Persistent effects of pre-Columbian plant domestication on Amazonian forest composition

In "Persistent effects of pre-Columbian plant domestication on Amazonian forest composition" (3 March 2017, p. 925), Levis et al. tested for relationships between current distributions of tree species in the Amazon basin and the distribution of pre-Columbian settlement sites, to assess how environmental and anthropogenic factors shape current distributions. Using multiple linear regression (MLR) they concluded that across the Amazon basin "the relative abundance and richness of domesticated species increase in forests on and around archaeological sites", and that past human settlement is responsible for roughly half the variation explained in the abundance and richness of domesticated tree species. Distance from archaeological sites and rivers served as proxies for intensity of human land use.
Levis et al.'s use of MLR is misspecified since it only accounts for linear relationships between explanatory variables and vegetation measures, and therefore inadequately controls for environmental gradients. Results showing that human land-use proxies were prominently and significantly related to domesticated tree species may have occurred because out of all explanatory variables tested, the relationship between human land-use proxies and vegetation is perhaps the only variable that is expectedly linear or at least monotonically decreasing (e.g. as distance from archaeological sites increases, the abundance of domesticate species decreases). Conversely, relationships with environmental variables that were tested (e.g. soil pH) may be non-linear and non-monotonic. Tree species often grow within environmental optima and do not exhibit monotonic relationships with environmental variables, and the use of linear functions in quantitative models is inconsistent with ecological theory (1). Scatterplots of vegetation measures versus environmental variables, which might exhibit such relationships, were absent from Levis et al.'s supplementary material.
More appropriate would be to apply advanced quantitative modeling approaches, particularly from the extensive literature on species distribution models (SDMs) (2). SDMs are statistical and machine-learning techniques that are typically used for predicting probabilities of species presence, but have also been adapted for modeling abundance and richness. They possess numerous advantages over MLR, including their ability to model non-linear relationships between environmental variables. Indeed, SDMs have detected monotonic relationships between vegetation and land-use proxies along with non-monotonic relationships between vegetation and environmental conditions (3), in regions where prior linear-based methods detected relationships between vegetation and land-use proxies but less so with environmental variables (4). We suggest that future research in the Amazon basin and elsewhere apply more rigorous quantitative modeling such as SDMs to separate environmental controls from anthropogenic signatures upon vegetation patterns.

References and notes:
1. M. Austin, Species distribution models and ecological theory: A critical assessment and some possible new approaches. Ecol. Model. 200 (1-2), 1-19 (2007).
2. A. Guisan, N.E. Zimmermann, Predictive habitat distribution models in ecology. Ecol. Model. 135 (2-3), 147-186 (2009).
3. S.J. Tulowiecki, C.P.S. Larsen, Native American impact on past forest composition inferred from species distribution models, Chautauqua County, New York. Ecol. Monogr. 85 (4),557-581 (2015).
4. B.A. Black, C.M. Ruffner, M.D. Abrams, Native American influences on the forest composition of the Allegheny Plateau, northwest Pennsylvania. Can. J. For. Res. 36, 1266-1275 (2006).

We agree that Amazonia is not as pristine as often imagined, but Levis et al. seem to have taken their interpretations to the other extreme. They argue that the effect of pre-Columbian humans on some aspects of Amazonian tree communities equals the effect of the modern environment (fig. 5, p. 930).
We argue that their analyses substantially underestimated the importance of soils. Amazonia is known to harbor remarkably diverse edaphic conditions (1), and numerous studies have shown that Amazonian forest composition and plant species distributions reflect soil properties at all spatial scales (2–4). The best predictor variables have generally been physiologically important plant nutrients, such as calcium, magnesium, potassium and phosphorus. However, none of these were included in the analyses of Levis et al., which only used pH and cation exchange capacity (CEC). These are poor surrogates, because soils of any given pH value or CEC can differ greatly in nutrient availability in Amazonia (5), so these variables were probably chosen because they are available as basin-wide digital raster layers rather than because of their ecological relevance. CEC reflects the amount of exchangeable surface charges, which is of interest in agriculture as it indicates the potential of the soil to respond to fertilizers. However, in natural conditions more than 90% of CEC is often occupied by aluminum (1) and, therefore, does not indicate actual nutrient availability. CEC can correlate with nutrient content in basic soils, but this correlation is expected to be weak in the generally acid soils of Amazonia.
Although Levis et al. focused on the local abundance and species richness of 85 putatively domesticated species, the human-related variables explained, on average, less than 10% of the variance. Nevertheless, their final conclusion was that "Domestication shapes Amazonian forests". This and other conclusions of Levis et al. seem to exaggerate the relative importance of pre-Columbian human effects across Amazonia and to underestimate the importance of environmental conditions, of which soils can be expected to be especially influential at the relevant spatial scales.
References and Notes:
1. C. A. Quesada et al., Soils of Amazonia with particular reference to the RAINFOR sites. Biogeosciences. 8, 1415–1440 (2011).
2. H. Tuomisto et al., A compositional turnover zone of biogeographical magnitude within lowland Amazonia. J. Biogeogr. 43, 2400–2411 (2016).
3. E. N. Honorio Coronado et al., Multi-scale comparisons of tree composition in Amazonian terra firme forests. Biogeosciences. 6, 2719–2731 (2009).
4. R. Cámara-Leret, H. Tuomisto, K. Ruokolainen, H. Balslev, S. Munch Kristiansen, Modelling responses of western Amazonian palms to soil nutrients. J. Ecol. 105, 367–381 (2017).
5. C. A. Quesada et al., Variations in chemical and physical properties of Amazon forest soils in relation to their genesis. Biogeosciences. 7, 1515–1541 (2010).

Methodological biases predispose Levis et al. (1) to conclude that ancient domestication has shaped Amazonia. Their analysis fails to consider four centuries of European influence, native population recovery in the 1600s, and the Rubber Boom (ca. 1850 - 1920 AD) that followed the pre-Columbian era. Ancient occupation sites have higher likelihoods of subsequent settlement (2); people today still cultivate on Amazonian Dark Earths formed in ancient times (3). The analysis also ignores that Amazonian plant distributions are patchy (4), and the domesticates are native species that were naturally abundant in some areas without human intervention. Levis et al. consider all occurrences of trees identified as domesticates as human-induced. Also, many domesticated species in the study have evidence of cultivation only in the modern era or in geographic regions outside Amazonia, with little to no evidence for their pre-Columbian use in Amazonia.
Their results, furthermore, did not support their conclusions. Environmental conditions and unknown factors explained a significantly higher proportion of the variance in the dataset compared with the 'human' metric (c. 20%)(1, Fig. 5). Domesticates significantly decreased at distances over 20 km from archaeological sites (1, Figs. S7-S8, S10). When forests sitting atop archaeological sites were excluded from the analyses, no relationship existed between humans and domesticates for the entirety of Amazonia, and it remained significant in only in two of six regions (1, Table S3). Also, four of five domesticates had increased abundances only at distances over 1000 km from assumed domestication origins, contradicting expected patterns of long-term domestication and dispersal, and potentially reflecting modern population distributions (1, Figs. 1-2). No analysis was made comparing domesticates with modern settlement.
"Domestication shapes Amazonian forests" is an unsupported overreach. A more realistic conclusion is that pre-Columbian domestication may have influenced some forests, especially those nearest known human occupation sites (5).

1. Levis C, et al. (2017) Persistent effects of pre-Columbian plant domestication on Amazonian forest composition. Science 355(6328):925-931.
2. McMichael CN, Matthews-Bird F, Farfan-Rios W, & Feeley KJ (2017) Ancient human disturbances may be skewing our understanding of Amazonian forests. Proceedings of the National Academy of Sciences:201614577.
3. Junqueira A, Almekinders C, Stomph T-J, Clement C, & Struik P (2016) The role of Amazonian anthropogenic soils in shifting cultivation: learning from farmers' rationales. Ecology and Society 21(1).
4. ter Steege H, et al. (2013) Hyperdominance in the Amazonian Tree Flora. Science 342(6156):1243092.
5. Bush MB, et al. (2015) Anthropogenic influence on Amazonian forests in prehistory: An ecological perspective. Journal of Biogeography 42(12):2277-2288.

In their Research Article Levis et al (1) provide important new evidence to resolve the controversial debate as to the extent humans, specifically pre-Columbian Societies have influenced tropical forest across the Amazon Basin. Levis et al. demonstrate that domesticated forest tree species are five times more likely to be hyperdominant in forests on or around known archaeological sites. To substantiation their findings Levis et al highlight that "species domesticated in one particular environmental setting had wide geographical distributions and tended to be more abundant in locations not associated with their known or hypothetical origins of domestication". While we agree that a disassociation of domesticated species hyperdominance with putative origins of domestication is one potentially important signal, it is important to highlight that for many hyperdominant species multiple origins of domestication may well be the case. This has been documented in economically and ecologically important forest tree species such as the Brazil nut and cacao, for which there are at least two independent origins of domestication (2, 3).

Evidence suggests that two of the three traditional cacao cultivars in the Amazon (the Ecuadorian Nacional cacao and the Amelonado from eastern Brazil), are likely to have been domesticated from different source populations in Western and Central Amazonia, respectively (2, 4, 5). Levis et al (2017) only identify one origin of domestication in Amazonian Ecuador, and consider this the "known" origin as opposed to the "hypothetical" origins of all other species in their Figure 1. While this region is likely to be one of the first centers of domestication of Nacional cacao (>5000 BP) (6) from where it was introduced to coastal Ecuador through indigenous trade routes across the Andes (5) , other research suggests that domestication may have started from source populations in southern Peru, from where preselected genotypes were spread along northward dispersal routes (2).

For Brazil nut, Levis et al (1) present three hypothetical domestication origins, but do not seem to embrace the possibility of multiple origins . Our research suggests that domestication is likely to have started in at least two independent areas, one in southwestern Amazonia and one in central to northeastern Amazonia north of the Amazon River (3). The putative southeastern origin identified by Levis et al (1) actually refers to the origin of Brazil nut as a species (7, 8), which may or may not coincide with a third origin of domestication. What is clear is that more genetic data is needed for many of the hyperdominant tree species across the amazon to resolve the multiplicity of origins of domestication. This would be valuable for management of the forest genetic resources and restoration of resilient forest landscapes across the Amazon.

References and Notes:
1. C. Levis et al.,. Science. 355, 925–931 (2017).
2. E. Thomas et al., PLoS One. 7, e47676 (2012).
3. E. Thomas, C. Alcázar Caicedo, C. H. McMichael, R. Corvera, J. Loo, J. Biogeogr. 42, 1367–1382 (2015).
4. J. C. Motamayor et al., PLoS One. 3, e3311 (2008).
5. R. G. Loor Solorzano et al., PLoS One. 7 (2012).
6. http://whc.unesco.org/fr/listesindicatives/6091/.
7. personal communication Scott Mori cited in J. W. Clay, C. R. Clement, "Selected Species and Strategies to Enhance Income Generation from Amazonian Forests" (Rome, Italy, 1993).
8. Y. Y. Huang, S. A. Mori, L. M. Kelly, Phytotaxa. 203, 085–121 (2015).