The initial search for sequence repeats was performed on the major intergenic regions of 29 different geminivirus or nanovirus genomic sequences (Figure 1 and Additional file 1). The search was arbitrarily halted after 105 candidate repeats were identified and was not intended to provide a comprehensive representation of all duplicated sequences within any of the viral sequences examined. Although generated using different search criteria than those employed by Arguello-Astorga et al 22, the resulting collection of geminivirus sequence repeats contains some sequences similar or identical to the described "iterons" (it should be noted that functional testing of nearly all of the "iterons" listed has not yet been reported in the literature).

Of the 105 repeats tested (Figure 1 and Additional file 1), 14 (13%) reproducibly resulted in increases of at least 25% above that of the 35S min construct (p < 5% by Student's T-test, the T-test was used only as a guide since by the nature of the assay used, individual data sets are small) (Figure 1 and Additional file 1). The remaining 91 (87%) failed to produce any measurable enhancement of reporter gene activity (see Additional file 1). All the positive elements identified by the in vivo assay were subsequently tested using an in vitro dual-luciferase^® system from Promega Corp. and produced levels of enhancement very similar to those obtained using the in vivo assay (the enhancement values and standard error reported in Figure 1 and Additional file 1 include both in vivo and in vitro data normalized to 35S min = 1.0). The observed enhancement of promoter activity (~2 fold) is relatively modest compared to other viral transcriptional enhancers that have been isolated and tested (e.g., G-box 23 and AS-1 24 type elements enhance 35S min activity 8–10 fold using this assay, data not shown). This outcome may reflect limitations of the original search parameters (only repeated elements were tested). However, several of the geminiviral elements identified in this screen have been subsequently found to display clear and unique synergistic effects when combined or multimerized (Cazzonelli, Burke and Velten, manuscript in preparation), supporting their potential to contribute to viral gene regulation during infection.

Since all assays were performed on tobacco plants that had been neither infected with any of the viruses screened, nor transfected with any viral components, it is unlikely that elements strictly dependent upon virally encoded regulatory factors, or factors not native to N. tobacum, would be identified. In addition, the screen was limited to those elements that increase gene expression, and no effort was made to confirm data suggesting that an element might be a 'repressor' (e.g., the 11 elements that show 'enhancement' values less than, or equal to, one third of the 35S min activity, see Additional file 1). Considering these limitations, the finding that 13% of the sequences tested produced measurable up-regulation of transcription supports the original assumption that basic transcription regulatory elements are enriched within repeated sequences from the viral intergenic regions. Despite having tested approximately equal numbers of inverted sequence repeats (IR) and direct sequence repeats (DR), 11 of 14 active elements were members of the DR set, with the remaining 3 positives being palindromic (inverted repeats with no sequence between the repeats). This is somewhat surprising since many of the iterated DNA sequence elements within geminivirus intergenic regions are found as both direct and inverted repeats 22, and as such could have been present in either the DR or IR set of elements. Although the numbers tested are small, and the screen was performed using a single plant species, these results suggest that directly repeated sequences within geminivirus and nanovirus intergenic repeats have a higher probability of positively influencing transcription levels than do the inverted sequence structures. It is possible that this bias may reflect the presence within the intergenic region of DNA elements responsible for viral replication 25, including a conserved inverted repeat structure with a ubiquitous central-loop sequence 26. Seven of the IR elements tested in this study are part of predicted replication hairpin structures (see Additional file 1) and did not, in this test system, result in any measurable enhancement of reporter gene expression.

Manual alignment of all the active DR sequences produced three classes of related elements and several unique individuals (Figure 3). Five of the 14 positive DR elements contain an already identified geminiviral transcription control element, the "conserved late element" or CLE {GTGGTCCC, 2227}. The CLE sequence had been previously shown to affect expression from a minimal 35S promoter, and to be up-regulated by the viral AC2 gene product 27. The two remaining grouped elements include a pair of "CT" rich repeats (DR08 and DR13) and two related, nearly-palindromic direct repeats from beet curly top virus (BCTV, elements DR19 and DR30). Despite the lack of an exact G-box core sequence {ACGT, 28}, the nearly palindromic structure of the DR19 and DR30 elements {aaACTTc} is reminiscent of duplicated G-box type geminiviral elements noted by Arguello-Astorga et al 22 and later proposed as functional components within tomato golden mosaic virus (TGMV) and subterranean clover stunt virus (SCSV) promoters 1120. When scanned against the online PlantCARE promoter element database {2930} no clear consensus emerges regarding similarity of the discovered viral elements with characterized plant cis regulatory elements (the most common hits were against light or stress responsive elements, although that may simply represent the distribution of plant elements contained within the database).

Short of directed mutagenesis of each identified viral element, followed by analysis of resulting 'mutant' virus function within infected plants, it is difficult to directly determine what contribution each of the identified enhancer elements makes to viral gene regulation. Computer analysis of an element's frequency of occurrence in defined DNA sequence databases provides an alternative mechanism for gaining insight into likely biological function for short sequence elements 31. For example, the occurrence frequency of functionally important promoter elements is higher within DNA sequences upstream from gene coding regions, compared to the frequency within non-regulatory sequences 31. Since the element enrichment approach works best when applied to relatively short, core consensus sequences 31, viral element searches were limited to those viral enhancers that showed a clear core consensus (CLE, BCTV DR19/30, CT-rich, Figure 3).

The viral enhancers identified in this work were found to function within un-infected test plants, indicating that the viral elements can make use of intrinsic plant transcription factors (not virally encoded) and may, therefore, be similar or identical to endogenous plant promoter elements. In order to test for enhancement of viral enhancer sequences within higher plant promoters, the PatMatch page of the TAIR web site 32 was used to access sub-datasets of the A. thaliana genomic sequence that are exclusive to annotated coding sequences {CDS} and three upstream sequence lengths {-3000, -1000, -500 bp, measured from each CDS start codon}. Each of the sub-datasets was searched for the viral elements (CLE, BCTV DR19/30, CT-rich) and, as controls, several well defined plant promoter element consensus sequences (the "G-Box" {CACGTG}, a common plant promoter element that is associated with members of the pZIP family of transcription factors 3334, and two less prevalent plant promoter elements, the drought response element ('DRE', RCCGAC 35) and abscisic acid response element (ABRE-like, ACGTGKM) 35).

Performing similar oligonucleotide frequency searches for element enrichment within viral promoters was complicated by the lack of comprehensive annotation of viral sequence entries within the GenBank database. Without clear annotation of intergenic and coding sequences within the viral GenBank entries, it was impossible to directly perform the same sort of 'upstream sequence' (in this case, viral intergenic regions) versus 'coding sequence' frequency comparisons that were possible using the fully annotated Arabidopsis genome sequence and PatMatch. As an alternative, screens were performed to determine frequencies of occurrence for viral enhancers (and control plant elements) within a sequence database consisting of all geminivirus or nanovirus GenBank entries as of May 13, 2004 36, and the results compared with those obtained scanning the same sequences against the Arabidopsis PatMatch datasets. The searched viral sequence database has the potential for bias due to the existence of a numerous entries containing only coding regions or only intergenic sequences, as well as some duplication of sequences in separate entries. Any such bias should, however, similarly affect the baseline frequency values resulting from searches using the 18 matched random oligonucleotides (in parenthesis, Table 1), thus all element enrichments are considered relative to the random oligo values. It was decided to perform the searches using the full geminiviral plus nanoviral database, since limiting the viral entries to only those containing fully annotated, complete viral sequences would have greatly reduced the number of different viruses examined.

The results of the searches are displayed in Table 1. Each frequency value (cHits/Mbp) represents the number of hits per million base pairs, corrected for the database base composition using empirically determined G/C and A/T ratios for each of the databases examined (see Materials and Methods). To facilitate comparison, the resulting cHits/Mbp from the Arabidopsis upstream databases (-3000 to -1001, -1000 to -501, and -500 to -1 bp) were normalized relative to the value obtained for each element's occurrence within the A. thaliana coding sequence database (CDS value set to 1.0). In addition to the predicted frequency values, in each case, the element's observed frequency was also compared to a value generated using the average of 18 random oligomers having the same length and base composition as the element tested (in parenthesis, Table 1). The test sequences for plant ABRE-like and G-box elements showed clear enrichment within the upstream Arabidopsis sequences, especially within the -1 to -500 region (ABRE-like element = 3.0 time the CDS value, vs 1.44 for random sequences and G-box = 4.35 vs 1.47 for random sequences, all as normalized cHits/Mbp). Results for the DRE element were less convincing (2.13 vs 1.46 in the -1 to -500 dataset) and likely reflect lower functional usage of this element within the Arabidopsis genome 35.

As expected, the CLE consensus sequence (GTGGNCCC) was found to be markedly enriched within the viral database, occurring 6 times more frequently than the mean of 18 random 8-mers of identical base composition (CLE = 17.36 normalized cHits/Mbp vs 2.81 from matched random sequences). This frequency is similar to that found (17.11 vs 3.42) using a short sequence of identical base composition and length that matches a highly conserved replication stem-loop sequence (CGCGNCCA), a component that is evolutionarily conserved within the geminivirus population 37. Enhancement of CLE within Arabidopsis promoters is less obvious (CLE = 3.9 in the -1 to -500 database vs 2.79 for random sequences). The observed relatively small CLE enrichment is consistent with reports of a low frequency of occurrence for a CLE-like "TCP domain" binding consensus sequence (Gt/cGGNCCC) within Arabidopsis promoters 38. It is possible that TCP domain-containing transcription factors contribute to the observed CLE enhancer activity since Arabidopsis promoters containing the TCP domain consensus binding element were found to function in transgenic tobacco and to show reduced activity after mutation of the element's core sequence 38.

The test sequences for plant element occurrence within the viral database (ABRE-like = 2.4 vs 1.28 and G-box = 3.81 vs 1.43, DRE = 1.09 vs 0.85) provide further indication of the technique's utility. The G-box viral frequency is consistent with a previous report that a G-box element contributes to transcriptional regulation from the major intergenic region of Tomato Golden Mosaic Virus {TGMV, (20}. The ABRE-like element enrichment in the viral database may indicate that viruses make use of biotic and abiotic stress-induced up-regulation 39 of genes driven by ABRE-containing promoters, a possibility open to additional research.

Of the remaining viral elements tested against the Arabidopsis and viral databases (Table 1), only the DR08/13 TC-rich sequence showed clear enrichment in both plant promoter and viral sequences (Arabidopsis -1 to -500 = 7.75 vs 0.35 and viral = 6.92 vs 0.53). Similar TC-rich regions have been reported within plant promoter regions 4041, but we are unaware of any published report that confirms enhancer activity associated with an isolated TC-rich element, either viral or plant in origin.

The search for repeated DNA sequences was performed by visual inspection of computer-generated dot matrix comparisons (criteria: ≥ 66% identity, 10 base window, GeneWorks v2.5.2, Oxford Molecular Group Inc.). Dot matrices generated using each viral plus strand plotted against itself were used to identify direct repeats while inverted repeats were found by plotting each plus strand against its complement.

The identified repeats were synthesized as DNA cassettes containing the duplicated elements in their original orientation, either directly repeated with spacer sequence ('DR', 41 elements), inversely repeated with spacer sequence ('IR', 45 elements), or palindromic inverted repeats without spacer ('PAL', 20 elements). In order to limit the tested component to only the repeated elements themselves, any sequence occurring between the viral repeats (ranging from 0 to 146 bp, median separation = 9 bp) was replaced with a 10 bp randomized stuffer sequence (GAAGATAATC). The resulting cassettes were inserted immediately upstream from a minimal promoter (-46 to +1 relative to transcription start, 35S min) reporter system derived from the cauliflower mosaic virus (CaMV) 35S promoter fused to an intron-modified firefly luciferase (FiLUC) gene (Figure 2, 45). The resulting test constructs were generated as part of a modified pPZP211 46 binary plant transformation vector (Figure 2) and were introduced into the Agrobacteria tumefaciens strain, EHA105 47 by electroporation 48. The final Agrobacteria strains each contain, in addition to the test plasmids, a second, compatible, binary transformation vector expressing an intron-modified version of the Renilla reniformis luciferase gene (RiLUC) 49 under control of the constitutive Super-promoter 50. The FiLUC and RiLUC enzymes can be independently assayed, making the co-transferred constitutive RiLUC gene a useful marker for gene transfer and for normalization of FiLUC values between individual elements 45.

Agrobacteria harboring the test and normalization binary plasmids were grown at 28°C in LB media containing the appropriate antibiotic selection (25 μg/mL kanamycin sulfate or 100 μg/mL spectinomycin) until an OD₆₀₀ of 0.8 was achieved. The resulting cultures were centrifuged at 3000 rpm for 15 minutes, washed and re-suspended in an equal volume of infiltration media (50 mM MES, 0.5% glucose, 2 mM NaPO₄, 100 μM Acetosyringone) before being mechanically infused (5 ml syringe) into multiple individual tobacco (N. tobacum, cv. SR1) leaves (2–4 leaves per test construct). Assays were performed in groups of 4–8 constructs and the resulting luciferase activities (both FiLUC and RiLUC) determined after 3–4 days using an in vivo floating leaf-disk assay developed for this enhancer screen 45. Test constructs were assayed from 1 to 6 times, with each assay consisting of 2–4 disks (3 mm diameter) per infusion. The disks used in vivo assays were each measured for light production in separate wells of a white-walled 96 well microtiter plate (FLUOstar Optima luminometer^® from BMG Lab Technologies Inc.) and all elements that tested positive in the in vivo assay were subsequently confirmed using the in vitro dual-luciferase^® from Promega Corp (assays performed according to the manufacturers instructions, separate leaf disks from the same leaf infusions were used for the in vivo assays). Each test group included an infusion containing the 35S min construct (lacking any viral test element). In order to compare the various assay systems, all activities were normalized to the activity of the 35S min construct included within each assay set (35S min activity arbitrarily set to 1.0).

Since the element enrichment approach works best when applied to relatively short, core consensus sequences 31, database searches were limited to those viral enhancers that displayed a clear core consensus (CLE, BCTV DR19/30, CT-rich, Figure 3). Results from the viral enhancer searches were compared to values obtained using previously reported plant promoter elements (DRE, ABRE-like, and G-box), and a short DNA sequence that is part of a highly conserved geminiviral replication loop stem sequence (CGCGNCCA) that is identical in base composition and length to the CLE consensus (Table 1). The short sequence elements were each tested for their frequency of occurrence within a set of DNA sequence databases. One database consists of all entries for geminiviruses plus nanoviruses (36, as of May, 2004) and all others are from the A. thaliana genomic sequence at the TAIR, PatMatch web site 32. The geminivirus/nanovirus BLAST searches were set for short exact matches (the statistical significance threshold set to 1000 and word size set at the element's length), returning the number of occurrences of exact matches for the full length element within the database. The TAIR PatMatch searches (default settings: Max hits, 7500; both strands; mismatch = 0; minimum hits/seq = 1; maximum hits/seq = 100) were performed against sub-datasets representing Arabidopsis coding sequences {"GI CDS (- introns, - UTRs)"}, and various lengths of upstream regions {"Locus Upstream Sequences", -1 to -500, -1 to -1000 and -1 to -3000}. Results from the -500 search were subtracted from the -1000 results, to generate hits from -501 to -1000 and -1000 results subtracted from the -3000 data to calculate hits from -1001 to -3000. In order to allow direct comparison between searches in different databases, using sequence elements of differing length and base composition, the number of database hits was corrected for the size of the database (number of hits divided by the database size in mega-base pairs {Mbp}) and base composition (hits/Mbp divided by the predicted number of hits per Mbp using upon the element sequence and base composition of each search database). The dataset base compositions were determined from downloaded sequence files and are: A. thaliana CDS: A/T = 55.8%, G/C = 44.2%; A. thaliana upstream (-1 to -500): A/T = 67.43%, G/C = 32.57%; A. thaliana upstream (-501 to -1000): A/T = 66.24%, G/C = 33.76%; viral: A/T = 56.2%, G/C = 43.8%. The resulting frequency of occurrence is a corrected number of hits per mega-base pairs (cHits/Mbp). For ease of comparison between elements, all of the cHits/Mbp values have been normalized to the corresponding cHits/Mbp number from the A. thaliana CDS database (set arbitrarily to 1.0). Correction of the element's frequency using the calculated random probability of occurrence does not account for the possible impacted by intrinsic base-order bias that may occur within each sequence database, specifically the coding region database. These biases can potentially shift cHits/Mbp numbers markedly from those calculated using simple random base composition frequencies. To help confirm the significance of any observed enhancement in an elements frequency, mean cHits/Mbp values for 18 randomly generated sequences that match each test sequence for base composition and length were determined to provide a baseline value for comparison to that of the test element (shown in parenthesis, Table 1). A total of 18 sequences were used to produce the reported baseline as mean cHits/Mbp values were found to routinely level off at n values of between 8–12 random sequences examined (data not shown).