| Nucleic Acids Research | Pages |
©1999 Oxford University Press |
Exhaustive mining of EST libraries for genes differentially expressed in normal and tumour tissues
Introduction
Materials And Methods
Creating a set of non-redundant ESTs
Identification of homologous sequences
Assembly of ESTs
AUTEX: a protocol for the automatic extension of partial DNA sequences
Electronic northerns
Statistical evaluation of EST occurrences
Results
Discussion
Acknowledgements
References
Exhaustive mining of EST libraries for genes differentially expressed in normal and tumour tissues
Received June 14, 1999;
ABSTRACT A four-step procedure for the efficient and systematic mining of whole EST libraries for differentially expressed genes is presented. After eliminating redundant entries from the EST library under investigation (step 1), contigs of maximal length are built upon each remaining EST using about 4 000 000 public and proprietary ESTs (step 2). These putative genes are compared against a database comprising ESTs from 16 different tissues (both normal and tumour affected) to determine whether or not they are differentially expressed (step 3; electronic northern). Fisher's exact test is used to assess the significance of differential expression. In step 4, an attempt is made to characterise the contigs obtained in the assembly through database comparison. A case study of the CGAP library NCI_CGAP_Br1.1, a library made from three (well, moderately, and poorly differentiated) invasive ductal breast tumours (2126 ESTs in total) was carried out. Of the maximal contigs, 139 were found to be significantly ([alpha] = 0.05) over-expressed in breast tumour tissue, while 13 appeared to be down-regulated.
INTRODUCTION
Expressed Sequence Tags (ESTs), single pass reads from randomly selected cDNA clones (1,2), have proved to be a valuable resource for genome research (3-7). In many branches of genetic research, EST libraries serve as a gateway for the detection and characterisation of new candidate genes. Virtually all EST libraries are tissue specific and their mode of preparation is documented. Database entries of EST sequences carry a label in their header which permits the identification of their source tissue, a feature which is of major interest in the present study. EST libraries of many human tissues of various developmental stages, both normal and diseased, have been and are still being established and partly made accessible to the public.
With the average size of EST libraries ranging typically between 1000 and 10 000 entries, an EST library cannot be regarded as faithfully representing the gene expression pattern of a tissue (8). It is estimated that, varying with the cell type, between 10 000 and 30 000 different genes are expressed in a cell, with an average of about 300 000 mRNA molecules per cell.
Therefore, an EST library cannot be more than a coarse grained snapshot of the mRNA composition of a certain tissue at a certain time. Especially, the representation of low abundance genes in an EST library cannot be taken for granted. However, the availability of many EST libraries derived from the same type and state of tissue mitigates this problem inasmuch as pools of equivalent EST libraries can be created. The EST numbers of such pools reaching tens of thousands, a proportional representation of all abundant and moderately expressed genes can be assumed.
Following the paradigm that a cell is, to a large extent, characterised by its transcript composition, and that the amount of a protein, the actual biochemical agent, is positively correlated with the abundance of its mRNA, such EST pools enable us to carry out a meaningful statistical expression analysis in silico (9,10). Of course, for mRNA expression analysis only non-normalised EST libraries are eligible for pooling; normalised or subtracted libraries have to be excluded.
It is the purpose of this work to present a method which permits exhaustive exploitation of the information relevant to biologists and pharmacologists hidden in EST libraries. We are especially interested in genes which are significantly up- or down-regulated in diseased tissue. The panel of genes resulting from such analysis may serve as candidates for therapeutic or diagnostic targets (11).
The basic method to hunt for differentially expressed genes would be to carry out a simple sequence comparison procedure with a standard sequence comparison program like BLAST (12). Each EST of a tumour library is searched against all other ESTs from that library and against a database of ESTs from the benign counterpart. Comparing the number of homologous sequences (hits) found in either of the two libraries, benign and tumour, (with consideration of the potentially different library sizes) would give a first clue if an EST represents a differentially expressed gene.
The protocol presented here goes beyond this first approach in two ways. (i) The counts obtained in the above sketched EST-EST comparison procedure will tend to remain low, because only direct matches of the query EST sequence can be detected. The observed hit number might not be sufficient to secure differential expression statistically. Since cDNA sequences can extend over several kilobases and since EST sequences rarely exceed a couple of hundred base pairs, many ESTs belonging to a gene will remain unnoticed. (For the results of a control study see Results.) In order to catch all available ESTs belonging to a gene it is desirable to use full-length sequences for the search process whenever possible. Therefore, we applied an iterative search and assembly procedure (see AUTEX: a protocol for the automatic extension of partial DNA sequences) to build in silico contigs of maximal length using all available ESTs. (ii) Building maximal contigs and pooling EST libraries still does not guarantee hit numbers high enough for easy and unequivocal identification of differential expression. The identification of changes in expression level based upon quotients of very small relative abundances is not very meaningful. Therefore, we compute P values by Fisher's exact test (see Statistical evaluation of EST occurrences), which allow statistical assessment of any observed hit distribution, even if the hit numbers are very low. A P value can be interpreted as the degree of certainty that an observed differential expression is not an artifact due to statistical fluctuations.
MATERIALS AND METHODS
Creating a set of non-redundant ESTs
Two similar EST sequences are very likely to match the same sequences in an EST database search, and, therefore, the in silico contigs built from their matches would also probably be almost identical. To save computer resources, it is advisable to extract a set of non-redundant sequences from the EST library under investigation. Only the sequences of this non-redundant set would serve as seeds (i.e. starting sequences) for the automatic extension procedure expounded below. We used an all-versus-all comparison to eliminate redundant ESTs from the initial set. The sequences of the library under investigation were sorted by decreasing lengths and were successively taken as query sequence for a BLAST search against all shorter sequences. Whenever the shorter sequence perfectly covered a third of the length of the longer one it was considered redundant and eliminated.
Identification of homologous sequences
We used the well-known basic alignment program BLAST (12) to find homologous sequences in a database. To increase sensitivity, we used a new BLAST implementation (v.2.0.5) which tolerates gaps in the sequence alignments (13). Our choice for the stringency parameters defining sequence homology were 10-4 for the E value and 95% sequence identity. For the characterisation of the contigs BLAST searches against the nucleotide database nt were run (14).
Assembly of ESTs
We used the program GAP (15) for the assembly of ESTs. In a first round, 2% mismatches between the EST sequences were allowed; this relatively stringent choice was made to guarantee a stable backbone to the assembly. Since the sequencing error rate of ESTs can amount to up to 5% towards their 3[prime]-ends, the stringency was lowered to 5% mismatches in a second round to collect as many ESTs as possible.
AUTEX: a protocol for the automatic extension of partial DNA sequences
We developed an iterative procedure, AUTEX, to build contigs of maximal length based upon arbitrary partial DNA sequences (seed sequences). An initial BLAST search of the seed sequence is performed against our repository of about 4 000 000 ESTs: the human division of dbEST (16) (~1 000 000) and a proprietary EST database (Incyte Pharmaceuticals, Palo Alto, CA) (~3 000 000). The sequences homologous to the seed sequence (parameters as in Identification of homologous sequences) are extracted from the databases and assembled as described in Assembly of ESTs. The consensus sequence is derived from the resulting multi-sequence alignment and taken as query sequence for another BLAST search against the EST databases. This procedure of alternating BLAST search and assembly of matching sequences is repeated until the consensus sequence has reached its maximal length. Approaches similar to AUTEX have been described by Ebeling et al. (20) and Gill et al. (21), and the feasibility of generating full-length gene sequences from ESTs using such a method was demonstrated by Prigent et al. (22).
Electronic northerns
Thanks to the specification of the source library in the headline of most EST database entries the source tissue of origin and related information, e.g. mode of cDNA library preparation or disease state of the source tissue, can be traced back. Therefore, a report generated in a BLAST search of a given DNA sequence against an EST database consisting of such `pedigree' ESTs allows the inference of that gene's tissue distribution. In analogy to the corresponding laboratory method, this analysis is called electronic northern. The paucity of ESTs in a typical library makes it necessary to lump together EST libraries of common origin to create EST pools.
Demanding for a given tissue a pool of ESTs from both tumour and normal tissue and setting a minimum of 10 000 ESTs in a pool, we could create pool pairs for 16 tissues. One pool pair thus consists of all available ESTs from a tissue in both the normal and diseased states. Two hundred and sixty EST libraries from normal tissues and 172 EST libraries from tumour or cancerous tissues contributed to the EST pools. We mention in passing that pool pairs can be generated from public EST libraries alone for the following eight tissues: brain, breast, colon, kidney, lung, ovary, prostate and uterus.
In our electronic northern blot the tissue-specific relative abundance of a gene is defined as the ratio of number of homologous ESTs and total number of ESTs in the corresponding pool. Relative abundance figures were determined for normal and tumour pools separately, and the ratio of normal and tumour relative abundances, the so-called expression ratio or fold change, is used as a measure for the up- or down-regulation of a gene in tumour tissue with respect to normal tissue. For convenience, we introduce a classification scheme for the degree of differential expression comprising three levels: moderate, strong and very strong (see Table 1).
Table 1. A general classification scheme for degrees of differential expression and listing of all cases of differential expression found in our case study
| Class | Degree | Expression ratio | [alpha] | Regulation in tumour | [Sigma] | |
| Down | Up | |||||
| I | Very strong | >10 | <0.001 | 0 | 15 | 15 |
| II | Strong | >5 | <0.01 | 2 | 17 | 19 |
| III | Moderate | >2 | <0.05 | 11 | 107 | 118 |
| [Sigma] | 13 | 139 | 152 | |||
Statistical evaluation of EST occurrences
In order to assess the distribution of hits between normal and affected tissues observed in a BLAST search, we applied Fisher's exact test (17), a statistical test widely used for the evaluation of 2 × 2 contingency tables, i.e. representations of yes/no outcomes obtained from two disjoint samples. The outcome of Fisher's exact test is a significance value P ranging between 0 and 1 which describes the likelihood of the null hypothesis being true: `The frequency of an event is the same in either of two samples' or, in our specific example `The frequency of a gene is the same in normal and in diseased tissue'. The closer the significance value is to 1.0 the more the observations are compatible with the null hypothesis. A P value close to 0, on the other hand, is indicative of significant differential expression of the gene under consideration. Fisher's exact test is a conservative test as compared to other statistical tests (18), therefore selection of genes for further investigation based upon the criterion of small P values can be considered restrictive.
RESULTS
We performed a case study of the CGAP library NCI_CGAP_Br1.1 (provided by B. Soares and M.F. Bonaldo), a non-normalised EST library made from three pooled invasive ductal breast carcinomas (19) comprising 2126 entries.
Upon removal of 661 redundant ESTs (see above) we were left with a set of 1465 unique ESTs, which served as seed sequences for the automatic elongation by the software protocol AUTEX (see above). Almost all (1456/1465) of the seed ESTs found homologous ESTs in our concatenated EST database (Incyte and dbEST), and could, hence, be assembled and elongated. Nine seed ESTs did not find any matching sequences; they remained singletons. In 50 cases, elongated sequences were found to overlap with other elongated sequences. In other words, 50 non-overlapping pairs of ESTs were each derived from the same mRNA sequence and collapsed during the elongation process into one and the same contig. The median length grew from 418 bp for the initial EST set to 721 bp for the set of elongated sequences (Fig. 1).
Figure 1. The length distributions for the initial set of ESTs (open bars) and for the contigs of maximal length (black bars) obtained in the automatic elongation procedure (see text) are shown. The median length nearly doubled from 418 to 721 bp. Binning width was 50 bp.
We performed a simple test to assess the validity of our assembly approach. Chimeric assemblies can be identified by inspection of the reports generated by a BLAST search against a database of known sequences. An assembly containing ESTs from two different known genes generates a contradictory BLAST report listing perfect matches to these two genes. Inspection of the BLAST reports suggests a rate of less than 5% cases of chimeric assemblies, a finding which is in accord with our experience gathered in other analyses.
Expression analysis was performed with the non-redundant set of 1406 contigs of maximal length. Of these, 152 proved to be differentially expressed in normal and tumour breast tissue, i.e. they exhibit an expression ratio of at least 2 and their P value computed by Fisher's exact test was below 0.05. Up-regulation in tumour tissue is found more than 10 times more frequently than down-regulation (139 versus 13 genes). This finding confirms our expectation that the EST library under investigation should be enriched with genes which are over-expressed in breast tumour tissue. The classification of the differentially expressed genes is listed in detail in Table 1 and the results of the nt database consultation for these sequences are given in Table 2.
 |
 |
 |
 |
Table 2. Database comparison of 152 assembled sequences differentially expressed in breast
The 152 elongated sequences which were found to be differentially expressed in normal and tumourous breast tissue were compared against the non-redundant NCBI database nt. The accession no., abbreviated description, BLAST score, and E value of and similarity to the highest scoring sequence are given. The order of presentation is: sequences with very strong (class I), strong (class II) and moderate differential expression (class III). There is no apparent order within the classes.
One hundred and twenty-two genes matched database entries with an E value of 10-100 or better. Twelve of the 152 differentially expressed genes are of mitochondrial origin (plus one marginal hit), while one is a ribosomal gene. While it is well known that EST libraries can contain mitochondrial genes (23,24), the number can vary due to the mode of preparation. It is striking that most (eight out of 12) mitochondrial genes were classified as very strongly differentially expressed (class I). Two EST libraries in the breast tumour EST pool, the seed library NCI_CGAP_Br1.1 itself and NCI_CGAP_Br3, were identified as generating most of the hits. Differential expression of these mitochondrial genes might thus be an artifact of EST library preparation. On the other hand, up-regulation of mitochondrial genes in cancerous tissues has been reported previously (25,26). Thirteen of the genes in class I are represented in public databases, while an additional two exhibit weak homologies to known genes.
Apart from some genes associated with the increased metabolic activity of tumour cells, class II contains several genes which are known to be associated with various kinds of solid tumours. Two of three members of the trefoil gene family are represented in class II. Each of these secreted molecules is abundantly expressed in a broad but specific set of tumours (27). TFF1 (formerly pS2) was found to be overexpressed, for example, in carcinomas of the breast, pancreas, endometrium, bladder, prostate and lung, while TFF3 (formerly P1.B) is overexpressed in carcinomas of the intestine and invasive lobular and ductal carcinoma (28). Further examples of prominent tumour-associated genes found in class II are the proliferation-associated gene involved in breast cell proliferation (29) and the creatin kinase B gene, which is abundantly expressed in small cell lung carcinoma (30).
Class III can be considered a suitable source to find new tumour-associated genes since about 20 of the 118 genes in this class exhibit no or only very faint similarity to known genes. Several genes from class III identify tumour-associated genes which have been identified by `wet experiments' only very recently. The transmembrane protein NET-4, for instance, belongs to the family of tetraspan proteins that have been implicated in the prognosis of different types of tumours such as lymphomas (31) and bladder cancer (32). The adult T cell leukemia-derived factor 1 (ATL-1), or thioredoxin, exhibits increased expression in gastric cancer cells that are resistant to the chemotherapeutic agent cis-diaminedichloroplatinum (33). Cathepsin D overexpression in breast cancer cells was shown to be associated with increased risk of early relapse (34) and metastasis (35). Finally, overexpression of the human H19 gene has been described in breast adenocarcinoma (36). Imprinting and maternal expression of H19 is lost in a variety of cancers (37). Interestingly, an important function of the untranslated H19 RNA is the regulation of IGF-II expression (37), which in turn seems to regulate the routing of the cathepsin D gene product (38) mentioned above. This combined finding of genes which may belong to the same biochemical pathway demonstrates the value of our in silico approach and supports the expectation that further important cancer-related genes may be present within the cohort of the remaining 20 new candidates. However, since class III includes genes exhibiting an expression ratio of as low as 2, a factor which is commonly believed to be compatible with normal fluctuations, some of the genes will certainly be false positives.
Electronic northerns for three of the genes obtained in our assembly procedure, the TFF3 (formerly P1.B or ITF) gene coding for a secreted protein (39,40), the mammaglobin gene and a gene homologous to the mouse JIP-1 gene (accession nos L15203, U33147 and AL021708, respectively), are depicted in Tables 3, 4 and 5, respectively.
Table 3. Electronic northern for the TFF3 gene
| Tissue | Benign | Tumour | P value | Expression ratio | Class | ||
| Abundance | Pool size | Abundance | Pool size | ||||
| Bladder | 0 | 25 643 | 0 | 42 556 | |||
| Brain | 2 | 178 865 | 0 | 100 254 | 0.540 | ||
| Breast | 11 | 120 733 | 37 | 67 587 | 7.4 × 10-9 | 6.0+ | II+ |
| Colon | 49 | 52 193 | 22 | 35 113 | 0.117 | 1.5- | |
| Endocrine tissue | 4 | 62 286 | 9 | 60 780 | 0.175 | 2.3+ | |
| Kidney | 0 | 44 687 | 0 | 20 743 | |||
| Liver | 3 | 21 521 | 6 | 15 763 | 0.181 | 2.7+ | |
| Lung | 2 | 102 767 | 14 | 54 087 | 1.8 × 10-5 | 13.3+ | I+ |
| Muscle/skeleton | 0 | 58 355 | 0 | 27 075 | |||
| Ovary | 13 | 33 707 | 9 | 41 943 | 0.200 | 1.8- | |
| Pancreas | 0 | 60 518 | 3 | 21 810 | 0.019 | ||
| Prostate | 10 | 106 113 | 43 | 76 774 | 9.0 × 10-9 | 5.9+ | II+ |
| Stomach/oesophagus | 0 | 13 801 | 2 | 12 120 | 0.219 | ||
| Testis | 0 | 24 906 | 0 | 16 900 | |||
| T lymphoma | 1 | 60 743 | 0 | 17 318 | 1.000 | ||
| Uterus | 2 | 67 669 | 3 | 21 740 | 0.096 | 4.7+ | |
DISCUSSION
We have presented an integrated procedure for the mining of EST libraries for genes differentially expressed in normal and tumour tissues. This procedure comprises four steps: normalisation of an EST library of interest yielding a set of non-redundant ESTs (seeds); iterative assembly of all available ESTs homologous to the seeds (AUTEX); generation of electronic northerns for the resulting elongated sequences; and characterisation of the assemblies through database searches. Almost 10% (139/1406) of the assembled sequences were found to be significantly overexpressed (expression ratio at least 2, [alpha] = 0.05) in breast tumour tissue and less than 1% (13/1406) were down-regulated in tumour tissue. This shows clearly that EST libraries obtained from tumour tissue are enriched with up-regulated genes. Both the set of up- and down-regulated genes contain candidate genes which could be involved in the development of tumours.
We introduced a classification scheme of differential expression comprising three levels, moderate (class III), strong (class II) and and very strong (class I), to describe the degree of differential expression. Interestingly, many genes of class I turned out to be involved in cell metabolism or tumour infiltration, whereas many well-known tumour-related genes were classified in classes II and III. Thus, on the one hand, the degree of overexpression does not seem to reflect the significance of a gene in tumourigenesis. On the other hand, there is good reason to believe that, apart from the known tumour-associated genes, our set contains further, tumour-associated genes.
Due to the enormous variance in size of tissue-specific EST pools (ranging from about 10 000 to almost 180 000 ESTs), absolute numbers of matching ESTs found in BLAST searches do not suffice to assess whether or not a gene is differentially expressed. Even the expression ratio, i.e. the quotient of the relative abundances in normal and tumour tissues, does not correct these circumstances entirely. Expression ratios derived from low hit numbers do not allow statistically solid conclusions, whereas the same expression ratio can be statistically significant if derived from higher hit numbers. For example, a hit ratio of 5:1 observed in two pools each of 10 000 ESTs is associated in Fisher's exact test (two-tailed) with a P value of 0.2187, while a hit ratio of 10:2 is statistically substantiated by a P value of 0.0224. Thus, one and the same expression ratio 5 is statistically non-significant by traditional criteria in the first case, whereas it is statistically significant in the second case at the level [alpha] = 0.05. We found the P value to be especially helpful when one of the two pools was completely devoid of hits. Then the expression ratio is not defined, but the P value computed by Fisher's exact test still allows statistical assessment and classification. For example, the P values for the two hit distributions 5:0 and 6:0 are 0.0625 (not significant, unclassified) and 0.0312 (significant at the level [alpha] = 0.05, class III), respectively (pools of 10 000 ESTs).
We were also interested in whether differential expression is detectable without prior assembly and to this end calculated electronic northerns from all 1465 ESTs of the non-redundant seed set. One hundred and fifty-five ESTs showed differential expression (expression ratio greater than 2, P value below 0.05) in breast tissue. However, only 100 of those differentially expressed ESTs were found among the seed ESTs which finally led to the set of 152 differentially expressed elongated sequences. This means that differential expression determined on the basis of non-elongated ESTs leads to rates of about 30% false positives and 30% false negatives.
For example, the EST which served as seed in the assembly resulting in the murine JIP-1 gene homologue did not collect any breast EST apart from itself (P = 0.40, unclassified), while it collected eight ESTs from normal brain libraries and one EST from a brain cancer library (P = 0.17, expression ratio 4.5, unclassified). This gene thus would not have been classified as differentially expressed by the stringent criteria used in this study. For comparison, the elongated sequence matched four ESTs in breast cancer and none in normal breast tissue (P = 0.017, class III+ ), and 25 ESTs in normal brain and five ESTs in brain cancer tissue (P = 0.036, expression ratio 2.8, class III-; see Table 3). As in Tables 3-5, the superscripts + and - indicate up- and down-regulation in tumour tissue, respectively.
We used for this case study our full repository of about 4 000 000 ESTs, 3 000 000 of which were from a proprietary database. The use of exclusively public ESTs permits the creation of EST pools large enough (at least 10 000 ESTs) for the calculation of electronic northerns for eight tissues, among them tissues of great pharmacological interest, like prostate, breast and lung. Since many (122 out of 152) of our assembled sequences represent known genes with an E value of 10-100 or better in BLAST searches against nt (Table 2), the procedure could be modified by restricting the assembly step to ESTs with no clear annotation and using the matching full-length gene sequence for the calculation of electronic northerns. Building assemblies for the remaining 30 unknown ESTs from only public ESTs is, according to our experience, possible, but results in poorer coverage of the assembly, hence in a somewhat poorer sequence quality and shorter sequence length.
The software for this analysis, essentially shell scripts controlling BLAST searches and EST sequence assemblies, was not optimised for speed; the time necessary for complete analysis of one EST depends very much on the number of cycling steps taken to build a contig of maximal length and on the number of ESTs involved in the assembly step, but was estimated to be of the order of 10 min on a 250 MHz Sun Ultra SPARC processor.
In future studies, various steps could be undertaken to accelerate and improve the suggested protocol, e.g. a filtering procedure at the very beginning to avoid housekeeping genes being analysed. Further characterisation of the set of assembled sequences could include Pfam searches and translating searches against protein databases.
ACKNOWLEDGEMENTS
We thank Ralf Herwig (Max Planck Institute for Molecular Genetics, Berlin) and Henrik Seidel (Schering AG, Berlin) for valuable contributions, and Daryl Faulds (Berlex Biosciences, Richmond, CA), Robert Grützmann and Birgit Weber (metaGen) for critical review of the manuscript.
REFERENCES
*To whom correspondence should be addressed. Tel: +49 30 8413 1668; Fax: +49 30 8413 1674; Email: armin.schmitt{at}metagen.de Present address: Georg Beckmann, Hoechst Marion Roussel, 102 Route de Noisy, F-93235 Romainville, France
This page is run by Oxford University Press, Great Clarendon Street, Oxford OX2 6DP, as part of the OUP Journals
Comments and feedback: jnl.info{at}oup.co.uk
Last modification:
Copyright© Oxford University Press, 1999.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  M. Tsujimoto, K. Nakabayashi, K. Yoshidome, T. Kaneko, T. Iwase, F. Akiyama, Y. Kato, H. Tsuda, S. Ueda, K. Sato, et al. One-step Nucleic Acid Amplification for Intraoperative Detection of Lymph Node Metastasis in Breast Cancer Patients Clin. Cancer Res., August 15, 2007; 13(16): 4807 - 4816. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. C. Georg and S. L. Gomes Transcriptome Analysis in Response to Heat Shock and Cadmium in the Aquatic Fungus Blastocladiella emersonii Eukaryot. Cell, June 1, 2007; 6(6): 1053 - 1062. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. H. Nagaraj, R. B. Gasser, and S. Ranganathan A hitchhiker's guide to expressed sequence tag (EST) analysis Brief Bioinform, January 1, 2007; 8(1): 6 - 21. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
E. Dahl, G. Kristiansen, K. Gottlob, I. Klaman, E. Ebner, B. Hinzmann, K. Hermann, C. Pilarsky, M. Durst, M. Klinkhammer-Schalke, et al. Molecular Profiling of Laser-Microdissected Matched Tumor and Normal Breast Tissue Identifies Karyopherin {alpha}2 as a Potential Novel Prognostic Marker in Breast Cancer. Clin. Cancer Res., July 1, 2006; 12(13): 3950 - 3960. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M Burkhardt, E Mayordomo, K-J Winzer, F Fritzsche, T Gansukh, S Pahl, W Weichert, C Denkert, H Guski, M Dietel, et al. Cytoplasmic overexpression of ALCAM is prognostic of disease progression in breast cancer J. Clin. Pathol., April 1, 2006; 59(4): 403 - 409. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
P. J. Wild, A. Herr, C. Wissmann, R. Stoehr, A. Rosenthal, D. Zaak, R. Simon, R. Knuechel, C. Pilarsky, and A. Hartmann Gene Expression Profiling of Progressive Papillary Noninvasive Carcinomas of the Urinary Bladder Clin. Cancer Res., June 15, 2005; 11(12): 4415 - 4429. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. J. Wild, R. C. Krieg, J. Seidl, R. Stoehr, K. Reher, C. Hofmann, J. Louhelainen, A. Rosenthal, A. Hartmann, C. Pilarsky, et al. RNA expression profiling of normal and tumor cells following photodynamic therapy with 5-aminolevulinic acid-induced protoporphyrin IX in vitro Mol. Cancer Ther., April 1, 2005; 4(4): 516 - 528. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. Qiu, L. Guo, T.-J. Wen, F. Liu, D. A. Ashlock, and P. S. Schnable DNA Sequence-Based "Bar Codes" for Tracking the Origins of Expressed Sequence Tags from a Maize cDNA Library Constructed Using Multiple mRNA Sources Plant Physiology, October 1, 2003; 133(2): 475 - 481. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Kristiansen, C. Denkert, K. Schluns, E. Dahl, C. Pilarsky, and S. Hauptmann CD24 Is Expressed in Ovarian Cancer and Is a New Independent Prognostic Marker of Patient Survival Am. J. Pathol., October 1, 2002; 161(4): 1215 - 1221. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Trendelenburg, K. Prass, J. Priller, K. Kapinya, A. Polley, C. Muselmann, K. Ruscher, U. Kannbley, A. O. Schmitt, S. Castell, et al. Serial Analysis of Gene Expression Identifies Metallothionein-II as Major Neuroprotective Gene in Mouse Focal Cerebral Ischemia J. Neurosci., July 15, 2002; 22(14): 5879 - 5888. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. J. Riggins and R. L. Strausberg Genome and genetic resources from the Cancer Genome Anatomy Project Hum. Mol. Genet., April 1, 2001; 10(7): 663 - 667. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. A. Prade, P. Ayoubi, S. Krishnan, S. Macwana, and H. Russell Accumulation of Stress and Inducer-Dependent Plant-Cell-Wall-Degrading Enzymes During Asexual Development in Aspergillus nidulans Genetics, March 1, 2001; 157(3): 957 - 967. [Abstract] [Full Text]
|
|
|
|
 |
|
 J. Muilu, P. Rodriguez-Tomé, and A. Robinson GBuilder---An Application for the Visualization and Integration of EST Cluster Data Genome Res., January 1, 2001; 11(1): 179 - 184. [Abstract] [Full Text]
|
|
|
|
 |
|
 D. Scheurle, M. P. DeYoung, D. M. Binninger, H. Page, M. Jahanzeb, and R. Narayanan Cancer Gene Discovery Using Digital Differential Display Cancer Res., August 1, 2000; 60(15): 4037 - 4043. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||