Online Supplement for
Aneuploidy in Arabidopsis (Huettel et al., 2008) 4 Feb 2008 Vienna

This section provides figures illustrating the observed expression changes (both trends and deviations from the trends), links to fully annotated tables of affected genes, and tables showing trends by GOslim category. Please scroll down for Figures S3 and S4.

Expression Changes – Figures

Figure 5 of our manuscript shows the systemic increase of expression for chromosome 5 genes in the trisomic plants, as a function of the average gene expression level (on the x-axis). Transcripts on chromosome 5 are coloured green, and the intensity dependent trend plus/minus standard deviation is plotted in magenta. The trend for transcripts on other chromosomes is shown in orange. The dashed vertical line marks the intensity A₁₊₁ where the lower magenta and the upper orange lines cross and the trends are separated by 1+1 standard deviations. Surveys of trends thus focus on the strongly expressed transcripts to the right of the dashed lines, where the assay will be most accurate (*). The dotted vertical line indicates the lowest expression intensity for which a statistically significant change could be detected with p < 5% (Holm FWER). We here provide the figure in different resolutions [ low | medium | high ].

(*) The improved separation of more strongly expressed chromosome 5 genes from the trend of genes on other chromosomes that is evident in Figure 5 of our manuscript affects our ability to detect expression differences between trisomic and disomic samples. By plotting the percentage of genes on chromosome 5 that could be identified as having higher expression in trisomics, we can show that the sensitivity of our assay improves considerably for genes with an average intensity larger than approximately A₁₊₁ and that the exact choice of the threshold makes little difference [ Athresh ]. To test the dosage-related increase in expression for genes on the trisomic chromosome 5 with lower expression levels, four moderately and five lowly expressed genes were selected for examination by high-sensitiviy qRT-PCR. Since the quantification of lowly expressed genes by real-time PCR can be non-trivial, genes were chosen with a well-documented exon/intron structure according to TAIR. The primers were designed to span the exon/intron junctions in order to ensure that PCR amplification did in fact report on differential gene expression rate and not, e.g., on residual genomic DNA. Consistent with the general chromosome 5 trend, a higher steady state transcript level in trisomics was indeed observed for the majority of these genes [ Fig. S1 and Fig. S2 ].

Figure 4 of our manuscript shows that all areas of the triplicated chromosome seem similarly affected [ cis.Mpos ] and that the observed trans effects appear to scatter randomly across chromosomes [ trans.Mpos ]. In these plots only strongly expressed genes, for which the effect could be assessed accurately are shown.

For a study of deviations from the cis chromosomal trend, this trend was subtracted from the data and tests performed on the zero-calibrated expression values. After successful calibration, the magenta trend line traces the x-axis; and significant deviations from the trend are shown as blue dots [ cal0M: low | medium | high resolution ]. We here also show the 100 genes most strongly deviating from the cis trend in the context of an M(A) plot and their chromosomal locations [ chr5to0.MA | chr5to0.Mpos ]. Also when all genes significantly differentially expressed relatively to the chromosomal cis-trend are included in the figure [ chr5to0.Mpos.plain ], effects appear randomly distributed over the chromosome. Rainbow colours indicate relative significance (red/yellow is highest, blue/magenta is lowest). Only a minority of genes is below the general cis-trend and therefore dosage compensated or down-regulated.
Figure S3 highlights in red the six epigenetic modifiers located on the triplicated chromosome that are discussed in our manuscript (see legend). All six fully follow the cis chromosomal trend of increased expression. Similarly, as an example of trans effects, Figure S4 surveys differential regulation on chromosome 2. Red highlights show the two epigenetic modifiers ROS1 and RDR5, prominently up-regulated in the trisomic plants.

Genes Affected – Trends: A discussion of Dosage Compensation

We considered two approaches for studying dosage compensation:

1. For the data shown in Figure 5, we could apply standard tests for differential expression. For the trend estimates in the manuscript, we used two independent statistical tests:
   1. a convex decreasing density estimate for the number of non-Null hypothesis, giving a lower bound of 3%, and
   2. an empirical Bayes regularized t-test with Benjamini-Yekutieli adjustment for multiple testing and a False Discovery Rate (FDR) threshold of 5%.
   Raising this threshold did not increase the estimated number of significantly different genes, as the percentage of False Positives expected increases fast with a higher FDR threshold. We can consider genes with an expression level that is significantly higher in trisomics than in disomics as not fully dosage compensated. Then we can designate genes with no significant change in expression as dosage compensated genes. With the FDR controlled at 5%, the actual percentage of False Positives can be anywhere between 0% and 5%. Test results therefore gave an upper bound of 15% for the ‘best case’ of an FDR of 0%, and 11% for the ‘worst case’ of an FDR of 5%. This analysis approach has the advantage that we can also discern ‘overshooting’ dosage compensation:- about 1% of genes in trisomic samples were actually more weakly expressed. The disadvantage of this method is that it does not consider partly dosage compensated genes.
2. A complementary approach therefore considers the deviation from the chromosome 5 trend. This fully considers partial dosage compensation. Under ideal measurement conditions, this would correspond to identifying genes with a significant deviation from 1.5× change. In realistic conditions, this test can be performed by subtracting the average chromosome 5 trend from the data, yielding a zero-centred distribution [ cal0M ]. We can then consider genes on chromosome 5 with an expression level that is significantly lower than the chromosome trend to be dosage compensated. This is again assessed by an empirical Bayes regularized t-test, Benjamini-Yekutieli adjustment for multiple testing, and an FDR threshold of 5%. Raising this threshold does not increase the estimated number of significantly different genes, as the percentage of False Positives expected increases with a higher FDR threshold. This gives an independent estimate of 14% of (somewhat) dosage compensated genes (and with a 5% FDR, the lower bound is 13%). Intuitively, how much gene expression needs to be lower to be considered significant is determined by the standard deviation of the random scatter of gene expression on the chromosome.

In summary, at least 13% of genes were dosage compensated, at least 3% of genes were fully dosage compensated, whereas at least 85% of genes were not dosage compensated. The manuscript text collects these summaries in a concise form, with test details given in the Methods section. Evidence from individual tests is quoted in parentheses in the main manuscript text.

Genes Affected – Annotated Tables

These tables are TAB-delimited text files that can be loaded into any spreadsheet program. The largest tables are about 20MB in size. Subsets considering only strongly expressed genes are marked as bright. Calibrated cis data tested for deviation from the chromosomal trend is labelled Chr5to0.

Group | A subset | p.adjust	Test subsets
Cis | bright | Holm FWER	All, Up, Down
Cis | all | Holm FWER	All, Up, Down
Cis | bright | B-Y FDR	All, Up, Down
Cis | all | B-Y FDR	All, Up, Down
Chr5to0 | bright | Holm FWER	All, Up, Down
Chr5to0 | all | Holm FWER	All, Up, Down, Diff ( = Up or Down )
Chr5to0 | bright | B-Y FDR	All, Up, Down
Trans | bright | Holm FWER	All, Up, Down
Trans | all | Holm FWER	All, Up, Down, Diff ( = Up or Down )

GOslim trends

The tables below are small TAB-delimited text files and can be loaded into any spreadsheet program or text editor. The first column shows the odds-ratio (OR) for the GOslim group being overrepresented in the test-set vs the entire chip, then follows a Holm adjusted FWER p-value, the relative (percentage) and absolute counts in test and reference sets, and the GOslim group being tested. Groups are sorted by significance of over- or underrepresentation in the test set. Only strongly expressed (bright) genes were considered, as our assay is most accurate for these.

Group	Test subsets
Cis	Down, Not diff, Not up ( = Down or Not diff )
Chr5to0	Diff
Trans	Diff

(... back to Micorarray Methods)