Correlation of microsynteny conservation and disease gene distribution in mammalian genomes

Microsynteny conservation of mouse autosomes

We evaluated the level of microsynteny conservation between the mouse genome and those of human, dog and rat. First, we obtained all protein-coding genes and their genomic locations on all mouse autosomes as annotated in the Ensembl mouse genome browser [3, 31] (release 50). We also obtained protein-coding genes from the human [32, 33], rat [2], and dog [34] genomes. These were chosen because they had a sufficient level of assembly and annotation to allow comparison. The use of the dog genome as an outgroup to human, rat, and mouse improves the stringency of the study [35]. To identify orthologs of mouse genes in the other genomes, along with their genomic locations, Ensembl BioMart homology filters were used to compile a list of orthologous genes.

Although the four mammalian genomes chosen are those with the best available annotation, the degree and quality of annotation may vary somewhat between species. In order to control for this we took additional steps to find the human, rat, and dog orthologs of mouse genes. Protein sequences of all mouse genes that did not have an annotated ortholog in another species were searched using BLAT [36] against the other genomes to identify orthologous sequences in the other genomes. To allow a moderately strict search with a limited number of false positives, all hits with E-value < 10^-5 were retrieved. The genomic location of the best BLAT match in the other genome was used for the evaluation of microsynteny conservation. We searched a total of 6173 genes in at least one other species, finding an ortholog for 1210 of the genes in other genomes. A sensitivity analysis demonstrates that the number of genes retrieved from the BLAT searches is relatively insensitive to the choice of E-value cut-off, as changing the cut-off point from 10^-3 to 10^-7 results in 1266 - 1169 ortholog annotations respectively. Therefore, utilizing alternate E-value cut-off points in this range would have changed the annotation of only 0.48% of the total mouse genes analyzed in our study.

Genes were defined as having conserved microsynteny if their orthologs had the same two orthologous neighboring genes in all four species examined. Each mouse gene was queried to determine whether it met these criteria for conserved microsynteny. We then assessed the level of conservation for segments of the mouse genome, determining the percentage of conserved genes in each segment. We examined 20 Mb regions of the mouse genome, as genomic regions of this size have been analyzed experimentally through region-specific mutagenesis [37–40]. Thus, the identification of additional genomic regions with conserved microsynteny will be useful for further experimental functional genomic annotation.

We found that the conservation of microsynteny varied throughout the mouse genome. The average percentage of genes with conserved microsynteny in a 20 Mb interval was 38.29%, with a standard deviation of 11.81%. The Shapiro-Wilk test was applied to non-overlapping 20 Mb windows to determine whether the distribution of gene microsynteny was normal. A P-value of 0.16 indicated that the null hypothesis (a normal distribution) should not be rejected. However, the use of non-overlapping 20 Mb windows restricted the resolution of the study. For example, chromosome 19 has only two observations from non-overlapping windows. To improve the resolution of our study, we next examined 20 Mb intervals staggered by 5 Mb. This sliding window analysis allowed more observations on each chromosome.

As the data conformed to a normal distribution, we therefore calculated Z-scores (number of standard deviations above or below the mean) for each 20 Mb sliding window. Windows with Z>1 were considered to have increased conservation, those with Z<-1 were considered to have decreased conservation, and windows with 1>Z>-1 had intermediate conservation. There were 51 sliding windows of the mouse genome found to have Z>1, indicative of higher microsynteny conservation, and 91 intervals found to have Z<-1, indicative of lower microsynteny conservation. Three hundred twenty-two genomic regions demonstrate intermediate microsynteny conservation with scores of 1>Z>-1. On individual chromosomes there is variation in the conservation of microsynteny, with most chromosomes containing both windows of increased microsynteny conservation and windows of decreased microsynteny conservation (Figure 1 black lines). However, mouse chromosomes 5, 7, 8, 10, 13, 14, 16 and 18 do not contain any regions of increased (Z>1) microsynteny conservation. We found that the percentage of genes with conserved microsynteny per chromosome also showed variation, with chromosome 13 having the lowest percentage of conserved genes at 25%, and chromosome 15 having the highest percentage of conserved genes at 48% (Table 1). Previous work has shown that syntenic conservation is not simply related to gene density [29].

Comparison with sequence-based synteny blocks

Conserved genes are located next to other conserved genes

Although there is variation in the density of genes with conserved microsynteny across the genome, it is possible that this variation merely represents random variation within a normal distribution. To determine whether the genomic arrangement of genes with conserved microsynteny is random, we calculated the likelihood that a gene with syntenic conservation is found next to another gene with syntenic conservation. We then compared this to the frequency of conserved-synteny neighbors in a set of 10,000 randomized genomes. In each of the randomized genomes the number and position of genes is maintained, but the annotation of conservation is randomly shuffled.

We find that the frequency of co-occurrence of genes with conserved microsynteny is significantly non-random (Figure 2A, P < 0.003). By our definition, for a gene to have conserved microsynteny it must have two orthologous neighbors in all genomes examined. Thus, a pair of genes with conserved micro-synteny represents a larger block of conserved synteny. The finding that there are conserved microsynteny blocks in the genome that extend beyond groups of several genes suggests that there are constraints on genome evolution that influence gene arrangement, as the placement of conserved genes significantly differs from a random distribution.

Distribution of mouse orthologs of human disease-related genes

As we found that conserved genes were more likely to have conserved genes as neighbors, we investigated whether any other groups of genes were found preferentially in regions of the genome with conserved microsynteny. One group of genes that are of interest is disease genes, as they are highly relevant to human health. We therefore performed a genome-wide analysis of the mouse orthologs of human disease-related genes to assess whether they were found at a greater density in conserved regions of the mouse genome. We identified human genes with a disease-associated mutant allele from the OMIM Morbid Map database, and cross-referenced them to the mouse genome using Ensembl BioMart to identify mouse orthologs. Using the genomic locations of the mouse orthologs of human disease genes from our study on microsyntenty conservation, we determined the proportion of human disease gene orthologs in each 20 Mb sliding window of the mouse genome. The mean percentage of disease-related gene orthologs as compared to the number of total genes in a 20 Mb interval is 7.68%, with a standard deviation of 2.72%. We found variation in the distribution of disease gene orthologs in the mouse genome, with the highest percentage found on chromosome 18, and the lowest on chromosome 7 (Table 1). However, on all other autosomes, 7 - 8% of the genes are orthologs of human disease genes.

Disease gene orthlogs are located next to other disease gene orthologs

As we had found that the distribution of genes with conserved microsynteny is non-random, we examined whether that was also true for genes with disease-related orthologs. Using a similar approach, we calculated the number of mouse orthologs of human disease genes with at least one disease gene ortholog as a neighbor. As a control, we randomized which genes were annotated as disease orthologs, keeping the same total number of disease gene orthologs. From 10,000 random trails we found that the mouse orthologs of human disease genes were significantly more likely to have other disease gene orthologs as neighbors (Figure 2B, P < 0.07). This finding demonstrates that the distribution of orthologs of human disease genes in the mouse genome is not random.

Correlation between microsynteny conservation and disease gene distribution

We next assessed whether the orthologs of disease genes were located in regions of the mouse genome with increased microsynteny conservation. We detect a correlation between regions of the genome with conserved microsynteny and the distribution of disease gene orthologs over the whole genome (Figure 3A, Pearson's R = 0.90, P < 1 × 10^-6). Such a representation over-estimates the true correlation between the two sets, since gene density varies considerably in different windows. This confounds the analysis as regions with high gene density would be expected to have both high numbers of disease orthologs and high numbers of genes with conserved microsynteny regardless of whether there is an additional correlation between microsynteny conservation and the presence of disease gene orthologs. When corrected for gene density, a significant correlation between microsynteny conservation and disease gene ortholog density is still observed (Figure 3B, Pearson's R = 0.40, P < 4.0 × 10^-4).

The density of disease gene orthologs for each genomic region is shown in Figure 1 (blue lines). Z-scores are displayed to allow direct comparison between microsynteny conservation and disease orthologs. The additional calculation of Z-scores does not change the overall correlation. There is a significant correlation (P = 0.05) between microsynteny conservation and the density of disease gene orthologs for 12 of the 19 mouse autosomes. Thus, genomic regions with a high percentage of genes with conserved microsynteny also have a high percentage of disease gene orthologs. The chromosome with the best correlation between conserved microsynteny and density of disease gene orthologs is mouse chromosome 13, while the chromosome with the worst correlation is mouse chromosome 10.

To demonstrate that this correlation was not an artifact of our analysis, we randomized the annotation of disease genes. We assigned alternate genes as orthologs of human disease genes, keeping the total number of disease genes per chromosome the same as the first analysis. We then recalculated the percentage of alternate disease genes as compared to total genes in each sliding window throughout the genome, and the average and standard deviation for each sliding window. We plotted the Z-scores for each window containing these alternate disease orthologs, and compared them to the Z-scores for microsynteny conservation (Figure 1 red lines). When the chromosomal positions of orthologs of disease genes are changed to random locations, the correlation with microsynteny disappears (Pearson's R = 0.02, P < 0.58). As an additional control, we also randomized disease genes while retaining the same number of observed disease gene pairs for each chromosome. Again, we found no correlation (Pearson's R = 0.004, P < 0.93). Should the correlation between observed disease gene ortholog distribution and microsynteny conservation be an artifact of our methodology, we would also expect the randomized annotations to be correlated. This is not the case, demonstrating that the link between microsynteny correlation and density of disease gene orthologs does not arise from an artifact of the methodology.

Robustness to changes in window size

Microsynteny conservation

All protein coding genes on mouse autosomes were retrieved using a BioMart search on the Ensembl Genome Browser (http://www.ensembl.org/index.html, release 50, NCBIM37 dataset). Orthologs of these genes in the human, rat, and dog genomes, along with their genomic positions, were retrieved using Ensembl BioMart homology filters. Any mouse genes that did not have an ortholog in another genome identified in BioMart were then subjected to a BLAT search http://genome.ucsc.edu/cgi-bin/hgBlat?command=start on the other genome using the mouse protein sequence retrieved with the BioMart sequences filter. The genomic location of the best match to the mouse protein sequence from the second genome was retrieved. Locally developed software was used to identify genes with conserved microsynteny. Each mouse gene is analyzed in turn. For each mouse gene m, its neighbors in the mouse genome (m-1 and m+1) are identified. The orientation of m-1, m and m+1 is determined from the start and stop positions. In the dog, human and rat genomes, the homologue of m is identified from both Ensembl BioMart homology filters and BLAT searches. We term these homologues d, h and r. For each of these genes their neighbors in their respective genomes (d-1, d+1, h-1, h+1, r-1 and r+1) are identified, and their orientation determined. The gene m is defined as having conserved microsynteny only if all of the following are true: (i) m, d, h, and r comprise a homologous set of genes (ii) m-1, d-1, h-1 and r-1 comprise a homologous set (iii) m+1, d+1, h+1 and r+1 comprise a homologous set (iv) the m, d, h, and r have the same orientation (v) m-1, d-1, h-1 and r-1 have the same orientation and (iv) m+1, d+1, h+1 and r+1 have the same orientation. Results are robust whether or not we limit the definition of conservation only to those genes close to each other in the genome (within 200 Kb). Results presented have conservation limited in this manner.

Once conserved microsynteny had been defined for each gene the percentage of genes with microsynteny was determined for 20 Mb intervals of the mouse genome, staggered by a 5 Mb sliding window. To define the percentage of microsynteny for a given window, the number of genes with conserved synteny was divided by the total number of genes in that window. The position of each gene was chosen as the start site listed in Ensembl. For genes with multiple transcripts, the position of the start site of the longest transcript was used. The Shapiro-Wilk test was applied to non-overlapping 20 Mb regions of the mouse genome to determine whether the distribution gene microsynteny was normal. Z-scores were calculated for each region of the mouse genome using the equation:

Where x is the proportion of genes with conserved synteny within each window, μ is the mean proportion of genes with conserved synteny in all windows and s is the standard deviation of genes with conserved synteny in all windows. Thus the Z-score represents the number of standard deviations above or below the mean.

Comparison with sequence-based synteny blocks

Each highly conserved mouse genomic region was compared to synteny maps based on sequence alignment. Maps of the mouse compared to dog, rat, and human were retrieved from the Ensembl genome browser (Comparative Genomics - Synteny, http://www.ensembl.org/Mus_musculus/Location/Synteny) database. Genomic positions of synteny blocks were retrieved from the database and tabulated.

Statistical analysis of conserved and disease gene pairs

Genes annotated as having conserved synteny or as disease orthologs were examined to determine whether the following gene on the chromosome was similarly annotated. The number of such gene pairs was determined per chromosome. By this definition a run of three genes with similar annotation would be counted as two pairs.

To determine the significance of the frequency of such gene pairs, we compared the observed value to a random distribution. The random distribution was calculated by keeping the both the total number of genes and the number of genes with a given annotation constant, but randomizing the assignment of those annotations over the total number of genes. The annotations were randomized 10,000 times, and the number of neighbors with similar annotations calculated. The significance of the observed value was determined from this simulation i.e. the likelihood of the observed value is the proportion of random values that are greater than or equal to the observed value.

Disease-related ortholog distribution

Human genes with a disease-associated mutant allele were identified from the OMIM Morbid Map database http://www.ncbi.nlm.nih.gov/Omim/getmorbid.cgi. These were cross-referenced to the mouse genome using Ensembl BioMart homology filters to identify mouse orthologs. The distribution of disease-related gene orthologs was analyzed by sliding window analysis, in the same manner as for conserved synteny. For each 20 Mb window the number of disease gene orthologs was divided by the total number of genes in a window to determine the proportion of disease gene orthologs in the window. Z-scores were then calculated for each window using the equation above.

Correlation analysis

The Pearson's correlation between the microsynteny conservation Z-scores and disease-related ortholog Z-scores was calculated, and significance calculated using the chi-squared test. As a control, we assigned new disease genes at random for each chromosome, keeping the total number of disease genes per chromosome the same. We then recalculated the density of disease genes in each sliding window throughout the genome. We plotted the Z-scores for each window containing these alternate disease orthologs, and compared them to the Z-scores for microsynteny conservation. The Pearson's correlation was then re-calculated, showing no significance for the alternate set of disease genes. Correlation analysis was also performed on window sizes of 10 Mb, staggered by 2.5 Mb, 5 Mb, staggered by 1.25 Mb, 2 Mb, staggered by 0.5 Mb, and 1 Mb, staggered by 0.25 Mb. Randomization trials were performed on 10,000 random annotations for each alternate window size as a control. To account for windows with no genes at the window sizes of 2 Mb and 1 Mb, the correlation analysis was repeated with all windows containing 0 genes removed from the actual and randomized datasets.