Nucleic Acids Research

Oversampling in differential display

One of the key determinants of differential display’s effectiveness is the process by which the total mRNA population is divided into subsets. This division is a necessary consequence of the limited number of discernible bands that can be displayed on a sequencing-length gel, compared with the total number of genes expressed in a particular cell type or tissue. In an ideal differential display system, the composition of each mRNA subset would be non-overlapping, preventing mRNAs from being sampled repetitively. This is important, because if the aim of an experiment is to compare the expression of all or most of the mRNAs in a system then oversampling is counterproductive, greatly increasing the total number of analyses that have to be made.

Two strategies are used in differential display to divide the total mRNA population into subsets. The first relies on the use of multiple reverse transcription reactions, with each reaction employing a different ‘anchored’ primer. The second involves the use of so-called ‘arbitrary’ primers during the PCR phase. Usually, two base anchored primers of the form T₁₂MN (where M = A, C or G and N = A, C, T or G) are employed to divide the mRNA population into 12 pools. Each of these pools is then PCR amplified using the same anchored primer used for reverse transcription, in combination with an array of arbitrary primers to yield the amplified subsets ready for gel analysis. Because one or more of the differential display PCR cycles are carried out under low-stringency annealing conditions, the arbitrary primer can initiate synthesis even when a limited number of primer-template mismatches occur, thereby enabling synthesis to proceed at a variety of sites. Theoretical models of arbitrary primer binding predict that this arbitrary primer mismatch tolerance will lead to the tendency for high abundance mRNAs to be amplified by more of the arbitrary primers than low abundance mRNAs (1). Consequently, high abundance genes will be oversampled by differential display, presumably at the expense of low abundance genes. Improvements to the original differential display method have been devised to increase the technique’s ability to sample low abundance genes and limit the repetitive sampling of high abundance genes. These improvements include increasing the length [and therefore the melting temperature (T_m)] of the primers, limiting the number of low stringency cycles to either one or four, increasing the dNTP concentration in the reaction and halting the reaction whilst still within the exponential phase of PCR amplification (i.e. 19-25 cycles, depending on the system, rather than the 40 cycles originally proposed) (2-4). These ‘second generation’ differential display methods generate very reproducible banding patterns and enable both qualitative and quantitative differences in gene expression to be detected and should theoretically reduce the tendency to repeatedly sample high abundance genes.

The low T_m of the anchored primers makes them particularly susceptible to promiscuous priming during PCR and, whereas ‘arbitrary mispriming’ is necessary for differential display, anchored primer mispriming is disadvantageous since it invariably results in repetitive sampling. Although mis-priming is a phenomenon generally associated with thermostable DNA polymerases like Taq, if anchored primer mispriming were to occur during reverse transcription, this would lead to a great deal of repetitive sampling during differential display. Degenerate priming by reverse transcriptase at the M position of T₁₂MN anchored primers has already been shown to occur during differential display (5). Such mispriming would be expected to generate differential display patterns with many bands in common when one particular arbitrary primer is used in combination with each of the primers T₁₂CX, T₁₂AX and T₁₂GX (where X is one specific base, either A, C, T or G). Yet, others have observed that the pattern of bands produced with these particular primer combinations are quite different and that few bands are common to all displays (2,6,7).

Anchored primers and polyadenylation sites

In their initial account of the differential display technique, Liang and Pardee (8) assumed that any one of the 12 anchored primers would, by probability, recognise one-twelfth of the total mRNA population. However, work by Linskens et al. (2) using an ‘enhanced’ differential display method, provided evidence that this was not the case. They found that (in combination with any particular arbitrary primer) some anchored primers produced differential display patterns containing many more bands than others and consequently yielded many more differentially expressed gene transcripts. Thus, they argued that the dinucleotides targeted by some anchored primers probably occurred in the mRNA population more frequently than others.The experimental results of Linskens et al. were supported by the theoretical analysis of polyadenylation site usage described by Sheets et al. (9). By analysing the genomic and cDNA sequences of 63 unrelated genes, Sheets and co-workers had searched for sequence motifs which poly(A) polymerase used in determining where to attach the poly(A) tail. Their results suggested that although this attachment position was typically located 15-25 bases downstream of the almost ubiquitous AAUAAA polyadenylation signal, it did not conform to any particular sequence motif. However, they did discover that the sequence preceding the poly(A) tail was not random and that, in accord with the experimental data of Linskens et al., some dinucleotides occurred more frequently than others (reviewed in 10-12).

Since Sheets et al. were able to compare cDNA sequences with their genomic counterparts, they were able to determine whether the first adenine base in a poly(A) tail could have been encoded by genomic DNA. If an adenine base at this position was encoded by the genomic sequence, they assumed that it arose as part of the pre-mRNA and thus was not strictly part of the poly(A) tail. Similarly, in cases where two adenine bases occurred, both of these bases were assumed to originate as part of the immature mRNA and it was hypothesised that this existing AA dinucleotide would serve as the attachment point of the poly(A) tail (see 13-15 for a review of the polyadenylation process).

Mammalian polyadenylation sites targeted by anchored primers

Because the theoretical analysis of polyadenylation sites carried out by Sheets et al. made the assumption that the first one or more adenine residues in a poly(A) tail could potentially have arisen from the transcription of genomic DNA, their estimation of the frequency distribution of the dinucleotide preceding the poly(A) tail cannot be directly related to differential display anchored primer binding. However, we have re-analysed the data of Sheets et al. to take account of this ‘genomic A’ phenomenon. The re-analysed data enables the frequency of anchored primer binding to be assessed, answering the question ‘Do each of the 12 anchored primers used in differential display target a similar number of mRNAs?’ In addition, we have conducted our own theoretical analysis of polyadenylation site usage (based on a very much larger sample of mRNA sequences contained in the EMBL/GenBank database and in a range of defined mammalian species). In conjunction with our own experience with a second generation differential display technique, this work raises concerns about the degree of oversampling that occurs during differential display and implicates reverse transcription as a key limiting factor in present methods.

Tissue preparation

We are using differential display to investigate retinal gene expression in the tree shrew (Tupaia belangeri) model of myopia (16). A tree shrew was monocularly deprived of focused vision for 24 h to induce myopia. After homogenisation of retinal tissue in a Teflon freezer mill, total RNA from each eye was isolated with TRI Reagent (Sigma) according to the manufacturer’s instructions. Phase separation of the TRI Reagent was facilitated by the use of ~100 µl of silicone vacuum grease (Dow Corning) during RNA isolation (17). Residual contaminating DNA was removed by incubating with 10 U RNase-free DNase (Stratagene) and 30 U RNase inhibitor (RNA Guard; Pharmacia) under standard buffer conditions (40 mM Tris, 6 mM MgCl₂, 2 mM CaCl₂, 1 mM DTT, pH 7.5) at 37°C for 60 min. The TRI Reagent extraction was repeated and the isolated RNA resuspended in 30 µl of DEPC-treated H₂O and quantified by absorbance at 260 nm. The RNA samples were then diluted with DEPC-treated H₂O into 2.5 µg aliquots and stored at -80°C.

Synthesis of first strand cDNA

Reverse transcription was performed in 12 independent reactions using the 3[prime] anchored primers described by Linskens et al. (2). Reactions contained 2.5 µg of total RNA, first strand buffer(1× final concentration), 2.5 µM anchored primer, 25 µM each dNTP, 10 mM DTT, 30 U RNA Guard and 200 U reverse transcriptase (Superscript II; Gibco BRL). The samples were incubated at 25°C for 20 min to anneal the primers, followed by 60 min at 37°C and then 15 min at 45°C. The RNA was degraded at 95°C for 10 min and then chilled on ice. The single-stranded cDNA was mixed with 230 µl ice-cold TE (10 mM Tris-HCl, pH 8.0, 1 mM EDTA), dispensed into six 40 µl aliquots and stored at -20°C until used.

PCR amplification of the cDNA

A 20 µl reaction mixture was prepared using 5 µl of diluted cDNA, 1× PCR buffer (50 mM KCl, 10 mM Tris-HCl, 0.1% Triton X-100), 1 µM both arbitrary and anchored primers, 5 µM each dNTP, 1 mM MgCl₂, 1 U Taq polymerase (Promega) and 0.7 µCi [[alpha]-³³P]dCTP (Amersham). The mixture was overlaid with 30 µl of mineral oil and subjected to PCR: 94°C for 5 min, 37°C for 5 min, followed by a slow temperature ramp of 7°C/min to 72°C for 5 min. This initial low stringency cycle was followed by 18 high stringency cycles of 94°C for 1 min, 55°C for 1 min, 72°C for 2 min and a final extension of 72°C for 5 min. After amplification, the aqueous sample was transferred to a fresh tube and mixed with 5 µl of 5× loading buffer (15% Ficoll 400, 10 mM EDTA, 0.13% xylene cyanol FF). Reaction products (5 µl) were resolved on a non-denaturing 6% polyacrylamide/1× TBE sequencing-length gel. Gels were dried onto Whatman 3MM paper, labelled with fluorescent tape (Radtape; Sigma) and exposed to BioMax MS X-ray film (Kodak) for 24 h.

Frequency of anchored primer binding in the mRNA population

The differential display anchored primers should, statistically, recognise one-twelfth of the mRNA population, assuming a random frequency distribution of the dinucleotide immediately preceding the poly(A) tail. To determine if the dinucleotide frequency has a random distribution, a group of unrelated mRNA sequences with known polyadenylation sites were examined. The EMBL/GenBank database was searched using the ‘Entrez’ browser (http://www3.ncbi.nlm.nih.gov/Entrez/nucleotide.html ) for full-length mammalian mRNA sequences with identified polyadenylation sites. The resulting group of nearly 4000 sequences was modified by excluding ESTs, commercial clones, splice variants and those transcripts described as being similar to other mRNA sequences. This modified search (see Appendix) identified a group of ~1400 mRNA sequences for analysis.

The mRNA sequences were downloaded from the EMBL/GenBank database as text files and the six bases immediately prior to the polyadenylation site of each sequence entered into an Excel spreadsheet along with their accession numbers, animal species and gene names. Sequences were sorted into separate subgroups according to animal species. Duplicate entries for the same gene product in each of the species groups were deleted along with any multiple members of the same gene ‘family’ to remove any sample bias, especially towards high abundance genes. The distinct species subgroups were also combined in a total group, which was further modified by deleting entries for the same gene product duplicated in the different species subgroups, along with multiple members of the same gene family. These distinct sequences were analysed to determine the frequency of each base in each of the six positions prior to the poly(A) tail, both for the total group and for the human and rodent animal species subgroups (other species subgroups contained too few entries to warrant individual analysis). The resulting frequency matrices were used to calculate the relative proportion of the total mRNA population that each individual differential display anchored primer would bind to.

Statistics

To assess whether the dinucleotide targeted by the ‘anchor’ region of each differential display anchored primer had a random frequency distribution, we tested the null hypothesis that each of the 12 possible target dinucleotides were represented in the total mRNA sample one-twelfth of the time. Thus, under the null hypothesis the expected proportion [P_exp] of the mRNA sample corresponding to each target dinucleotide would be predicted to be 0.083 (1 divided by 12) and the standard error of this expected proportion [se(p)] would be 0.0093 (for n = 889 sequences) (18). The test statistic Z was calculated for each of the actual target dinucleotide frequencies using the formula Z = (P_obtained - P_exp)/se(p) and from this value a probability estimate was established.

Mammalian polyadenylation site usage with respect to anchored primers

An analysis of several hundred unique mRNA sequences obtained from the EMBL/GenBank database enabled the nucleotide frequency distribution at each of the six positions prior to the poly(A) tail to be determined, for both a total group and individual human and rodent animal species subgroups (Table 1). The nucleotide frequency distribution values for each of these three groups appeared to be similar.

Table 1. Frequency distribution of nucleotides immediately prior to the polyadenylation site (+1) for each of the animal species groups
‘Human’ sample, n = 407; ‘Rodent’ sample, n = 463; ‘Total’ sample, n = 889.

These data allowed us to calculate the relative proportion of the total mRNA population that each individual differential display anchored primer would bind to (Table 2 and Fig. 1). However, unlike the study of Sheets et al. (9), we could not define the exact site at which poly(A) polymerase most commonly adds the poly(A) tail, because of the ‘genomic A’ phenomenon. It can be seen from Table 2 that, as first noted by Linskens et al. (2), anchored primer binding is far from random. Nevertheless, the difference in the proportion of the mRNA population targeted by the primers with the greatest and least share, respectively, is only 2.5-fold. This differs markedly from the results of Sheets et al., when their data has been re-analysed to account for the ‘genomic A’ phenomenon, in that it predicts a near 10-fold difference between primer extremes.

Table 2. The proportion (%) of the mRNA population (in a sample of 889 unrelated mammalian sequences) targeted by each of the anchored primers, in comparison with the re-analysed data of Sheets et al. (9) based on a sample of 63 unrelated vertebrate sequences
Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Figure 1. Proportion of the mRNA population targeted by each of the 12 differential display anchored primers based on a theroretical analysis of 889 distinct mammalian transcripts (dark shaded bars) and a re-analysis of the 63 vertebrate sequences investigated by Sheets et al. (9) (light shaded bars).

Experimental evidence of reverse transcriptase mispriming

The frequency distribution analysis of the dinucleotide preceding the poly(A) tail suggests that some differential display anchored primers should produce a more diverse range of first strand cDNA transcripts than others. In the most extreme case, for example, primer K would be expected to yield a 2.5-fold more varied transcript repertoire than primer A. This in turn would be expected to result in more complex differential display patterns from reactions using primer K when compared with primer A. The kinetics of differential display PCR reactions are likely to be complex, however, there is evidence that: (i) some tissues yield simpler banding patterns than others; (ii) all transcripts appear to be amplified exponentially during the exponential phase of PCR (6; M.R.Frost and J.A.Guggenheim, unpublished observations). Thus, if a 2.5 times more diverse set of first strand cDNA templates were available, one would expect a 2.5-fold more diverse set of PCR products. However, this would not necessarily lead to a 2.5-fold increase in the number of bands visible on a differential display autoradiograph as many of the products might be below the abundance threshold required for band visualisation, although one would expect to see a significant increase. This is not the case. In fact, we found the number of discernible bands produced in reactions with primers A and K to be almost identical (primer A, 141 ± 4 bands; primer K, 145 ± 4 bands; P = 0.10 paired t-test; n = 20 reactions, each with a different arbitrary primer) (Fig. 2).

Figure 2. Representative differential display autoradiographs comparing the two different anchored primers A and K in combination with identical arbitrary primers. Note that although the banding patterns are quite distinct, the number of bands produced with each of the anchored primers is similar. Each panel shows differential display reactions from four different retinal mRNA samples.

The experimental results of Linskens et al. suggested to them that some of their anchored primers were more ‘effective’ than others, as judged by the number of bands they displayed and the number of differentially expressed gene transcripts they yielded. As predicted from the data of Sheets et al., Linskens found that anchored primers targeting dinucleotides ending with a C were much more efficient than those ending with either a G or T (note that no anchored primers end with an A). The share of the mRNA population targeted by primers ending with a G, C or T predicted by our data and that of Sheets, and the ‘primer share’ determined experimentally by Linskens et al. are shown in Table 3. It is apparent that the ‘efficiency’ results obtained by Linskens et al. are inconsistent with both our theoretical analysis of primer share and that of Sheets, although the latter results provided by far the better fit. As an explanation, we suggest that the number of differentially expressed bands yielded by a particular anchored primer may in fact be a poor indicator of its primer share. The more direct measurements of primer share we have made from differential display autoradiographs (such as those described above for primers A and K) suggest that the proportion of mRNAs targeted by each anchored primer does not differ greatly. This result is also in accord with our theoretical analysis of primer share (Table 2), which is based on a very much larger sample than has been investigated previously. Despite the fact that the ‘genomic A’ phenomenon prevents us from directly assessing the sequence preference of the pre-mRNA cleavage and poly(A) addition site, our results suggest that other than adenine bases, this site most commonly ends T > C > G. This same order of site preference has also been described by Chen et al. (19) in their analysis of cleavage site determinants in the mammalian polyadenylation signal.

Table 3. Theoretical and experimental evaluations of the proportion of the total mRNA population targeted by differential display anchored primers

We conclude that the similarity in the number of bands produced by each anchored primer is most likely due to promiscuous priming of the reverse transcriptase. Two types of mispriming event can be envisaged. First, mispriming of one or more bases in the anchor region of the primer and, second, priming at an internal site on the mRNA rather than to the poly(A) tail. There is experimental evidence that both types of mispriming event are common (see below).

To explore the reverse transcriptase mispriming phenomenon in more detail, we tested the ability of a [beta]-actin forward primer to amplify the 3[prime]-region of tree shrew retinal [beta]-actin mRNA when used in conjunction with each of the 12 differential display anchored primers. A series of 12 3[prime] RACE (rapid amplification of cDNA ends) reactions were performed, with each reverse transcription reaction being carried out exactly as for differential display, i.e. each reaction containing a different anchored primer (the RACE paradigm is analogous to differential display except that a specific forward primer is substituted for the non-specific arbitrary primer). Following reverse transcription, 19 cycles of high-stringency PCR were performed. (Note that the first, low-stringency cycle adopted in our differential display experiments was replaced in these RACE experiments by a high-stringency cycle to reduce mispriming by the [beta]-actin primer. Otherwise, reaction conditions were identical to those for differential display.) A second set of PCR reactions was conducted as described above, but with an annealing temperature of 65 instead of 55°C to increase the stringency of priming.

Using the 3[prime] RACE paradigm, we expected not only to be able to identify reverse transcriptase mispriming, but also to be able to differentiate between the two types of mispriming event mentioned above, namely mispriming in the anchor region adjacent to the poly(A) tail and internal priming. [beta]-Actin was chosen as a target since it is abundantly expressed in most mammalian tissues and because its mRNA is likely to have only one possible 3[prime]-end sequence. We based this assumption on the evidence that [beta]-actin is usually transcribed from a single gene in mammals and without splice variants. Thus, although numerous [beta]-actin pseudogenes exist in most mammalian genomes, they are not normally expressed (20,21).

Figure 3. Comparative 3[prime] RACE reactions, each employing one of the 12 different anchored primers (A-M) in combination with an upstream [beta]-actin primer (5[prime]-TGGAGAAGAGCTACGAGCTGCCTG-3[prime]), using an annealing temperature of 55 (A) or 65°C (B). The expected product of 1067 bp is evident in several lanes at both annealing temperatures.

If reverse transcriptase mispriming were never to occur, we would expect only one of the 12 anchored primers to yield a [beta]-actin 3[prime] RACE product. However, if mispriming did occur, we would expect several of the anchored primers to yield products. Whereas mispriming in the anchor region would generate products of equal length, internal priming would be expected to yield truncated products. Our results (Fig. 3) suggest that both types of reverse transcriptase mispriming have occurred. Full-length (~1 kb) 3[prime] RACE products were produced by several anchored primers (subsequent DNA sequencing and high-stringency hybridisation with a [beta]-actin probe showed that 11 of the 12 anchored primers gave rise to the full-length product (see Fig. S1 in supplementary material). Truncated products were also synthesised (a minority of which were homologous to [beta]-actin). Even raising the annealing temperature of the subsequent PCR reaction above the optimal T_m of the primers in an attempt to increase stringency, and therefore further limit mispriming due to Taq polymerase rather than reverse transcriptase, did not prevent this effect.

In conclusion, our results appear to substantiate one of the key theoretical principles of Liang and Pardee’s differential display paradigm (8), namely that anchored primers provide a mechanism for dividing the total mRNA population into 12 subsets of roughly equal size. In an analysis of several hundred unrelated mammalian mRNA sequences, we confirmed that although the frequency of the dinucleotide targeted by the anchor region of anchored primers does vary, the extent of this variation is <3-fold. Of equal importance, however, we found that the number of bands displayed with each of the respective anchored primers was not affected by these variations in dinucleotide frequency, suggesting that anchored primer promiscuity permits mispriming during the reverse transcription stage of differential display. Experiments in which we characterised reverse transcription mismatch tolerance in detail for one high abundance gene ([beta]-actin) supported this view. Furthermore, both mispriming at the anchor region and internal mispriming were evident. This implies that many genes will be represented in several of the first strand cDNA pools produced after reverse transcription, at least for high abundance genes, and that this will lead to repetitive sampling in subsequent differential display PCR reactions. Consequently, it is impossible to predict the proportion of the mRNA population that has been screened in differential display experiments, unless the extent of repetitive sampling has first been quantified. Methodological modifications designed to reduce the promiscuity of reverse transcriptase priming, such as increasing the reaction temperature, should greatly improve the efficiency of differential display as a gene expression screening technique.

ACKNOWLEDGEMENTS

The authors are grateful to Dipak Ramji, David Millar and two anonymous reviewers for their expert appraisal of this manuscript and to Corey Baxter for help with the processing of sequence data. This work was supported in part by a grant from the National Eye Research Centre.

See supplementary material available in NAR Online (29.3 KB PDF file).

REFERENCES

1. Bertioli,D.J., Schlichter,U.H.A., Adams,M.J., Burrows,P.R., Steinbiß,H.-H. and Antoniw,J.F. (1995) Nucleic Acids Res., 23, 4520-4523. MEDLINE Abstract

2. Linskens,M.H.K., Feng,J., Andrews,W.H., Enlow,B.E., Saati,S.M., Tonkin,L.A., Funk,W.D. and Villeponteau,B. (1995) Nucleic Acids Res., 23, 3244-3251. MEDLINE Abstract

3. Ito,T., Kito,K., Adati,N., Mitsui,Y., Hagiwara,H. and Sakaki,Y. (1994) FEBS Lett., 351, 231-236. MEDLINE Abstract

4. Zhao,S., Ooi,S.L. and Pardee,A.B. (1995) BioTechniques, 18, 842-850. MEDLINE Abstract

5. Liang,P., Averboukh,L. and Pardee,A.B. (1993) Nucleic Acids Res., 21, 3269-3275. MEDLINE Abstract

6. Bauer,D., Müller,H., Reich,J., Riedel,H., Ahrenkiel,V., Warthoe,P. and Strauss,M. (1993) Nucleic Acids Res., 21, 4272-4280. MEDLINE Abstract

7. Pardinas,J.S., Combates,N.J., Prouty,S.M., Stenn,K.S. and Parimoo,S. (1998) Anal. Biochem., 257, 161-168. MEDLINE Abstract

8. Liang,P. and Pardee,A.B. (1992) Science, 257, 967-971. MEDLINE Abstract

9. Sheets,M.D., Ogg,S.C. and Wickens,M.P. (1990) Nucleic Acids Res., 18, 5799-5805. MEDLINE Abstract

10. Birnstiel,M.L., Busslinger,M. and Strub,K. (1985) Cell, 41, 349-359. MEDLINE Abstract

11. Wahle,E. and Keller,W. (1992) Annu. Rev. Biochem., 61, 419-440. MEDLINE Abstract

12. Wahle,E. and Keller,W. (1997) In Krainer,A.R. (ed.), Eukaryotic mRNA Processing. IRL Press, Oxford, UK, pp. 280-309.

13. Wahle,E. and Keller,W. (1996) Trends Biochem. Sci., 21, 247-250. MEDLINE Abstract

14. Colgan,D.F. and Manley,J.L. (1997) Genes Dev., 11, 2755-2766. MEDLINE Abstract

15. Hirose,Y. and Manley,J.L. (1998) Nature, 395, 93-96. MEDLINE Abstract

16. Frost,M.R., Guggenheim,J.A. and McBrien,N.A. (1997) Invest. Ophthalmol. Vis. Sci., 38, S761.

17. Mukhopadhyay,T. and Roth,J.A. (1993) Nucleic Acids Res., 21, 781-782. MEDLINE Abstract

18. Altman,D.G. (1991) Practical Statistics for Medical Research. Chapman and Hall, London, UK.

19. Chen,F., MacDonald,C.C. and Wilusz,J. (1995) Nucleic Acids Res., 23, 2614-2620. MEDLINE Abstract

20. Ng,S.-Y., Gunning,P., Eddy,R., Ponte,P., Leavitt,J., Shows,T. and Kedes,L. (1985) Mol. Cell. Biol., 5, 2720-2732. MEDLINE Abstract

21. Elder,P.K., French,C.L., Subramaniam,M., Schmidt,L.J. and Getz,M.J. (1988) Mol. Cell. Biol., 8, 480-485. MEDLINE Abstract

APPENDIX

Entrez search parameters:

(mammals [Organism] AND mRNA [All Fields] AND complete [All Fields] AND (polya site [All Fields] OR poly a site [All Fields])) BUTNOT (clone* [All Fields] OR splice* [All Fields] OR exon [All Fields] OR related [All Fields] OR unknown [All Fields] OR pseudogene [All Fields] OR like [All Fields]).

---

*To whom correspondence should be addressed. Tel: +44 1222 874000; Fax: +44 1222 874859; Email: guggenheim@cf.ac.uk

---

This page is run by Oxford University Press, Great Clarendon Street, Oxford OX2 6DP, as part of the OUP Journals
Comments and feedback: www-admin{at}oup.co.uk
Last modification: 11 Feb 1999
Copyright©Oxford University Press, 1999.
CiteULike    Connotea    Del.icio.us    What's this?

This article has been cited by other articles:

				C. Y. Chan, J. A. Guggenheim, and C. H. To Is active glucose transport present in bovine ciliary body epithelium? Am J Physiol Cell Physiol, March 1, 2007; 292(3): C1087 - C1093. [Abstract] [Full Text] [PDF]

				D. Meintrup and E. Reisinger A statistical model providing comprehensive predictions for the mRNA differential display Bioinformatics, October 15, 2005; 21(20): 3880 - 3886. [Abstract] [Full Text] [PDF]

This Article Abstract Print PDF (169K) Supplementary Data Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Search for citing articles in: ISI Web of Science (10) Request Permissions Commercial Re-use Guidelines for Open Access NAR Content Google Scholar Articles by Frost, M. R. Articles by Guggenheim, J. A. Search for Related Content PubMed PubMed Citation Articles by Frost, M. R. Articles by Guggenheim, J. A. Related Collections Miscellaneous/other Monitoring gene expression RNA characterisation and manipulation Social Bookmarking          What's this?