Plant ecological genomics at the limits of life in the Atacama Desert

Significance

In the current changing climate, it is essential to improve crop production and resilience under dry and nutrient-poor conditions. Desert plants have naturally evolved to flourish under such conditions. Therefore, understanding the underlying mechanisms for their adaptation can potentially help to ensure food security. The Atacama Desert, the driest nonpolar place on Earth, offers a unique opportunity to explore plant adaptations to extreme environmental conditions. Here, we reveal how the adaptive strategies common or specific to the major plant lineages in the Atacama include enrichment of plant growth-promoting bacteria near their roots and positive selection of genes that are associated with key beneficial processes for plant survival. These strategies can potentially direct the molecular breeding or engineering of resilient crops.

Abstract

The Atacama Desert in Chile—hyperarid and with high–ultraviolet irradiance levels—is one of the harshest environments on Earth. Yet, dozens of species grow there, including Atacama-endemic plants. Herein, we establish the Talabre–Lejía transect (TLT) in the Atacama as an unparalleled natural laboratory to study plant adaptation to extreme environmental conditions. We characterized climate, soil, plant, and soil–microbe diversity at 22 sites (every 100 m of altitude) along the TLT over a 10-y period. We quantified drought, nutrient deficiencies, large diurnal temperature oscillations, and pH gradients that define three distinct vegetational belts along the altitudinal cline. We deep-sequenced transcriptomes of 32 dominant plant species spanning the major plant clades, and assessed soil microbes by metabarcoding sequencing. The top-expressed genes in the 32 Atacama species are enriched in stress responses, metabolism, and energy production. Moreover, their root-associated soils are enriched in growth-promoting bacteria, including nitrogen fixers. To identify genes associated with plant adaptation to harsh environments, we compared 32 Atacama species with the 32 closest sequenced species, comprising 70 taxa and 1,686,950 proteins. To perform phylogenomic reconstruction, we concatenated 15,972 ortholog groups into a supermatrix of 8,599,764 amino acids. Using two codon-based methods, we identified 265 candidate positively selected genes (PSGs) in the Atacama plants, 64% of which are located in Pfam domains, supporting their functional relevance. For 59/184 PSGs with an Arabidopsis ortholog, we uncovered functional evidence linking them to plant resilience. As some Atacama plants are closely related to staple crops, these candidate PSGs are a “genetic goldmine” to engineer crop resilience to face climate change.

Human societies are facing a global environmental crisis. Climate simulations show increased aridity in the subtropics from 2016 to 2035, relative to 1986 to 2005 (1), leading to increased global desertification from 48 to 65% of total land surface. Abiotic stresses such as drought, elevated radiation, salinity, and extreme temperatures all adversely affect crop production and pose a significant threat to global food security (2). Identifying the genes and molecular processes that enable plant resilience to extreme abiotic conditions is crucial for sustaining crop production in an era of accelerated climate change.

Most molecular knowledge of plant stress responses and tolerance has been gathered by traditional laboratory-based studies using a few model species. While beneficial, such molecular-genetic studies likely miss the ecological context in which species have evolved, revealing only a limited portion of a plant’s environmentally modulated genetic repertoire (3, 4). Furthermore, community interactions such as plant–microbe associations cannot be discerned in studies that isolate organisms from their natural habitats (3).

Studying entire ecosystems in their natural environments can facilitate identification of adaptive genes and processes common or specific to the major plant lineages. Species that face a common harsh environment may independently evolve similar mitigation strategies through parallel or convergent evolution. Thus, the goal of our study was to identify these common adaptive processes and genes—across or within the major plant lineages—that have evolved to enable them to thrive in marginal environments. Our focus on identifying shared adaptive strategies in the Atacama species fits with our long-term goal of transferring such knowledge to enable crops and biofuel species to thrive on marginal soils.

To uncover key genes and processes associated with adaptation to the Atacama Desert, the most arid nonpolar environment on Earth, we studied plants in the Atacama Desert, which stretches from the western Andean slopes to the coastal range near the Pacific Ocean, with a strong altitudinal influence on temperature and precipitation (5). The desert’s well-known hyperaridity is complex in origin and results from coupled ongoing ocean–atmosphere–orographic processes that likely originated in the Middle to Late Miocene, 10 to 15 million y ago (6). In addition to water limitation, living organisms in the Atacama cope with low-nutrient availability and extremely high radiation (7, 8).

To understand how plants survive in the Atacama, we sampled the climate, soil, plant, and soil–microbe diversity at 22 sites (located every 100 m of altitude) along the Talabre–Lejía transect (TLT) (Fig. 1). The TLT spans a pronounced elevational cline where macroscopic life exists between extremes of low rainfall and low temperatures (5, 7, 9). Plant communities along the TLT can be clearly separated into three vegetation belts: prepuna, a sparsely vegetated xeric belt (2,400 to 3,300 m above sea level; masl); puna shrubland (3,300 to 4,000 masl); and steppe, high Andean altitude (4,000 to 4,500 masl). We sequenced the transcriptomes of the 32 most abundant and ecologically relevant plant species as well as the DNA of their root zone–associated soil microbiomes. To identify the genes undergoing positive selection in the Atacama, we conducted phylogenomic analysis of the protein-coding sequences of 32 plants spanning the major plant clades in the Atacama to their taxonomically closest publicly available sequenced species from other worldwide locations. This strategy enabled us to identify shared genetic changes, as well as plant lineage-specific ones, associated with plant adaptations to the harsh environment of the Atacama Desert, with potential relevance to engineering crops or biofuel species to thrive in marginal environments.

Fig. 1.

Atacama Desert TLT study site and experimental strategy. (A) Regional context of northern Chile, location on the Salar de Atacama, and digital elevation model indicating the sampling sites in the TLT. Red arrowheads show meteorological stations. Maps were built using the Elevation Derivatives for National Applications (EDNA) elevation model and Landsat 8 satellite images (data: Environmental Systems Research Institute [ESRI], Scripps Institution of Oceanography, National Oceanic and Atmospheric Administration [NOAA], US Navy, National Geospatial-Intelligence Agency, and General Bathymetric Chart of the Oceans). (B) Sample collection strategy. Weather parameters were collected in meteorological stations at 3,060 and 4,090 masl. Soil samples were collected at each station from the uppermost 10 cm for soil chemical analysis and ∼100 g from 5 to 10 cm deep from bare soil and from root-zone soil for microbiome analysis. Plant tissue samples of 32 species were collected and snap-frozen on site. Plant species richness and soil coverage were recorded at the TLT sites to assess plant diversity. (C and D) Sample processing and analyses. Environmental characterization was constructed using meteorological data and soluble soil fraction and elemental analysis data. Total DNA was extracted from soil samples and the 16S ribosomal and nifH gene barcodes were amplified and sequenced for TLT soil microbiome characterization. Plant RNA was extracted from the frozen samples, libraries were prepared and sequenced, and transcriptomic and phylogenomic analyses were performed.

Results and Discussion

Extreme Living Conditions for Atacama Plants.

To characterize the precise environmental conditions along the TLT, we installed two meteorological stations that recorded hourly local climate parameters over a 3-y period (Fig. 1). One station was located at the prepuna–puna junction at 3,060 masl, and the other at the puna–steppe junction at 4,090 masl (Fig. 1A). Temperature and solar radiation levels peaked during January to February, coinciding with the plants’ growing season (SI Appendix, Fig. S1 A–D). Temperatures were consistently higher at the prepuna–puna than the puna–steppe (average difference at surface: 9.14 ± 1.08 °C; soil 10-cm depth: 7.73 ± 1.68 °C; soil 50-cm depth: 7.47 ± 1.00 °C). The maximum temperature was recorded in the prepuna–puna, at 10-cm soil depth (26 January 2019, 37.81 °C), while the minimum temperature was recorded on the surface of the puna–steppe (29 August 2016, −15.55 °C). Large diurnal temperature variations (∼14 °C) occurred at both elevations. The average daytime solar radiation levels, 621.1 ± 4.6 W/m² in the prepuna–puna and 590.27 ± 5.5 W/m² in the puna–steppe, are approximately three times higher than the global average of 188 W/m² (8), and the maximum radiation recorded was 1,276.9 W/m² at both elevations (SI Appendix, Fig. S1D). Overall, these data indicate that plants growing in the Atacama cope with high temperatures, both above- and belowground, extremely high solar radiation during the day, and chilling to freezing temperatures during the night, especially at higher elevations.

Average annual rainfall over the 3 y of our study was very low, with large interannual variations at both elevations (149.8 ± 45.0 and 121.9 ± 95.8 mm annual rainfall, respectively). Rainfall was concentrated in summer (January to February), with almost no rainfall recorded during the rest of the year (SI Appendix, Fig. S1E). It is noteworthy that most of the annual rain fell within a few days, where 50% of the annual precipitation was concentrated within 2 to 3 d in the prepuna–puna, and 3 to 5 d in the puna–steppe stations. Soil water content patterns reveal that the soil quickly reaches full capacity, and most of the water runs off as flash floods (SI Appendix, Fig. S1F). The soil also dries out faster in the prepuna–puna, imposing longer drought periods at lower elevations (SI Appendix, Fig. S1F). Overall, these results indicate that plants in the Atacama need to cope with prolonged droughts, while taking advantage of very short and unpredictable periods of increased water availability.

Soil properties and composition also have a predominant role in plant growth. The soil in the TLT was 82% sand on average and dominated by silicon and aluminum (SI Appendix, Fig. S2 A and B). Organic material content was very low and less than 1% in most of the transect, whereas salinity is high in the prepuna (SI Appendix, Fig. S2C). Soil pH measurements revealed a pronounced pH gradient along the transect, with strongly acidic soils at high elevations and markedly alkaline soils at the lower end of the transect (SI Appendix, Fig. S2C). Such pH dissimilarities throughout the transect can affect mineral nutrient availability. Indeed, we found an elevational gradient for most nutrients, where metals tend to accumulate in the acidic steppe, and salts tend to increase in the prepuna (SI Appendix, Fig. S2 D–F). Nitrogen (N), the quantitatively most essential macronutrient, was extremely low in all tested soil samples (SI Appendix, Fig. S2D). Total N content fluctuated between 8.51 and 22.56 g/m³ (5.32 and 14.1 mg/kg, respectively), which is about five times lower than the 106.5 g/m³ global mean N content for warm temperate deserts (10), and 1.4 times lower than the minimal agricultural reference range for this nutrient (11) (SI Appendix, Fig. S2D). We also observed a phosphorus (P) deficiency in prepuna soils and a gradual increase in P levels along the puna, reaching agricultural reference values only in the steppe (11) (SI Appendix, Fig. S2D). In contrast, potassium was found at high levels in the prepuna and decreased in availability, reaching ∼50% of reference value at the puna and steppe elevations (SI Appendix, Fig. S2 D–F). Iron was deficient at lower elevations, whereas zinc was deficient throughout the transect (SI Appendix, Fig. S2D). Taken together, we found soil differences that can greatly affect plant growth. N was low throughout the TLT and is a key determinant that together with climatic factors challenges plant survival in the Atacama.

Atacama Plant Diversity.

We surveyed plant species richness and coverage along the TLT for 9 consecutive years (2011 to 2019) (12), during the month of April, the month when floral tissues enabled species identification. We found that only 7% of the soil area was covered with plants (SI Appendix, Fig. S3). We identified and collected a total of 78 plant species throughout the transect. The prepuna showed a higher proportion of annual plants, while perennial grasses predominated in the steppe (Fig. 2A). The puna belt had the greatest species richness, with a total of 47 species with mixed life strategies (Fig. 2A). All plant species found along the transect, except for Ephedra americana, are angiosperms (Fig. 2B and SI Appendix, Fig. S4B). Some orders, like Poales (grasses) and Caryophyllales, were represented throughout the entire transect, while other orders were constrained to a specific vegetation belt, such as the Zygophyllales (prepuna) and the Apiales (steppe) (SI Appendix, Fig. S4B).

Fig. 2.

Atacama plant species in the TLT span 10 major plant clades and life forms. Species richness and distribution across the TLT were recorded for 9 consecutive years (2011 to 2019). (A) Number of species (species richness) and life strategies across different elevations. (B) The distribution of 32 selected plant species along the TLT elevations. Colors represent the number of years each species was observed at each location (persistence). Plant life form, photosynthesis type, and endemism status are defined for each species. Evolutionary relationships among species are depicted based on the phylogeny described in Fig. 4A. Dashed lines represent the limits between vegetation belts.

We selected a subset of the 32 most abundant and persistent plant species along the TLT that span 83% of the total area covered by plants (SI Appendix, Fig. S3) for molecular characterization (Fig. 2B and Dataset S1). They represent 14 plant families across 10 major taxonomic orders, including crop relatives, with different life-history strategies, C3/C4 photosynthesis, and endemic/nonendemic geographical distributions (Fig. 2B). These 32 selected Atacama species thus represent independent evolutionary trajectories exhibiting convergent evolution in response to the Atacama’s hostile growth environment that we seek to uncover.

Enrichment of Growth-Promoting Bacteria near Plant Roots.

Associations between plants and microorganisms in their environment are essential for survival. Several mechanisms by which microbes can act beneficially on plant growth have been described (13). To gain insight into the contribution of soil bacteria to plant survival in the Atacama Desert, we collected samples from the soil surrounding the plant’s root system (root zone), and from a nearby bare soil patch at the same depth, for 30 plant species (Fig. 3A). In order to have enough soil sample to analyze the microbiome, we collected the root-zone soil that includes the rhizosphere and has been shown to be a niche for bacterial interactions in Atacama soils (14). To quantify the composition of bacterial communities, 16S ribosomal gene sequencing was performed. We identified 10,143 bacterial operational taxonomic units (OTUs) from 28 phyla (SI Appendix, Fig. S5A). Pairwise comparisons of root zone with bare soil identified that 68% of OTUs, on average, are exclusive to one soil compartment (SI Appendix, Fig. S5B and Dataset S2). Specifically, 2,252 OTUs are found exclusively near plant roots, while 2,103 OTUs are found exclusively in bare soil (Fig. 3B). The five most dominant phyla in root-zone soil samples were Proteobacteria (45%), Bacteroidetes (16%), Actinobacteria (12%), Acidobacteria (11%), and Verrucomicrobia (3%). By contrast, the five most dominant phyla in bare soil samples were Proteobacteria (26%), Acidobacteria (22%), Actinobacteria (22%), Firmicutes (8%), and Planctomycetes (6%). We compared the OTU abundance between bacteria exclusive to root zone versus bare soil and found bacterial genera that are consistently (in at least 10 plant species) more abundant (≥2-fold) in plant root zones (Dataset S3). Interestingly, among those genera are Pseudomonas, Sphingomonas, Flavobacterium, Devosia, Kaistobacter, Mycoplana, Janthinobacterium, Variovorax, and Methylibium, which contain known plant growth-promoting bacteria that contribute to N fixation (15–20), protection against pathogens (21–23), drought resistance (24, 25), and plant hormone production (17, 26–28). These substantial differences in bacterial taxonomic composition between the two soil compartments potentiate plant selection or recruitment of beneficial bacteria.

Fig. 3.

Root-zone soils associated with Atacama plants in the TLT are enriched in plant growth-promoting bacteria and plant pathogens. (A) For microbiome analysis, soil was collected near the roots (root zone) of 30 plant species, or from close by bare soil for DNA analysis. Plant family classification and the vegetation belt from which soil samples were collected are indicated for each plant species. Evolutionary relationships among plant species are depicted based on the phylogeny described in Fig. 4A. (B) The bacterial taxonomic composition, based on 16S ribosomal RNA, of 2,252 root zone–exclusive and 2,103 bare soil–exclusive OTUs. (C) Predicted relative abundance of bacterial functions using FAPROTAX. Values represent the log2 ratio between root-zone and bare soil OTU relative abundances, for each plant species and a given function. Functions that consistently (in at least 10 plant species) differentiate (log–fold change ≥2 or ≤−2) between root zone and bare soils are shown. (D) Abundance of nitrogen-fixing bacteria was quantified using the nifH marker. Values represent the log2 ratio between root-zone and bare soil nifH abundances. Data for both 16S and nifH represent the average of three individuals per plant species.

Next, we used FAPROTAX to predict the potential ecosystem functions of the root zone– and bare soil–exclusive bacteria (Dataset S4). We found that 31% of the root zone–exclusive bacteria and 26% of the bare soil–exclusive bacteria are chemoheterotrophs that rely on organic matter, which is found to be extremely low throughout most of the TLT (SI Appendix, Fig. S2C). Functions that consistently (in at least 10 plant species) differ between root zone and bare soils suggest that plant pathogenic bacteria are associated more with plant roots, which may contribute to biotic stress (Fig. 3C). We also found associations with arsenate detoxification and hydrocarbon degradation near plant roots, which suggest Atacama plants recruit these pollutant-degrading bacteria to improve growth (29, 30). However, neither the root-zone nor bare soil samples showed concentrations beyond the normal range (31) (Dataset S5).

We also found enrichment of C- and N-cycling functions of bacteria associated with plant roots (Fig. 3C). We observed that bacterial N fixation and ureolysis (which each produce ammonium) increase near the root zone, while nitrification seems to be reduced. These results suggest that plants in the Atacama may prefer ammonium over nitrate, which may also explain low foliar δ¹⁵N values relative to soils observed in the prepuna (7). To validate the hypothesis that plants recruit N-fixing bacteria near their roots, we analyzed the abundance of the nitrogenase marker gene nifH in the root-zone versus bare soil samples (Fig. 3D). We found N-fixing bacteria exclusively in the root zone of 10 species, while overall, we found more than a twofold enrichment in N-fixing bacteria near plant roots, for 18 plant species. This result indicates that some Atacama plant species developed an active N-fixing bacteria recruitment/facilitation mechanism to optimize N acquisition under the extremely low N Atacama soils.

Top-Expressed Genes in Atacama Plants Are Involved in Stress, Energy Production, and N Metabolism.

To gain insights into the molecular mechanisms underlying the successful adaptation of plants to the Atacama Desert, we deep-sequenced the transcriptomes of the 32 most predominant/ecologically relevant plant species from field-collected samples (SI Appendix, Fig. S6 and Dataset S6). These plant species were collected from 15 sites along the TLT. For each plant species, a pooled sample, composed of all available tissue types (including shoots and roots) at time of collection, was used for RNA extractions (32). To identify genes and processes associated with their response to the environment, we identified the 10% most expressed transcripts under field conditions from each of the 32 Atacama species and conducted a Gene Ontology (GO) overrepresentation analysis. The number of top-expressed transcripts per plant varied between 1,374 and 3,752, of which ∼76% had GO annotations. Among the overrepresented GO terms found in most species across the transect are processes involved in primary metabolism and its regulation, including biosynthesis of lipids and N-, P-, and S-containing metabolic compounds, such as amino acids, chlorophyll, and adenosine triphosphate (SI Appendix, Fig. S7 and Dataset S7). In addition, photosynthesis and cytokinin response are commonly overrepresented functions, suggesting that Atacama plants are invested in growth. In addition, various stress-related processes were overrepresented in the top-expressed genes, including response to oxidative, salt, and endoplasmic reticulum stresses, DNA damage, abscisic acid (ABA), as well as heat and cold acclimation. Interestingly, response to salt was overrepresented in all the prepuna plants, coinciding with the high salinity at this elevation (SI Appendix, Fig. S7).

Because the top-expressed genes among different Atacama species pertain to similar biological processes, we next tested whether the same genes are orthologous across species (using the orthology assignment described in the phylogenomic section below). We found that 50% of the total top-expressed genes have at least one ortholog among other Atacama species, and that closely related species share most of their top-expressed genes (SI Appendix, Fig. S8). To identify genes whose high-level expression is Atacama-specific, we compared the number of top-expressed genes in each orthologous group between the 32 Atacama species and their respective sister species (17 for which we had raw sequencing data) using the Fisher exact test (Dataset S8). This analysis uncovered 120 ortholog groups that were more frequently found in the top-expressed genes of the Atacama species, compared with their sister species. Two of these top-expressed ortholog groups specific to the Atacama plants are also positively selected genes (PSGs), as described in the section below. One such top-expressed PSG is specific to the Poaceae and another is shared among several Atacama plant lineages—the Fabaceae, Solanaceae, Verbenaceae, and Apiaceae (Dataset S9). A Sungear plot shows the overlap of these 120 ortholog groups top-expressed in the Atacama plant families, which reveals that 14 top-expressed ortholog groups are shared between all families, while 48 are specific to the Poaceae, the most dominant family in the Atacama (SI Appendix, Fig. S9). These results suggest that high-level expression of genes common across Atacama plant lineages, as well as plant family-specific genes (e.g., in the Poaceae), may play a role in adaptation of these species to the stressful environment in the Atacama.

Phylogenomic Reconstruction Uncovers 265 Candidate Genes under Positive Selection in Atacama Plant Lineages.

To uncover the molecular basis of adaptation to the environmental conditions in the 32 most abundant and persistent species in the Atacama Desert spanning 10 major plant lineages, we performed evolutionary analysis. First, we identified a set of 32 species that are a phylogenetic relative to these Atacama plants, using publicly available transcriptome data. This set comprises sister species and nearest relatives (Datasets S1 and S6). These selected species are from diverse worldwide locations, and occupy hydric, mesic, or xeric habitats, and therefore differ in drought tolerance capabilities. The predicted proteomes of 32 Atacama plants (786,237 sequences), 32 sister/nearest relatives (685,914 sequences), and 6 fully sequenced model plant species, including Arabidopsis thaliana and Zea mays (214,799 sequences)—comprising 70 taxa and 1,686,950 proteins—were analyzed using a phylogenomic pipeline (33), optimized for this large-scale analysis (Materials and Methods). This phylogenomic pipeline identified 15,972 ortholog groups, each represented by a sequence in at least 10 taxa. Protein alignments for these ortholog groups were concatenated into a supermatrix of 8,599,764 amino acid characters, used for phylogenomic reconstruction. The evolutionary relationships depicted by the maximum likelihood (ML) total evidence (TE) tree successfully separate all 14 plant families which include Atacama Desert plants (Fig. 4A) and are congruent with the taxonomic classification defined by the Angiosperm Phylogeny Group (APG IV) (34).

Fig. 4.

Phylogenomic analysis uncovered PSGs common to Atacama plants and/or specific lineages. (A) An ML TE tree inferred for the 32 most abundant Atacama species, 32 sister species from other worldwide locations, and 6 fully sequenced model plants, based on a supermatrix (8,599,764 amino acid characters) derived from 15,972 ortholog groups. Red circles, Atacama plants; white circles, sister or model plants. Model plants are marked with an asterisk. Species background colors correspond to plant family membership. The CodeML branch-site model was used for positive selection detection, under two complementary scenarios: (A, Left) Species-independent adaptation. Tested branches are marked with red stars. (A, Right) Ancestral adaptation, where ancestral state reconstruction was first applied, to predict possible ancestral adaptation origins (ancestral probabilities are depicted for each internal node as red pie charts). Tested branches are labeled with blue stars. (B) Tree-independent positive selection analysis using SNPGenie. Four between-group comparisons were performed. Species were grouped based on their drought and low N–tolerance capacity (Dataset S8), while 10-codon SWs were scanned for positive selection. (C) The light blue circles with numbers represent the number of PSGs found for each MRCA, based on the presence of the sequence in each plant family. (D) Sungear plots for the number of PSGs found (based on sequence presence) in each vegetational belt or plant life form type (vertices) and their overlap (vessels within triangles pointed with arrows). (E) Number of PSGs with Pfam domains, and whether positively selected dN/dS sites are within a domain.

We further compared orthologous sequences from Atacama and sister/nearest relative species (Datasets S1 and S13) to identify genetic modifications associated with adaptation to the extremely dry and N-poor Atacama Desert conditions. We used two codon-based nonsynonymous and synonymous substitution rate ratio (dN/dS) analysis methods to identify candidate PSGs: 1) the branch-site model in CodeML dN/dS analysis (Fig. 4A) and 2) between-group dN/dS comparisons in SNPGenie (Fig. 4B) (Materials and Methods). CodeML tests for episodic dN/dS–positive selection in a collection of lineages leading to Atacama species (Fig. 4A). CodeML also predicts potential positively selected sites using the Bayes empirical Bayes (BEB) method. Adaptation to Atacama Desert conditions might have occurred through independent events in each plant species. However, in some instances, such as the case of Baccharis and Adesmia (Fig. 4A), genetic changes might have taken place in the ancestor of these plants allowing for their preadaptation. We thus performed two complementary CodeML analyses to account for both scenarios: one that tests only the branches leading to the Atacama terminals (“independent adaptation”), and another that tests ancestral state reconstruction to take into account “ancestral adaptation” events (Fig. 4A). Next, in a complementary tree-independent approach, we used the SNPGenie “counting” method to identify positive selection in candidate genes between distinct groups of sequences, using 10-codon sliding windows (SWs) (SI Appendix, Fig. S10). To do this, our groupings compared drought-tolerant (Atacama + xeric sisters) against drought-sensitive (hydric + mesic sisters) species, as well as the xeric Atacama species adapted to extremely low N with nonadapted xeric sister/nearest relative species (Fig. 4B and Dataset S8).

The tree-based CodeML dN/dS analysis predicted 152 (1,359 BEB sites) and 117 (670 BEB sites) ortholog groups to be under positive selection pressure in the Atacama independent or ancestral origin scenarios, respectively (Fig. 4A). The non–tree-based SNPGenie approach predicted 85 ortholog groups to be under positive selection in the identified groupings (476 SWs) (Fig. 4B). Most of the PSGs were identified when Atacama species were compared with all sister/nearest relative species (77 PSGs), or only against the drought-tolerant–related species (15 PSGs) (Fig. 4B), suggesting that these differences are specific to the evolutionary adaptation of the Atacama species. In total, we predicted 265 PSG candidates that could underlie plant adaptation to Atacama Desert conditions (SI Appendix, Fig. S11 and Dataset S10).

We next aimed to determine if the PSGs were common across the Atacama species or specific to a certain plant family. To this end, we explored plant species composition (based on the presence of the sequence) for the identified 265 PSGs. For each PSG, we identified the most recent common ancestor (MRCA) on the phylogenomic tree (Fig. 4C). We found that 109 of the PSGs are shared by diverse plant lineages (where their MRCA is closer to the root), while 156 are specific to a particular plant lineage (Fig. 4C). For the most abundant plant families in the Atacama transect, we detected 74 Poaceae-specific and 21 Asteraceae-specific PSGs. Further comparative analysis revealed that most of the PSGs are common adaptations in the Atacama shared by species from all three vegetational belts and life forms (Fig. 4D).

To evaluate whether the positively selected sites impact protein function, we tested their occurrence within functional (Pfam) domains. We identified Pfam domains in 177/265 PSGs, and discovered that 113 (64%) contained at least one positively selected BEB site or SW (Fig. 4E and SI Appendix, Fig. S12). This result indicates that the adaptive changes in the Atacama species most likely affect their protein structure and/or function.

Adaptive Genes in Atacama Plant Species Are Functionally Linked to Selective Pressures of Their Environment.

We next used GO term annotation to characterize the potential functions of these PSGs in Atacama plant species (Fig. 5 and Dataset S11). Some of the biological processes predicted to be under positive selection in the Atacama species can be linked to the selective pressures of this environment. For example, genes involved in response to light stimulus, reactive oxygen species and DNA-damage stimulus, photosynthetic light reaction, and plastid organization can be related to the adaptation to the extreme high-light radiation in the Atacama. Similarly, genes involved in the regulation of stress response, salt stress, detoxification, and metal ions, as well as in energy, N, P, and S metabolism, could be related to the adaptation of these plants to their stressful, nutrient-poor environment.

Fig. 5.

PSGs in Atacama plants are involved in biological processes essential for plant development and survival. A GO term biological process network, for terms that are annotated in at least two PSGs, was constructed using Cytoscape. For simplicity, only level 5 GO terms (more informative) and their level 2 ancestral GO terms (more general) are presented. Node colors represent the number of PSGs annotated for each level 5 GO term. The percentage of PSGs containing positively selected sites within their Pfam domains is indicated for each GO term. Overrepresented GO terms (Fisher’s exact test, P ≤ 0.05) are labeled with red stars.

To further understand how these PSGs may have contributed to plant species adaptation in the Atacama, we reviewed the literature for functional characterization of these PSGs based on their closest ortholog from the model plant Arabidopsis, where most experimental information for gene function exists. We identified 184/265 PSGs with a closely related Arabidopsis sequence (SI Appendix, Fig. S12 and Dataset S11). We found functional evidence in Arabidopsis for 59/184 PSGs linking them to key physiological and molecular processes that can enhance plant resilience under extreme environmental conditions (Fig. 6). Specifically, candidate genes which show positive selection in Atacama species have been shown in Arabidopsis to play a role in high radiation and temperature stress tolerance via radical reactive oxygen scavenging and DNA repair [e.g., APX2 (35) and PRXIIE (36)] and in chloroplast development and movement, and for the integrity of the photosynthetic apparatus [e.g., CP24 (37) and MDA1 (38)]; regulate floral development and flowering time [e.g., AP2 (39) and SOC1 (40)], which are highly selective, given the short growth period (41, 42); function in defense response and pathogen resistance [e.g., HIR3 (43)]; regulate root-hair formation [e.g., LRX1 (44)], and therefore critical for water and nutrient uptake; maintain turgor pressure and water capture [e.g., DTX35 (45)]; and play a central role in nutrient sensing, uptake, assimilation, and metabolism [e.g., PAP26 (46)]. Many of these genes have confirmed stress tolerance or sensitivity when they were mutated or overexpressed in Arabidopsis (more information can be found in Dataset S12), and many of them are highly expressed in multiple Atacama species (Fig. 6 and Dataset S10). We also identified positively selected sites within the functional domains of these 34/59 genes, indicating a potential impact on their function (Fig. 6). These 59 characterized genes, representing 22% of the 265 identified PSGs, revealed the key traits both below- and aboveground under a strong selective pressure in the Atacama natural setting.

Fig. 6.

Genes associated with adaptive strategies for plant survival in the Atacama Desert environment. Both above- (A–C) and belowground (D–G) strategies for plants to cope with the Atacama harsh conditions are proposed, based on experimental evidence in Arabidopsis for orthologs or closest paralogs of PSGs identified in Atacama species (see SI Appendix, Fig. S13 and Dataset S12 for full information). PSGs (gray ellipses) are tagged with additional information, including evidence for stress tolerance, top expression in multiple Atacama species (within the 10% most expressed genes), and existence of positively selected sites within a Pfam domain. ETI, effector-triggered immunity; LD, long day; N-aA, nitrogen- and amino acid–related; PTI, pathogen-associated molecular pattern–triggered immunity; SD, short-day; SNARE, soluble NSF attachment protein receptor; UV-B, ultraviolet B.

Overall, this work has established an unparalleled natural laboratory study in the TLT in Chile, which is unique from other studied arid transects in China and the United States (47, 48), as both high altitude and extreme aridity converge in the Atacama. Further, our study is unique in its goal to uncover the genetic basis of plant adaptation to these extreme environmental conditions. Using a high-level ecosystem and phylogenomics overview, we were able to identify the common and lineage-specific adaptive strategies most relevant for plant resilience to one of the harshest environments in the world. Some of these extraordinarily resilient plants are closely related to staple crops, such as cereals, legumes, and the potato family, and therefore can provide invaluable genetic material for crop breeding (49). For example, Solanum chilense is phylogenetically close to Solanum lycopersicum, and possible use of these genes in engineering such agronomically important species may not be far off. Additionally, S. chilense—which produces small fruits in the desert—has the potential to be domesticated and studied as a food to grow in marginal soils. In addition to food crops, domestication of biofuel crops may also be possible, using some of the Poaceae species that are very abundant in the Atacama. Thus, the uncovered candidate genes and sites, underlying the successful adaptation of these plants to their environment, can be targets to engineer crop resilience to face climate change. Such new crops would be key to sustain agricultural productivity in a scenario of increased desertification of our planet.

Materials and Methods

Further details are provided in SI Appendix, Materials and Methods.

Measurement of Weather Parameters.

During the 2016 and 2017 field seasons, meteorological stations with Onset data loggers (HOBO U30/NRC) were installed in two monitoring sites at 3,060 and 4,090 masl. Soil temperature sensors (S-TMB-M006) were placed at depths of 10 and 50 cm. A tipping bucket rain gauge (S-RGB-M002) and air temperature/relative humidity (S-THB-M002) and insolation sensors (S-LIB-M003) were arranged 2 m above the ground. Data were recorded at 1-h intervals by the U30/NRC units.

Sample Collection.

Samples of 32 transect plant species were collected in situ, representing the most dominant species (83% of relative coverage) and covering 14 out of 21 (66.7%) plant families. All tissue types and developmental stages present were sampled for each species and pools of at least three individuals were stored after being snap-frozen in liquid nitrogen immediately after collection in the field. Frozen samples were transported in dry ice to the laboratory for RNA extraction. Soil samples were collected at each station by pooling samples from the uppermost soil layer (10 cm deep) from 10 different 30 × 30-cm squares per station. For bacteriome analysis, ∼100 g of soil was collected at 5- to 10-cm depth, from the root zone (soil incorporated into the fine fibrous root system) of three individuals per plant species, and from the bare soil (1 m away from selected individual plants).

Soil DNA Collection, Extraction, and Sequencing.

Soil samples for microbiome analysis were collected mostly in April 2012. For most of the species, the soil samples were collected from the same site where the plant tissues were collected. In a few cases, we could not find the plant on the same site. Thus, in those cases, we collected the soil samples from the nearest site where we could identify the species. Soil DNA was extracted from each soil subsample with cetyltrimethylammonium bromide–based methods (50, 51) combined with a Qiagen DNeasy Blood & Tissue Kit. DNA integrity was assessed by electrophoresis with a 2200 TapeStation (Agilent Technologies). Sequencing was performed at the Molecular Research DNA laboratory (Shallowater, TX) on an Illumina MiSeq platform in an overlapping 2 × 300-bp configuration, with a minimum throughput of 20,000 reads per sample.

Plant RNA Extraction and Sequencing.

RNA was extracted from the plant tissues with the PureLink RNA Mini Kit (Ambion; 12183-018A). Twenty-six Atacama species were sequenced by the Servicio de Secuenciación y Tecnologías Omicas (Sequencing and Omics Technology Service) at the Facultad de Ciencias Biológicas, Pontificia Universidad Católica de Chile. Library preparation for these samples was performed using the TruSeq RNA Sample Preparation Kit (Illumina; RS-122-2102), using at least 0.5 µg of total RNA, according to the manufacturer’s instructions. Sequencing was performed using an Illumina HiSeq 2500 instrument, generating 125-bp paired-end reads. Sequencing for six species (Senecio puchii, Chuquiraga atacamensis, Ambrosia artemisioides, Adesmia erinacea, and Exodeconus integrifolius) was performed at either the Eastern Sequence and Informatics Hub, University of Cambridge (Lupinus subinflatus) or at Macrogen, using an Illumina HiSeq 2000, generating 100-bp paired ends.

De Novo Transcriptome Assembly.

De novo transcriptome assembly was performed with Bridger using non–strand-specific paired-end sequences, with default k-mer size and considering a minimum k-mer coverage of five reads (52). The BUSCO v.2.0 (53) tool, with the embryophyta_odb10 1,375-single-copy gene set, was used to assess the completeness of gene coverage per sequenced species. Assembled transcripts were translated into proteins using the Transdecoder tool, integrated within Trinity software (54). The minimum protein size allowed was 100 amino acids. Pfam domains were identified for the final protein set using the “hmmsearch” search function in HMMER v.3.1b2 (http://hmmer.org/) against the HMM-based Pfam-A entries (55). Identified Pfam domain matches were filtered based on significance (i-Evalue 1e-5), and based on overlap, keeping domains with 40% or less overlap.

Phylogenomics Reconstruction and Orthology Assignment.

To study the evolution of the 32 Atacama plant species, we identified the 32 species which are the most closely related (“sisters” and nearest relatives) to the Atacama plants that have transcriptome sequences available in public databases (a full description is in SI Appendix, Materials and Methods). Publicly available sets of predicted protein sequences from six model species were also included (A. thaliana, Brachypodium distachyon, Glycine max, Physcomitrella patents, S. lycopersicum, and Z. mays). Phylogenomic analysis of the above 70 taxa was conducted using an optimized version of the phylogenomic pipeline previously described (33), where code optimization, parallelization, and checkpointing were necessary for completing the ortholog inference step of the huge input data size (1,686,950 protein sequences, 8,599,764 amino acid characters). The optimized code is available on GitHub (https://github.com/coruzzilab/PhyloGeneious). Protein alignments of ortholog groups with a sequence representation of at least 10 taxa (15,972 ortholog groups) were concatenated into a supermatrix. ML phylogeny reconstruction was performed with RAxML, using a partition model, where each ortholog group was modeled using the JTT protein substitution model and the GAMMAGTR model of heterogeneity (56).

PSG Analysis.

Genome-scale positive selection analysis was performed with the CodeML program from the PAML v.4.8 package (57), as an ML-based approach for tree-based dN/dS estimations. The branch-site model was used in order to detect episodic positive selection at branches along the phylogenetic tree that lead to Atacama Desert adaptation. Two complementary CodeML analyses were performed. 1) Species-independent adaptation: All the branches that led directly to the Atacama tips (32 branches) were grouped as foreground, and compared with all the other branches in the tree as background. 2) Ancestral adaptation: The ancestral state reconstruction program BayesTraits v.3 (58) was used, with a binary trait (“Atacama,” “non-Atacama” states) to detect the origin of adaptation to the Atacama in the ancestral branches. The QVALUE R package (59) was further used for multiple testing correction with a false discovery rate–corrected P value cutoff of 0.05. The BEB test implemented in CodeML was used to infer sites under positive selection (posterior probability > 0.95) (60).

A complementary approach, consisting of a between-group dN/dS analysis, was performed using the Nei–Gojobori method as implemented by the SNPGenie algorithm (https://github.com/chasewnelson/snpgenie) (61, 62). To identify genes underlying drought and low-N tolerance, four between-group analyses were conducted for each ortholog group. To identify genes under positive selection in drought-tolerant species, we compared the codon-aligned sequences from Atacama species (drought- and low N–tolerant) and sister/nearest relative drought-tolerant species with the drought-sensitive species. We also compared the drought-tolerant and drought-sensitive sister/nearest relative species (to exclude the N effect). Similarly, to identify positive selection in low N–tolerant species, we compared the Atacama sequences (low N–tolerant) with all sister/nearest relative sequences, and with only the drought-tolerant sister/nearest relative species (to exclude the drought effect). We used an SW of 10 codons to scan for positive selection. Windows were considered under positive selection if they met the following criteria: Both groups had a minimum of six defined (nongap) codons at all positions; and dN of the window was greater than the overall ortholog group dS, and the Z score of dN − dS was >1.96 (1,000 bootstrap replicates). No correction for multiple hits was used, making the test for dN > dS conservative. Ortholog groups containing positive selected windows in at least one of the four between-group analyses were further considered.

GO Overrepresentation Analysis.

For Atacama and sister/nearest relative plants (nonmodel species), InterProScan v.5.39-77.0 (63) was used for GO prediction. GO annotations of four model species—A. thaliana, B. distachyon, S. lycopersicum, and Z. mays—were obtained from the Plant Regulation Data and Analysis Platform database (http://plantregmap.gao-lab.org/download.php#go-annotation). For a more comprehensive annotation of the nonmodel species, we used the ortholog group assignments identified by the phylogenomic pipeline, and reassigned GO annotations to each ortholog group member, based on the annotations of all ortholog group members, including the model plant genes. GO overrepresentation analysis was performed using a custom Python script (https://github.com/GilEshel/OverRep).

Data Availability

The Atacama plant species barcode tRNA-Leu (trnL) gene intron has been deposited in GenBank (accession nos. MH115328–MH115388). Raw reads for all Atacama plant species discussed in this publication are available through the National Center for Biotechnology Information Sequence Read Archive (NCBI SRA) (BioProject accession no. PRJNA687835). All bacterial sequence data are also available in the NCBI SRA (BioProject accession no. PRJNA358231). The code for the optimized phylogenomic pipeline is available on GitHub (https://github.com/coruzzilab/PhyloGeneious). The Python script used for testing GO overrepresentation is available on GitHub (https://github.com/GilEshel/OverRep).