The notion of pristine rainforests has been questioned by increasing archaeological and ecological evidence suggesting long-term human activities across even the most intact forests worldwide (Denevan, 1992; Van Gemerden et al., 2003; Willis et al., 2004; Ross, 2011; Boivin et al., 2016; Roberts et al., 2017). Amazonia is no exception — over thousands of years with humans living in the region, forest composition has been altered significantly (Clement et al., 2015; Levis et al., 2017b). Many dominant species in Amazonian forests are widely used as food resources by native indigenous peoples (ter Steege et al., 2013), and at least 85 tree and palm species were domesticated to some degree during pre-Columbian times (Clement, 1999; Levis et al., 2017b). Plant domestication is a long-term process that results from the capacity of humans to overcome environmental selection pressures with the purpose of managing and cultivating useful plants (Kennedy, 2012; Boivin et al., 2016; Levis et al., 2017b), leading to significant changes in natural ecosystems and plant communities across landscapes (Clement, 1999; Terrell et al., 2003). First, useful individuals are managed in situ (Rindos, 1984; Wiersum, 1997a) and later humans select the best varieties with more desirable morphological traits for cultivation (Darwin, 1859; Rindos, 1984; Clement, 1999). Over time, humans create a mosaic of domesticated landscapes to favor numerous useful plant populations, each domesticated with different intensities and outcomes (Wiersum, 1997b). In modern Amazonian forests, legacies of past human societies are evident in the surroundings of archaeological sites, where humans enriched the forest with useful, especially edible, and domesticated plants (Balée, 1989; Erickson and Balée, 2006; Junqueira et al., 2010; Levis et al., 2017b). These pre-Columbian legacies suggest that Native Amazonians interacted with natural ecological processes and shaped the distribution of plants and entire forest landscapes across the region (Balée, 2013).

In Amazonia, as in any other ecosystem, natural ecological processes drive the formation of plant assemblages and communities (Keddy, 1992; Zobel, 1997; Lortie et al., 2004; ter Steege et al., 2006). The first ecological process described to structure plant communities is the plant's capacity to disperse its seeds across landscapes (Ricklefs, 1987; Lortie et al., 2004), which depends on the regional species pool and multiple dispersal strategies, including occasional events of long distance dispersal (Ricklefs, 1987; Nathan et al., 2008). In wet Neotropical forests, animal dispersal is used by 75–98% of the tree species (Howe and Smallwood, 1982; Muller-Landau et al., 2008) and mammals disperse large-seeded species over long distances (Jordano, 2017). Once a propagule arrives in a given location, the second ecological process is related to how plants are able to overcome local environmental filters to successfully germinate and survive (Lortie et al., 2004). Plants compete with their neighbors for limited amounts of resources, such as light, nutrients and water (Moles and Westoby, 2006). The understory of a tropical forest is typically light-limited, forcing trees to either grow tall or survive in shady conditions (Poorter et al., 2003). Soils are also limited in water and nutrients, and plants need to compete in the rooting zone (Barberis and Tanner, 2005; Schnitzer et al., 2005). The third ecological process structuring plant assemblages is interaction with other organisms, such as herbivores and pathogens (Lortie et al., 2004; Bagchi et al., 2014). These multiple environmental and biological filters act simultaneously, resulting in trade-offs. For instance, species that grow fast under high light conditions tend to produce leaves that are less protected from herbivores, compared to the tougher and more resistant leaves of shade-tolerant species (Coley, 1983). In the long run, these ecological processes result in the selection of numerous adaptive plant traits (Reich et al., 2003), allowing species to thrive in complex and highly diverse systems, such as Amazonian forests. The high diversity of tropical ecosystems is in part maintained by natural disturbances and local biotic interactions, sometimes promoted by herbivores and pathogens that reduce the abundance of the most effective competitors, creating space for other species (Connell, 1978; LaManna et al., 2017).

Nonetheless, a few tree species often dominate plant assemblages forming oligarchic forests in diverse tropical forests (Connell and Lowman, 1989; Peh et al., 2011), including Amazonia (Peters et al., 1989; Pitman et al., 2001, 2013; ter Steege et al., 2013), Africa (Hart et al., 1989; Hart, 1990; Peh et al., 2011), Mesoamerica (Campbell et al., 2006), and Asia (Connell and Lowman, 1989; Peh et al., 2011). Natural and anthropogenic origins for the hyperdominance of tree species in Amazonian forests have been proposed. Aggregated patches of a few pioneer species occur after human or natural disturbance, while aggregated patches of a few shade-tolerant species may occur due to dispersal limitations (Valencia et al., 2004). Other hypotheses to explain why some species dominate large areas of Amazonian forests include: the species' ability to tolerate multiple environmental conditions, and to disperse over long distances (Pitman et al., 2001, 2013); and, in the case of useful species, the intentional or non-intentional enrichment promoted by past and contemporary human societies (Balée, 1989, 2013; Peters et al., 1989; ter Steege et al., 2013; Levis et al., 2017b).

During the Holocene, useful plant populations benefited from a new set of interactions when humans started to transform landscapes (Denevan, 1995; Smith, 2011; Boivin et al., 2016), and manage plant populations, consciously or not (Rindos, 1984; Wiersum, 1997a,b; Peters, 2000). Indigenous management practices were formally defined by Wiersum (1997a, p. 7) as “the process of making and effectuating decisions about the use and conservation of forest resources within a local territory.” When humans consciously manage forest resources, the underlying intention of their actions is not to domesticate forests, but to achieve certain short-term objectives, for instance to favor individual plants in the forest and promote their regeneration. Although changes in forest composition may not be the main goal of human actions, management practices also modify forest composition and structure beyond the targeted species in a long-term process. In tropical and subtropical forests worldwide, native societies have managed plants and landscapes, promoting oligarchic forests dominated by useful plant species, also defined as cultural or domesticated forests (Balée, 1989, 2013; Peters et al., 1989; Campbell et al., 2006; Michon et al., 2007; Reis et al., 2014; Morin-Rivat et al., 2017).

Today, many indigenous and traditional peoples recognize the handprints of their ancestors in the landscape (Frikel, 1978). Indigenous people are defined here as the descendants of native ethnic groups and members of an indigenous community that retains historical and cultural connections with the social organization of pre-Columbian indigenous societies (https://pib.socioambiental.org)¹. Traditional peoples can be understood as culturally differentiated and recognizable groups that have their own forms of social organization using knowledge, innovations and practices generated and transmitted by tradition, but they are not recognized as a member of indigenous communities (Brazilian Federal Decree No. 6.040)². In Amazonia, traditional peoples are generally descendants of migrants who intermarried with local indigenous peoples and they often exchange practices, objects and knowledge with members of indigenous communities. Although contemporary indigenous and traditional societies both cultivate fruit trees in their territory, they also take advantage of the aggregated patches of fruit trees created by the practices of previous generations (Frikel, 1978; Balée, 1989, 2013). These ancient cultivated landscapes were probably created by integrated agroforestry systems that included homegardens, swiddens and managed fallows in which tree and non-tree crops were intertwined (Denevan et al., 1984; Stahl, 2015). Such integrated systems were likely more efficient, in terms of food production, than long-fallow shifting cultivation systems when only stone axes were used to clear the forest in the past (Denevan, 1992). This is supported by the fact that past indigenous tree cultivation (arboriculture) was a common and widespread practice covering large areas of forest-savanna transition zones in Amazonia (Frikel, 1978).

Because trees persist in the forest following management (Levis et al., 2017b) and annual crops disappear after human abandonment (Clement, 1999), contemporary indigenous and traditional people commonly attribute the aggregated distribution of useful perennial plants to the action of their ancestors. Based on this knowledge, they sometimes select a new place to settle in the forest (Frikel, 1978; Politis, 2007; Rival, 2007; Zurita-Benavides et al., 2016). For instance, the Nukak Indians in Colombian Amazonia prefer camping around sororoca plants (Phenakospermum guyannense), because they believe that these plants were brought by their ancestors to “their living world,” and they discard a large quantity of seeds around their temporary camps, contributing to form new patches (Politis, 2007). Given that multiple human generations have moved around through time, places like riverine settings and archaeological sites were frequent dispersal routes of people and their cultures, and consequently of useful plants in pre- and post-Columbian times (Denevan, 1996; Hornborg, 2005; Guix, 2009; Heckenberger and Neves, 2009; Clement et al., 2010; Levis et al., 2017a,b). The intimate connections between Native Amazonians, their ancestors and their plants can reveal how persistent pre-Columbian forest management practices (Balée, 2000) contributed to the large-scale vegetation patterns we observe in modern forests (Pitman et al., 2011; Levis et al., 2017a,b).

Our study aimed to unravel how people interacted with natural ecological processes to transform pristine forests into domesticated forests with different degrees of human intervention through unintentional and intentional management practices. How indigenous and traditional peoples have used and shaped Amazonian forests is described in ethnographical, ethnobotanical, archaeological, paleoethnobotanical, paleoecological, and ecological publications. Here we used a historical-ecological perspective to evaluate the available information about how Native Amazonians have affected the distribution of plant species used mainly as food resources. Based on the information gathered from the literature, we developed an interdisciplinary conceptual model of how multiple management practices transformed pristine forests into domesticated forests, considering temporal and spatial contexts. In the field, we collected data about management practices and the composition of forest patches dominated by useful plants surrounding 30 contemporary villages, settled on or near archaeological sites. We compared field and literature data by documenting the multiple management practices known by 33 informants from two villages along the lower Tapajós River, and by relating these practices to the distribution and composition of the forest patches surrounding all 30 villages.

Based on our multidisciplinary approach, we provide a framework for understanding how human practices have led to the formation of patches of useful perennial plant species across Amazonian forests. Our conceptual model portrays how Amazonian peoples manage forests in multiple ways through eight categories of management practices that interfere with natural ecological processes and promote domesticated forests around human settlements. The similarities between ethnographic descriptions of management practices across the basin and our field observations of two villages indicate the commonness of these practices, suggesting that pre-Columbian and contemporary peoples transformed forest composition at varying distances from their settlements by multiple management practices. In the field, we confirmed that multiple diverse patches of useful species, currently managed by indigenous and traditional peoples, occur mainly near these settlements. Overall, our results support the view that these diverse patches of useful plant species were created and maintained by human actions.

Our conceptual model also reflects positive long-term interactions between humans and plants (Smith, 2011), as described in other tropical regions worldwide (Wiersum, 1997a; Michon, 2005; Kennedy, 2012; Reis et al., 2014; Boivin et al., 2016; Roberts et al., 2017). Previous models had suggested that the plant and forest domestication processes are associated with the cultivation of domesticated tree crops (Wiersum, 1997a,b). Although our model is inspired by previous studies (Harris, 1989; Wiersum, 1997a,b), we present a new framework to understand the domestication of Amazonian forests that simplifies the complex network of interactions between human actions and natural ecological processes. Because these interactions cannot be understood by separately assessing only individual management practices or species, the intricate groups of management practices shown in our model illustrate how multiple human actions interact to shape Amazonian forests. Species-specific details are scattered in the literature, and here we synthesized this information into a single model that can be tested with individual site-specific situations.

In our model, forest domestication is defined as an open-ended process (Rival, 2007; Kennedy, 2012), in which domesticated forests can originate through varying degrees of human intervention from settlements and swiddens, and also from old-growth forests. This perspective makes the typical distinction between hunter-gatherers vs. farming groups inappropriate for the Amazonian context (Terrell et al., 2003; Kennedy, 2012), as most ancient Native Amazonians (often characterized as hunter-gatherers) were actually practicing many activities, including planting tree species (Frikel, 1978). Amazonian forests that were once cultivated and domesticated are often transformed into swiddens or settlements as a cyclic pattern that has also been observed in Indonesian forests (Michon, 2005). Because early successional species usually depend on forest gaps for recruiting, they are maintained with management practices, similar to fully domesticated plant populations that require human care for survival and reproduction (Clement, 1999).

Although it is likely that current management practices maintain the legacy of past societies (Junqueira et al., 2017), the effects of past forest domestication have been detected in forests even without recent management activities (Van Gemerden et al., 2003; Dambrine et al., 2007; Ross, 2011; Levis et al., 2017b). The persistent effect of pre-Columbian plant domestication on modern forest composition has been revealed in Amazonian old-growth forests (Junqueira et al., 2017; Levis et al., 2017b), secondary forests (Junqueira et al., 2010) and even in highly dynamic homegardens growing in archaeological sites (Lins et al., 2015). Domesticated species adapted to stable soil conditions created by management practices, such as TPI, may persist for a long time after abandonment (Quintero-Vallejo et al., 2015). This may explain why domesticated palms dominate modern forests growing on pre-Columbian mounds, anthropogenic soils and geoglyphs abandoned more than 400 years ago (Erickson and Balée, 2006; Quintero-Vallejo et al., 2015; Watling et al., 2017b). Another possible explanation for this persistence is the continuous recruitment of useful and domesticated plants present in the forest seed bank (Lins et al., 2015). Pre-Columbian peoples may also have played a major role in disseminating large multi-seeded fruits within and across Neotropical biomes during the Holocene, resulting in the spread of diverse patches of useful plants associated with human settlements and trails (Guix, 2005). Human-mediated dispersal of invasive plants is well-documented (Hodkinson and Thompson, 1997; Nathan et al., 2008); however, ecological studies frequently overlook this mechanism when considering native species (Levis et al., 2017a).

Modern Amazonian peoples who live on pre-Columbian settlements seem to have inherited indigenous knowledge, including these management practices that benefit useful and domesticated plant populations. Our field data show that most useful species dominant in forest patches occur in more than one sub-basin visited, suggesting a widespread use and management of forest resources by past and contemporary peoples. The forest domestication process was assimilated by contemporary societies through the transmission of indigenous knowledge from one generation to another, as described for indigenous groups from Ecuadorian Amazonia (Zurita-Benavides et al., 2016) and traditional people in Brazilian Amazonia (Alves et al., 2016). Villages with homegardens that were occupied by several pre-Columbian cultures contain a higher beta diversity of useful plants compared to villages with homegardens occupied by a single culture (Lins et al., 2015), suggesting that previously existing useful plants were incorporated into new agroforestry systems when old villages are re-occupied (Miller and Nair, 2006). Some practices, however, have changed in intensity and extension through time. Slash-and-burn agriculture, for instance, has increased since the arrival of European societies that introduced metal tools to cut down the forest (Denevan, 2001). In pre-Columbian times, sedentary societies frequently improved soil conditions by managing fire in their habitation and cultivation zones (Denevan, 2001; Neves et al., 2003; Woods et al., 2013). Sedentary societies with high human population densities were responsible for the formation of anthropogenic soils that are no longer being created on a broad scale (Neves et al., 2003). These same anthropogenic soils, however, are widely used by modern societies to cultivate crops, allowing the diversification and intensification of food production in Amazonia (Woods et al., 2013; Junqueira et al., 2016b).

Amazonian societies managed fire, planted useful species and improved soils that resulted in substantial transformation in forests close to their homes. Although some scholars argue for a localized impact involving these three practices in pre-Columbian Amazonia, associating them with the margins of the main rivers (McMichael et al., 2012, 2014; Bush et al., 2015; Piperno et al., 2015), the impact of long-term management practices has been detected in the forests of interfluvial areas (Levis et al., 2012; Franco-Moraes, 2016; Watling et al., 2017b) and across the Amazon basin (Levis et al., 2017b). These findings suggest that even in remote areas, far from known archaeological sites, contemporary people also manage the forest, protecting useful species and removing the non-useful, which are the most frequent practices reported by contemporary societies. Logistical limitations constrain our ability to detect the long-term effects of these practices away from current human settlements (Stahl, 2015), and even the participatory techniques used in this study are based on current knowledge about the forest, requiring ethnographic projection to infer the impact of past peoples. For instance, patches of rubber tree (Hevea brasiliensis) have been managed by modern societies driven by economic interest since the mid-nineteenth century (Schroth et al., 2003), but were probably managed differently before that time. Although several socio-economic factors push contemporary peoples to concentrate their activities on market-oriented forest resources (Jakovac et al., 2016b), they occasionally use and manage forest patches located up to 40 km from their villages for hunting animals and gathering fruits (Figure 5; Franco-Moraes, 2016). As an alternative approach, the abundance and richness of useful plants, especially of domesticated species, might be used to predict the location of ancient human settlements in these remote Amazonian areas (Levis et al., 2017b).

Future multidisciplinary studies that combine alternative methods may help to reconstruct forest composition dynamics (Stahl, 2015), as Watling et al. (2017b) did in the geoglyph region of Acre, revealing more details of the influence of past peoples in Amazonian forests. The integration of paleoecology, archaeology, archaeobotany and forest ecology is a promising combination (Mayle and Iriarte, 2014; Iriarte, 2016; Watling et al., 2017a,b). In southwestern Amazonia, archaeobotanical remains have revealed that past peoples consumed a rich diet, including many palm fruits (Dickau et al., 2012). The increase in palm abundance is also visible in soil profiles of archaeological sites across the region (McMichael et al., 2015; Watling et al., 2017b), suggesting that past societies enriched the forest with useful palms to improve food production. Today, useful and domesticated palms are dominant in southwestern Amazonian forests (Levis et al., 2017b), growing on abandoned pre-Columbian mounds, anthropogenic soils and geoglyphs created by past management practices (Erickson and Balée, 2006; Quintero-Vallejo et al., 2015; Levis et al., 2017b; Watling et al., 2017b). Many palm species were found in most of the forest patches investigated here, suggesting long-term human management. Regional contrasts in palm and other plant species composition across Amazonia may reveal different human practices or specific environmental conditions that should be investigated in detail.

We conclude that our literature review, conceptual model and field results contribute to explain how domesticated forests were formed in Amazonia, in part by revealing how integrated categories of management practices interfere with natural ecological processes that shape plant communities in tropical forests. Different degrees and types of management, cultural preferences and environmental conditions may lead to a wide variety of outcomes and explain why diverse combinations of useful species were found in Amazonian forest patches. Insights from agroforestry systems in tropical and sub-tropical regions confirm that indigenous management practices have been used worldwide to domesticate plant species and entire forest landscapes (Wiersum, 1997a,b; Michon, 2005; Kennedy, 2012; Reis et al., 2014). Learning about indigenous knowledge of forest management is important not only to understand the plant and landscape domestication processes, but also to guide policies for forest conservation, local people's empowerment, and food production (Michon et al., 2007; Roberts et al., 2017). In Amazonia today, millions of people live in rural landscapes, with partial dependence on forest resources for their well-being, and with profound local knowledge that should be incorporated in environmental conservation and management plans.

CL conceived the study; CL, BF, PM, BL, RA, JF-M, JL, and EK collected data; CL, BF, PM, BL, RA, JF-M, JL, EK, FB, MP-C, FC, and CC designed the analyses; CL, BF, PM, BL, RA, JF-M, JL, and EK performed the analyses; CL, BF, PM, BL, RA, JF-M, JL, EK, FB, MP-C, FC, and CC discussed further analyses; CL, BF, PM, BL, RA, JF-M, JL, FB, MP-C, FC, and CC wrote the manuscript.