Identification of differentially expressed genes
We used the A-C statistical test to identify differentially expressed genes in 94 normal human and 99 normal mouse tissues (see website) using p value thresholds of 5E-2, 5E-3, 5E-4, 5E-5, and 1E-6, which are used hereafter to measure the extent of differential expression. As expected, the number of differentially expressed genes decreased as we lowered the p value threshold (Fig. 1). Table 1 further shows that most genes were expressed in only a few tissues: at p
Comparison with microarray and published results
To evaluate our results, we compared the human genes identified as differentially expressed using different p values with the microarray data provided by Su et al. [23] for 17 tissues (Fig. 2). For the comparison, we first identified, for each tissue, differentially expressed genes under a specified p value cut-off, and extracted those that were also included in the microarray experiments. Of these extracted genes, we identified those showing over-expression (i.e., 3-fold higher than the median expression) in the microarray experiment. We then calculated the percentage of genes identified as differentially expressed under various p-value thresholds in our data that were also over-expressed in the microarray experiment. The results showed that, for all 17 tissues, this percentage increased as more genes without significant tissue specificity were filtered out, i.e. as the p value threshold was set lower. Furthermore, at a threshold of 1E-6, this percentage varied significantly across tissues, ranging from less than 10% for ovary and skin to ~60% for liver and three brain tissues. These results are consistent with an earlier analysis showing that the correlation between microarray and EST data for genes differentially expressed in brain (r2 = 0.43) is much higher than that for genes differentially expressed in the pancreas (r2 = 0.02) or ovary (r2 = 0.03) [24].
To further evaluate the usefulness of our work, we compared our results with published data for several known tissue-specific genes. KLK3, TMEM10, and AMBP are three notable examples. KLK3, a member of the kallikrein gene family, is prostate-specific [25]. In our analysis, KLK3 was identified in the prostate with a very high specificity (p TMEM10, a recently reported novel human brain-specific gene [26], was also found to be specifically expressed in the forebrain (p = 2.57E-27), whole brain (p=9.49E-20), hippocampus (p = 8.77E-10), and hypothalamus (p = 2.43E-08). Alpha-1-microglobulin/bikunin precursor°(AMBP) is a well known gene exclusively expressed in liver both in human and mouse [27,28], and our data showed that AMBP was expressed with very high specificity in the liver (p [29] and all of the human brain-specific genes reported by Huminiecki et al. [24] were found to show the same tissue specificity (p
Correlation analysis of human and mouse orthologous genes
Of the 10,307 human and mouse orthologous gene pairs downloaded from the NCBI HomoloGene database, 7,853 contained sufficient EST data to qualify for the A-C test; 1,268 of these were expressed in fewer than 3 tissues and were therefore excluded from the p value correlation analysis, as described in the Methods. For the remaining 6585 gene pairs, the average p value correlation coefficient was only 0.20 (Table 2). Of the 6585 gene pairs, we further extracted genes differentially expressed, as defined by a given p value threshold, in at least one tissue in human and also one tissue in mouse. That is, for example, when the threshold was set at 1E-6, genes expressed in at least one human tissue and one mouse tissue with p 2, as the threshold for defining tissue-specific orthologs was lowered, the correlation became better, and orthologs differentially expressed in at least one tissue with p
The correlation coefficient measures the extent to which the human and mouse orthologous genes show the same tissue specificity. We further dissected the strength of this association for those orthologs with significant tissue specificity, i.e., those expressed in at least one tissue with p 3, the results demonstrated that 40% of the qualified gene pairs showed a very high positive correlation (r ≥ 0.8), indicating that these orthologs exhibited very similar expression patterns in human and mouse. The pair showing the strongest correlation (r = 0.92) was human PCDH8 and its mouse counterpart, Pcdh8, both of which were found to be expressed in cerebrum, forebrain, hypothalamus, hippocampus, and whole brain (Fig. 3A). This result is in agreement with previous experimental findings that this gene is expressed predominantly in the brain in both human and mouse [30].
Another example is IL2RG, which is reported to be essential for the development of T and NK lymphocytes and mutation of which can cause severe combined immunodeficiency disorder (SCID) [31]. We found that human IL2RG and its mouse ortholog Il2rg were both expressed in 13 tissues with highly similar tissue specificities (r = 0.9), and, in accordance with their function [31], were preferentially expressed in T cells, lymphocytes, leukocytes, and whole blood (Fig. 3B).
In contrast, very strong negative correlations of orthologous genes (r ≤ -0.8), indicative of different tissue specificities, were found in three pairs (Table 3). The strongest negatively correlated pair (r = -0.99) was human KIAA0748 and its mouse counterpart, 5830405N20Rik. As shown in Fig. 4A, they were both expressed in blood, leukocytes, lymphocytes, thymus, and T cells, but a large discrepancy was seen in terms of thymus specificity (Fig. 4A), which accounted for the negative correlation. Interestingly, this thymus discrepancy is supported by microarray data [32], which show significantly upregulated expression of 5830405N20Rik (Fig. 4C), but not of KIAA0748 (Fig. 4B). The second most strongly negatively correlated ortholog pair was human MS4A1 and its mouse counterpart, Ms4a1 (r = -0.96), both of which were preferentially expressed, though to different extents, in B cells, lymphocytes, and leukocytes (Fig. 5A), these preferred tissues were in agreement with reported expression results [33] and with microarray data [32]. The main discrepancy was in the blood vessels, for which mouse Ms4a1 exhibited a much higher specificity (p = 2.12E-30) than human MS4A1 (p = 6.42E-02) (Fig. 5A). This discrepancy could not be checked by microarray data, as the blood vessel was not examined in the microarray experiments [32], nor could we find any previous reports on the expression of these two genes in blood vessels. The third most strongly negatively correlated pair was human SLC2A6 and mouse Slc2a6 (r = -0.87), both of which were preferentially expressed in macrophages, but their brain specificity differed significantly (Fig. 5B). In this case, the data from the two reports on expression of human SLC2A6 in brain [34,35] are contradictory and thus could not be used to assess the EST results.
Answers to all your Biology Questions
