Incidence of "quasi-ditags" in catalogs generated by Serial Analysis of Gene Expression (SAGE)
© Anisimov and Sharov; licensee BioMed Central Ltd. 2004
Received: 27 May 2004
Accepted: 18 October 2004
Published: 18 October 2004
Serial Analysis of Gene Expression (SAGE) is a functional genomic technique that quantitatively analyzes the cellular transcriptome. The analysis of SAGE libraries relies on the identification of ditags from sequencing files; however, the software used to examine SAGE libraries cannot distinguish between authentic versus false ditags ("quasi-ditags").
We provide examples of quasi-ditags that originate from cloning and sequencing artifacts (i.e. genomic contamination or random combinations of nucleotides) that are included in SAGE libraries. We have employed a mathematical model to predict the frequency of quasi-ditags in random nucleotide sequences, and our data show that clones containing less than or equal to 2 ditags (which include chromosomal cloning artifacts) should be excluded from the analysis of SAGE catalogs.
Cloning and sequencing artifacts contaminating SAGE libraries could be eliminated using simple pre-screening procedure to increase the reliability of the data.
Serial Analysis of Gene Expression (SAGE) is a rapid method to study mRNA transcripts in cell populations . Two major principles underline SAGE: (1) short expressed sequenced tags (ESTs) are sufficient to identify individual gene products, and (2) multiple tags can be concatenated and identified by sequence analysis [1, 2]. With the ever-expanding sequence information available in public databases, identification of gene transcripts with SAGE tags has greatly facilitated transcriptome comparisons and gene identification .
SAGE data are usually analyzed with software packages like "SAGE300" or "SAGE2000". The majority of SAGE libraries use NlaIII or Sau3A (SalAI) as anchoring enzymes (AE) to create SAGE tags. Both of these enzymes have 4-bp palindromic recognition sequences (CATG for NlaIII and GATC for Sau3A) that flank individual ditags within concatemers. A major component of the software analysis is the identification of anchoring enzyme recognition sequences (AERS) that flank target sequences (SAGE ditags). After finding the first AE recognition sequence, the software continues reading the sequence until it finds the next one. The software then compares the distance between these recognition sequences with predicted ditag lengths (20–24 bp in the case of NlaIII or Sau3A), and ditags that are too short (<20 bp) or too long (>24 bp) are excluded. However, if the length of the AERS-flanked sequence satisfies the size criteria, it is identified as a ditag. This algorithm relies on the assumption that all sequences have correctly organized ditag concatemers; however, the cloning efficiency of SAGE rarely reaches 100%. In this report, we show that up to 5% of ditags from some SAGE libraries should be omitted from the final analysis. These false ditags (termed "quasi-ditags") result from genomic contaminants and apparently random combinations of nucleotides generated by cloning or sequencing errors. Using a mathematical model to simulate the frequency of quasi-ditags in DNA, we propose a method to exclude quasi-ditags from SAGE catalogs.
Probability to find one or more "quasi-ditag" in the nucleotide sequence of the given length (P(20–24))
Sequence Length (L)
Frequency of ≥ 1 quasi-ditags in sequence
S. Cerevisiae chromosome 3
Computer simulation 1
In vivo 2 simulation (S. Cerevisiae)
SAGE is an important tool of modern molecular biology widely used in a number of applications. We hypothesized that actual SAGE catalogs could be contaminated by false ditags ("quasi-ditags") of various origins. Although SAGE software packages are designed to ignore sequences that lack 20–24 bp sequences flanked by two anchoring enzyme recognition sites, it does not exclude quasi-ditags originating from genomic contaminants or unknown sequences that may arise as cloning or sequencing artifacts (Figure 2). Negative controls (self-ligated vector) do not produce any colonies after Zeocin selection and cannot account for the appearance of background clones and quasi-ditags in Zeocin-resistant bacteria. Since some quasi-ditags, however, originate directly from E. Coli, we suggest that one probable source for these contaminanting tags is from recombination events that occur in E. Coli. Indeed, such a mechanism has already been documented  and has led to the development of Stbl2 bacteria that are mcrA-/mcrBC-hsdRMS-mrr-. Since pZErO-1 was not translated into recombination deficient bacteria (DH10B), large-scale amplifications of this plasmid within bacteria would be expected to lead to some random recombinations, and the generation of quasi-ditags (e.g. Figure 2A).
Some of the ditags derived from the clones that had produced a least number of ditags (1–2 per clone) do not match genomic sequences and thus might be originated from sequencing errors. We therefore suggested a model that provides a mathematical basis for the hypothesis that such a possibility exists. The mathematical model presented in the manuscript is an attempt to predict the frequency distribution of quasi-ditags in random sequences. The phenomenon itself is rather complex and there is no simple model that would capture it in full complexity. We, however, believe that we have selected a reasonable level of model complexity that captures the major pattern of frequency distribution.
Using the computer simulation we show that random combinations of nucleotides generated could be indeed recognized by SAGE software as valid SAGE ditags. We also demonstrate that quasi-ditags may constitute a non-negligible proportion of SAGE catalogs. Our model, which simulates the frequency of quasi-ditags in DNA (equations (1–6)), suggests that single or double ditags may represent quasi-ditags; however, the results of the in silico experiments show that the probability of finding more than two quasi-ditags in the same sequence converges effectively to zero (Table 1 and Figure 4). Based on these findings, we suggest that additional steps be performed with SAGE libraries. We recommend removing clones with sequences containing ≤ 2 ditags at a pre-processing step ("clean-up"). The removal of clones containing 1 or 2 ditags can effectively remove bacterial genomic sequences and potential sequencing artifacts from SAGE libraries. The overall number of SAGE tags excluded by this additional step (authentic and quasi-ditags) is usually low, and generally does not exceed 1.0–1.8% of the total number of sequenced SAGE tags [5, 6, 9, 10]; however, the frequency of potential quasi-ditags could be high (>5%) in some SAGE libraries. In AMH-I library, for example, the fraction of clones lacking appropriate ditag concatemers was >20%. In these instances, quasi-ditags significantly contribute to the final SAGE tag count, and should be removed.
Chart in Figure 6 plots values for ditag distribution from both the model-based simulations (L = 800 bp) and actual clones from the SAGE libraries that had sequences of the same mean length (L ≈ 800 bp). The expected maximum frequency of 1–2 quasi-ditags in the plotted model data approximated the observed frequency of clones with 1–2 total ditags detected in the pool of the actual SAGE clones. Contrary to that, the frequency of occurrence of three or more quasi-ditags predicted by the model is extremely low, demonstrating a divergence in the distribution of expected quasi-ditags and valid SAGE ditags for higher number of ditags per clone. Note that owing to the gel-purification of concatemers the majority of clones in the representative samples belong to the clusters of higher ditag numbers (AMH-II and ES R1 libraries, 13–26 total ditags; AMH-I library, 4–11 total ditags).
Comparing values of observed frequencies of the actual SAGE clones that produce 1–2 total ditags with those of expected quasi-ditag frequencies for the sequences of given length might be indicative on the possible contribution of cloning and sequencing artifact-derived quasi-ditags (Figure 6). The possible contribution of quasi-ditags to the final tag yield in SAGE libraries cannot be accurately predicted in advance but a failure to report the cloning efficiency and the number of clones with 1 or 2 ditags precludes an evaluation of potential false tags present in SAGE catalogs. Current SAGE protocols do not ensure 100% accurate size fractioning of concatemers: some of the smallest concatemers could therefore be cloned and sequenced. We recognize that some authentic tags (representing valid, but extremely short inserts that were not extracted during gel-purification of concatemers) will be excluded by removing all clones containing only 1 or 2 ditags. Nevertheless, we suggest that any potential loss of authentic ditags in the clean-up procedure is negligible compared to the advantage of having more reliable SAGE results.
SAGE protocols are extremely complex technologically and every possible mean should be employed to ensure qualitative and quantitative accuracy of catalogs on both the experimental and analytical steps. Evaluation of the cloning efficiency and precision (e.g. with RAST-PCR ) and sequencing accuracy are therefore essential on the stage preceding large-scale sequencing of the clones. Nonetheless, introduction of the simple pre-processing step eliminating false ditags would further improve the accuracy of the method resulting in its wider application.
We have hypothesized that actual SAGE catalogs could be contaminated by false ditags (termed "quasi-ditags") of various origins and employed a mathematical model to predict the frequency of quasi-ditags in random nucleotide sequences. Cloning and sequencing artifacts contaminating SAGE libraries could be eliminated using simple pre-screening procedure to increase the reliability of the data.
Serial analysis of gene expression (SAGE) was performed according to the original protocol  with minor modifications [5, 12]. Human (PC3) and mouse (P19, R1, D3, EG-1, MEF) cells and tissues (adult and old heart) have been employed for construction of SAGE libraries and sequence analysis to illustrate the "clean-up" process. SAGE tags were generated with NlaIII and BsmFI restriction enzymes (New England Biolabs, Beverly, MA, USA). Sequencing was performed by Perkin-Elmer Applied Biosystems / Celera Genomics (Foster City, CA, USA) and Agencourt Bioscience Corporation (Beverly, MA, USA).
Anchoring enzyme recognition sites (AERS) are 4 bp long. Assuming for simplicity that all 4 nucleotide bases (A, T, C, and G) have equal frequencies, a probability that a random combination of 4 nucleotides would match the AERS is 4-4 = 1/256. In a sequence of length L, the expected number of AERS (e.g. CATG for NlaIII anchoring enzyme) is L/256. Thus, the probability of finding k tags CATG in a random sequence of length L is determined by the Poisson distribution:
If two CATG sequences (AERS(CATG)) are located within the sequence of length L, then the probability that they are separated by a 20–24 bp distance (P(20–24)) is approximately:
where 10 is the number of possible relative positions of two AERS(CATG) that yield a quasi-ditag and 24 is the mean distance from the center of one SAGE tag to the end of the sequence that does not leave enough space for another tag to form a quasi-ditag.
If >2 AERS are present in the sequence, then there is a chance that additional AERS would appear within the quasi-ditag formed by first two AERS. A probability that additional AERS will not appear within the quasi-ditag is approximately:
where 30 is the average length of a nucleotide string outside of the ditag.
If the total number of AERS(CATG) equals k, then the number of possible AERS pairs is:
Taken together, a probability of at least one quasi-ditag in the sequence that has exactly k AERS(CATG) is:
Then, a probability (Q1) to find at least one quasi-ditag in a sequence of given length L is:
where p(k) is given by equation (1).
There is also a probability that more than one quasi-ditag exists within the sequence. In some cases the same AERS(CATG) could serve as a portion of the two neighboring quasi-ditags (...CATG-(N)20–24-CATG-(N)20–24-CATG...). In other cases, two or more quasi-ditags can be located independently in the sequence. If a sequence with k tags already has one quasi-ditag bounded by two tags, then other (k-2) tags may form additional quasi-ditags. The probability of existence of additional quasi-ditags on condition that one ditag is already present is approximately q(k-2). Then the total probability that any random sequence has at least two quasi-ditags is:
In the same way,
and so on.
The probability that a random sequence has exactly n quasi-ditags is:
R n = Q(n) - Q(n + 1) (9).
Software and analysis
A random nucleotide generator (for L = 600–1200) and analysis program that mimics "SAGE300" or "SAGE2000" software algorithms was written in Visual Basic and is available upon request. Genomic DNA sequences of Saccharomyces cerevisiae that lack SAGE ditag concatemers were also employed for in vivo / in silico model validation. Randomly selected S. cerevisiae chromosomes were downloaded from GenBank, fragmented to create a minimum of 300 sequences (L = 600–1200) and searched for quasi-ditags using "SAGE2000" software (available at SAGE website ). Frequency distribution of the number of ditags was analyzed in raw sequences from 3 randomly chosen 96-well plates from AMH-I, AMH-II and ES R1 SAGE libraries (285 sequences for each library) using the same software.
We would like to thank Dr. Paul Pullen (NIA/NIH, USA) for writing a code for software effecting computer simulations and Dr. Kenneth Boheler (NIA/NIH, USA) for the valuable help in preparing this manuscript.
- Velculescu VE, Zhang L, Vogelstein B, Kinzler KW: Serial Analysis of Gene Expression. Science 1995, 270: 484–487.View ArticlePubMedGoogle Scholar
- Velculescu VE, Madden SL, Zhang L, Lash AE, Yu J, Rago C, Lal A, Wang CJ, Beaudry GA, Ciriello KM, Cook BP, Dufault MR, Ferguson AT, Gao Y, He TC, Hermeking H, Hiraldo SK, Hwang PM, Lopez MA, Luderer HF, Mathews B, Petroziello JM, Polyak K, Zawel L, Zhang W, Zhang X, Zhou W, Haluska FG, Jen J, Sukumar S, Landes GM, Riggins GJ, Vogelstein B, Kinzler KW: Analysis of human transcriptomes. Nat Genet 1999, 23: 387–388. 10.1038/70487View ArticlePubMedGoogle Scholar
- Boheler KR, Stern MD: The new role of SAGE in gene discovery. Trends Biotechnol 2003, 21: 55–57. 10.1016/S0167-7799(02)00031-8View ArticlePubMedGoogle Scholar
- Angelastro JM, Ryu EJ, Torocsik B, Fiske BK, Greene LA: Blue-white selection step enhances the yield of SAGE concatemers. Biotechniques 2002, 32: 484–486.PubMedGoogle Scholar
- Anisimov SV, Tarasov KV, Tweedie D, Stern MD, Wobus AM, Boheler KR: SAGE identification of gene transcripts with abundances unique to pluripotent mouse R1 embryonic stem cells. Genomics 2002, 79: 169–176. 10.1006/geno.2002.6687View ArticlePubMedGoogle Scholar
- Anisimov SV, Tarasov KV, Stern MD, Lakatta EG, Boheler KR: A quantitative and validated SAGE transcriptome reference for adult mouse heart. Genomics 2002, 80: 213–222. 10.1006/geno.2002.6821View ArticlePubMedGoogle Scholar
- Tatusova TA, Karsch-Mizrachi I, Ostell JA: Complete genomes in WWW Entrez: data representation and analysis. Bioinformatics 1999, 15: 536–43. 10.1093/bioinformatics/15.7.536View ArticlePubMedGoogle Scholar
- Bhat MB, Hayek SM, Zhao J, Zang W, Takeshima H, Wier WG, Ma J: Expression and functional characterization of the cardiac muscle ryanodine receptor Ca(2+) release channel in Chinese hamster ovary cells. Biophys J 1999, 77: 808–816.PubMed CentralView ArticlePubMedGoogle Scholar
- Anisimov SV, Tarasov KV, Riordon D, Wobus A, Boheler KR: SAGE Identification of Differentiation Responsive Genes in P19 Embryonic Cells Induced to Form Cardiomyocytes in vitro. Mech Devel 2002, 117: 25–74. 10.1016/S0925-4773(02)00177-6View ArticleGoogle Scholar
- Potapova OU, Anisimov SV, Gorospe M, Dougherty RH, Gaarde WA, Boheler KR, Holbrook NJ: Identification of targets of JNK2 signaling involved in regulation of human tumor cell growth. Cancer Research 2002, 62: 3257–3263.PubMedGoogle Scholar
- van den Berg A, van der Leij J, Poppema S: Serial analysis of gene expression: rapid RT-PCR analysis of unknown SAGE tags. Nucleic Acids Res 1999, 27: e17. 10.1093/nar/27.17.e17PubMed CentralView ArticlePubMedGoogle Scholar
- Kenzelmann M, Muhlemann K: Substantially enhanced cloning efficiency of SAGE (Serial Analysis of Gene Expression) by adding a heating step to the original protocol. Nucleic Acids Res 1999, 27: 917–918. 10.1093/nar/27.3.917PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.