigenvalues obtained in the Discriminant Analysis (32.5, 21.1, 18.8, 11.7, 4.9, 4.1, and 3.5%) explain 96.7% of the genetic variation among populations, with the first three eigenvectors explaining a greater percentage of the variation (72.4%). The results obtained are different from those verified by the Bayesian analysis carried out in the Structure program. However, DACP groupings I, IV, V, VI, and VII are closely related to the hydrographic basins where the populations were sampled (Table 5). In DACP group I, the populations of Urucara and São Sebastiao do Uatumã are grouped within the Uatumã River. In group IV, the populations of Tabatinga, Benjamin Constant and Atalaia do Norte are grouped at the headwaters of the Solimões River. In group V, the populations of Anori, Codajas, and Manacapuru are at the center and end of the Solimões River basin. In group VI, Borba and Manicore are located on the Madeira River. In group VII, Nova Mamoré, Guajará Mirim and Porto Velho are part of the Mamoré River that originated the Madeira River.

This is the first study to report on diversity and genetic structure of the natural populations of E. precatoria. The results will help define domestication and in situ and ex situ conservation strategies, as well as the management of natural or planted populations for the benefit of traditional Amazonian farmers and future enterprises. The E. precatoria plants sampled in the 19 populations under study revealed high levels of genetic diversity, mainly confirmed by the high content of genetic information observed in each index analyzed. The results on genetic diversity presented for this species are close to those of E. oleracea Mart. (Oliveira et al., 2010) and E. edulis (Novello et al., 2018). The negative values for the fixation indices (f) and the observed (H_O) and expected (H_E) heterozygosity coefficients indicate high genetic diversity (Nazareno and Reis, 2014), as confirmed by the excess of heterozygotes in the populations. The low f values and high H_O also indicate that the result may be influenced by the reproductive system of the species (Ramos et al., 2011) which is allogamy (Ramos et al., 2018, 2019). Similar results were found in other palm species, such as Geonoma schottiana (Silva et al., 2011), Phoenix dactylifera (Arabnezhad et al., 2012) and Oenocarpus bataua (Ottewell et al., 2012), which could indicate a behavior common to plants of the family Arecaceae.

The high genetic diversity observed for the species is an important information for the domestication process of E. precatoria, even though the anthropic extraction may affect the genetic diversity of the species in the future, in the process of recruiting new plants, in the seedling bank. However, domestication generally begins with the exploration of wild plants, proceeds with the cultivation of plants selected in nature, not genetically different from wild plants, and ends with the fixation of morphological and genetic characteristics carried out by human selection (Pickersgill, 2007). Thus, the domestication process must be part of any proposals for genetic conservation and management in order to contribute adequately to the process of evolution and adaptation throughout the region of occurrence of the species and the planting zones (Kalinowski, 2004).

Traditionally, genetic analyses in natural populations may not present equilibrium conditions when based on the assumption of panmixia that is not always met (Oosterhout et al., 2006). We also observed a great geographical distance among many of the studied populations, and different evolutionary events may have taken place in each location, causing the absence of HWE for some of the initiators of certain populations.

The 19 populations of E. precatoria distributed in the Legal Amazon showed genetic differentiation when Wright’s F statistics (Hartl and Clark, 2010) was analyzed, indicating the existence of a population genetic structure according to the F_ST values. The results also confirm that a large part of the genetic variability observed for the set of populations sampled is mostly within populations, owing to the reproductive system (F_IS). The populations also showed significant differences when analyzed in pairs (F_ST), also confirming the existence of genetic structure among the sampled populations (Wright, 1951). This result may be strongly influenced by the reproductive characteristics of the species, as well as the dispersion capacity of its genetic material, the degree of isolation of the population, the self-incompatibility system and allelic diversity that it presents (Leducq et al., 2011; Ramos et al., 2018).

The genetic structure observed also indicates that the sampled populations would be grouped according to the geographical distances within the hydrographic basins. Moreover, the occurrence of gene flow among populations is inversely proportional to the geographical distances among them. Isolation by distance has been previously identified to occur among populations of E. oleracea (Freeland et al., 2011). Thus, considering the geographical distance among populations of E. precatoria and the results of the comparisons of the F_ST analysis pair by pair, this result, hypothetically, shows that the gene flow that would be shared among these populations is proportional to the geographical distances among them because individuals are more likely to disperse to nearby locations (isolation by distance), confirming the population structure of the species. The positive and significant genetic correlation between geographical distance (as a dispersion limiter) and genetic distance supports this hypothesis and the dissimilarity in the composition of E. precatoria populations (Freeland et al., 2011; Kristiansen et al., 2012). Therefore, the allelic frequencies observed in the studied populations are related to geographic distances. This geographic differentiation information is confirmed by the groupings obtained in the analyses by the Structure program and the DACP function. The connectivity of the populations, according to the group they are in, is similar between both analyses (Figures 2, 3 and Table 5). For example, in the three municipalities of Nova Mamoré, Guajara Mirim and Porto Velho in the state of Rondonia, the same group was observed.

The results also suggest that the populations of E. precatoria would have started the process of formation of subpopulations in different parts of the Amazon, which could be related to the domestication process of E. precatoria, likely initiated by the Amerindians (Clement, 2001). This event may be closely related to their behavior because the Amerindians had a traditional subsistence system with high species diversification and building complex agroecosystems, including wood and non-wood products (Caviglia-Harris and Sills, 2005). Secondary dispersion is also carried out by birds and monkeys and also by man, transporting the fruit to consume, sell or plant in other areas. This seed dispersal will determine the colonization of new sites and migration among neighboring populations, especially if it is zoochoric because the range of seed dispersion can be substantially greater (Barluenga et al., 2011). However, the Acai-do-Amazonas, as well as the production of agricultural products already domesticated, incipiently domesticated, or collected in the forests by farmers, is traded among these local populations, and they mainly use river transport for this purpose. This is a dispersion vector that influences the movement of seeds, affording the plant with the capacity to persist, expand and colonize new habitats (Liu et al., 2013). This was also observed in the population of Presidente Figueiredo, which comes from fruits obtained by different government projects that were transported from Anori and Codajas.

It is believed that the genetic structure of populations may be related to the dispersion of seeds carried out among the extractive farmers themselves, who use river transport among the nearest communities. Therefore, sampling of seeds or propagules to compose tests of origins and progenies and/or commercial plantations, both for conservation and improvement, must consider the three groups identified in the analysis of genetic structure, as well as the estimated diversity among and within populations. This means that seed collections must be carried out across the three main hydrographic basins.

There is a concern about the intensification of fruit collection by extractive exploitation which is driven by the market (Instituto Brasileiro de Geografia e Estatística – IBGE, 2019) as it could be having significant impacts on genetic erosion in the near future, and, consequently, the reduction of fruit availability for frugivorous vertebrates, as well as the reduction of gene flow among populations (Ramos et al., 2018). To address this, restrictions on fruit harvesting should be carried out, such as avoiding fruit collection in years of low productivity and maintaining at least minus 50% of the fruits on mother palms for later consumption of fauna (Novello et al., 2018) and regeneration of new remnants.

In the process of ecological and dynamic succession of the forest with respect to the species E. precatoria, the plants of the populations may be showing high levels of genetic diversity at different ontogenetic ages (Ramos et al., 2018). The seeds are expected to represent a mixture of half sibs and full sibs (Bittencourt and Sebbenn, 2007). However, it is important to adopt a management plan that includes the destination of some E. precatoria fruits to the ecological community to maintain forest dynamics. Studies aimed at characterizing populations based on morpho-agronomic characters related to fruit production, as well as plant resistance and adaptation, will be important to identify the most productive individuals and progenies that can be used in different production systems, as well as enrich natural populations with the species. Information related to phenotypic, genetic and environmental characterization of populations is important to guide conservation and management strategies and the formulation of public species management policies in Amazonian watersheds.