Characterizing the expression of the human olfactory receptor gene family using a novel DNA microarray.

BACKGROUND: Olfactory receptor (OR) genes were discovered more than a decade ago, when Buck and Axel observed that, in rats, certain G-protein coupled receptors are expressed exclusively in the olfactory epithelium. Subsequently, protein sequence similarity was used to identify entire OR gene repertoires of a number of mammalian species, but only in mouse were these predictions followed up by expression studies in olfactory epithelium. To rectify this, we have developed a DNA microarray that contains probes for most predicted human OR loci and used that array to examine OR gene expression profiles in olfactory epithelium tissues from three individuals.
RESULTS: We detected expression of 437 (76%) human OR genes in these olfactory epithelia. Interestingly, we detected widespread expression of OR pseudogenes, an observation that may shed light on the mechanism of OR gene choice in the olfactory sensory neurons. To address the hypothesis that OR genes may carry out additional functions, we also characterized the expression of OR genes in a number of non-olfactory tissues.
CONCLUSION: While our results corroborate the functional annotation of the majority of predicted human odorant receptors, we find that a large number of putative human OR genes are expressed in non-olfactory tissues, sometimes exclusively so. Our evolutionary analysis of ectopically expressed human OR genes does not lend support to the hypothesis that these genes have alternative functions.

Background

Buck and Axel [1] identified the odorant receptor (OR) gene family based partly on the observation that OR genes were expressed in olfactory epithelium, but were not detected in lung, liver, spleen, kidney, retina, and brain. Subsequently, additional OR genes were recognized in genomic sequences by their similarity to the first set of identified OR genes [2,3], and by the presence of certain predicted protein motifs [1,4]. Recently, the complete genomic sequence of a number of mammalian species became available, permitting interspecies comparisons of complete OR gene repertoires. The analyses of published mammalian genomes suggested that the original estimate of the size of the mammalian OR gene repertoire, approximately 100 genes, was a severe underestimate. Indeed, it is now thought that mammalian genomes carry 800-1,400 OR genes [5-10], which are typically organized in gene clusters and are found on many chromosomes. With roughly 3% of all genes coding for odorant receptors, OR genes are by far the largest gene family in mammalian genomes.

To date, however, mammalian OR genes have remained largely orphan receptors. In fact, until recently, there was no systematic study of putative mammalian OR gene expression in olfactory epithelium [7,11], such that the functional annotation of OR genes remained unclear. Moreover, expression of several predicted mammalian OR genes was detected only in non-olfactory tissues, notably in testis [12-14]. These observations raised the possibility that a subset of predicted OR genes may not be odorant receptors at all but have other functions, with important implications for functional studies in olfaction and comparisons of mammalian OR gene repertoires. Alternatively, OR genes may have a function beyond odor recognition, for example, in sperm chemotaxis [15].

Recently, Zhang et al. [16] studied the expression of nearly all predicted OR genes in mouse using a newly developed DNA microarray [16]. Most (approximately 80%) predicted mouse OR genes were confirmed to be expressed in olfactory epithelium, but a subset were found to be expressed only in non-olfactory tissues and, consequently, their functional annotation is now in question [16]. In humans, it is not known how many of the predicted OR genes are expressed in the olfactory epithelium, and hence how many are likely to participate in odorant binding. Moreover, the predicted human OR gene repertoire includes nearly 600 pseudogenes [5] and it remains unknown how often they are expressed. Since olfactory sensory neurons are believed to express only a single functional OR gene, if these pseudogenes are routinely expressed in the olfactory epithelium, a large proportion of neurons may either express a single non-functional gene, or co-express a functional and non-functional OR genes [17].

A recent study [14] used expressed sequence tag (EST) data and results of genome-wide microarrays to survey human OR gene expression in olfactory epithelium and several non-olfactory tissues. However, that analysis was limited by shortcomings of the available data, including biases and inaccuracies in the EST databases and incomplete sampling of OR genes on the human genome-wide microarray (which includes probe-sets for only 356 predicted OR genes and pseudogenes). Moreover, the genome-wide microarray was not optimized specifically to measure OR gene expression, so many of the probes may be susceptible to cross-hybridization by other OR genes [14]. Indeed, the authors' analysis of the probe-set sequences suggested that the expression of only 217 human OR genes and pseudogenes could be estimated with confidence using the genome-wide microarray data [14].

To comprehensively and reliably assess expression of predicted human OR genes, we designed a new microarray with probes for nearly all human OR genes. We used this microarray to characterize the expression of human OR genes in olfactory epithelium as well as in a number of other tissues.

Results and discussion

To measure the expression of human OR genes, we extracted total RNA from three samples of human olfactory epithelium tissues collected by the National Disease Research Interchange)[18] within eight hours of the donor's death. We confirmed that RNA was extracted from olfactory epithelium tissues by amplification of the odorant binding protein 2B (OBP2B) gene (Figure 1), which is expressed exclusively in olfactory epithelium [19]. In addition, we tested for the presence of olfactory sensory neurons in each sample by amplifying the olfactory sensory neuron marker gene, the olfactory marker protein (OMP) [20]. Once we confirmed the source of the RNA, we proceeded by labeling and hybridizing each olfactory epithelium RNA sample, in two independent technical replicates, to a custom human OR gene microarray (see Materials and methods). Similarly, we hybridized RNA from human liver, lung, kidney, heart, and testis (purchased from Ambion (Austin, TX, USA)) to the microarray in two technical replicates each.

Figure 1

Gel electrophoresis of PCR amplification results using cDNA from three olfactory epithelium tissues, heart, testis, liver, lung, and kidney as template. (a) The 440 base-pair (bp) OBP2B product was only amplified from the olfactory epithelium samples. (b) The 562 bp cathepsin C (CTSC) product was successfully amplified from all three samples. Both primer pairs were designed to amplify multiple exon products and hence are expected to yield a much larger product (1,686 bp) if genomic DNA was used as template. (c) The 378 bp product of the OMP gene was amplified from the olfactory epithelium samples to confirm that these samples contain neutrons.

Expression of OR genes in human olfactory epithelium

Our first goal was to detect which of the predicted human OR genes are expressed in olfactory epithelium, thereby lending support to their functional annotation as odorant receptors. One way to examine this is to rely on the absent/present calls that the Affymetrix software provides for each probe-set. However, microarrays were not designed to detect expression but rather to compare levels of expression between samples or treatments and, as a result, existing algorithms are not well suited for our application [21,22]. In particular, as probes vary in their specificity and sensitivity, using cut-offs for absolute hybridization intensity as a detection tool is unreliable [23-25]. An alternative is to use a comparison of gene expression levels across the studied RNA samples in order to detect genes that are expressed [16]. The rationale of this approach is that genes with significantly higher expression in sample A compared to sample B are clearly expressed in A. Thus, OR genes with significantly elevated expression in olfactory epithelium compared to other tissues can be considered 'detected'. Accordingly, we compared expression of OR genes in the olfactory epithelium samples with their expression levels in the five non-olfactory tissues (see Materials and methods).

We found 437 predicted OR genes on the array to be expressed in human olfactory epithelium at P < 0.05 (Figure 2a; Additional data file 1), when at most 29 are expected by chance, given the statistical cutoff and the number of genes on the array (see Table 1 for results using alternative statistical cutoffs). These results were validated by performing RT-PCR on ten randomly chosen OR genes whose expression was detected in olfactory epithelium (Additional data file 1); in all ten cases, the RT-PCR confirmed the arrays results. Thus, we confirmed that the vast majority (76%) of predicted human OR genes are indeed expressed in olfactory epithelium. In contrast, the functional annotations of 141 predicted human OR genes are now in question, as these were not detected as expressed in olfactory epithelium. We note, however, that 109 of the above 141 OR genes did not have significantly elevated expression levels in any tissue (Additional data file 1). Since our detection criterion is based on differential expression, we cannot exclude the possibility that a subset of the 109 OR genes were not detected because they are expressed at similar levels in all tissues, including olfactory epithelium [16].

Figure 2

Expression profile of human OR genes across tissues. The log transformed detection P values for OR genes in all tissues (from Additional data file 1) were standardized to have mean 0 and standard deviation 1 and are color coded (red and blue shades indicate values above and below the mean, respectively). The dendrograms on top of each panel illustrate the clustering (by hierarchical clustering in dchip [48]) of tissue samples based on the profile of OR gene expression. (a) All 578 predicted OR genes are included in a comparison between olfactory epithelium (OE) and the non-olfactory tissues (see Materials and methods). (b) Shown are the data for only the 147 OR genes with significantly elevated expression in non-olfactory tissues.

Table 1

Number of expressed OR genes in human olfactory epithelium

P value*	No. of detected OR genes and pseudogenes	Only intact†	Only pseudogenes†
0.001	192 (33%)	131 (36%)	61 (29%)
0.01	342 (59%)	232 (63%)	110 (52%)
0.05	437 (76%)	295 (80%)	142 (67%)

*The statistical cutoff used to identify OR genes as expressed. †The number and percentage (in parenthesis) of intact OR genes and pseudogenes on the array detected as expressed. The array includes probe-sets for 578 predicted human OR genes.

Previous studies also noted anecdotal expression of individual human OR pseudogenes [11,14]. Of the 212 human OR pseudogenes on the microarray, 142 (67%) were found to be expressed in olfactory epithelium (Additional data file 1). While considerable, this fraction is significantly lower than the fraction of intact OR genes, 295 out of 366 (80%), found to be expressed in olfactory epithelium (Fisher Exact test, P < 10-3; see Table 1 for results using alternative statistical cut-offs). Moreover, intact OR genes appear to be expressed at a higher level on average than OR pseudogenes (assuming that the probe effect is canceled in large samples, such that we can compare hybridization intensity across groups of probe-sets, the mean difference in normalized intensity is +16%; Mann-Whitney-U, P = 0.005). The more frequent and greater expression of genes relative to pseudogenes is consistent with recent observations in mice (XZ and SF, manuscript in preparation). The observation of widespread OR pseudogene expression in this study, as well as a number of others [11,14], suggests that a nonsense-mediated decay RNA system (reviewed in [26]) does not efficiently remove OR pseudogene mRNAs.

On our array, we included only OR pseudogenes with one or two premature coding region disruptions. These ORs are likely to be recent pseudogenes, so that expression in olfactory epithelium may not be unexpected. This said, the observation of widespread OR pseudogene expression has implications for the outstanding question of how OR gene expression is regulated in olfactory sensory neurons. A common model is one in which mature olfactory sensory neurons are assumed to have a cellular mechanism that restricts the expression to only one OR gene [27,28]. This model predicts that a neuron expressing a pseudogene would eventually switch to express a functional gene [27,28], while a functional and an OR pseudogene would rarely be co-expressed. Thus, if OR pseudogene expression is widespread, as our observations suggest, this implies that, at any given time, a large proportion of neurons will not express functional genes and, thus, will not contribute to the sense of smell. This prediction is consistent with the small numbers of OR pseudogenes found in species that rely heavily on their sense of smell [29].

An alternative model is that expression of OR pseudogenes in olfactory sensory neurons occurs with the same probability as expression of intact OR genes, but that the neurons expressing only a non-functional OR gene do not converge in the olfactory bulb, never reach maturation and are removed [17], while neurons that co-express both a functional gene as well as a pseudogene will converge in the olfactory bulb [17]. This model does not rely on a cellular mechanism that restricts the expression to only one OR gene. Accordingly, this model predicts that pseudogenes can be either co-expressed with a functional gene, or are limited to young neurons that are removed. This prediction is consistent with our observations of a lower proportion and weaker expression of OR pseudogenes compared to intact genes. If true, it would suggest that a considerable proportion of human olfactory sensory neuron cells co-express a functional as well as an OR pseudogene.

Inter-individual variation in OR gene expression

Recent work suggested that the extensive genetic variation in human OR protein coding regions (in particular, segregating null mutations) may account for inter-individual variability in the sense of smell [30,31]. Here, we also find evidence that the repertoire of expressed OR genes varies across individuals (Figure 3). Although our sample of only three olfactory epithelium tissues is insufficient to quantify the variability in human OR gene expression, it suggests that such variability is abundant. Regardless of the statistical cutoff (10-5