The development of agriculture is one of humankind’s most pivotal achievements. Questions about plant domestication and the origins of agriculture have engaged scholars for well over a century, with implications for understanding its legacy on global subsistence strategies, plant distribution, population health and the global methane budget. Rice is one of the most important crops to be domesticated globally, with both Asia (Oryza sativa L.) and Africa (Oryza glaberrima Steud.) discussed as primary centres of domestication. However, until now the pre-Columbian domestication of rice in the Americas has not been documented. Here we document the domestication of Oryza sp. wild rice by the mid-Holocene residents of the Monte Castelo shell mound starting at approximately 4,000 cal. yr bp, evidenced by increasingly larger rice husk phytoliths. Our data provide evidence for the domestication of wild rice in a region of the Amazon that was also probably the cradle of domestication of other major crops such as cassava (Manihot esculenta), peanut (Arachis hypogaea) and chilli pepper (Capsicum sp.). These results underline the role of wetlands as prime habitats for plant domestication worldwide.
More than half of the world’s population depend on rice for >20% of their daily calories1. Modern global consumption is dominated by varieties of the domesticated Asian (Oryza sativa L.) and African (O. glaberrima Steud.) species2, which were respectively domesticated in the early Holocene in the Yangtze River, China3, and approx. 2,000 cal. yr bp in West Africa4. In North America, Zizania wild rice was so important to the subsistence economy of several Upper Great Lakes Native American tribes that some early-twentieth-century ethnologists designated this region as a distinct ‘wild rice culture area’5. Wild rice was already a seasonal staple of indigenous subsistence in South America long before the introduction of Old World species in the eighteenth century6. Growing in seasonally flooded areas that compose up to 10% (1.4 × 106 km2) of lowland South America (Fig. 1), wild rice is a particularly important resource during the rainy season when flooding causes other resources to be dispersed and scarce7. Historical and ethnographic accounts (sixteenth–nineteenth centuries) report extensively on the consumption of wild rice species by indigenous groups in this region. Similar to the traditional North American canoe-and-flail harvesting method, native South American people were reported to harvest wild rice by beating the grains of mature inflorescences into their canoes with wooden poles8,9,10,11. South American accounts hint towards the importance and culinary practices involving wild rice. For example, De Azara12 mentions the consumption of an unknown type of rice in southern Paraguay that “… feed a nation of approximately seventy warriors”. Cardim13 mentions that wild rice was mixed with maize to make bread, and Acosta14 describes its consumption in the form of a fermented brew, similar to wine. Locally known as arroz-de-pato (duck rice) or arroz-do-brejo (swamp rice) today, wild rice is stilly consumed as a valuable source of carbohydrates when other food resources are scarce by riverine communities across the Amazon. It is still gathered and consumed in various modern localities close to the study site along the Guapore River such as Costa Marquez and Santo Antônio, where the communities used to manage wild rice stands until the first half of the twentieth century. This kind of landscape management can still be observed in other parts of the Amazon, such as in wild rice fields of the municipality of Manaquiri, in the lower Solimões River basin15 (Supplementary Fig. 1b). In the Pantanal, the native Guató communities consume the wild native species O. glumaepatula and O. latifolia by sun drying the seeds, peeling them, and boiling them16. However, despite the occasional reference to its potential role in pre-Columbian diets17,18, the domestication of rice has not yet been investigated in this region. Increasingly large Oryza sp. husk phytoliths recovered from mid-Holocene levels of a shell mound in southwestern Amazonia (Fig. 1) dating to ca. 4,000 cal. yr bp show the progressive selection of larger wild rice seeds by its pre-Columbian residents, who were already engaged in the cultivation of maize (Zea mays) and squash (Cucurbita sp.)19.
Taxonomy and domestication phytoliths of Oryza
Oryzoideae (syn. Ehrhartoideae) is a subfamily of the true grass family Poaceae that includes around 120 species in 20 genera. The Oryzeae tribe within the Oryzoideae subfamily consists of twelve genera and is distributed in tropical and temperate regions worldwide. Five of these twelve genera occur in South America: Leersia, Luziola, Rhynchoryza, Zizaniopsis and Oryza20,21. The Oryza genus comprises 22 known wild species. Four of them are endemic to Latin America with a tropical–subtropical distribution from Cuba 23° N to the Paraná River delta 34° S, including the diploid (2n = 24, AgpAgp) O. glumaepatula, and three tetraploids (2n = 48, CCDD) O. alta, O. grandiglumis and O. latifolia22 (Fig. 1). Oryza spp. rice are all aquatic emergent macrophytes that grow along rivers, lakes and wetland margins. O. alta, O. grandiglumis and O. latifolia are perennial species, while O. glumaepatula can be annual, biannual or perennial depending on the geographical location23,24. Oryza spp. have a nutty flavour and firm consistency. Preliminary studies on O. glumaepatula show that it has high levels of total protein, albumin and glutelin fractions, which compares favourably with O. sativa commercial cultivars25. Wild rice can also be stored and can be rather productive. Although, not directly comparable to Oryza, the traditional canoe-and-flail harvesting of Zizania wild rice in North America yield about 125 kg ha–1 (ref. 26), while for modern domesticated shattering resistant cultivars, yields have been reported as high as 1,680 kg ha–1 in Minnesota and twice that amount in California27.
The Oryzoideae subfamily produce four distinct phytoliths associated with different parts of the plant. The Oryzeae tribe produce: (1) cuneiform keystone bulliform cell phytoliths exhibiting fish-scale decorations on the fan edges are produced in the leaves (Fig. 2d) and (2) ‘scooped-shaped’ bilobates in the leaves and stems (Fig. 2e). The Oryza genus produce: (3) double-peaked glume cells (Fig. 2a,b,f–i); and (4) deeply serrated phytoliths both derived from the epidermis of the Oryza seed glume (husk; Fig. 2c)28,29,30. The presence of diagnostic Oryza phytoliths produced in the different parts of the plant has allowed the detection of crop processing stages31 and different agricultural techniques32 in Asia. For example, the distinctive bulliform and bilobate phytoliths from Oryzeae leaves and stalks are representative of the early stages of harvesting and processing, while the Oryza husk double-peaked glumes represent later stages of processing, such as pounding, winnowing and storage.
Domestication is a process that causes genetic changes in populations such that the average phenotype diverges from the range found in wild populations18. Domestication causes a gradual increase in plant size from wild to domesticate as a result of selective exploitation33. As the plant become larger, so do the phytoliths. The increase in phytolith size has been documented in Zea mays34, Cucurbita35 and Musa bananas36, where larger fruits and seeds often yield considerably larger phytoliths. A clear correlation has been demonstrated29,30 between increasing phytolith size and domestication in Asian rice based on analysis of 27 accessions of domestic rice, originated from China, and 79 specimens from the nine wild rice species considered ancestral to rice distributed geographically in South and Southeast Asia. A discriminant function was devised30 to differentiate assemblages of wild from domesticated Oryza rice using five different size measurements of the double-peaked glume cells including: (1) top width (TW), the distance between the two peaks of the projecting hairs; (2) maximum width (MW), the width at the point where the glume projection attaches to the base; (3,4) heights of each hair (H1, H2), length from the tip to the base of the hair, where H2 is defined as the smaller measurement; and (5) the curve depth (CD), distance from the tip of H1 to the lowest point of the curve (Fig. 3m). Further comparative research including hundreds of grass species from China37 and including wild and domesticated rice species from East Asia28 have confirmed their results.
Archaeological background of the Monte Castelo shell mound
Dating back to approx. 10,000 cal. yr bp, a diversity of coastal and freshwater38 shell mounds represent some of the oldest forms of human occupations across lowland South America, some of which are associated with the earliest ceramics on the continent39. Our study site, the Monte Castelo residential shell mound is located in the Upper Madeira basin of southwest Amazonia, Rondônia state, Brazil. The region is characterized by a seasonally flooded tropical wetland exhibiting gallery forest along the larger streams, which are dotted with anthropogenic shell mounds38. Monte Castelo is a 6.3-metre-high platform-shaped freshwater shell-mound, exhibiting a 160-metre-long elliptical base (Fig. 1c) and dating from 9,400 cal. yr bp40,41. The first excavation of Monte Castelo in 198442, revealed a seven-metre-deep stratigraphy bracketing a long-term occupation from 9,130 to 667 cal. yr bp (Supplementary Table 1). Miller defined three major and one transitional occupation phases based on stratigraphy, artefact content and sixteen radiocarbon dates including: Cupim phase (700–685 cm; 9,130–7,701 cal. yr bp), Sinimbu phase (670–275 cm; 7,701–4,822 cal. yr bp), Sinimbu–Bacabal transitional stratum (275–220 cm; 4,862–4,388 cal. yr bp) and Bacabal phase (220–30 cm; 4,388–689 cal. yr bp)42. Renewed excavations at Monte Castelo in 2014 and 2016 by the Laboratory of Tropical Archaeology of the University of São Paulo expanded the previous excavation, reaching a depth of 640 cm. They uncovered ten archaeological strata across the Sinimbu to Bacabal phases dating from 5,310 cal. yr bp. to 689 cal. yr bp (Fig. 3k; Supplementary Fig. 3; Supplementary Note 1)40. The stratigraphy shows a sequence of construction events evidenced by unburnt entire Pomacea shell layers, occupation floors marked by lenses of crushed shells, primary burials and human-created dark soils. Sample collection for microfossil analysis was carried out in undisturbed sectors of each of the layers and targeted samples were collected from particular features such as burials (Supplementary Note 1; Supplementary Fig. 3; Supplementary Table 3).
Results and discussion
To investigate the use and potential domestication of wild rice by the Monte Castelo residents we analysed both archaeological samples and modern wild rice reference material. A total of 16 archaeological sediment samples, from across all ten levels uncovered during the 2014 Monte Castelo excavations (Fig. 3; Supplementary Table 3), and 19 modern specimens from the four wild species of rice occurring in South America (Supplementary Table 2), were analysed for phytoliths following standard procedures34 (Methods, Supplementary Tables 2 and 3). Each slide was scanned until the first 20 double-peaked glume cells were encountered. Following ref. 30, the five metric attributes (Fig. 3m) were measured from 20 Oryza double-peaked glume phytoliths from each of the archaeological (16) and modern samples (19) totalling 700 phytoliths.
Phytolith preservation was excellent in all context analysed. All archaeological sediment samples analysed yielded phytoliths of wild rice. Our analysis shows a clear increase in the proportion of rice morphotypes in the total phytolith assemblage from 6.4% on average in the Sinimbu phase occupation (layers J–H) to 14.4% in the more recent Bacacal phase, suggesting that rice may have played a larger role in diet over time (Fig. 3f).
At Monte Castelo, there is also an increase in the proportion of Oryza seed phytoliths from the lower to the upper levels of the mound reflected in the husk:(leaf + stem) ratio. For example, during the Sinumbú phase (layers J–I; 280–460 cm) Oryza sp. seed phytoliths represent on average 3.4% of the total assemblage and Oryzeae leaf and stem phytoliths constitute on average 3%, a 1:1 ratio. During the Bacabal occupation (layers F–A; 30–210 cm) Oryza seed phytoliths constitute on average 12% of the total assemblage and leaf phytoliths constitute on average 3.5%, a ratio of 3.4:1, over three times the relative proportion of seed husks as occur in the Sinimbu occupation (Fig. 3g). The collection and flailing of wild rice in canoes in the Americas should leave leaf and stem bulliform and bilobate phytoliths in the place of harvest while double-peaked and deeply serrated glume phytoliths should be more abundant at residential sites where the grain is brought for consumption. Therefore, the increase in the ratio of husk:(leaf + stem) Oryzeae phytolith morphotypes suggests that the Monte Castelo residents became more efficient harvesters over time, bringing more grain and fewer leaves to the site.
The analysis of the average size of the attributes measured on the Oryza glume phytoliths (Fig. 3 and Supplementary Fig. 5) shows a gradual increase in height (H1, H2) and width (TW, MW) through time. Mean H1 values increase approx. 8 μm (17–25 μm) and H2 increases approx. 7 μm (15–22 μm) from layers J to A. MW increases 9 μm (48–57 μm) through the stratigraphy. Mean CD values are larger in the upper occupation layers (A–H) compared to its initial dimensions in layers I–J (Fig. 3). We used principal component analysis (PCA) of modern reference wild species to determine the variables that best explained phytolith shape differences among specimens, which are the two highly correlated height and width measurements (Supplementary Note 2, Supplementary Figs. 6–9). Following ref. 30, therefore, we created a simple model of phytolith size to characterize the changes in phytolith morphology through time. Results of a one-way ANOVA show that mean phytolith size varies significantly among layers and pairwise comparison (with Bonferroni-corrected p-value) shows phytoliths in the upper archaeological layers (A–D) are significantly larger than those in layer J and wild reference specimens (Supplementary Table 4). Figure 4 illustrates mean height and width of all Oryza phytolith specimens, showing an increase in phytolith size through time. The data show a significant shift towards bigger phytoliths compared to wild specimens began in layers D–E (Fig. 3k) around 4,000 cal. yr bp. Phytolith size in lower archaeological layers were not significantly different from some botanical specimens (O. latifolia, O. alta) (Supplementary Table 4). The gradual increase in Oryza husk phytolith dimensions since the basal layers of the Monte Castelo shell mound suggest that the Monte Castelo residents may have been manipulating Oryza by at least 5,000 cal. yr bp. Phytolith data also show that subsistence strategies of the Monte Castelo residents were based on a mixture of wild and domesticated resources including cultivars such as maize and squash as well as other plants of economic importance including palm fruits and possibly soursop (Annona sp.) (Fig. 2j–m).
Our results indicate a significant increase (Fig. 4 and Supplementary Table 4) in the size of double-peaked glume phytoliths across the Monte Castelo occupation starting around 4,000 cal. yr bp. Wild rice constituted an important seasonal resource for the Monte Castelo residents, who began to husband wild rice stands at lake or river edges. The phytolith data show that wild rice was modified by human intervention to produce larger grains, exceeding the range of variation found in the lower levels of the Monte Castelo shell mound and the modern populations of wild rice. The possibility that the increase in dimensions of husk phytoliths may be a result of selection for large seeds during collection from wild plant stands is countered by fact that no husk phytoliths with larger dimensions than the domesticated ones have been found on the modern wild rice specimens.
Oryza alta, O. grandiglumis and O. latifolia are perennial species, whereas O. glumaepatula can be annual, biannual or perennial depending on the geographical location23,24. Although we cannot distinguish specific Oryza species using phytoliths, it is likely that the Monte Castelo residents were targeting the annual varieties of O. glumaepatula due to their generally larger-scale seed production compared to perennials, as seen with other cereal grains43. The specific husbandry practices that led to this process of domestication are unknown; however, native North Americans increased natural Zizania wild rice stands by mixing wild rice seeds into clay, rolling it into a ball and dropping the clay ball into the water27. Monte Castelo residents may have seeded the Guapore basin wetland margins with a similar practice. With this technique, larger seeds might have been indirectly selected because they would germinate better from the clay balls, eventually leading to domestication. In addition, like traditional societies in India today, they may have practised burning of enriched rice patches during the dry season to remove competing vegetation after rice grains were embedded safely in the soil. To what extent the selection of non-shattering types contributed to the fact that the Monte Castelo residents became more efficient harvesters, as shown by the increase in husk:(leaf + stem) ratio, is something we cannot directly detect with phytolith analysis, since phytoliths cannot document the presence/absence of this key domestication syndrome trait.
It is interesting to note that the apparent major role of rice in the diet of the Monte Castelo residents, as well as the beginning of its domestication, coincides with a rapid increase in precipitation in the Amazon. As summarized previously44, the palaeoclimate records from southern Amazonia and adjacent regions influenced by the South American Low Level Jet show a consistent long-term trend of increasing precipitation starting during the mid-Holocene (~6,000 cal. yr bp), showing a rapid rise up to 4,000 cal. yr bp, and then continued to increase slightly towards the present. This higher precipitation would probably have expanded the spatial extent of wetlands across the basin and possibly made the flooding season longer. Because wild rice is a particularly important resource during the rainy season in wetlands and floodplains when flooding causes other resources to be disperse and scarce, the increase precipitation would have probably made wild rice a critical seasonal resource, which may have, in turn, led populations to focus on its manipulation, which ultimately led to its domestication. Further work is needed on this hypothesis.
The presence of phytoliths from known cultigens, such as the wavy-top rondels of maize and scalloped spheres from squash, in the strata analysed shows that both crops were commonly grown in the region from at least 5,300 cal. yr bp onwards (Fig. 3, Supplementary Fig. 4). This in turn, indicates that the Monte Castelo shell mound residents began to systematically select larger rice seeds when they were already engaged in the cultivation of maize and squash. While in other regions of the Americas, wild grasses such as Setaria45 or marsh elder46 decrease in importance or are replaced by maize, the opposite trend is apparent in the Monte Castelo record. Wild rice was domesticated and increased in importance a considerable time after Monte Castelo residents had become engaged in farming practices.
The arrival of Europeans to the American continent in ad 1492, with the consequent population decimation and impact on cultural practices, caused the domesticated traits to gradually disappear. The loss of domesticated varieties is a phenomena that has also occurred for other indigenously domesticated species in both South18 and North America46. A case in point similar to Oryza is the ‘extinct cultigen’ marsh elder (Iva annua), a member of the Asteraceae family greatly appreciated for its achene oil content, which was originally domesticated in southeastern North America and then abandoned with the introduction of maize46. As in our case study, the achenes of marsh elder from the earlier archaeological sequences are not much larger than the modern ones, but the achenes from the more recent archaeological contexts are much larger than any existing races of Iva annua today. In the case of rice, some varieties are in the process of de-domestication today; modern studies of Californian weedy rice show how reversions to non-domestic or wild-traits (such as seed shattering, presence of awns) can occur following abandonment47. In our case study, it is likely that the wind-pollinated wild rice progressively hybridized with the domesticated one, with the consequent return to the wild characteristics seen today.
Our study highlights the importance of wetlands for the adoption and intensification of agriculture48,49. The results contribute to a broader understanding of how wetlands and the seasonal tropical forests of the Amazon may have been critical for early human settlement and the origins of food production in the Americas. This domestication process took place in a region that was probably the cradle of domestication for cassava, peanuts and chilli peppers pointing to the importance of this region of South America19.
Our research has implications for sustainable Amazonian futures. Modern intensive breeding for high yield and pest resistance has narrowed the genetic diversity of cultivated rice leaving crops more susceptible to disease and less adaptable to the effects of climate change. Understanding the process of rice manipulation by ancient Native Americans and the role of South American native varieties could help provide more resistant high-yielding varieties, and provide further knowledge for plant breeders interested in the introgression of genes from wild Oryza species into modern rice varieties22.
Phytoliths were identified and counted under a Zeiss Axioscope 40 light microscope at 500× magnification. Phytolith identifications were made using published material for the Neotropics and the Oryzoideae family29,30,34 and by direct comparison with the phytolith reference collection of the Archaeobotany and Palaeoecology Laboratory in the Department of Archaeology of the University of Exeter. A minimum of 200 phytoliths were counted per slide. Following ref. 30, the five metric attributes (Fig. 3m) were measured from 20 Oryza double-peaked glume phytoliths from each of the archaeological (16) and modern samples (19) totalling 700 phytoliths.
The dataset analysed is available from corresponding author upon request.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The research was funded by the European Research Council project ‘Pre-Columbian Amazon-Scale Transformations’ (ERC-CoG 616179) to J.I. L.M.H. was funded by CAPES (Ministry of Education, Brazil) and Monte Castelo fieldwork was funded by grants from the Brazilian National Science Development Council (CNPq-307179/2013-3) and The National Geographic Society (W243-12) to E.G.N.
Electronic supplementary material
Supplementary notes, tables, figures, and references.