The size of a polysome is determined by the number of associated ribosomes per mRNA transcript and, for a given mRNA, can be estimated from the position of the peak of gradient fractions containing that mRNA. The polysomes with fewer than 10 ribosomes per mRNA are well separated on 20% to 50% sucrose gradients, and the large polysomal complexes (> 10 ribosomes per mRNA) were assigned to fractions 11 and 12 by a logarithmic extrapolation 11. As expected, we observed that the average ORF length of the mRNA species sedimenting in a given fraction increases from the low molecular weight fractions to the high molecular weight fractions (Figure 3a). The size of polysomes associated with a given mRNA can be affected by changes in the efficiency of ribosomal initiation, elongation and termination. In most cases studied, ribosomal initiation on a transcript is the rate-limiting step, and so the average distance between two adjacent ribosomes on a transcript is mainly determined by ribosomal initiation efficiency. With a same initiation rate, the number of ribosomes per unit length of mRNA, which we call the mRNA ribosomal density, correlates with the length of the ORF of an mRNA 13.

To estimate the average ribosomal density, we used the ratio of the assigned number of associated ribosomes at each fraction to the median ORF length of all the mRNAs sedimenting in that fraction (Figure 3b). It is known that many maternal mRNAs in Drosophila early embryos are sequestered in mRNP complexes and sedimentation of these mRNAs is not simply a function of ORF length. Because mRNA-ribosome associations are typically magnesium-dependent, polysome-associated mRNAs can be released from high molecular weight fractions in the presence of EDTA. Therefore, we excluded the mRNAs that cannot be released from high molecular weight fractions by EDTA. However, we found a significant number of mRNAs with long ORFs sedimenting in the disome (2 ribosomes per mRNA) in fraction 6 and trisome (3 ribosomes per mRNA) in fraction 7 (Figure 3a). We cannot distinguish between ribosome-containing mRNP complexes and translating mRNA-ribosomal complexes with the EDTA treatment assay. Furthermore, it is difficult to accurately estimate the number of ribosomes for the extremely long ORFs sedimenting in fraction 12. Therefore, we have less confidence in the accuracy of the average ribosomal density for the mRNAs sedimenting in fractions 6, 7 and 12 and excluded these three fractions from ribosomal density estimation. The average number of associated ribosomes in fractions 8 to 11 shows a linear relationship with the median ORF length of mRNAs (Figure 3b). Therefore, the average ratio of ORF lengths to the number of their associated ribosomes is the slope of this trend line. We estimate the average spacing of ribosomes on the majority of transcripts in Drosophila embryos to be about 30-32 amino acid codons, giving an average ribosomal density of about 90-100 nucleotides per ribosome. Most mRNAs have a ribosomal density in the range of 25-40 codons per ribosome at all three developmental stages.

Arava et al. 11 reported an unexpected inverse correlation in S. cerevisiae, namely that ribosomal density decreases with increasing ORF length. In a subsequent study, they concluded that the less frequent initiation of translation of mRNAs with longer ORF is responsible for the observed inverse correlation between the ORF length and ribosomal density 14. When we analyzed our data using the methods described in Arava et al. 11, we also found an inverse correlation (Figure 3c). However, the inverse correlation may simply result from the wide range of ORF lengths seen for the transcripts sedimenting in a given fraction, while one single number of associated ribosomes is assigned to all these transcripts. Instead, we found a fairly stable ratio between the median ORF length of transcripts sedimenting within a given fraction and the assigned number of associated ribosomes to that fraction (Figure 2b). Because the exact number of associated ribosomes to individual transcripts is not clearly defined by the polysome gradient analysis used here and by Arava et al., the assertion that ribosomal density decreases with increasing ORF length may not be completely justified.

The consistency of the average ribosomal density during the three developmental time periods we examined indicates that translational regulation of polysome size during early embryogenesis is exerted in a gene-specific manner, rather than at a general level. Based on available ORF lengths in Flybase, about 1.5% of transcripts have a ribosomal density lower than 200 nucleotides per ribosome at each time period and most of them stay in low-density polysomes throughout all three development periods. These mRNAs might have a lower ribosomal initiation rate due to modulation by certain cis-regulatory elements in the mRNA UTRs. However, we could not identify by computational methods any consensus features among those transcripts with a significantly low ribosomal density. Although long UTRs are likely to contain cis-regulatory RNA elements, we did not observe any correlation between the length of the ribosomal density and 5' UTR or 3' UTR length. This may result from both the limitation of computational analysis and the mixture of various cis-RNA regulatory elements within these UTRs. In addition, a prevailing feature of the polysomal profiles of individual transcripts at each of the three time periods examined is that the mRNAs are either sequestered from polysomes or fully loaded with polysomes to an extent correlated with their ORF lengths. This bimodal pattern suggests that most cis-RNA elements in the UTRs are likely to regulate the amount of a transcript associated with polysomes, instead of controlling the ribosome density of a transcript. Furthermore, we did not find significant over-presentation of any particular Gene Ontology (GO) terms among the mRNAs in the lowest 5% density. Our analysis of the 50 mRNAs with highest densities of associated ribosomes revealed that their calculated densities are derived from incorrectly predicted ORF lengths, or a possible cross-hybridization signal from their long alternatively spliced isoforms. These issues make it difficult to pinpoint the highest or lowest ribosomal densities from our analyses. However, these analyses do identify mRNAs whose ORF predictions warrant re-examination.

A key feature of translation status is the proportion of mRNAs associated with polysomes. The relative amounts of free ribosomes (fractions 1-5) to ribosomes engaged in polysomes (fractions 6-12) in the global polysomal profiles are similar at each of the three developmental time periods we examined (Figure 4). Since rRNAs account for more than 95% of total cellular RNAs in most known cell types, and a similar level is seen in cellular RNAs recovered from embryonic lysates, it is likely that rRNA levels or assembled ribosomes remain steady during early embryogenesis 9. Thus, the number of ribosomes incorporated into polysomal complexes is fairly constant during early development. Furthermore, the appearance of significant amounts of free ribosomes at fractions 3, 4 and 5 also indicates that the availability of ribosomes does not limit the global translation efficiency. To rule out the possibility that the polysomes are arrested at the elongation step, we pulse-labeled in vitro translation lysates from 0-2 hour, 4-6 hour and 8-10 hour old embryos with [³⁵S]-methionine/cysteine and did not detect any dramatic difference of [³⁵S] incorporation (data not shown). Therefore, global translation efficiency appears to be fairly constant throughout early embryogenesis.

The percentage of an individual RNA species in the polysomal fractions (fractions 6-12) is defined as the transcript's ribosomal occupancy 11. The ribosomal occupancies of individual transcripts in each time period covered a wide range, from 20% to 100%, with most mRNA species only partially loaded on polysomes (Figure 5). Thus, the availability of ribosomes and transcripts are excluded as rate-limiting factors for global translational control, and amplitude of translation of individual transcripts is modulated by gene-specific translation factors. The distribution of ribosomal occupancies at 0-2 hours appear more dispersed than that at the later stages, which reflects the diverse mechanisms of translational regulation of specific maternal transcripts (Figure 5).

We classified mRNA species into three statistically defined groups reflecting their translational status: a preferentially translated group, a preferentially untranslated group, and the general group of remaining transcripts. Selected groups of mRNA species were defined by using a logit (see Materials and methods and Additional data file 1). Due to the dispersed distribution of ribosomal occupancy at 0-2 hours, more transcripts were found to be either preferentially translated or preferentially untranslated mRNAs than at the later stages. To understand the biological significance of translational control at each developmental stage, selected groups of mRNAs were clustered by the GO terms of biological process, molecular function and cellular component using the Affymetrix NetAffx analyses tool 15. We identified the representative significant GO terms among the selected groups, which are listed in Tables 1 and 2 (0-2 hour old embryos), and Tables 3 and 4 (4-6 hour old embryos).

Some nuclear proteins, such as factors with general RNA polymerase II transcription activity (GO0016251) and transcription regulator activity (GO0030528) are the essential components of early zygotic transcription and embryonic pattern formation. It is perhaps unsurprising that most of their mRNAs are associated with polysomes, preferentially synthesizing their protein products in 0-2 hour old embryos (Table 2). At two hours after fertilization, the process of cellularization starts, forming mononucleate blastoderm cells and zygotic transcription begins. The active translation of proteins involved in RNA processing and metabolism in the 4-6 hour old embryos may facilitate the transition to active zygotic transcription. At this stage, nuclear proteins continue to be preferentially translated (Table 4). In contrast, synthesis of ribosomal proteins (rp) is highly inefficient in spite of the high abundance of their transcripts in 0-2 hour old and 4-6 hour old embryos (Tables 1 and 3). The selective silencing of rp-mRNAs during early embryo development is also observed in Xenopus in that mRNAs encoding ribosomal proteins are initially in mRNP particles and start to become mobilized to polysomes at a later stage 16, reflecting a need for new ribosomes 17. A sufficient number of maternal ribosomes are stored in early Drosophila embryos before zygotic control begins (Figure 4) and the priority of ribosomal protein synthesis is low. Furthermore, many components involved in macromolecular metabolism, including lipid membrane and protein metabolism, are not observed to be actively engaged in polysomes, and we speculate that their products are also abundantly supplied in the maternal cytoplasm and are not a priority for protein synthesis. Another interesting group of preferentially untranslated mRNAs encode products involved in cell cycle progression (for example, cyclin G and cyclin T) that may also be due to the maternal storage of these products to support the rapid early mitotic cycles. At up to 8-10 hours, when organogenesis starts, we observed no significance in the molecular function or biological process among the selected groups of translated transcripts, which may indicate the complexity of differentiation at this development stage. Sixteen mRNAs encoding proteins involved in RNA binding (for example, Staufen and Smaug) are not found associated with translating polysomes. These mRNAs are known to play critical roles during oogenesis and early embryogenesis. In situ expression staining suggests these transcripts are restricted to germ cells. Due to the rapid degradation of these mRNAs after zygotic transcription begins, it is unclear whether these mRNAs are zygotic or maternal transcripts that are protected in the germ cells.

From the GO cluster analyses, we found that mRNAs involved in a common process or cellular components can be specifically co-regulated over the stages of development. Sufficient maternally deposited ribosomes may result in the selective translational repression of rp-mRNAs in early Drosophila embryogenesis, and rp-mRNAs gradually move into polysomes when embryos require more ribosomes at the later developmental stages. In comparison to the general polysomal association profile, both the group of large ribosomal proteins (GO0005842) and the group of small ribosomal proteins (GO0005843) were gradually increased from early development (0-2 hour embryos) and finally reached the average level of translational activity at 8-10 hours. We performed a two-sample Kolmogorov-Smirnov test of the null hypothesis of equal distribution to confirm the significance of the polysomal-association increase of rp-mRNAs as groups (GO0005842 and GO0005843) over the three development periods. In contrast, the group of mRNAs with transcription coactivator activity (GO0016563) was preferentially associated with polysomes at 0-2 hours and their association decreased to the average level at 8-10 hours (Figure 6b). There is no significant difference in the overall polysome association of all the probe sets and the probe set including mRNAs with the top 50% abundance among the three time periods (Figure 6a). In contrast, polysome-association of the highly abundant mRNAs (top 25%) in 0-2 hour old embryos is, on average, less than at the later time (Figure 6a, right-hand panel). This suggests that, in general, those abundant maternal transcripts are present in great excess over their requirements for translation.

The rp-mRNAs of Drosophila and Xenopus embryos always appear either in the mRNPs or fully loaded with ribosomes 1718. In Xenopus, the translation efficiency of mRNAs encoding several protein components of the translational apparatus, including ribosome proteins, is predominantly dependent on the status of cellular growth. This mode of regulation strictly depends on the 5' terminal location of the oligopyrimidine (5' TOP) tract and sequences immediately downstream of the 5' TOP 16. A characteristic oligopyrimidine tract starting at a C residue has been found in transcripts encoding ribosomal proteins; the 5' and 3' UTRs of these genes are significantly smaller than the genome average 19. Ribosomal subunit 40S protein S6 phosphorylation has been implicated in an up-regulation of translation of mRNAs encoding components of the protein synthesis machinery that contain a TOP in their 5' UTR. The concomitant activation of translation of TOP-containing mRNA led to the notion that rpS6 phosphorylation increases the affinity of ribosomes for TOP-containing mRNAs and thus facilitates their initiation 1620. S6 knockout mice show decreased growth and cell size 21 and disruption of the Drosophila gene encoding S6 kinase leads to small body size and growth rate. Because S6 kinase regulates ribosomal protein production in mammals, loss of Drosophila S6 kinase function may have a direct impact on cell growth and proliferation 22. Therefore, selective translational control of TOP-containing mRNAs might be a translational repression mechanism, which has been evolutionarily conserved in early embryogenesis. However, we could not confirm the existence of the TOP sequence motif in Drosophila rp-mRNA 5' UTRs due to lack of the complete 5' UTR sequences of most of these mRNAs.

The selective translational repression of particular mRNA species in response to a reduced cellular need for their protein products might apply to other mRNAs sequestered in 0-2 hour embryos, particularly mRNAs encoding the components of the macromolecule biosynthesis machinery. However, there is no evidence of these mRNAs carrying 5' TOP elements or responding to TOP signaling regulation. Co-regulation of these transcripts implies that translation of these mRNAs is controlled by some shared features, which remain to be defined.

At none of the three developmental stages we examined did we find any notable general correlation between polysome-association and mRNA abundance (data not shown). Thus, translation activity is not simply determined by level of transcript accumulation, but more likely reflects dynamic cellular requirements for particular polypeptides. Our study did reveal trends in the relationship between changes in transcript levels and ribosomal occupancies and shows that these trends can vary over the course of development.

A comparison of 4-6 hour and 0-2 hour embryo samples shows that where mRNA levels decrease during very early embryogenesis, there is a tendency for ribosomal occupancies to increase (Figure 7b, left). This change may simply reflect the rapid degradation of excess maternal transcripts after zygotic control begins, and not necessarily any upturn in the level of protein synthesis. For example, a gene cluster including 169 maternal mRNAs whose steady state level quickly drops two hours after egg deposition was identified by Tomancak et al. 23. The average ribosomal occupancy of these 169 mRNAs is much lower than that of the majority of other transcripts at 0-2 hours, whose levels change less between the 0-2 hour stage and the 4-6 hour stage (Figure 7a). It is likely that the general inverse correlation of changes in mRNA abundance and ribosomal loading is simply a consequence of clearing maternal transcripts at the zygotic transition.

On the other hand, comparing ribosomal occupancies in 8-10 hour and 4-6 hour embryos reveals a positive correlation with changes in transcript abundance (Figure 7b, right). This correspondence of elevation of mRNA accumulation with increased ribosomal occupancies indicates some coordination of steady state RNA levels and translation activity; the levels of gene regulation underlying this phenomenon and how directly and generally they are integrated remain questions for future investigation.

Using a multivariate empirical Bayes (MB) statistic 24, we ranked all the transcripts according to the difference between ribosomal occupancy between 0-2 hours and 4-6 hours, and between 4-6 hours and 8-10 hours (Additional data file 2). The higher a gene's rank, the more dramatic is the observed change in its polysomal profile. The percentile polysomal profiles of representative mRNAs with high rankings are shown in Figure 8. Expression of these genes is likely to be translationally regulated. We were not able to identify any significantly shared features of the primary sequences among mRNAs with high MB rankings, although they are expected to be modulated in response to certain common development signals.

For the spatially localized mRNAs that have been studied, such as nanos mRNA, their spatial localization and translation control are often closely linked, with translation being repressed during mRNA translocation and activated on reaching its destination 32526. Until this study, biochemical analysis of ribosomal association to estimate mRNA translation status has been completed for only few Drosophila mRNAs. We evaluated the polysome association profiles of several known localized mRNAs during embryogenesis.

Our observation that only a small proportion of nanos (5% in fraction 12) is associated with polysomes is consistent with the published estimate; it has been reported that about 4% of nanos mRNA is localized to the posterior and actively translated 27 (Figure 9a). However, Clark et al. 27 found a comigration of 53% of nanos mRNA associated with polysomes and suggested, based on their study, that nanos mRNA translation is regulated in a novel manner. We do not understand the basis of this difference between our observations and those of Clark et al.

Consistent with immunocytological observations that translation of hunchback transcripts and caudal transcripts is regionally repressed in the early embryo 326, the experiments presented here indicate that only a small proportion of these transcripts are associated with polysomes (Figure 9b). The unique polysomal profiles of maternal transcripts, for example, the transcripts of nanos, oskar and bicoid, suggest the existence of multiple mechanisms controlling gene-specific translation of mRNAs during early pattern formation.

Canton S embryos were collected in 2-hour intervals and aged to generate animals 0-2, 4-6 and 8-10 hours old. To confirm that the embryos were collected at the desired developmental stages, we examined the morphology of a small aliquot of the synchronized embryos as described previously 23. In addition, we also validated the synchronization by comparing the variation of RNA abundance of representative mRNAs over the three time periods with the previous microarray measurements performed by Tomancak et al. 23 (data not shown). The embryos were then dechorionated and transferred to Eppendorf tubes.

Unfractionated RNA was prepared by homogenization of dechorionated embryos with a motorized plastic pestle in RNAwiz solution (Ambion, Austin, TX 78744-1832), followed by chloroform extraction and ethanol precipitation. To prepare the polysome-associated RNAs, the dechorionated embryos were first incubated with 0.1 mg/ml cycloheximide in PBS for 10 minutes on ice, then homogenized with a motorized plastic pellet pestle in a lysis buffer (20 mM Tris-HCl, pH 7.4, 140 mM KCl, 5 mM MgCl₂, 0.5 mM DTT, 1% Triton X-100, 0.1 mg/ml cycloheximide, 1 mg/ml heparin, 50 unit/ml RNasin) and incubated for 10 minutes on ice. The debris were removed by centrifugation at 12,000 × g for 10 minutes at 4°C, and supernatants were loaded onto 20% to 50% sucrose gradients with the same extraction buffer without Triton X-100. The extracts were sedimented at 35 k rpm for 160 minutes in a SW41 rotor at 4°C. Twelve fractions were collected from the tops of the gradients using an ISCO fraction collection system. RNAs were precipitated from each fraction with guanidine hydrochloride and ethanol followed by a second precipitation in 1.5 M LiCl at -20°C overnight. The RNA precipitate was washed with 70% ethanol and resuspended in an equal volume of Tris-HCl buffer (1 mM Tris-HCl, pH 8.0). Purified RNAs from individual fractions were quantified with a spectrophotometer and visualized on formaldehyde agarose gels 2729.

EDTA-treated embryos were lysed in an EDTA extraction buffer (20 mM Tris, pH 7.4, 140 mM KCl, 15 mM EDTA, 0.5 mM DTT, 1% Triton X-100, 0.1 mg/ml cycloheximide, 1 mg/ml heparin, 50 unit/ml RNasin) and sedimented through 20% to 50% gradients prepared with the same EDTA lysis buffer, but without Triton X-100.

Although most cytoskeleton-associated and endoplasmic reticulum-associated RNAs as well as ribosomes are released into the soluble extract under this buffer condition (X Qin, unpublished data), it is possible that some mRNAs are sequestered in insoluble complexes and excluded from polysomal gradient analysis. RNAs potentially in the insoluble debris were not characterized in this study.

Either unfractionated total RNA or equal proportions of RNA from each polysomal fraction were reverse transcribed into cDNAs with a High Capacity cDNA Archive kit (Applied Biosystems, Inc. Foster City, CA 94404). Gene-specific TaqMan^® probes were designed and manufactured through Assay-by-design (Applied Biosystems). Equal proportions of cDNA samples mixed with TaqMan^® Universal PCR Master Mix and gene-specific TaqMan^® probes were quantified in a 96-well plate on ABI PRISM^® 7000 Sequence Detection Systems as described by Applied Biosystems.

RNAs from the first five gradient fractions were pooled. An equal volume of the pooled RNAs and RNAs from the remaining seven fractions was used for cRNA labeling. Thus, the pooled RNA sample used for array labeling was the average amount of RNA from the first five fractions and each RNA sample contained at least 10 μg of RNAs. cRNA was hybridized to a GeneChip Drosophila Genome 2.0 Array using standard protocols. Thus, we collected eight GeneChip array scans of each polysomal gradient and the success of the experiments was determined by the reproducibility of the two independent replicates. Similarly, we prepared four pools of RNA from the first five fractions, from fractions 6 and 7, from fractions 8 and 9 and from fractions 10, 11 and 12. These four RNA samples were used for microarray analysis of the EDTA treated samples. Total RNA (20 μg) from unfractionated cell lysates at each time point was used for one-step labeling and GeneChip hybridization. Gene expression measures were normalized and computed using the robust multichip average (RMA) method described in 30 and implemented in the Bioconductor R package. Statistical analyses were all performed with the open-source software R, version 2.2.0 and Bioconductor 1.7 packages 31. The following R packages were used mainly; Affy (version 1.8.1), limma (version 2.2.0) and Drosophila 2 (version 1.10.0) 32.

Moderated t-statistics were used to determine whether a transcript is released from polysomes by EDTA filtering 33. If the amount of a particular gene's mRNA in the first fraction of EDTA-treated profiles is higher than that in the first fraction of non-EDTA-treated profiles, this transcript is releasable by EDTA since the materials dissociated by EDTA are expected to sediment at the pooled non-polysomal fraction. The false discovery rate (FDR) was controlled at p = 0.05. All data for such genes were removed for further analyses. Among all the 18,952 probes on the Drosophila Genome 2.0 GeneChips, 16,513 genes at 0-2 hours, 16,519 genes at 4-6 hours and 14,593 genes at 8-10 hours were left for further data analysis. In addition, we excluded mRNAs with low signal intensity to exclude the background noise as well as possible signal saturation of those mRNAs with extremely high intensity as described in individual analyses.

To determine the peak fraction of each mRNA for ribosomal density estimation, we used m as the measure for selecting genes with a sharp peak in their polysomal profiles. We first averaged the two normalized replicates of polysomal gradients on the logarithmic scale. Next, we removed the genes whose transcripts were not releasable by EDTA and whose intensities were below the median of all the probe sets. Then, we calculated m as the following:

m = % [peak fraction] - average% [adjacent two fractions], if peak fraction is not 6 or 12

m = % [peak fraction] - % [fraction 7] if peak = 6

m = % [peak fraction] - % [fraction 11] if peak = 12

The 3,000 genes with highest m values were expected to have a distinct peak and were selected to estimate the ribosomal density at each time interval.

The number of ribosomes per transcript in fractions 6-10 was obtained directly from the peaks in the average of multiple OD254 profiles (Figure 3a). For fractions 11 and 12, which lacked single ribosome resolution, the number of ribosomes per transcript was estimated by a logarithmic extrapolation from the clearly defined peaks as described by Arava et al. 11. The assigned number of associated ribosomes for fractions 6-12 is in the order of 2, 3, 5, 7.5, 11, 17 and 26, respectively. The R² value of the logarithmic curve over the defined region is 0.9983 (data not shown).

We used two parameters to describe the polysomal profiles: the logarithmic ratio of the polysomal fractions to non-polysomal fractions (Logit):

Logit = log [percent(fraction 6-12)/percent(fraction 1-5)]

and the standard deviation (SD) of the expression levels among all the gradient fractions.

The preferentially translated mRNAs are expected to have a high Logit and a high SD, while unpreferentially translated mRNAs have low Logit and low SD. We examined the distribution of Logit and SD of each time point (data not shown) and decided to use a cutoff of Logit > 1.5 and SD > 0.5 to define actively translated genes, while those with log ratio < -0.15 and SD < 1 are defined to be translationally inactive genes. Any other genes not included in these two clusters were defined to be in the general translation group.

Clustered genes were analyzed using the NetAffx Gene Ontology Mining Tool provided by Affymetrix 34. The goal of GO analysis was to find statistically overrepresented GO terms within a group of genes 35. Also of interest is comparison of the distribution of a statistic such as the Logit among genes associated with a certain GO term. The GOstats package of R was used to get GO-filtered data. The distributions of Logit for a set of genes such as those associated with a given GO term can be compared across time points. For this purpose, a two-sample Kolmogorov-Smirnov test of the null hypothesis of equal distribution was performed.