We monitored mRNA decay in exponentially growing cell cultures by measuring transcript abundance at 3, 6, 9, 12 and 15 minutes after transcriptional arrest by actinomycin D, previously demonstrated to block transcription in human cells 1617, the archaea Haloferax mediterranei 15 and S. solfataricus 12. Accurate half-life determinations require rapid and quantitative inhibition, and the analysis is based on the demonstration by Bini et al. 12 that these effects are obtained by actinomycin D treatment of S. solfataricus. The analysis was focused on the S. solfataricus transcriptome, but S. acidocaldarius was also surveyed for general trends, providing an independent biological replicate. As we had little a priori knowledge on RNA decay rates in these species, we applied a normalization procedure based on the assumption that a given proportion of the transcripts were stable (see Materials and methods). After filtering out genes with missing data points and/or noisy decay profiles, we obtained half-life data for 2,064 and 1,582 genes in S. solfataricus and S. acidocaldarius, respectively (Additional data file 1). When the proportion of stable transcripts was set to 10%, the median half-life was estimated to 5.3 and 5.1 minutes for S. solfataricus and S. acidocaldarius, respectively (changing the proportion to 5% did not alter the results significantly; median half-lives 4.4 and 4.5 minutes). Similar half-life distributions were obtained in the two species (Figure 1a, b), with approximately 50% of the transcripts within the 4 to 8 minutes range, and with approximately 8% displaying half-lives longer than 20 minutes. No transcripts with a half-life shorter than 2 minutes were observed. Transcripts with lower stability could potentially be hidden among those with missing data, but there was no bias in missing data points toward the end of the time series that would indicate such a relationship.

The technical reproducibility of the assay was monitored by comparing the half-lives obtained in independent labeling/hybridization series (Figure 2a). The average variation was 10%, as calculated by dividing (absolute) half-life difference with half-life average, and 87% of the transcripts displayed less than 20% variation. By comparing the results obtained in independent cultures, the biological variation could be measured. This was only slightly higher, with 12% average variation and 84% of the transcripts displaying less than 20% variation (Figure 2b). Genes expressed in the same polycistronic messages are expected to display similar half-lives. Few operons have been experimentally verified in Sulfolobus, but occurrences of putative transcription initiation sites together with intergenic distance distributions 1819 indicate that genes on the same strand separated by less than 10 nucleotides are likely to belong to the same operon. We observed a significant correlation (Pearson r = 0.72, P < 10^-15) between the half-life of the upstream and downstream gene for such gene pairs in S. solfataricus (Additional data file 2).

To confirm the robustness of the data, we performed quantitative real-time PCR (qPCR) on cDNA from all time points for nine transcripts that represented a wide range of half-lives. The half-lives calculated from the qPCR results correlated well (Pearson r = 0.94, average half-life variation 17%) with the microarray-derived data (Additional data file 3).

Earlier studies have revealed that transcripts encoding proteins with related functions tend to decay at similar rates 810. To address this issue in Sulfolobus, we compared the stability of S. solfataricus genes belonging to different functional categories in the COG (clusters of orthologous groups of proteins)database (Figure 3). All genes with measured half-lives were split into four equally sized groups according to half-life, and each group was tested for over-representation of any functional category. Although these categories are rather broad, and include many potential subgroups (for example, anabolic and catabolic genes in the same group), the distributions of half-lives for certain categories deviated significantly from the average. Over-represented categories in the group of shortest half-lives (0 <t_1/2 ≤ 4) were 'translation', 'amino acid transport and metabolism' and 'energy production and conversion'. No functional category was significantly over-represented in the other three groups, but genes not present in the COG database (and thus uncommon in other species) were over-represented within the group with longest half-lives (t_1/2 > 9). The most enriched category in the group of longest half-lives was 'inorganic ion transport and metabolism', although not significantly (P = 0.14). Although the 'transcription' category was not over-represented in any half-life group, the components of the basal transcription machinery displayed significantly shorter half-lives than average (see below). On the whole, transcripts encoding proteins involved in growth-related processes, such as transcription, tRNA synthesis, translation and central metabolic pathways, generally decayed rapidly, whereas those encoding products necessary for maintaining cellular homeostasis, for example, ion transporters, were more stable. Similar results were obtained in S. acidocaldarius (Additional data file 4).

Many growth-related genes are highly expressed. To investigate the relationship between expression level and half-life further, we measured relative transcript abundance in a sample harvested immediately before actinomycin D was added to the culture. This was done by hybridizing cDNA derived from this culture together with genomic DNA from a stationary phase culture. As all genes are present in equal copy-number in stationary phase cells 20, the cDNA/DNA ratios serve as relative estimates of transcript abundance. Similar to what was demonstrated using the same approach in Escherichia coli 8, a negative correlation between transcript abundance and stability was observed in Sulfolobus (Spearman rank order ρ < -0.4, P < 10^-15 in both species; Figure 4a and Additional data file 5).

Among the 163 mRNAs in S. solfataricus with t_1/2 > 20 minutes, 63% were either not present in the COG database (and lacked functional annotation) or belonged to one of the categories 'function unknown' or 'general function prediction only'. This is significantly more than the 33% in the entire mRNA decay data set, and uncharacterized genes are thus over-represented among those with stable transcripts. Among those with computationally predicted general functions, we found a few groups enriched in stable genes. These included ATPases of the AAA+ superfamily, for which 12 out of 14 transcripts had half-lives above the median, membrane proteins (10 out of 12 above median) and nucleotidyltransferases (7 out of 7). As a final example, within the group of predicted nucleic acid-binding proteins containing a PIN domain, 11 out of 12 transcripts displayed half-lives above median. In S. solfataricus, these genes constitute the 'toxin' in the 'toxin-antitoxin' loci of vapBC-type that recently were found to be abundant in archaeal species 21. The cellular target(s) of the VapC toxins is not yet clear, but in eukaryotes PIN domain proteins are ribonucleases involved in nonsense-mediated decay and RNA interference pathways 22, and may thus be involved in related processes in archaeal cells 2324.

To investigate the relationship between transcript half-life and protein function in more detail, we focused on the transcriptional and translational apparatuses of S. solfataricus. The archaeal transcription machinery is a simplified version of the eukaryal system, in which RNA polymerase, TATA-box-binding protein (Tbp) and transcription factor B (Tfb) are sufficient for initiation from most promoters in vitro 25. Thirteen out of the fourteen subunits of the S. solfataricus RNA polymerase 26, as well as Tbp, displayed transcript half-lives below median. Interestingly, the one polymerase subunit that was markedly more stable than the others, RpoG (SSO0277), has only been identified in Sulfolobus, indicating a unique function for this protein in an otherwise highly conserved transcriptional apparatus.

Similar to several other archaea, Sulfolobus encodes multiple Tfb homologues, and it has been proposed that differential use of these may be a mechanism for gene regulation 27. Consistent with this hypothesis, one of the Haloferax volcanii Tfbs was found to increase in abundance relative to its paralogs in response to heat shock 28. Interestingly, and in agreement with an earlier study 12, the three transcripts predicted to encode Tfbs in S. solfataricus displayed differential half-lives (t_1/2 = 3.3, 7.8 and 9.1 minutes for SSO0946, SSO0280 and SSO0446, respectively).

In contrast to most transcripts encoding the basal transcription machinery, we found mRNAs encoding transcriptional regulators generally to be long-lived in S. solfataricus (Figure 3). Of the 48 genes known or predicted to encode transcriptional regulators, 37 displayed half-lives above median. Some groups, for instance regulators belonging to the Ars family, all showed half-lives above median.

In a growing S. solfataricus cell, ribosomal protein mRNAs are among the most abundant (Figure 4a). The regulation of ribosomal protein synthesis, as well as ribosome biogenesis, differs significantly between E. coli and budding yeast. Whereas yeast regulation primarily is conducted at the transcriptional level, it mainly occurs at the translational level in E. coli 29. How regulation is conducted in archaea is not known, but since the half-lives of the ribosomal protein transcripts were among the shortest in both transcriptomes, the regulation appears to be conducted at the transcriptional level in Sulfolobus.

Two RNAs deviated from the uniform half-life distribution of ribosomal protein transcripts in S. solfataricus: rpl13E (SSO2442) and rpl34E (SSO6374) (t_1/2 = 12.1 and 8.7 minutes, respectively). The Rpl13E protein belongs to COG4352, which is restricted to eukaryotes and crenarchaea. Interestingly, the gene has been reported to be induced upon cold stress in the plant Brassica napus 30, and E. coli cells expressing a homologous protein derived from the green algae Chlamydomonas were reported to display enhanced tolerance against salt and cold stress 31. The long transcript half-life in S. solfataricus indicates a specialized function for Rpl13E in crenarchaea as well. There are, to our knowledge, no published data indicating a specific role for Rpl34E but COG2174, representing this protein, is absent in several archaeal lineages.

Although numerous cis-elements and transacting factors affecting half-lives of individual transcripts have been identified in a variety of organisms 3233, few general determinants of RNA stability are known. Previous genome-wide surveys did not reveal any strong correlations between half-life and GC content, open reading frame length or codon usage 810. Since some rapidly decaying transcripts, encoding, for instance, ribosomal proteins, are expressed as polycistronic messages in Sulfolobus, we wished to investigate whether operon length was related to half-life. To find putative operons of defined sizes, we identified contiguous stretches, 'runs', of one, two or three genes encoded on the same strand, separated by less than ten nucleotides, and bounded by genes encoded on the other strand. Half-life was found to decrease with the number of genes in a run (Mann-Whitney test P < 0.0001 for one- versus two-gene runs, and P < 0.02 for two- versus three-gene runs; Figure 4b). Since this could be a consequence of a correlation between half-life and transcript length (counting in nucleotides) we controlled for this by splitting, based on transcript lengths (counting from first start to last stop codon), all one- and two-gene runs into two sets of equal number of runs. No significant half-life differences were obtained between one- and two-gene runs in either of the two sets (P > 0.8 in both). Moreover, half-life was correlated with transcript length within each group of one-, two- or three-gene runs (P < 0.02 in all groups; Figure 4c). Hence, transcript length per se seemed to be the important factor. The correlation was independent of the correlation between expression level and half-life: transcript abundance and length is not correlated in one-gene runs (ρ = 0.01, P > 0.6), while in this group half-life is negatively correlated with transcript length (ρ = -0.415, P < 10^-15). Similar results were obtained in S. acidocaldarius (Additional data file 5).

Our experimental approach allowed us to compare the mRNA half-lives of orthologous genes in the two Sulfolobus species. We observed a highly significant correlation between the log transformed mRNA half-lives of the two organisms (Pearson r = 0.51, P < 10^-15; Figure 2c), although many genes displayed differential half-lives. This can be compared with the correlation obtained from different cultures of the same species (Figure 2b; Pearson r = 0.93). Thus, although the mRNA half-lives of the two species were correlated, there was substantial deviation that could not be explained by experimental noise.

S. solfataricus DSM1617 and S. acidocaldarius DSM639 (Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH, Braunschweig, Germany) cultures were grown at 79°C in modified Allen mineral base medium containing 0.2% w/v tryptone 40. Growth was monitored by optical density (OD) measurements at 600 nm, and doubling times of 3.5 and 6 h were obtained for S. acidocaldarius and S. solfataricus, respectively (not shown). Flow cytometry analysis was performed as described previously 20, and cell size and DNA content distributions typical for exponentially growing Sulfolobus populations were obtained (Additional data file 6). For extraction of RNA, 200 ml cultures were grown in 1 l Erlenmeyer flasks to an OD of 0.1. For extraction of stationary phase genomic DNA, 200 ml cultures were grown in 1 l Erlenmeyer flasks to an OD of 0.7.

Actinomycin D, dissolved in DMSO, was added to a final concentration of 15 μg/ml and samples (25 ml) were collected for half-life analysis at 3, 6, 9, 12 and 15 minutes after addition, and rapidly chilled on ice by addition of an equal volume of ice-cold medium. Samples were also extracted from untreated cultures to generate reference RNA. At each time point, samples for flow cytometry analysis were also collected and analyzed. No effects were seen on cell and DNA integrity during this time interval (Additional data file 6).

RNA from each time point (5 μg) was reverse transcribed into Cy3-labeled cDNA using aminoallyl-modified nucleotides, as described 41. Cy5-labeled cDNA was prepared from the reference culture using the same protocol. To minimize the variation in the measurements, the reference cDNA was pooled and aliquoted. Genomic DNA from the stationary phase culture was purified and labeled with Cy3-dUTPs as described 42. The microarray data have been submitted to ArrayExpress (E-MEXP-894).

Microarrays with gene-specific tags (GSTs) for the two Sulfolobus species were produced as previously described 43, and augmented with GSTs for additional protein-encoding genes. Probes for 2,484 and 1,946 S. solfataricus and S. acidocaldarius genes, respectively, were printed in duplicate on Ultra GAPS glass slides (Corning Life Sciences, New York, USA) with a QArray2 microarray printer (Genetix, New Milton, Hampshire, UK).

For mRNA half-life analysis, Cy5-labeled cDNA from each time point was co-hybridized for 40 h with an aliquot of Cy3-labeled reference cDNA, as described 41. For transcript abundance analysis, 5 μg of Cy5-labeled cDNA from an untreated culture was co-hybridized with 2 μg of Cy3-labeled genomic DNA. The slides were scanned with an Agilent G2565BA microarray scanner (Agilent Technologies, Palo Alto, CA, USA), and data were collected with GenePix 5.0 software (Axon Instruments, Foster City, CA, USA). Low-quality spots were excluded as described 42. Cy5/Cy3 log₂ ratios of background-subtracted foreground intensities were extracted and, for each gene, averaged over the spot replicates. Only genes with at least one measurement for each time point were included. Two labeling/hybridization series were analyzed for S. solfataricus, and two independent cultures were analyzed for S. acidocaldarius.

The half-life calculations were based on two assumptions: first, that RNA decay proceeded at a constant rate over time; and second, that a given proportion (for example, 10%) of the transcripts were stable. According to the first assumption, log ratios should decrease linearly over time. Thus, we sought to normalize the log ratios for each time point (microarray) such that the decay profiles (log₂ ratios over time), overall, were as linear as possible. This was achieved by iterating the process of: fitting a line to each decay profile by the least-squares method; and, at each time point, calculating the average (over all genes) deviation from the lines, and adjusting the log ratios at this time point with the average deviation. The process converged within five iterations. Subsequently, new lines were fitted to the adjusted decay profiles, and for each line the slope k, as well as the sum of the squared residuals, was calculated. To exclude noisy data, the 20% measurements that displayed the highest sum of squared residuals were excluded. To fulfil the second assumption, each k was normalized by subtracting the average k for the 10% of genes with the largest k values. After normalization, more than 95% of the remaining genes displayed decay profiles with a Pearson correlation coefficient of ≤ 0.95. For genes with multiple measurements, the k was averaged over the replicate time series. The half-life (t_1/2) of a transcript was finally calculated as t_1/2 = -1/k. As the precision in the estimates was limited for extremely long-lived transcripts, all transcripts that displayed half-lives of > 20 minutes were assigned a half-life of 20 minutes. Software for normalization is available from the authors on request.

Primer pairs for independent confirmation of microarray data were designed against 10 S. solfataricus genes (SSO0071, SSO0946, SSO0708, SSO6374, SSO2652, SSO1300, SSO0446, SSO2442, SSO0277, SSO2688) selected to cover the entire range of decay rates. Primers were designed using Primer Express 2.0 (Applied Biosystems, Foster City, CA, USA) with default settings, and purchased from Operon Technologies (Cologne, Germany). Half-volume SYBR Green PCR Core Reagents (Applied Biosystems) reactions were performed using an ABI Prism 7000 Sequence Detection System (Applied Biosystems) with the following settings: 2 minutes at 50°C and 10 minutes at 95°C followed by 40 cycles of 15 s at 95°C and 1 minute at 60°C. The cDNA template was synthesized as for microarray experiments with the exception that only unmodified dNTPs were used. Three replicate reactions were performed per time point and primer pair. Microarray data-comparable log₂expression values (M-values) for a given gene and timepoint t were calculated as:

M(t min) = CT(3 min) - CT(t min)

qPCR data were normalized against the transcript with the longest half-life, as for the microarray data.

We applied the Inparanoid software 44 for identifying S. solfataricus orthologs in S. acidocaldarius. Only orthologs with a one-to-one relation were considered, to avoid genes that had undergone duplication after the species diverged. The S. acidocaldarius genes were assigned the same COG (clusters of orthologos groups of proteins) functional category as their S. solfataricus orthologs.

The genes were separated into four equally sized groups according to half-life: 0 <t_1/2 ≤ 4, 4 <t_1/2 ≤ 5.5, 5.5 <t_1/2 ≤ 9 and t_1/2 > 9. Each group was tested for over-representation of genes belonging to different functional categories in the COG database 45, relative to all genes with measured half-lives. P values were calculated using Fisher's exact test. To account for multiple testing, only P values below 0.005 were considered significant.