Tutorial

Yeast Case Study

The data were obtained from the study of Jerison et al. 2017 contained S. cerevisiae populations that were evolved for 500 generations in optimal temperature. 273 different founder strains were used (different genomic context). The founders selected for this sequencing were chosen to ensure approximately equal representation of each parental KRE33 allele, but were otherwise random. For each founder 4 parallel populations were obtained that were evolved under ordinary temperature (OT). Per population the variants with an allele frequency >60% were selected, 68% were at > 90% frequency. Most mutations in the given list were nonsynonymous. The list of de novo mutations in each of the populations evolved under OT conditions was obtained from the authors of the paper and can be downloaded here.

The figure below gives an overview of the different genes containing de novo mutations as were detected in the original paper.

The list below recapitulates the genes described in the paper, (*) indicates that the genes were detected with IAMBEE-run1; (not in strict network) indicates the genes were not in the network used in run and hence could not be retrieved.

• YNL132W,KRE33 *
• YGR145W,ENP2 (*, not in strict network)
• YDR299W,BFR2 (*, not in strict network)
• YPL217C,BMS1 (*)
• YBL004W,UTP20 (*)
• YBL072C,RPS8A (*)
• YPL090C,RPS6A (*)
• YGR218W, CRM1
• YMR128W,ECM16 (*)
• YBR140C,IRA1 (*, not in strict network)
• YOL081W,IRA2 (*, not in strict network)
• YOR360C,PDE2 (*, not in strict network)
• YDL042C,SIR2
• YLR442C,SIR3 (*, not in strict network)
• YDR227W,SIR4 (*, not in strict network)

Run1: Strict network (edge probability cutoff of 0.95)

In this setting we run IAMBEE with a S. cerevisiae STRING derived network (stringency of the edges larger >0.95). The network contains about 16000 edges. We used as input all variants excluding the synonymous variants and the variants that were in the input file not mapped on genes (intergenic etc). A default parameter sweep was performed (note that in the default setting the parameters of the cost used during the sweep are determined automatically and are dataset specific. The parameter values used are available in the results folder. Only compensation for the mutator phenotype was used as the functional information and the increase in frequency during the sweep were not available (make sure these sliders are switched off).

This run took about 30 minutes.

The result can be visualized in the browser. However to assess the relevance of the results, download the results folder and unzip the folder. Go to the folder "\opt\resulting_networks\d3js_visualization"

You can open the file ‘highestScoringSubnetwork.html’ and visualize the network in the browser. The edge width is here replaced by the opacity of the edges. Edges which are more dark in color represent the more significant signals in the data. A print screen of this network is given below.

From the folder \opt\resulting_networks\ open the file rankedMutations.txt. You will see the set of genes that were prioritized together with their rank (based on the sweep on the cost parameter). A lower rank means that the gene was already prioritized at a higher cost value and hence represents a more significant signal in the data.

To interpret the visualization results it is important to focus on the genes indicated in this file. These are the genes that carry the actual mutations. In the network visualization many more genes are indicated some of which are ‘connector genes’. Genes that are used to find paths that connected the mutated genes. For each mutated gene the visualization provides in the circle around the gene a colored section which indicates the population in which the gene was mutated (see below as indicated in the figure). Genes with a colored section are the ones that are mutated and can be retrieved in the text file.

Example of the text file with prioritized genes for run1.

To focus on genes or sub-network, start with the highest ranked genes from the list is possibility. Usually these genes are also the most frequently mutated (indicating that the optimization does its job correctly). For instance, YNL132W (KRE33). The figure shows the location of KRE33 and its neighboring mutated genes (indicated figure with red and blue box respectively). These are the genes prioritized because they are part of a recurrently mutated sub-network (despite the fact that most of them are only mutated once or twice over all populations. The network in its entirety is frequently mutated.

1. YNL132W (KRE33 - frequently mutated, KRE33 was present in the background genome of certain founders of the population): Protein required for biogenesis of the small ribosomal subunit; heterozygous mutant shows haploinsufficiency in K1 killer toxin resistance; essential gene; NAT10, the human homolog, implicated in several types of cancer and premature aging
2. YGL078C (DBP33 - mutated once): RNA-Dependent ATPase, member of DExD/H-box family; involved in cleavage of site A3 within the ITS1 spacer during rRNA processing; not essential for growth, but deletion causes severe slow-growth phenotype
3. YNL308C (KRI1 - mutated once): Essential nucleolar protein required for 40S-ribosome biogenesis; associate with snR30; physically and functionally interacts with Krr1p-kri1
4. YNL182C (IPI1 - mutated once): Component of the Rix1 complex and pre-replicative complexes (pre-RCs). It is required for processing of ITS2 sequences from 35S pre-rRNA, component of the pre-60S ribosomal particle with the dynein-related AAA-type ATPase Mdn1p; required for pre-RC formation and maintenance during DNA replication licensing; highly conserved protein which contains several WD40 motifs; IPI3 is an essential gene; other members include Rix1p, Ipi1p, and Ipi3p, IPI3

Another frequently mutated gene is YPL217C (BMS1). This gene together with its neighbor are also located on the network (red boxed). These genes together with its mutated neighbors from the prioritized list forms a second recurrently mutated subcluster.

1. YPL217C (BMS1 - mutated frequently): GTPase required for ribosomal subunit synthesis and rRNA processing; required for synthesis of 40S ribosomal subunits and for processing the 35S pre-rRNA at sites A0, A1, and A2; interacts with Rcl1p, which stimulates its GTPase and U3 snoRNA binding activities; has similarity to Tsr1p
2. YBL004W (UTP20 - mutated twice): Component of the small-subunit (SSU) processome; SSU processome is involved in the biogenesis of the 18S rRNA
3. YLR409C (UTP21) (mutated once): Subunit of U3-containing 90S pre-ribosome and SSU processome complexes; involved in production of 18S rRNA and assembly of small ribosomal subunit; synthetic defect with STI1-Hsp90 cochaperone; human homolog linked to glaucoma; Small Subunit processome is also known as SSU processome.
4. YPL090C (RPS6A - mutated twice): Protein component of the small (40S) ribosomal subunit; homologous to mammalian ribosomal protein S6, no bacterial homolog; phosphorylated on S233 by Ypk3p in a TORC1-dependent manner, and on S232 in a TORC1/2-dependent manner by Ypk1/2/3p; RPS6A has a paralog, RPS6B.
5. YBL072C (RPS8A - mutated once): Protein component of the small (40S) ribosomal subunit; homologous to mammalian ribosomal protein S8, no bacterial homolog; RPS8A has a paralog, RPS8B, that arose from the whole genome duplication
6. YMR128W (ECM16- mutated once): Essential DEAH-box ATP-dependent RNA helicase specific to U3 snoRNP; predominantly nucleolar in distribution; required for 18S rRNA synthesis.

The sub-networks indicated with the blue and red boxes are also the sub-networks that receive the highest edge width (in the online visualization) or lowest opacity in the html offline visualization.

Some of the infrequently mutated genes (those that received e.g. rank 6), that appear to have been mutated only once in the population and that are not located in a cluster of frequently mutated genes are likely passengers. At the least stringent cost, passengers can get connected spuriously.

The example shows that IAMBEE finds as a recurrently mutated sub-network a set of genes that is involved in 90s preribosomal pathway which is in line with the results described in the original paper. Note that some of the genes described in the paper of Jerison et al could not be recovered in this run, mainly because they were not connected in the network. Therefore it is important to ensure that the ‘mutated genes’ are connected in the network, otherwise one might not be able to recover them.