Evidence of neutral transcriptome evolution in plants

Authors


Author for correspondence:
Martin Broadley
Tel:+44  115 9516382
Fax:+44 115 9516334
Email: martin.broadley@nottingham.ac.uk

Summary

  • • The transcriptome of an organism is its set of gene transcripts (mRNAs) at a defined spatial and temporal locus. Because gene expression is affected markedly by environmental and developmental perturbations, it is widely assumed that transcriptome divergence among taxa represents adaptive phenotypic selection. This assumption has been challenged by neutral theories which propose that stochastic processes drive transcriptome evolution.
  • • To test for evidence of neutral transcriptome evolution in plants, we quantified 18 494 gene transcripts in nonsenescent leaves of 14 taxa of Brassicaceae using robust cross-species transcriptomics which includes a two-step physical and in silico-based normalization procedure based on DNA similarity among taxa.
  • • Transcriptome divergence correlates positively with evolutionary distance between taxa and with variation in gene expression among samples. Results are similar for pseudogenes and chloroplast genes evolving at different rates. Remarkably, variation in transcript abundance among root-cell samples correlates positively with transcriptome divergence among root tissues and among taxa.
  • • Because neutral processes affect transcriptome evolution in plants, many differences in gene expression among or within taxa may be nonfunctional, reflecting ancestral plasticity and founder effects. Appropriate null models are required when comparing transcriptomes in space and time.

Introduction

The transcriptome is the set of transcripts (mRNAs) of an organism at a defined spatial and temporal locus. The abundance of thousands of transcripts varies markedly in response to environmental and developmental perturbations, affecting protein translation and activity, and thus organism phenotypes. Consequently, it is widely assumed that variation in the abundance of transcripts among individuals within a population will ultimately lead to divergence in transcript abundance among populations and taxa through adaptive phenotypic selection. Pääbo and colleagues recently proposed a neutral theory of transcriptome evolution which challenges this belief (Khaitovich et al., 2004, 2005, 2006). In their theory, divergence in transcript abundance among taxa is driven not by adaptive selection, but by stochastic processes in which, by analogy to neutral genome evolution (Kimura, 1968; Gould, 2002), most differences in transcript abundance among individuals are likely to be selectively neutral or nearly neutral. Thus, within stabilizing constraints, variation in the expression of a transcript among individuals will drive expression divergence among populations and taxa as a consequence of drift. Empirical evidence of neutral transcriptome evolution is strongest among primate taxa of small population size (Khaitovich et al., 2004, 2005, 2006). In these studies, transcriptome divergence among taxa accumulates monotonically over time, correlating positively with evolutionary distance among taxa and with variation in gene transcript abundance among individuals. However, a neutralist interpretation of these data has been challenged (Gilad et al., 2006), based on inherent uncertainties in quantifying and normalizing the transcriptomes of taxa whose genome sequences are polymorphic, and because of sampling constraints.

The aim of this study was to determine if there is evidence of neutral transcriptome evolution among plants. We chose to study taxa of the Brassicaceae family, which has a basal age of c. 40 Myr (Bailey et al., 2006). Despite rapid speciation in several clades and a complex genome, which has undergone several duplications, the Brassicaceae phylogeny is well resolved based on chloroplast gene and internal transcribed spacer (ITS) sequences (Koch et al., 2005; Bailey et al., 2006; Beilstein et al., 2006; Schranz et al., 2007). Furthermore, Brassicaceae genome sizes are small among angiosperms (Johnston et al., 2005) and several are currently being fully sequenced: those of Arabidopsis lyrata, Capsella rubella, Brassica rapa and Eutrema salsuginea (previously Thellungiella halophila). These efforts are hastening our understanding of complex genome evolution, making the Brassicaceae a tractable model family for evolutionary studies (Schranz et al., 2007). However, we chose this family specifically because it contains Arabidopsis thaliana, for which there are > 4500 transcriptome data sets obtained on a single array type (ATH1-121501 GeneChip, ATH1; Affymetrix, Santa Clara, CA, USA), curated in the public domain (Craigon et al., 2004). This unique public resource supports the meta-analyses of transcriptomes from tissue-specific, developmental, mutant vs wild type, and environmental-response studies (Schmid et al., 2005; Brady et al., 2007). Furthermore, ATH1 – which comprises 11 perfect-match (PM)/mismatch (MM) pairs of short (25-mer) oligonucleotide probes per probe-set – has proven an effective platform for robust cross-taxa transcriptome analyses, using genomic DNA-based masking strategies and normalizations to remove the effects of DNA polymorphisms among species on RNA expression estimates (Hammond et al., 2005, 2006; Graham et al., 2007). These techniques have been applied to the study of gene regulation in Brassica crops of economic consequence (Hammond et al., 2005) and in extremophile species adapted to metalliferous and saline habitats (Broadley et al., 2007). In this study, we use robust cross-taxa transcriptomics and meta-analyses to show that neutral processes profoundly affect transcriptome divergence among plant taxa.

Materials and Methods

Seeds were obtained of Arabidopsis thaliana (L.) Heynh. (Columbia (Col-0) and Landsberg erecta (Ler-0); donor: Nottingham Arabidopsis Stock Centre, Nottingham, UK), Arabidopsis lyrata ssp. petraea (L.) O’Kane & Al-Shehbaz and Arabidopsis halleri (L.) O’Kane & Al-Shehbaz (donor: Mark Macnair, University of Exeter, Exeter, UK; Filatov et al., 2006), Capsella bursa-pastoris (L.) Medik. (720-81 and 798 accessions) and Capsella rubella Reut. (donor: Michael Lenhard, John Innes Centre, Norwich, UK), Alyssum murale Waldst. & Kit. and Alyssum lesbiacum (Candargy) Rech.f. (donor: Alan Baker, University of Melbourne, Melbourne, Australia), Brassica cretica Lam. and Brassica oleracea L. (donor: Genetic Resources Unit, University of Warwick, Warwick, UK), and Thlaspi arvense L. and Noccaea caerulescens (J. Presl & C. Presl) F.K. Mey. (previously Thlaspi caerulescens) (from Ganges, southern France (Hammond et al., 2006) and from Youlgrave, Derbyshire, UK (collected by Martin Broadley and Steven Whiting)). Cleome pinnata Pursh. (Cleomeaceae; seed from B and T World Seeds, Olonzac, France) was used as an outgroup.

For cross-taxa normalizations, a two-step strategy before transcriptome analyses was adopted using established methods (Hammond et al., 2005, 2006; Graham et al., 2007). For the first physical step, 250 ng of genomic DNA was extracted from the 15 taxa and hybridized to separate ATH1 GeneChips. Noccaea caerulescens (Viviez) genomic DNA hybridizations were reported previously (Hammond et al., 2006). To estimate evolutionary distance among taxa, for each of the 16 taxa, a custom ‘chip definition file’ (taxon1.cdf ... taxon16.cdf) was produced using Xpecies V2.0 scripts (available from http://www.affymetrix.arabidopsis.info/xpecies; Hammond et al., 2005, 2006; Graham et al., 2007). For each taxon, the custom *.cdf file was based on A. thaliana, but with 100 000 of the least informative PM probes excluded, that is, PM probes with the lowest DNA hybridization signal intensities for each taxon. Subsequently, a similarity matrix of ‘least-informative’ probe pairs common to each taxa pair was constructed. Branch lengths (based on proportional units) and topology for the 16-taxon phylogeny were calculated using Neighbor-Joining in phylip (Version 3.67; Felsenstein, 1989). For the second in silico step, we developed a universal custom chip definition file (‘universal.cdf’), retaining only those PM/MM probe pairs whose corresponding PM probe had DNA hybridization signal intensities > 100 (Fig. 1b), as described previously (Broadley et al., 2007). In the present study, the universal.cdf defines 41 370 probe pairs (16.3% of the total), representing 18 494 ATH1 probe sets (81.3% of the total), at 2.2 probe pairs per probe set (Fig. 1b).

Figure 1.

DNA hybridizations to ATH1 GeneChips resolve a consensus Brassicaceae phylogeny. (a) Unrooted topology of 16 taxa of Brassicaceae based on DNA similarity (i.e. the branch-length scale is a proportional estimate of dissimilarity). (b) The distribution of perfect-match (PM) probe pairs retained on the universal chip definition file (universal.cdf) used for subsequent transcriptome analyses. Black bars represent the distribution of 41 370 probe pairs among 18 494 ATH1 probe sets (out of 22 746), retained at a DNA hybridization signal intensity of 100 across the 14 taxa. Chloroplast genes (grey bars) represent 79 retained ATH1 probe sets, annotated as chloroplast genes, out of a total of 80 on the ATH1 array.

For transcriptome analyses, all plants were grown simultaneously on 0.8% agar with 0.25× Murashige and Skoog (MS) salts, adjusted to pH 5.6 with NaOH, under conditions described previously (Hampton et al., 2004). At 19 d after sowing, nonsenescent, fully expanded rosette leaves were pooled from eight to 22 plants and snap-frozen at −70°C. The experiment was performed in triplicate in three sequential blocks. RNA was extracted, and 2–4 µg was reverse-transcribed, labelled in an in vitro transcription assay, fragmented and hybridized to ATH1 as described previously (Hammond et al., 2005). Thus, 42 RNA .cel files were analysed in GeneSpring (Version GX7.3; Agilent Technologies, Santa Clara, CA, USA) using the universal.cdf and a Robust Multi-array Average (RMA) prenormalization routine. Raw probe-set (gene) expression values were subsequently normalized to the median gene expression value across all 42 samples, and loge-transformed. The arithmetic mean and variance of the normalized, loge-transformed gene expression was calculated for each taxon (n = 3). Transcriptome divergence was calculated for 91 pair-wise comparisons of taxa. Transcriptome divergence between members of a pair of taxa is the square-root of the sum of the squared differences in mean expression value for all genes. Expression diversity for each taxon was the variance among triplicate samples; genes were ranked by their arithmetic mean variances across all 14 taxa.

For meta-analyses, three independent data sets were used, all of which were re-analysed in GeneSpring (Agilent) after normalizing the data using our universal.cdf and RMA algorithms. First, we used 34 RNA .cel files generated by Benfey and colleagues (Brady et al., 2007). These data comprised 12 cell types (n = 3 unless stated), sampled by cell sorting or dissection: atrichoblasts; columella; cortex; endodermis-cortex-quiescent centre (QC); endodermis beyond mature hair zone QC; epidermis; epidermis-lateral root cap; pericycle; phloem (n = 2); QC (n = 2); root stele to elongation zone; xylem. Transcriptome divergence among cell types was calculated as the square-root of the sum of the squared differences in mean expression value for 66 pair-wise comparisons. Transcriptome divergence was calculated for all genes, and for genes with the highest and lowest expression diversities. Expression diversity for each cell type was the variance among samples; genes were ranked subsequently by their arithmetic mean variances across all 12 cell types. The second and third data sets were combined with our 42 cross-taxa RNA .cel files. The second data set comprised 126 RNA .cel files representing a ‘developmental-baseline’ of A. thaliana Col-0, generated by the AtGenExpress Consortium (Schmid et al., 2005). These data include shoot, root and floral material from a variety of tissue ages and growth conditions (Supporting Information Fig. S1). The third data set comprised 24 RNA .cel files generated by Macnair and colleagues, representing shoot and root tissue from A. lyrata ssp. petraea and A. halleri grown at high and low zinc supplies (Filatov et al., 2006). Thus, we analysed 192 RNA .cel files simultaneously. Principal components analysis (PCA) was performed on normalized, loge-transformed gene expression values, using GenStat (Version 10.1.0.72; Lawes Agricultural Trust, VSN International, Hemel Hempstead, UK).

Results and Discussion

We have found evidence that stochastic variation in gene expression has a role in plant transcriptome evolution. Our study required the use of robust cross-taxa transcriptomics, which involves a two-step physical and in silico-based normalization procedure. For the physical step, we hybridized genomic DNA from each taxon to the ATH1 array. This provided a robust estimate of evolutionary divergence among taxa. Thus, a similarity-based topology of ATH1 PM probes hybridizing to DNA from Brassicaceae taxa approximates the Brassicaceae phylogeny based on chloroplast and ITS sequences (Fig. 1a; Koch et al., 2005; Bailey et al., 2006; Beilstein et al., 2006). The populations of N. caerulescens from southern France – which included an in-group control from Viviez – resolve as expected. Capsella does not resolve according to taxonomic characters which are notoriously difficult to interpret within this group (Ceplitis et al., 2005; Hawes et al., 2005). Thus, it may be possible to use genomic DNA hybridizations to microarrays to support phylogenetic reconstruction within certain groups of organisms. For the in silico step, we developed a universal chip definition file (‘universal.cdf’), retaining only those PM/MM probe pairs whose corresponding PM probe had raw DNA hybridization signal intensities > 100 for all 14 taxa (Fig. 1b). In this study, the universal.cdf defines 41 370 probe pairs (16.3% of the total), representing 18 494 ATH1 probe sets (81.3% of the total), at 2.2 probe pairs per probe set (Fig. 1b). The 80 chloroplast genes on ATH1 are represented by 79 genes on the universal.cdf at an average of 7.8 (SEM ± 0.28) probe pairs per probe set, consistent with sequence conservation among chloroplast genes. Of 3889 A. thaliana genes annotated as pseudogenes (TAIR7.0), 744 have probe sets on ATH1, of which 511 are retained in the universal.cdf with one probe per probe set, consistent with greater sequence divergence among pseudogenes.

We quantified gene transcripts in nonsenescent leaves sampled from 14 Brassicaceae taxa cultivated in vitro. The mean expression and variance of a transcript were estimated using loge-normalized data, derived from three biological replicates. Expression diversity for each gene was calculated as the mean variance across all 14 taxa; this value represents variation which can be attributable to genetic, environmental or technical error components. Transcriptome divergence between members of each pair of taxa (n = 91) was represented by a single numeric descriptor, calculated as the square-root of the sum of the squared differences in mean expression value across (1) all 18 494 genes, (2) 511 pseudogenes, and (3) 79 chloroplast genes. Transcriptome divergence accumulated monotonically as a function of evolutionary time for all genes (Fig. 2a), for pseudogenes (Fig. 2b) and for chloroplast genes (Fig. 2c). The significance of associations between transcriptome divergence and evolutionary distance was determined using Mantel tests, based on 10 000 random permutations for all genes, pseudogenes and chloroplast genes (GenStat Version 10.1.0.72). For all genes and pseudogenes, no permutation produced a stronger association between transcriptome divergence and evolutionary distance than was observed in this study. For chloroplast genes, only 1% of permutations produced a stronger association between transcriptome divergence and evolutionary distance. Transcriptome divergence was greater for the 25% of genes with the highest ranked expression diversity than for the 25% of genes with the lowest ranked expression diversity, both for all genes (Fig. 2a) and for pseudogenes (Fig. 2b). These observations are consistent with a neutralist interpretation (Khaitovich et al., 2004).

Figure 2.

Transcriptome divergence among Brassicaceae increases monotonically with evolutionary distance and correlates positively with expression diversity. Transcriptome divergence is shown for (a) all genes (n = 18 494; black circles), including those with the highest (n = 4624; red) and lowest (n = 4624 blue) expression diversities, respectively; (b) pseudogenes (n = 511), with labels as for (a); (c) 39 randomly selected chloroplast genes (n = 79; green), and all genes with eight probes per probe set (n = 39; yellow) among 14 taxa; and (d) 500 randomly selected genes represented by one (black), two (green), three (red), four (blue), and five (cyan) probe pairs per probe set, and for pseudogenes (green triangles).

We tested if the physical properties of mRNAs affected estimates of transcriptome divergence using information for 13 012 A. thaliana Ler-0 genes (Narsai et al., 2007), 10 454 of which were represented on the universal.cdf. These genes were ranked, in turn, according to mRNA stability, length, number of introns, or GC content. Within each category, genes were grouped into two sets representing the 1000 highest and 1000 lowest values. There were no significant differences in the rate of transcriptome divergence among taxa in terms of stability (P = 0.96), length (P = 0.38), number of introns (P = 0.40), or GC content (P = 0.19) of the mRNA. We also tested whether hybridization artefacts affected estimates of transcriptome divergence. When two or more probe pairs were retained in a probe set, estimates of transcriptome divergence as a function of evolutionary time were unaffected (Fig. 2d).

If stochastic variation drives transcriptome divergence among taxa, then it will also drive transcriptome divergence among tissues (Khaitovich et al., 2004, 2006), albeit subject to evolutionary developmental and stabilizing constraints operating at several selection loci (Gould, 2002). As the A. thaliana root transcriptome has been mapped at high resolution, we sought evidence of neutral transcriptome properties among 18 454 genes from 34 individual ATH1 arrays, representing cell-sorted or dissected cell types from 12 root zones (Brady et al., 2007). There is strong evidence that stochastic variation is an important component of the transcriptome divergence among root cell types. Thus, genes with the highest expression diversity among samples diverged among cell types at a much faster rate than genes with the lowest expression diversity, based on 66 pair-wise comparisons (Fig. 3). Remarkably, there was a significant positive correlation between expression diversity in the 12 root cell types of Brady et al. (2007) and leaf-transcriptome divergence among the 14 Brassicaceae taxa from the present study (r = 0.28, df = 18 492, t = 29.3, P = 1.6 × 10−181). Tissue-specific data from several taxa are now required to disentangle cell type-specific functional insights from evolutionary and developmental constraints.

Figure 3.

Genes whose expression varies the most among individual samples have greater expression divergence among Arabidopsis thaliana root cell types. Data are from 34 source data sets representing 12 root cell types (Brady et al., 2007). Transcriptome divergence was calculated for 66 pair-wise comparisons of cell types and ranked for all genes. Transcriptome divergence between members of a pair of cell types is the square-root of the sum of the squared differences in mean expression value for all genes (n = 18 494; black circles), and genes with the highest (n = 4624; red) and lowest (n = 4624; blue) expression diversities, respectively.

A profound consequence of neutral transcriptome evolution is that a difference in the expression level of a gene between two taxa may reflect an ancestral plasticity and/or a founder effect of a small population size, rather than a functional adaptation (Khaitovich et al., 2004, 2006). Here we observed that functionally distinct tissues (e.g. leaves, roots and floral organs) sampled from A. thaliana at different tissue ages and under different growth conditions have transcriptomes that are more similar to each other than to the transcriptomes of corresponding, functionally homologous tissues sampled from different species under identical growth conditions (Fig. 4). This observation is consistent with a neutralist interpretation of transcriptome divergence among taxa, and with evidence that relatively small numbers of genes are likely to control organ identity (Soltis et al., 2007). Thus, principal component (PC) 1 separates the transcriptomes of A. thaliana (Col-0 and Ler-0) from all other taxa, irrespective of growth conditions or plant part. In PC1, leaf transcriptomes of A. thaliana Col-0 and Ler-0 accessions from our current study were more similar to green and nongreen tissue (roots and floral organs) transcriptomes of A. thaliana Col-0 from previous studies (Schmid et al., 2005) than to the leaf transcriptomes of close relatives from the Arabidopsis genus or tribe Camelineae (which contains both Capsella and Arabidopsis; Schranz et al., 2007), despite the obvious conservation of leaf function and identical growth conditions. Similarly, in PC1, the leaf transcriptomes of A. lyrata ssp. petraea and A. halleri from our current study were more similar to shoot and root transcriptomes of A. lyrata ssp. petraea and A. halleri measured previously (Filatov et al., 2006) than to the leaf transcriptomes of other close relatives. Arabidopsis lyrata ssp. petraea and A. halleri (2n = 16) are more closely related to each other than either species is to A. thaliana, and are inter-fertile. PC2 separates green and nongreen tissue transcriptomes and higher resolution phylogenetic and anatomical separation of transcriptomes occurs in PC3–PC10 (Fig. S1).

Figure 4.

Transcriptome divergence among functionally homologous tissues sampled from congenerics is greater than divergence among functionally specialized tissues from conspecifics. Data are principal components (PC) 1 and 2 from a PC analysis of transcriptome data from (1) 42 ATH1 data sets reported in this study, (2) 126 ATH1 data sets of ‘developmental-baseline’ tissue for Arabidopsis thaliana Col-0 from the AtGenExpress Project (Schmid et al., 2005); (3) 24 ATH1 data sets of shoot and root tissue from Arabidopsis lyrata ssp. petraea and Arabidopsis halleri grown at high and low zinc supplies (Filatov et al., 2006). Black symbols, A. thaliana Col-0; yellow, A. thaliana Ler-0; light green, A. halleri; blue, A. lyrata ssp. petraea; dark red, Capsella bursa-pastoris 798; dark green, C. bursa-pastoris 720; dark yellow, Capsella rubella; red, Alyssum murale; pink, Alyssum lesbiacum; cyan, Brassica oleracea; light grey, Brassica cretica; dark grey, Thlaspi arvense; dark cyan, Noccaea caerulescens Youlgrave; white, N. caerulescens Ganges. Circles, leaf/shoot material; cross symbols, roots (colours as before); triangles, floral tissues, including flowers, carpels, petals, sepals, stamens and pedicels; squares, stems, hypocotyls and shoot apex tissues. PC1 and 2 are highlighted. A full 45-panel figure representing PCs 1–10 is available as Supporting Information Fig. S1.

In conclusion, we have found compelling evidence for neutral transcriptome evolution, from new cross-taxa studies and from examination of Arabidopsis transcriptomes mapped previously. Transcriptome divergence among taxa and tissues is clearly heritable. It will be intriguing to discover how much of this is attributable to simple sequence polymorphisms, genome rearrangements, transposon insertions or epigenetic modifications. As sequencing costs decline and more genomes and transcriptomes are sequenced, it will be possible to further explore the causes and effects of neutral transcriptome evolution. Whilst many differences in transcript abundance among tissues and among taxa will have functional consequences, integrated ’omics and phenotypic data are required for several species before these can be interpreted correctly in space and time.

Acknowledgements

All .cel files and normalization scripts are available from http://affymetrix.arabidopsis.info/xpecies. Funding for this work was provided in part by the UK Biotechnology and Biological Sciences Research Council (BBSRC), and UK Engineering and Physical Sciences Research Council (EPSRC) through the UK Centre for Plant Integrative Biology (CPIB; MRB, NJG, RGF and STM). Funding was also provided by the Scottish Executive's Environment and Rural Affairs Department through the SCRI Innovation Fund (SEERAD; PJW, JWM and PPMI), and the UK Department for Environment, Food, and Rural Affairs (Defra; JPH and HCB). We thank Malcolm Bennett (University of Nottingham) for comments on the manuscript.

Ancillary