Genomic Analysis of Differentiation between Soil Types Reveals Candidate Genes for Local Adaptation in Arabidopsis lyrata Authors: Thomas L. Turner, Eric J. von Wettberg, Sergey V. Nuzhdin Published: 2008, PLOS ONE DOI: 10.1371/journal.pone.0003183 ABSTRACT Serpentine soil, which is naturally high in heavy metal content and has low calcium to magnesium ratios, comprises a difficult environment for most plants. An impressive number of species are endemic to serpentine, and a wide range of non-endemic plant taxa have been shown to be locally adapted to these soils. Locating genomic polymorphisms which are differentiated between serpentine and non-serpentine populations would provide candidate loci for serpentine adaptation. We have used the Arabidopsis thaliana tiling array, which has 2.85 million probes throughout the genome, to measure genetic differentiation between populations of Arabidopsis lyrata growing on granitic soils and those growing on serpentinic soils. The significant overrepresentation of genes involved in ion transport and other functions provides a starting point for investigating the molecular basis of adaptation to soil ion content, water retention, and other ecologically and economically important variables. One gene in particular, calcium-exchanger 7 , appears to be an excellent candidate gene for adaptation to low Ca∶Mg ratio in A. lyrata . INTRODUCTION Serpentine soil mosaics are a classic context for ecological adaptation [1] , [2] . These widespread soils occur in small patches along fault lines where igneous rocks such as serpentinite are exposed. This environment is characterized by a suite of challenging abiotic factors such as low calcium-to-magnesium ratios, increased heavy metal concentrations, nutrient deficiency, and low moisture retention [3] , leading to sharp transitions in abiotic conditions at the boundaries of serpentine patches [4] . The ecological community is differentiated along these boundaries as well: serpentine soils support an considerable array of endemic plant species, with serpentine specialists comprising 12.5% of native California plants, despite these soils comprising only 1% of the land area in the region [5] . Some widespread species are also found on serpentine, with several documented cases of local adaptation to soil conditions (e.g. Collinsia [6] , Cerastium [7] , and Pinus [8] ). Populations of Arabidopsis lyrata , a perennial self incompatible crucifer, grow on serpentine outcroppings, interspersed with populations on other soil types such as granitic outcroppings and sand dunes [9] . Though information regarding local adaptation is lacking for A. lyrata , its proximity to the genetic model organism A. thaliana provides an opportunity to locate polymorphisms which are associated with the serpentine soil habitat. The role of these polymorphisms in serpentine adaptation, if any, can then be experimentally investigated. By hybridizing genomic DNA from A. lyrata to A. thaliana Affymetrix tiling arrays, we can measure genetic differentiation between soil types at 2,853,369 probes throughout the A. lyrata genome. When DNA is hybridized to the array, probes which overlap a polymorphic SNP or indel will hybridize poorly in individuals with the mismatched allele [10] – [14] ; when hybridization intensity is significantly different between populations, a differentiated polymorphism can be mapped to a specific location in the genome [15] – [17] . This method facilitates the discovery of adaptive variation in several complementary ways. First, we can discover candidate genes for adaptation to important environmental conditions, such as the low Ca∶Mg ratio which is characteristic of serpentine soil. In A. thaliana (which cannot normally survive on serpentine soils), an induced loss of function mutation in the calcium-proton antiporter cax1 enhances survival on soils with a low Ca∶Mg ratio [18] : differentiation of natural variation at related genes in A. lyrata would provide clear candidate loci for adaptation to low Ca∶Mg ratio. Second, we can use the natural distribution of genetic variation between environments to investigate the functions of unannotated genes and non-coding features. If an uncharacterized locus consistently assorts with soil ion content, for example, then it can be hypothesized to interact with this environmental variable to determine fitness. Third, we can use genetic differentiation at genes with known function to form hypotheses about other environmental differences which may be important in nature. Although abiotic factors are thought to be the most important drivers of serpentine adaptation, if differentiation is also found in genes coding for anti-microbial compounds, anti-predatory compounds, or proteins which mediate competitive interactions, then parasitism, predation, or competition can be inferred to be important. In the current work, we have hybridized genomic DNA from two serpentine and two granitic populations of A. lyrata to the A. thaliana tiling array. This has allowed us to locate many polymorphisms which are differentiated between soil types, including excellent candidates for adaptation to soil conditions. We do not mean to imply that selection is the only force which leads to differentiation between populations, as stochastic demographic forces may also lead to correlated distributions of genetic polymorphisms and environmental conditions, especially in this small sample of populations [19] – [21] . Indeed, there has long been a debate as to the relative importance of selection and demography in creating population differentiation [19] , [22] , [23] . This debate has only intensified in the genomic era, as it is now clear that selection can be rampant, and lead to adaptive change at loci across the genome [24] – [26] . Our paper is not intended to resolve this debate–rather, we provide a genomic portrait of differentiation which is necessary but not sufficient to determine which polymorphisms are influenced by spatially varying selection. This data can then inform functional investigation of candidate genes and processes. RESULTS AND DISCUSSION To map differentiated polymorphisms between serpentine and granitic A. lyrata localities, DNA from three individuals from each of the four localities ( figure 1 ) was fragmented, labelled, and hybridized to an A. thaliana tiling array (12 total arrays). On the array, each perfect match (PM) probe that matches the genome is adjacent to a mismatch (MM) probe, which has a mismatched base at the middle base pair. Comparison of normalized Σ(PM i ) and Σ(MM i ) intensities over the i chips for each probe indicates that the probes matching the draft A lyrata genome are the most sensitive markers of DNA differentiation, as expected ( figure 2 ). For these 371,642 probes, the Σ(PM i )−Σ(MM i ) is positive for 98.8% of probes. For the other 2,481,727 probes on the array, the Σ(PM i )−Σ(MM i ) is positive at 65% of probes. This indicates that probes matching the draft genome are the most sensitive markers for detecting DNA differentiation, but also that the 2.48 million other unique probes contain useful information. Probes with a perfect match in the draft A. lyrata genome have the highest mean intensity (red), and the corresponding mismatches for these probes have the next highest mean (orange). For probes without a match in the draft genome, intensities are still higher at perfect match (grey) than mismatch (purple) probes. Though some of these probes might have a matching DNA sequence in our populations, but not in the reference genome, many of these probes are probably comparisons between a single mismatch and a double mismatch, yet are still providing valuable information. Note that extreme values are not shown for both axes. To determine which probes overlapped differentiated polymorphisms, we computed a t -test p -value for between soil types at each array probe; the distribution of p -values for probes with a perfect match in the draft A. lyrata genome is shown in figure 3 . There is a considerable excess of probes with low p -values: 2402 probes have p <0.001, whereas only 371 are expected by chance. For comparison, t -tests were computed between the other two possible combinations of our four collections. Dividing the samples along these additional axes revealed a small excess of probes with low p -values in both cases ( figure 3 ), but substantially fewer than in the serpentine vs. granitic comparison, indicating that most differentiation between these samples assorts with soil type. Three combinations of our four populations were tested: test 1 is the soil type comparison (blue line) = serp1+serp2 vs. granite1+granite2. For comparison, we also computed the other two combinations of the four localities: test 2 (red line) = serp1+granite1 vs. serp2+granite2; test 3 (green line) = serp2+granite1 vs. serp1+granite2). After correcting for multiple tests, 39 probes are significantly differentiated between soil types at a bonferroni p <0.05 (for this and all further analyses, the full 2.48 million probes were used; see Materials and Methods for details). Using a less conservative false discovery (FDR) rate criterion, 168 probes are significant at a FDR<0.01, and 751 probes are significant at a FDR<0.05. The 39 bonferroni significant probes overlap 33 genes, including two genes with multiple significant probes. When the FDR<0.01 probes are included, 112 genes overlap at least one significant probe; adding the FDR<0.05 set of probes increases this number to 455 genes including 81 genes with 2–12 significant probes each ( Data Set S1 ). To locate additional differentiated genes, we used a sliding window analysis to determine if probes with low, but not individually significant FDR are clustered in the genome. Using permuted data to generate a null distribution, we mapped windows of 5, 10, and 20 probes that are significantly enriched for probes with FDR<0.30. Seventy-one small genomic regions were significantly differentiated between serpentine and granitic samples (windows with a permutation-based FDR<0.001 were considered significant, see methods ). These regions are mostly a few kb in size (mean = 6.6 kb), with 11 regions larger than 10 kb and 3 larger than 30 kb. These 71 regions overlap 184 total genes, including 52 genes already significant from the analysis of individual probes. The combined list of significantly differentiated loci (probes with FDR<0.05 and windows with FDR<0.001) includes 586 genes, 545 of which are annotated with at least one function in A. thaliana (based on gene ontology assignments, hereafter referred to as GO terms). When the probes with 0.01FDR>0.01, which is expected to include some false discoveries. Including these probes also increases the sensitivity of the analysis, however: the significant probe in cax7 had a FDR = 0.011, and would have also been excluded had we only used a threshold FDR of 0.01. Despite reduced power when probes with 0.010.30 were considered to have an FDR = 1.00, as this was previously found to make the analysis more conservative [17] . Because FDR estimates for windows were found to be less accurate than FDR estimates for individual probes [17] , we considered windows significant only if they had FDR<0.001 when compared to the permuted chromosomes. DIFFERENTIATION OF FUNCTIONAL CATEGORIES Using the TAIR7 annotation of the A. lyrata reference genome, we determined which genes overlapped our differentiated probes and windows; a gene was considered to overlap a differentiated locus if any portion of its transcript overlaps a significant probe or window. Significance of gene ontology (GO) terms among our significant genes was determined in two complementary ways: a binomial sampling test, and a permutation test (GO terms are a controlled vocabulary which classifies genes based on all known functional associations determined computationally or experimentally). First, we compared the observed proportion of significant genes in each GO term to an expected number, which is the binomial sampling probability of sampling an equal or larger number of genes in each category given the number of significant genes in the genome. For the permutation test, we simulated our discovery procedure by randomly sampling probes and genomic regions of the same size and number as our significant probes and regions. For example, in the primary analysis presented below, there are 751 significant probes and 71 significant regions. To generate a null distribution of genes for this analysis, we sampled this same number of probes and regions (of an equal size distribution), and determined which genes overlap the random sample. We considered a GO term to be significantly overrepresented if the observed sample has more genes in the given term than 5% of 500 simulated data sets, and less than 5% probability of occurring by chance in the binomial sampling test. COPY NUMBER VARIANTS If serpentine and granitic populations are equidistant to the reference genome, then any given differentiated probe should have higher intensity on serpentine half the time, and higher intensity on granite half the time. In some significant regions, however, all probes are more intense on one soil type, consistent with differentiated copy number variants (CNVs). Differentiated regions are considered candidate CNVs if the ratio of mean hybridization intensity for that region is more extreme than expected. We extracted the raw (unnormalized) intensity values for all probes in each significant region, and computed a mean intensity for serpentine and granitic populations. These means were minimally normalized by dividing by the median intensity for all probes for each soil type, and the ratio of the normalized means was used as the test metric. An expected distribution of ratios was created by randomly sampling the number of probes in the test region from the set of probes from all significant regions. Regions were considered candidate CNVs if their ratio was more extreme than 9,990/10,000 random sets ( p <0.001). This permutation controls for many possible confounding effects. For example, if all derived SNPs were found on serpentine, then granitic populations would have higher average intensities in significant regions; because our permutation samples only from significant regions, the candidate CNV are significantly beyond any such effect. SEQUENCING AND PCR Three loci were amplified via long PCR using Phusion polymerase (Finnzymes), and two were directly sequenced. Population genetic analysis of sequenced loci was done in DNAsp [40] ; see table 1 for summary statistics. The third locus, AT5G27100 , is polymorphically deleted. The presence/absence of this gene was investigated using primers within the deletion as mapped by the arrays. These primers amplified a fragment in 11/12 and 9/12 individuals in the serpentine1 and serpentine2 populations, but only 3/12 and 5/12 individuals in the granitic1 and granitic2 populations; 24 individuals were typed twice with complete reproducibility. Deletion frequency was calculated assuming Hardy-Weinburg equilibrium, as this is a dominant marker. Primers are available from the authors upon request. SUPPORTING INFORMATION Complete lists of differentiated probes, regions, genes, and GO terms at multiple significance threshold. (Note that this is a large data set formatted as an .xls file with multiple spreadsheets. We have converted the initial sheet, of GO terms into a .pdf, but cannot easily submit the entire data set as .pdf.) (0.03 MB PDF) REFERENCES [1] Kruckeberg AR 1951 Intraspecific variability in the response of certain native plant species to serpentine soil. American Journal of Botany 38(6) 408 419 [2] Kruckeberg AR 1984 California Serpentines: Flora, Vegetation, Geology, Soils, and Management Problems Berkeley, CA University of California Press [3] Brady KU Kruckeberg AR Bradshaw HD 2005 Evolutionary ecology of plant adaptation to serpentine soils. Annual Review of Ecology Evolution and Systematics 36 243 266 [4] Kruckeberg AR 2002 Geology and Plant Life: The Effects of Landforms and Rock Types on Plants Seattle, WA University of Washington Press [5] Safford HD Viers JH Harrison SP 2005 Serpentine endemism in the California flora: a database of serpentine affinity. Madrono 52 222 257 [6] Wright JW Stanton ML Scherson R 2006 Local adaptation to serpentine and non-serpentine soils in Collinsia sparsiflora . Evolutionary Ecology Research 8(1) 1 21 [7] Berglund ABN Dahlgren S Westerbergh A 2004 Evidence for parallel evolution and site-specific selection of serpentine tolerance in Cerastium alpinum during the colonization of Scandinavia. New Phytologist 161(1) 199 209 [8] Wright JW 2007 Local adaptation to serpentine soils in Pinus ponderosa . Plant and Soil 293(1–2) 209 217 [9] Rhoads AF Klein WH 1993 The Vascular Flora of Pennsylvania: Annotated Checklist and Atlas Philadelphia, PA American Philosophical Society [10] Ronald J Akey JM Whittle J Smith EN Yvert G 2005 Simultaneous genotyping, gene expression measurement, and detection of allele-specific expression with oligonucleotide arrays. Genome Research 15(2) 284 291 [11] Winzeler EA Richards DR Conway AR Goldstein AL Kalmant S 1998 Direct allelic variation scanning of the yeast genome. Science 281(5380) 1194 1197 [12] Borevitz J Liang D Plouffe D Chang H-S Zhu T 2003 Large scale identification of single-feature polymorphisms in complex genomes. Genome Research 13 513 523 [13] Werner JD Borevitz JO Warthmann N Trainer GT Ecker JR 2005 Quantitative trait locus mapping and DNA array hybridization identify an FLM deletion as a cause for natural flowering-time variation. Proceedings of the National Academy of Sciences of the United States of America 102(7) 2460 2465 [14] Gresham D Ruderfer DM Pratt SC Schacherer J Dunham M 2006 Genome-wide detection of polymorphism at nucleotide resolution with a single DNA microarray. Science 311(5769) 1932 1936 [15] Turner TL Hahn MW Nuzhdin SV 2005 Genomic islands of speciation in Anopheles gambiae . Plos Biology 3(9) e285 [16] Nuzhdin SV Harshman LG Zhou M Harmon K 2007 Genome-enabled hitchhiking mapping identifies QTLs for stress resistance in natural Drosophila. Heredity 99(3) 313 321 [17] Turner TL Levine MT Eckert ML Begun DJ 2008 Genomic analysis of adaptive differentiation in Drosophila melanogaster . Genetics 179 1 19 [18] Bradshaw HD 2005 Mutations in CAX1 produce phenotypes characteristic of plants tolerant to serpentine soils. New Phytologist 167(1) 81 88 [19] Robertson A 1975 Gene frequency distributions as a test of selective neutrality. Genetics 81 775 785 [20] Akey JM Eberle MA Rieder MJ Carlson CS Shriver MD 2004 Population history and natural selection shape patterns of genetic variation in 132 genes. Plos Biology 2(10) e286 [21] Thornton KR Jensen JD Becquet C Andolfatto P 2007 Progress and prospects in mapping recent selection in the genome. Heredity 98 340 348 [22] Lewontin RC Krakauer J 1973 Distribution of gene frequency as a test of theory of selective neutrality of polymorphisms. Genetics 74(1) 175 195 [23] Ehrlich PR Raven PH 1969 Differentiation of populations. Science 165(3899) 1228 1232 [24] Hahn MW 2008 Towards a selection theory of molecular evolution. Evolution 62(2) 255 265 [25] Barreiro LB Laval G Quach H Patin E Quintana-Murci L 2008 Natural selection has driven population differentiation in modern humans. Nature Genetics 40(3) 340 345 [26] Hancock AM Witonsky DB Gordon AS Eschel G Pritchard JK 2008 Adaptations to climate in candidate genes for common metabolic disorders. Plos Genetics 4(2) e32 [27] Takano J Wada M Ludewig U Schaaf G von Wiren N 2006 The Arabidopsis major intrinsic protein NIP5;1 is essential for efficient boron uptake and plant development under boron limitation. Plant Cell 18(6) 1498 1509 [28] Shigaki T Rees I Nakhleh L Hirschi KD 2006 Identification of three distinct phylogenetic groups of CAX cation/proton antiporters. Journal of Molecular Evolution 63(6) 815 825 [29] Tajima F 1989 Statistical method for testing the neutral mutation hypothesis by DNA polymorphism. Genetics 123(3) 585 595 [30] Fu YX Li WH 1993 Statistical tests of neutrality of mutations. Genetics 133(3) 693 709 [31] Lynch M 2007 The Origins of Genome Architecture Sunderland, MA Sinauer Associates, Inc [32] Perry GH Dominy NJ Claw KG Lee AS Fiegler H 2007 Diet and the evolution of human amylase gene copy number variation. Nature Genetics 39(10) 1256 1260 [33] Hanikenne M Talke IN Haydon MJ Lanz C Nolte A 2008 Evolution of metal hyperaccumulation required cis-regulatory changes and triplication of HMA4. Nature 453(7193) 391–U344 [34] Begun DJ Holloway AK Stevens K Hiller LW Poh Y-P 2007 Population genomics: Whole-genome analysis of polymorphism and divergence in Drosophila simulans . Plos Biology 5(11) e310 [35] Kim S Plagnol V Hu TT Toomajian C Clark RM 2007 Recombination and linkage disequilibrium in Arabidopsis thaliana . Nature Genetics 39(9) 1151 1155 [36] Hinds DA Stuve LL Nilsen GB Halperin E Eskin E 2005 Whole-genome patterns of common DNA variation in three human populations. Science 307(5712) 1072 1079 [37] Latham RE 1993 The serpentine barrens of temperate eastern North America: critical issues in the management of rare species and communities. Bartonia 57 Suppl. 61 74 [38] Gautier L Cope L Bolstad BM Irizarry RA 2004 affy-analysis of Affymetrix GeneChip data at the probe level. Bioinformatics 20(3) 307 315 [39] Storey JD Tibshirani R 2003 Statistical significance for genomewide studies. Proceedings of the National Academy of Sciences of the United States of America 100(16) 9440 9445 [40] Rozas J Sanchez-DelBarrio JC Messeguer X Rozas R 2003 DnaSP, DNA polymorphism analyses by the coalescent and other methods. Bioinformatics 19 2496 2497