INTRODUCTION

TOP ABSTRACT INTRODUCTION RESULTS Methods REFERENCES

The Human Genome Project now is at a historic turning point, from genomic DNA sequencing to functional genomics. According to the announcement from both public domain and private sector sequencing efforts, a "working draft" of the human genome sequence was just obtained, and the completion of the sequence will be achieved before the end of 2001 (Collins et al. 1998; Venter et al. 1998; Marshall 1999, 2000). The gene discovery and understanding of genetic information will require annotation of the sequence data using bioinformatic tools (Burge and Karlin 1997). Meanwhile, cloning of full-length cDNA has been listed as one of the major tasks of the current phase of genomic science (Collins et al. 1998). The integration of cDNA sequences with the genomic ones will greatly ease the identification of transcriptional units, as well as their mRNA levels and specificities in cells/tissues as a result of a fine regulation of the transcriptional expression at genomic level (Dunham et al. 1999; Hattori et al. 2000). Moreover, the cDNA project links directly to protein structural biology and exerts significant impact on the medical genetics and biotechnology/pharmaceutical industries.

Hematopoietic stem/progenitor cells (HSPCs) possess important roles for the physiological and pathological hematopoiesis, one of the essential areas in biomedicine, and the molecular basis of hematopoiesis remains to be better understood (Morrison et al. 1995, 1997). Over the last 3 years, we have been undertaking to catalog the expressed sequence tags (ESTs) from cDNA libraries of CD34+ HSPC populations from both umbilical cord blood (Mao et al. 1998) and adult bone marrow (Gu et al. 2000). This approach turned out to be very successful in terms of both gene expression profiling and discovery of novel genes in an efficient way. More recently, we have been extending this work to the cloning and sequencing of full-length cDNAs for previously undefined genes and to investigate their functions.

In this work, we report on the characterization of structural/functional features, chromosomal localization, and transcriptional expression patterns in different hematopoietic cell lines of 300 cDNAs with putatively entire open reading frames (ORFs) isolated from CD34+ cells. We also tried to integrate these data with the genomic sequence information and to propose some strategies to deal with the major challenges in expression genomics facing the completion of the human genomic sequences in the coming 1 or 2 years.

	RESULTS

TOP ABSTRACT INTRODUCTION RESULTS Methods REFERENCES

Primary Gene Expression Profiles of CD34+ HSPCs

RT/PCR-based Capfinder cDNA libraries were constructed using mRNA from highly purified CD34+ HSPCs of cord blood and adult bone marrow, using methods described previously (Mao et al. 1998). In total, 1 × 10⁶ recombinant clones were obtained from CD34+ cell library of cord blood origin (CB) and 0.5 × 10⁶ clones were acquired from that of bone marrow (BM). The average size of inserts in both libraries was 1.2 kb. Among 9866 and 4142 EST sequences obtained from CB and BM CD34+ cell libraries, respectively, the repetitive DNA elements, rRNA, and mitochondrial DNA sequences accounted for 11.7% and 17.3% of the total, respectively. After eliminating these sequences, the meaningful ESTs were classified into known gene, dbEST, and novel EST groups by database search. For useful ESTs from both origins, the known and named gene groups occupied the largest portion (5377 out of 7376 from cord blood and 2265 out of 3424 from bone marrow, respectively); the list of all ESTs corresponding to known genes from both origins is now available at http://www.chgc.sh.cn. The ESTs representing undefined genes (dbEST and novel EST groups) were assembled into 2060 clusters, which then served as candidates for cloning of full-length coding sequences.

Cloning of cDNAs with Putatively Entire ORF for Previously Undefined Genes

Sequences of cDNA clones representing 2060 EST clusters of undefined genes were obtained. Those clones with continuous sequences encoding at least 100 amino acids (with an exception of a few smaller ORFs bearing very high homology to the known small genes) were checked for the presence of putatively entire ORFs using the following criteria. First, when a sequence had high homology to a known gene, its ORFs were compared with each other. If the amino acid sequences of both ORFs initiated by an ATG codon could be reasonably aligned, the ORF contained in the novel gene cDNA was defined as a putatively complete one. Second, those sequences without homology to known genes were searched for in-frame stop codons upstream of an ATG codon-initiated ORF of >100 amino acids. If no such stop codon was found ahead of an ORF, the nucleic acid sequence flanking the first ATG should bear similarity to the well-conserved KOZAK motif (Kozak 1986). The above analysis revealed that 222 of our clones might contain an entire ORF. In 78 EST clusters, an obvious but incomplete reading frame was present. Different methods were employed to prolong the ORF in these 78 clones until complete ones were considered to be reached according to the aforementioned criteria. In silico cloning with dbEST extension allowed us to obtain 69 putative entire ORFs, which were then confirmed by sequencing of material cDNA clones obtained by appropriately designed RT-PCR. Finally, for those sequences that could not be extended properly with an electronic approach, rapid amplification of cDNA ends (RACE) was applied to get the 5' or 3' ends with Marathon-ready cDNA libraries from appropriate tissue origins. Another nine entire ORFs were cloned and sequenced this way. In total, 300 cDNAs with putatively entire ORFs were obtained. Their nucleic acid sequences were 360-3496 bp in length and their ORFs coded for peptides of 58-752 amino acids. The major features of each gene are summarized in Table 1. It is worth pointing out that, although a 3' poly(A) sequence or a polyadenylation signal was found in most (214/300) cDNAs as evidence of containing the complete 3' UTR, the integrity of the 5' UTR could not be certain in the majority of the cDNAs.

In the remaining 1760 EST clusters corresponding to previously undefined genes, 512 clusters contained partial reading frames, 806 represented 3' UTRs as they had no obvious reading frames but presented polyadenylation signal and poly(A) tails, and the remaining 442 contained sequences of which the features should be further analyzed.

View larger version (373K): [in this window] [in a new window]

Functional Significance Indicated by Homology Comparison with Genomic Sequences through Evolution

It is well accepted that homologous genes often share similarities at sequence and/or functional levels (Henikoff et al. 1997). Hence, sequence similarity acquisition is an efficient method to predict the function of a novel gene. Members belonging to the same gene families could be assumed/determined with this strategy and conserved genes often show conserved sequence elements within the important functional domains or motifs. Based on this consideration, putative ORFs from model organisms with completed genome sequence, including bacteria, Saccharomyces cerevisiae, Caenorhabditis elegans, and Drosophila, and ORFs of identified genes from Arabidopsis and mammals (excluding primates) were retrieved to compare the amino acid sequence similarities with those of ours (Table 1; Fig. 1). Sequences with similarity >25% over a region of 50-100 amino acids were considered here to have some homology (Russell et al. 1997). Among our 300 cDNA sequences, 21 share similarity to the coding sequences in all species examined, indicating that they are well-conserved genes and important for cell life. In fact, 16 of these 21 genes have assigned functions. A total of 204 cDNAs contained ORFs with >25% similarity to the sequences in at least one species. Functional clues have been available in 105 of these 204 genes. Taken as a whole, at least 225 genes identified in the current work are evolutionarily conserved. Interestingly, as shown in Figure 1, an increased gradient of similarity in terms of both number of related genes and the degree of homology is present from bacteria to Drosophila. In the case of Arabidopsis, only part of the genomic sequence is available in the public database. However, 66 of our cDNAs found their homologs in this plant. As expected, the number of genes with high homology (>50%) was great in mammals. The fact that 75 cDNAs had so far no obvious similarity to any genes across different species implied that they might be functionally specific genes acquired relatively late during evolution.

View larger version (36K): [in this window] [in a new window]   Figure 1   Homology comparison of ORFs contained in our cDNAs to known genes from different model organisms. The horizonal blocks represent numbers of the ORFs bearing homology to genes in a given species; different colors indicate the degree of homology. Each number listed at right indicates the total number of our ORFs having homologous genes in that organism. (B) Bacteria; (Y) yeast; (C) C. elegans; (D) Drosophila; (A) Arabidopsis; (M) mammals not including primates.

Structural and Functional Assignment with Bioinformatic Prediction

Basic structural motifs predicted by some algorithms on the primary structure in the ORFs are listed in Tables 1 and 2, including leucine zipper, C2H2 zinc finger, and C3HC4 ring finger. Some consensus patterns of protein kinase, growth factor, and cytokine receptor-associated protein were also found by such methods. However, caution should be taken in interpreting these data. For instance, leucine zipper motif was predicted on primary structure in 12 ORFs using the Motifs software in the GCG package. Further analysis with Coilscan and Peptidestructure programs also provided by the GCG package revealed, nevertheless, that only 1 of these 12 leucine zippers was located in a coiled-coil structure. Because a typical leucine zipper should be included in a coiled-coil domain, this result indicates the importance of integration of information generated by different prediction methods, including those for conserved motifs at primary sequence level and those for secondary or higher structures. In analyzing the signal peptide, two different approaches, Spscan (in GCG package) and signalP (http://www.cbs.dtu.dk/services/Signalp/) were applied to our ORFs. The former algorithm is based on the weight matrix method in concert with McGeoch's discrimination of a minimum signal peptide, whereas the latter is based on two neural network methods for recognition of signal peptides and their cleavage sites. Of note, only cleavable signal peptides, but not the uncleavable ones like signal anchor and internal signal, can be detected with these algorithms. Interestingly, the two approaches gave quite coherent results in predicting putative amino-terminal potential signal peptides in 11 ORFs, including 8 with -helix transmembrane domains outside the signal peptide region. One such example was an ORF with both signal peptide and 6-transmembrane domains (HABC7, GenBank accession no. AF038950), which contains an ABC transporter family signature. We therefore speculated this ORF encodes a putative transmembrane transporter protein.

Genomic Organization and Alternative Splicing Identification

Of our genes, 243 were preliminarily characterized in terms of exon-intron organization after comparison of cDNA sequences with the genomic sequences in the database (Table 1). The estimated genomic sizes of these genes spanned 384 bp to 144 kb, containing 1 to >17 exons, and correspondingly 0 to >16 introns. The size distribution of the exons was from 20 bp to 2023 bp, whereas that of characterized introns ranged from 77 bp to 86 kb. Of note, 17 genes composed of only 1 exon varied in sizes from 384 bp (HSPC016, accession no. AF077202) to 1346 bp (P47, accession no. AF078856). On the other hand, cDNAs of short length could contain multiple exons. For example, HSPC245 (accession no. AF151079), consisting of 5 exons, and HSPC024 (accession no. AF083241), consisting of 7 exons, were only 497 bp and 581 bp in length, respectively.

During the characterization of the genome organization of our genes, some alternative splicings were determined. A 453-bp sequence in hSC2 (accession no. AF038958) was deleted in an isoform (accession no. AF038959), which was only found in CD34+ cells so far, whereas LYPL-A1 (accession no. AF077198) used a 48-bp stretch that did not exist in the short form transcript (accession no. AF077199) (Fig. 2A). The fact that these alternatively used sequences are located in ORFs in an in-frame way supports the idea that these are physiologically existing isoforms and not artifacts in cDNA library construction. Indeed, the isoforms of the two genes were further confirmed by RT/PCR assay (data not shown). Interestingly, the cDNA sequence of HSPC070 (accession no. AF161555) located on chromosome 3p25 was found to share a 105-bp stretch in the 3' UTR including the polyadenylation signal with that of RAF oncogene (accession no. X03484) (Bonner et al. 1986) in reversed orientation (Fig. 2B). This was further confirmed by the draft genome sequence from GenBank (AC018494, AC018500, AC026153, and AC026170) (see legend for Fig. 2B).

View larger version (35K): [in this window] [in a new window]   Figure 2   (A) Alternative splicing present in lysophospholipase gene transcripts as long (L, accession no. AF077198) and short (S, accession no. AF077199) forms. The numbers indicate the amino acid positions of deduced proteins. Note that the ORF is maintained in the alternatively spliced S isoform. (B) Overlapping of HSPC070 (accession no. AF161555) and RAF genes located on opposite DNA strands at the same locus. Both genes are mapped to the same region on chromosome 3p25. The comparison of sequences between cDNAs and genomic DNA has allowed the exon-intron structure of both genes to be established, with exons represented by boxes and their numbers indicated. Note that a stretch of 105 bp is shared by the 3' UTRs of both genes. Arrows indicate the orientations of transcription.

Chromosomal Mapping

Chromosome localization is an important aspect of a gene's general information. Combining strategies of bioinformatics acquisition from both UniGene and other databases, and radiation hybrid (RH), a total of 230 genes were mapped to proper chromosome positions (Fig. 3). Among 55 genes mapped with G3 or GeneBridge 4 RH panels, 38 had not been mapped previously, whereas the remaining 20 RH results showed good concordance with those by electronic mapping. The detailed mapping results are available at http://www.chgc.sh.cn. Of note, the 5 C2H2 zinc finger genes are all located on chromosome 19. 

View larger version (32K): [in this window] [in a new window]   Figure 3   Chromosome localization of 230 previously undefined genes by applying both STS searching and radiation hybrid (those marked with #). Detailed mapping information can be obtained from http://www.chgc.sh.cn

Expression Patterns in Different Tissues and in Distinct Hematopoietic Cell Lines

Among the 300 cDNAs, 270 could be analyzed using electronic Northern because their dbEST hits were available from UniGene resource. As shown in Table 1, most (207/270) genes showed ubiquitous transcriptional expression patterns as their corresponding ESTs were found in >10 tissues. The expression was found in a more selective way (<10 tissues) in 63. Only 13 showed relatively restricted expression in hematopoietic organs/tissues (bone marrow, foetal liver, spleen, lymph nodes, etc.). To explore the biological meanings of our genes in hematopoiesis, 285 cDNAs from the CB CD34+ cell library were also examined using cDNA macroarray for their expression levels in hematopoietic cell lines (the array membrane used in this work did not include the 15 cDNAs from the BM CD34+ cell library). The cDNA probes were prepared with mRNAs isolated from NB4 (granulocytic), HL60 (granulocytic), U937 (monocytic), K562 (erythro-megakaryocytic), and Jurkat (T lymphocytic) cell lines representing distinct lineages of hematopoietic cells. The RNA quality was ensured with appropriate ratio between 18S and 28S rRNA bands on agarose gel electrophoresis, and the labeling efficiencies of cDNA probe were confirmed to be >50%. To evaluate the expression levels, the membranes were exposed to Phosphor screen and the relative intensity of each gene was quantified with FLA-300 detection system. Hybridization signals in separate experiments with different membranes and/or probes were calibrated using housekeeping genes including GAPDH and total amount of signals on the membrane as reference. The feasibility of the technology system was confirmed by reproducible results of the paralleled duplicate spots on the same membrane (Fig. 4A) and with independent tests on different membranes (Fig. 4B). The comparison of expression levels in different cell lines for 285 genes examined is shown on Table 1 (normalization with different references revealed similar results though only those based on GAPDH control are shown). Although most genes exhibited expression in all five cell lines, 35 of them displayed restricted expression in only one or two lineages. Northern blot analysis was performed for three genes, HSPC070, ZNF254, and HSPC135. According to the UniGene data, HSPC070 has a ubiquitous expression pattern, whereas the expression of ZNF254 and HSPC135 could be restricted to hematopoietic system (Table 1). Indeed, Northern blot analysis showed that HSPC070 was expressed in a variety of tissues (Fig. 4C) whereas no obvious transcriptional expression of ZNF254 and HSPC135 was detected in these tissues (data not shown). However, the three genes were all found expressed in most of the hematopoietic cell lines examined in this work.

View larger version (61K): [in this window] [in a new window]   Figure 4   Regression analysis of the cDNA array results (A,B) and Northern blot analysis of three cDNAs (C). (A) The scatterplot of detected signal intensity for duplicate spots on the same membrane. (B) Scatterplot of detected signal intensity for the corresponding dots in two membranes with independent tests from RNA of same cell origin. All signals are normalized by using GAPDH gene as internal control. The figures were made with Microsoft Excel spread sheet and the correlation line was indicated. (C) Northern blot analysis of HSPC070, ZNF254, and HSPC135. (Top) HSPC070 with a ubiquitous tissue expression pattern. GAPDH or 28S/18S ribosomal RNAs were used as sample loading control. (Bottom) Expression of ZNF254 and HSPC135 in hematopoietic cell lines, with GAPDH as control.

DISCUSSION

Because tissue- or development stage-related differential expression exists for many genes, cloning of full-length cDNA based on EST analysis in different tissues represents a useful approach for gene identification, especially for those subject to temporo-spatial regulation. In strict sense, a full-length cDNA should cover both the ORF and the complete 5' and 3' UTR. Although a number of methods have been used to surmount the technical obstacles for getting the 5' end of cDNA (Carninci et al. 1996), it is still difficult to reach the transcription start site in many cases. However, as the most important functional information of an mRNA is contained in the ORF, cDNAs containing entire ORFs are often considered as being full-length. By combining several technologies including construction of full-length cDNA enriched libraries, in silico cloning, and RACE, a relatively efficient working system has been established to obtain full-length cDNAs, or more precisely, cDNAs including entire ORFs, in a cost-effective way. This system has enabled the first resource of cDNAs with putatively entire ORFs to be generated for previously undefined genes whose expression is found in human CD34+ HSPCs.

One strong challenge to genomic science presently is to elucidate the functions of the newly discovered huge amount of genes. In this work, we tried to apply the currently available bioinformatic tools to the analysis of the structural and functional characteristics of each ORF. Using BLAST search, 121 out of 300 ORFs were found to share homology to genes with functional information, offering important clues for the choice of appropriate functional assays in further study. The difficulty was how to deal with the majority of the ORFs without obvious functional information. We therefore attempted to evaluate the conservation of the sequences through evolution. As a result, 225 ORFs show >25% similarity at amino acid level to those identified in organisms including bacteria, S. cerevisiae, C. elegans, Drosophila, Arabidopsis, and nonprimate mammals, whereas 75 have so far no similarity. It is quite possible that the 21 ORFs well-conserved across a wide range of species may be derived from the "essential genes." Although a large proportion of these evolutionarily conserved genes are of unknown function, this analysis can provide at least the following information: On the one hand, they are most likely to exert important biological functions; and on the other, the lower organisms containing homologous sequences can be used as models in the functional study with gene knockout or other methods. Moreover, efforts have been made to approach the gene function by search of distinct motifs and domains with combined use of algorithms based on different methods and taking into consideration not only the primary sequence but also the secondary structure of the proteins. Of note, in addition to those well-known functional motifs such as zinc finger and leucine zipper, a putative signal peptide was found in 11 ORFs with or without transmembrane motif in proper location. This information may lead to future works to identify possible secreted proteins and transmembrane proteins, and hence may allow recognition of new regulatory pathways involved in the self-renewal and/or differentiation of HPSCs.

Characterization of gene expression with regard to tissue distribution is another way to approach the gene function. Genes with ubiquitous expression are more likely "housekeeper" genes, whereas genes whose expression shows tissue specificity may exert functions related to the development and differentiation of a given tissue or cell population. In this work, both electronic Northern and macroarray screening were carried out to study gene expression patterns. Because the majority of the genes presented in this work had been already hit by dbESTs and relevant information was available in UniGene (Boguski and Schuler 1995; Shi et al. 1999), the electronic Northern could give an approximate estimation of the tissue distribution patterns. Of note, among 270 genes thus analyzed, 207 were hit by ESTs from >10 tissues while only 13 were mainly hit by ESTs of hematopoietic tissues. On the other hand, the macroarray system with relatively high efficiency and throughput was used in this work to study gene expression within the hematopoietic systems. Probes prepared from five hematopoietic cell lines were applied to cover granulocytic, monocytic, erythro-megakaryocytic, and lymphoid lineages. Of 285 genes expressed in CD34+ cells of cord blood origin, 35 were picked that showed relatively restricted or preferential expression along with a given orientation of differentiation. Therefore, combination of the two methods allowed us to find genes which may play a role in hematopoiesis-related functions.

In this work, we have also tried to take the opportunity of ever-increasing genomic mapping and sequence data to promote the understanding of structural organization of our genes discovered by cDNA approach. Application of bioinformatic information from public database, including sequence tag sites (STS) map (Stewart et al. 1997) and UniGene database (Boguski and Schuler 1995), allowed us to assign the chromosomal localizations for 192 novel genes. Retrieving genomic sequences from the "working draft" corresponding to our cDNAs obtained the exon-intron organizations in 243 genes, and the characterization of genomic structure of all genes can be expected in the near future with the accelerated schedule of the Human Genome Project. Although our work is only a small part in the international effort to establish a detailed whole genome transcription map, it may give some suggestions to the future study. Now, the gene discovery in genomic DNA sequencing depends largely on annotation but the successful rate based on theoretical prediction is not high enough. Hence, full-length cDNA cloning projects will provide the definitive evidence to the predicted transcription units. In contrast, genomic DNA sequences can also offer unique information for the full-length cDNA cloning. For instance, obtaining the 5' ends of genes with large coding sequence is often difficult. Exon prediction may lead experimental work to help their cloning. Besides, genes with very low expression levels or extremely narrow expression windows may be absent or poorly represented in most of the cDNA libraries. Annotation of genomic sequences may facilitate the identification of these genes. Moreover, comparison of cDNA and genomic sequences can reveal some complex mechanisms of genomic organization and expression. To this end, it is interesting to note the overlapping in reversed orientation of our HSPC070 gene and the known RAF gene located on chromosome 3p25, as well as the alternative splicing patterns in some genes. According to the comparative analysis between the whole genome sequence data from C. elegans (The C. elegans Sequencing Consortium 1998) and Drosophila (Adams et al. 2000), the functional complexity of a genome is determined not only by the number of the genes, but even more importantly by the alternative splicing as well as complex regulatory mechanisms of the genome at transcriptional level. Finally, the chromosomal distribution of genes bears not only evolutionary meaning, such as the mapping of all five C2H2 zinc finger genes on chromosome 19 suggestive of recent duplication events, but also indicates candidate genes in disease-related loci.

	Methods

TOP ABSTRACT INTRODUCTION RESULTS Methods REFERENCES

EST Sequencing and Data Analysis

Mononucleated cells were harvested from cord blood and bone marrow with gradient centrifugation and CD34+ populations were separated with anti-CD34 MAb-conjugated MACS system (Miltenyi Biotec, Germany). After two rounds of separation, CD34+ cells were of 96%-99% purity according to flow cytometry analysis (Gu et al. 2000). RNA extraction, ZAPII cDNA libraries construction, Bluescript phagemid templates preparation, sequencing strategy, and data management were manipulated as before (Mao et al. 1998; Gu et al. 2000). The sequencing primers were universal primers including M13 Reverse and/or Forward, T3 and/orT7 primers, and sequencing mix was BigDye Terminator (Perkin Elmer). 5' or 3' end ESTs generated were categorized into known gene, dbEST, and novel EST groups by searching against GenBank database with BLAST and FASTA programs in GCG package.

Cloning of Full-Length cDNA

The EST clones corresponding to previously undefined genes were candidates for full-length cDNA cloning. The clone inserts were sequenced with end sequencing, primer extension, and sequencing after partial deletion/subcloning. AutoAssembler (Perkin Elmer) was applied to assemble the sequences into contigs. DNA Strider (Version 1.0) was employed to analyze the ORF. For those clones containing partial reading frames, in silico EST assembly and RACE were performed. Proper Marathon-ready cDNA libraries (Clontech) were chosen as RACE template, and the gene-specific primers were generated according to the clone sequence. The ORFs thus obtained were confirmed with RT-PCR.

Structure and Function Analysis with Bioinformatics

Sequence Similarity Comparison

Genomic Organization Determination

Fundamental Structural and Functional Elements Searching

Chromosomal Mapping

Electronic Mapping

Radiation Hybrid

Gene Expression in Different Tissues

In silico Northern Blot

Northern Blot

Screening of Gene Expression in Different Hematopoietic Cell Lines with Macroarray

Membrane Preparation

Preparation of the Probes

Hybridization

Signal Detection and Gene Expression Quantification