Canine OR sequences were cloned following PCR amplification of dog genomic DNA with a set of primers designed to amplify various regions between TM3 and TM7. Using five different primer pairs, 774 clones were obtained. Sequencing showed a certain level of redundancy, with only 190 unique sequences, of which 10 correspond to OR genes previously identified 5923. A consensus sequence was derived from an alignment of all known human OR sequences, and was used to screen a database of sequences derived from 1.2-times coverage of the dog genome. Fragments of 737 unique OR genes were identified, of which 100 corresponded to PCR-screened ORs. This combination of approaches has provided sequence information for 838 unique canine OR genes.

Specific markers could be defined for 562 OR sequences that were mapped on the RHDF5000 panel. Two-point analysis, computed using the MultiMap software on the latest RH map containing 3,270 markers 29, provided lod-scores with neighboring markers. Details are available on the website 30 and can also be found in Additional data file 1. As shown in Figure 1, the dog OR sequences are spread over 24 of the 40 chromosomes (see Additional data file 3). Three RH groups, containing 2, 2 and 11 OR sequences, were not attributed to any chromosome, and eight OR markers remained unlinked. The mapped OR genes are distributed over 37 chromosomal regions, in 33 clusters of 2-124 genes, and four isolated OR genes. Two canine chromosomes (17 and 29) contain only one OR gene, 20 canine chromosomes contain 3 to 65, while two canine chromosomes (18 and 21), contain 124 and 109 OR genes respectively, and constitute the two largest clusters of OR genes in the dog genome. These two chromosomes alone contain 41% of the mapped canine OR sequences.

Following careful visual inspection of sequencing traces, translation of OR nucleotide sequences showed an uninterrupted open reading frame (ORF) for 417 of 661 fully characterized OR sequences. For the other 244, apparent insertions, deletions or substitutions caused frameshift mutations and/or in-frame stop codons. However, by sequencing PCR products of 142 putative pseudogenes, each derived from several DNA sources, 75 were confirmed as pseudogenes, and 67 were reclassified as genes. For the remaining 102 sequences, the apparent mutations could not be checked owing to their proximity to the sequence ends. Therefore, the status of these sequences could only be inferred.

Among the sequences harboring a single mutation, only 30% were confirmed as pseudogenes. Sequences with two, or more than two, mutations were confirmed as pseudogenes in 72% and 97% of cases respectively. For the 102 unverified sequences, 81 have a single mutation, 12 have two mutations, and 9 have more than two mutations. Considering the percentages found on checked sequences, the number of true pseudogenes in these 102 sequences is estimated to be (81 × 30%) + (12 × 72%) + (9 × 97%) = 42. This figure is likely to be an overestimate, as apparent mutations that are close to the ends of sequences are more likely to be due to sequencing errors than those that are detected in the middle of the sequencing traces.

Nevertheless, a conservative estimate for the number of pseudogenes in the fraction of analyzed ORs is 117 (75 + 42), or approximately 18%. The OR sequences that were screened for mutations represent only half of the complete ORFs. Consequently, the presence of additional mutations outside the sequenced regions cannot be excluded. However, as a large proportion of pseudogenes tend to contain several mutations that are evenly distributed along their ORF, the probability of mutation(s) located only in the non-sequenced regions is low. Indeed, for 54 full-length OR sequences that were checked by resequencing, we found only six pseudogenes (11%), with numerous mutations spread throughout the ORF. In summary, the value of 18% pseudogenes is considered to be a conservative estimate.

For 403 OR canine sequences, the regions encompassing TM2 and EC2 (140 amino acids) were aligned, and a phylogram was constructed (30 and see Additional data file 4). This identified 51 families and 202 subfamilies. In common with human and mouse, the canine OR families are separated into two classes. Classes I and II are composed of 10 and 41 families respectively. Each family contains between 1 and 35 genes, representing between 1 and 11 subfamilies.

We retrieved 906 human OR sequences from the HORDE38 version 20. Of these, 714 sequences, spanning the TM2 to EC2 regions, were aligned and used to construct a phylogram (30 and see Additional data file 5) using the same method as for the canine OR sequences. We identified 61 families, separated in two classes as expected 19; 10 families for class I, and 51 families for class II. Families contain between 1 and 93 OR sequences, and are divided into a total of 285 subfamilies. Interestingly, 111 subfamilies contain only pseudogenes.

The present study showed that, in the fraction of canine ORs analyzed, 18% are estimated to be pseudogenes - a similar percentage to that found in the mouse OR repertoire (20%) 22 but much lower than the 63% calculated from the HORDE 39 version 20 for the human repertoire (our present work). This feature could at least partly explain the difference in olfaction between macrosmatic and microsmatic species.

A detailed analysis of the distribution of the pseudogenes throughout the canine genome showed that they are spread over all chromosomes harboring OR sequences. In contrast to a rather even distribution of pseudogenes within most of the clusters, the two largest clusters on canine chromosomes 18 and 21 harbor only 4% and 10% pseudogenes, respectively. In comparison, such a low fraction of pseudogenes was not observed in the largest orthologous mouse clusters 22. At the family level, a rather uneven distribution of pseudogenes was noted. Three families (named 16, 26 and 38) contain up to 43% pseudogenes, while eight other families (named 6, 11, 18, 27, 32, 41, 42 and 51) harbor only 5% pseudogenes.

In humans, an overall even distribution is observed: small clusters (< 10 genes) and isolated genes (totaling approximately 300 sequences) have a high level of pseudogenes (78%), whereas the largest clusters on human chromosome 11 contain 56% pseudogenes (compared with a mean value of 63% for the whole repertoire). The pseudogene distribution within families is relatively uniform, except for family 41 (named family 7 by Glusman et al. 19), which contains 87% pseudogenes.

The higher number of human subfamilies observed in the phylograms (285 in human versus 202 in dog) is partly due to the higher number of human pseudogenes that are evolving more rapidly than functional genes, and tend to be fractionated into specific subfamilies (see Additional data file 5). In addition, the number of canine subfamilies is likely to increase as the full repertoire of canine genes is identified.

For comparison of the two classifications, we constructed a phylogram containing both human sequences (714) and canine sequences (403). Generally, human and canine families are intermingled with class I and class II branches well separated (see Additional data file 6). For class I, no human or dog families appear to be species specific. In contrast, for class II, one dog family (number 38), which contains 26 sequences and is spread over chromosomes 3, 8, 10, 27 and 29, appears to be dog specific as no human counterpart has been found. On this phylogram, one human family (number 41, named 7 by Glusman et al. 19) is particularly large, containing 93 OR sequences as compared to 35 in the dog. This family was shown to have expanded in humans and to consist of a large number of pseudogenes 19.

Apart from class I genes, which are all located on canine chromosome 21, gene families of more than five genes are in most cases (17 families out of 22) scattered over several genomic locations; for example, members of the canine-specific family 38, of which 17 genes are mapped, are dispersed over eight chromosomal regions. In contrast, the 32 mapped sequences of family 41 are all located on canine chromosome 20 (in two clusters). At the level of subfamilies, and taking a threshold of three mapped genes per subfamily, 77% are chromosome specific, and even cluster specific. This is similar to the 73% found with the human classification made in the present work.

Thirty-three OR clusters, and four isolated genes, have been identified in the canine genome. This compares with 80 clusters, and 62 isolated genes, in the human genome. The difference in cluster numbers is likely to be at least partly due to the difference in the resolution of the two methodologies used for their identification. Human clusters were defined after genome sequencing, and any genes or group of genes separated by more than 1 megabase (Mb) were considered as independent clusters 19. In contrast, the identification of canine clusters is based on RH-mapping data. Consequently, in many cases, a canine cluster can correspond to several human clusters. In addition, owing to the complete identification of the human OR repertoire, several small human clusters presently have no canine counterpart. This is likely to reflect the fact that only half of the canine repertoire has been identified and the probability of having mapped isolated genes or small clusters is lower than for larger clusters. As already noted for the mouse OR repertoire 22, it seems clear that there are far more isolated genes in the human genome than in the mouse and canine genomes.

On the basis of synteny data and nucleotide sequence similarity, dog and human clusters were paired whenever possible, resulting in the pairing of 20 canine clusters to 29 human clusters. Surprisingly, no human orthologous clusters were found for 13 canine clusters, containing 2-16 genes. Apart for the three canine clusters attributed to unlinked RH groups, and for which no synteny information could be used, the other 10 canine clusters appear to be dog specific. Two of these are located on each of canine chromosomes 10 and 15, and one on each of chromosomes 1, 3, 6, 8, 30 and X. Interestingly, there are no OR clusters in the human orthologous regions of canine chromosomes 1, 3, 8, 15, 30 and X, or in the mouse orthologous regions of canine chromosomes 1, 3, 6, 8, 10, 30 and X. This indicates locations where canine-specific expansion of the OR repertoire has occurred. Among those, clusters on chromosomes 3, 8 and 10 contain members of canine-specific family 38.

Despite the fact that only half the canine OR genes have been identified, we frequently observed the presence of an equal or higher number of genes in the canine clusters compared to their human orthologous clusters (Table 1). In common with mouse, this result reflects the existence of a larger canine repertoire relative to human. Two large canine clusters, located on canine chromosome 20, are good examples of the human/canine OR repertoire evolution. These two clusters, containing 21 and 44 sequences, have a similar composition, with genes from families 16, 40 and 41, suggesting a cis-duplication. These two clusters correspond to two human orthologous clusters (19@12.2 and 19@19.8 20). However, these human and canine clusters differ in two respects. First, the two canine clusters contain more genes than their orthologous human clusters (44 and 21 in dog versus 14 and 14 in human), another example of OR gene expansion in the canine genome. Second, whereas canine gene family 41 is restricted to canine chromosome 20, the orthologous human family is scattered in 13 different chromosomes, with most of them pseudogenes (see above). Thus, whereas in dog it appears that only a cis-duplication event gave rise to two paralogous clusters, in human there appear to have been several additional trans-duplication events. As in the dog genome, murine OR genes of family 41 are all located on mouse chromosome 9, in the region syntenic to that on canine chromosome 20.

Finally, it is noticeable that, in spite of the fragmentation of the canine karyotype in 38 autosomal chromosomes, the canine olfactory repertoire is no more fragmented than in humans, with only 24 chromosomes harboring OR genes. This can be seen as a consequence of the fact that the same cluster organization exists in the human and dog genomes, and when several clusters are localized on the same chromosome they tend to be in close proximity, thus forming a region of limited size whose synteny was not disrupted during evolution. In all, the present study shows a similar clustered genomic organization of OR genes in dog, mouse and human, raising the question of the relative importance of gene number, pseudogenization and gene transcription regulation to explain the greater olfactory capacity of macrosmatic mammals (mouse and dog) versus a microsmatic mammal (human).

Dog DNA was extracted from the MDCK cell line as described previously 25. A PCR amplification mix was made from 25 ng dog DNA, 0.5 U AmpliTaq Gold polymerase (Applied Biosystems, Foster City, CA, USA), 1.2 μM of each degenerate primer (OR3p 5, OR5A, OR5B, OR3B 6, OR3.1, OR7.1 12), 0.25 mM of each dNTP (Amersham Biosciences, Piscataway, NJ, USA), 1x buffer and 3 mM MgCl₂ for OR3.1 and OR7.1 primers or 2 mM MgCl₂ for the others primers in a final volume of 50 μl. PCR cycling conditions were 40 cycles at 94°C for 30 sec, 45°C (or 35°C) for 45 sec, 72°C for 1 min, with a final extension at 72°C for 2 min. Amplification products (10 μl) were electrophoresed in 2% agarose gels in 0.5x TBE.

PCR products were purified using the QIAquick PCR purification kit (Qiagen, Valencia, CA) and cloned using the pCR 2.1-TOPO Vector (Invitrogen, Carlsbad, CA). Clones were transferred from agar plates into 96-well microplates containing PCR mix. PCR of inserts was performed in a final volume of 20 μl using PU and PR primers (1.2 μM each), 1U Taq polymerase (Promega, Madison, WI), 0.25 mM of each dNTP, 1x buffer and 2 mM MgCl₂. Amplification was checked by electrophoresis in a 2%-agarose gel in 0.5x TBE. Inserts were sequences using PU and PR primers and the Dye-T chemistry on an ABI prism 377 DNA sequencer (Applied Biosystems, Foster City, CA). The sequences were analyzed by the Phred/ Phrap and Consed software 323334.

A database of canine genomic sequence is maintained by The Institute for Genomic Research (TIGR). Sequence data was originally obtained from plasmid libraries of small (2 kb) and medium-sized (10 kb) genomic DNA inserts prepared from a male standard poodle, and sequenced at Celera Genomics, as described previously for the human genome 35. The sequence data consists of 6.2 million reads (average read length, 576 bases), representing approximately 1.2x sequence coverage of the haploid canine genome (3 Gb 36). The reads were assembled with Celera Assembler 35, and the assembly output consisted of 1.09 million contigs (mean length 1,393 bases; mean content 4.9 reads per contig) and 0.85 million singletons. These were searched for sequences that encode peptides that share similarity with the conserved peptide sequence, MAYDRYVAICXPLHY (single-letter amino-acid code), using tblastn (p < 0.3). Selected assemblies, or unassembled reads, were then searched against the nr sequence database of GenBank. Canine sequences that were more similar to known ORs than to any other classes of gene were selected for further analysis.

Program Fasta3 37 was used to align a query sequence against all the OR canine genes and primers were chosen manually in specific regions of the query sequence (see Additional data file 2) and were then submitted to the program Primer3 38. RH mapping was performed on the RHDF5000 radiation hybrid panel as described previously 24. To ensure that a given mapped marker actually corresponds to the gene sequence from which the primers had been derived, PCR products from two positive hybrid cell lines, as well as from the positive dog DNA control, were sequenced and compared with the original clone sequence. Statistical analysis of the RH data was performed using the MultiMap package 39 as described in Mellersh et al. 40.

Sequences were translated using the program 'Traduction multiple' at the Infobiogen site 41. Pseudogenes were scored when frameshift(s) and/or in-frame stop codon(s) were observed. When possible, mutation(s) were verified on different DNA samples (MDCK cell line, fibroblasts used to construct the RH panel, and mongrel dog lymphocytes).

Alignment of dog OR protein sequences between TM2 and EC2 was performed with CLUSTALW 42 using standard parameters. Pseudogenes were also included after artificial elimination of the frameshift mutations to restore the ORFs. The dog b₃-adrenergic receptor sequence (GenBank U92468) was used as outgroup and for rooting. The phylogram was constructed using the neighbor-joining method implemented in CLUSTALW. The sequences were classified in families and subfamilies using the criteria of Ben-Arie et al. 6; that is, a family is defined as a group of sequences with greater than 40% amino-acid sequence identity (ASI), and a subfamily with greater than 60% ASI.

OR sequences are named 'CfORxxx' for Canis familiaris olfactory receptor. One hundred and seventy-nine OR sequences have been deposited in GenBank (access numbers: AJ431391-AJ431569).