• No results found

Although the strategies for elucidating the emergence of complex traits discussed in this thesis are promising, there are practical limitations that cannot be circumvented, at least not easily. In cases where natural populations are studied, we are dependent on capturing the right moment when the biochemical process of interest is active. If we do not sample at the critical time point, we risk

missing the processes that actually define the trait of interest. Not only is the sampling time point critical for capturing e.g. a developmental stage, it is also important to consider daily fluctuations that can have an effect on gene expression. This is clearly demonstrated in a recent study in rice where gene expression was quantified in the field, and it was shown that even short term variations in temperature and sunlight levels modified gene expression in a reproducible manner [74].

When it comes to geographical range, we are limited to the range of sampling, and cannot do inferences beyond that. For example, in papers III and IV, we were limited to the range of Sweden, while aspens are spread out across more or less the whole northern hemisphere. In that context, Sweden is a very small part of the total distribution range, and it might not be very surprising that we did not see any pronounced population structure in the data.

In order to have enough statistical power in association studies, large sample sizes are needed. A problem with using forest trees is that it is very expensive and time consuming to maintain a large population of trees. Ideally, we want to grow them in a controlled environment in order to minimise environmental effects, but this is clearly not a practical approach. In the study of human height that has been mentioned several times in this thesis, they used more than 250,000 individuals, but this was a meta-study [3], i.e. a study collecting data from previous studies, avoiding the hassle of collecting the data themselves.

Even so, a meta-study of the same magnitude inP. tremula is not possible today, since the amount of data generated is not even close to that of human studies. The importance of good annotations such as previous efforts to elucidate regulatory mechanisms also play a big role [36]. As of yet, most of these efforts have been directed towards human studies (e.g. ENCODE [14]), quite understandably. While the results from these annotation efforts are not directly transferable to other species, information regarding general characteristics of genome structure and function can most likely be transferred to other species. This has to some extend been done already, but in the other direction. For example, the fruit flyDrosophila melanogaster has been used as a model organism for human genetics and disease for over a century [75].

The results of associations studies are just that—associations. A genetic variant that is associated with a particular phenotype is not necessarily the causative variant. It might be that it in turn is associated with the causal variant through linkage disequilibrium (LD). In the case of plants, a variant in LD with the causal variant might be good enough in many cases where marker assisted selection can be employed in breeding. However, if more control of the phenotype is needed, a variant in LD is not of much help. If the variant is not causal, it is likely that mutating this position will not result in a corresponding change in phenotype. Strategies to filter out the causal variants from association studies include integrating different types of data in order to single out the most likely candidate genes or loci, but there are several challenges associated with this kind of data integration. The individual data sets themselves have their own issues to begin with. There are systematic biases, normalisation issues, and correlation structures that are not trivial to deal with, and that can eat up a considerable portion of resources available to a project. Something a bit more abstract that could help with finding causal variants is transparency when it comes to publication. This could potentially help minimise confirmation bias and consequently the number of false positives in circulation [76].

2

Paper summaries

This chapter will give a short summary of each paper included in the thesis.

Paper I deals with gene regulation in a cyanobacterium, while papers II–IV considers aspects of gene expression, genotype, and phenotype in the deciduous treePopulus tremula(European aspen).

2.1 Paper I — Gene regulation in a cyanobacterium

Synechocystisis a fresh water cyanobacterium and it is one of the most studied cyanobacteria to date, being used as a model system for nitrogen fixation and photosynthesis amongst other things. Even though the genome ofSynechocystis was sequenced already in 1996, most of the genes in its genome are still annotated as having unknown function. In this paper, we created the web applicationSynergy to enable researchers working withSynechocystisto explore the gene expression and the gene regulation of this organismin silicoin order to find potential candidate genes for e.g. knock-out experiments. We collected 371

microarray experiments from public sources and constructed a co-expression network. As mentioned in section 1.3.3, a co-expression network is simply a manifestation of the underlying regulatory network, so in order to form a link between co-expression and co-regulation, potential regulatory motifs were identified using phylogenetic footprinting. This method is based on the alignment of regulatory regions of orthologous genes from related organisms.

In this case, 22 genomes from the Chroococcales taxon were used for the phylogenetic footprinting, and this resulted in a set of 4,977 potential regulatory motifs. In the paper we show that co-expression network neighbourhoods of regulatory proteins were enriched for regulatory motifs, thus providing a possible regulatory link between these regulators and the co-expressed genes. The user of the web application can then investigate whether their gene set of interest is co-expressed, and whether this to some extent can be explained by shared regulatory motifs. In order to make the application as useful as possible, the gene identifiers used were the well established identifiers from CyanoBase (http://genome.microbedb.jp/cyanobase/; [77]).

As part of a sanity check of the integrated data, a couple of case studies were conducted where both previously published results were confirmed, and also potentially novel regulatory relationships were presented.

Synergy is publicly available at http://synergy.plantgenie.org.