The breakdown of unique spore transcripts by functional categories closely matches that of the E. cuniculi genome as a whole, suggesting that there is not a strong bias in the kinds of mRNAs present in spores (data not shown). Unexpectedly, however, a large proportion of transcripts with similarity to known sequences are non-protein-coding, which fall into several classes (Fig. 1A). Two classes were unremarkable: 7 transcripts encoded rRNA fragments, and another 24 were highly similar (but not identical) to intergenic regions of the A. locustae catalase. This region is repeated in the genome (13), and these mRNA are inferred to be from expressed repetitive elements. More interesting is that 144 transcripts (17% of total matches) encoded the antisense of identifiable protein-coding genes. This large proportion of antisense transcripts may represent a high level of posttranscriptional control as found in other eukaryotes (14, 15). However, among the coding and antisense transcripts, several mRNA encode fragments of more than one gene (see below), suggesting another explanation.

Of the 871 clones found to encode recognizable genes, 97 transcripts (11%) from 70 distinct loci encoded sequence from more than one gene (Fig. 1 A; see also Table 1, which is published as supporting information on the PNAS web site). The polyA sites of these clones do not correspond to polyA tracts in the genome, so they are unlikely to derive from DNA contamination (see also below), but instead come from polyA RNA. In prokaryotes, polycistronic mRNAs commonly code for multiple proteins (11), but with few exceptions (12) eukaryotic mRNAs encode a single gene. A. locustae multigene transcripts encode two or three genes or gene fragments in various orientations (Fig. 1 B–I), but they cannot all be polycistronic messages because there is no bias for genes being on the same strand. Indeed, in the majority of multigene transcripts coding regions appear on both strands (and rarely also include tRNAs). Antisense sequences cannot be translated, so, although we cannot rule out polycistronic mRNAs for certain individual cases, multigene transcripts probably direct the translation of a single protein in general and reflect the movement of transcription control elements beyond their canonical intergenic location.

Examining the structure of various multigene transcripts shows that transcription termination and probably also transcription initiation are affected. The first coding region is most often in the sense strand (Fig. 1 B, C, and F). These mRNAs can be interpreted as run-on transcripts where termination signals are absent from the downstream intergenic regions and are instead found within or between downstream genes. Transcriptional termination and promotion within adjacent genes have been observed in other eukaryotes, including fungi, but these situations appear to be relatively rare and in some cases are known to have a deleterious effect on the transcription of adjacent genes (16–19). In A. locustae they are abundant. In other A. locustae cDNAs, the first coding region is in the antisense strand (Fig. 1 D, E, and G–I), presenting two possible explanations. On one hand, these cDNAs may represent 3′ run-on transcripts that are substantially 5′ truncated (many cDNAs in this library are 5′ truncated, which is common in cDNA library construction) but in some cases would suggest that transcripts are very large (in Fig. 1 G–I, for example, the transcripts would contain at least four genes). On the other hand, transcription initiation may also have been moved from the upstream intergenic region into the adjacent gene. The latter possibility has been shown to be the case for the A. locustae photolyase gene, where transcription initiates 430 bp into the upstream gene (20). The EST data suggest that such initiation within upstream genes is not restricted to one or two genes in A. locustae but could be found throughout the genome.

An interesting implication of both of these possibilities is a significant uncertainty as to which gene is actually being expressed in microsporidian transcripts. One could assume that the first sense-strand coding region is the target of expression, but, in many cases where transcripts include two sense genes, transcripts terminate downstream of the stop codon of the following gene (15 of 70 cases). Moreover, when single gene transcripts are also considered, many genes are represented in more than one transcript type (Fig. 2). It is sometimes possible to put forward a credible hypothesis for which gene is actually expressed. In many other cases, however, apparently different transcripts overlap substantially. In Fig. 2E, for example, T-complex protein transcripts might read into thioredoxin peroxidase, or thioredoxin peroxidase transcription might initiate within the T-complex protein. In other cases, transcripts from two adjacent genes appear to terminate within or beyond one another (Fig. 2B).

To provide additional evidence that multigene transcripts are derived from mRNA and to examine the nature of the transcripts, we used cap-dependent 5′ RACE and polyA dependent 3′ RACE. The most abundant multigene transcript encodes DNA polymerase α and ribosomal protein L24 (rpl24). These genes are separated by an intergenic region of only 23 bp, and in the 10 ESTs representing this fragment both genes are at least partially represented (Fig. 3A). Using primers within rpl24, cap-dependent 5′ RACE revealed two distinct transcripts: one starting one base upstream of the ATG of rpl24 and a second starting at the ATG of the DNA polymerase α gene. Overall, it appears that both genes have unique promoters, but DNA polymerase transcripts terminate downstream of rpl24, possibly using the same transcription termination signals. 3′ RACE similarly suggests overlap due to transcription termination points within genes. Using primers specific to the frataxin gene, we amplified transcripts terminating before the end of frataxin and within the downstream gene coding for a coatomer complex protein (Fig. 3B). These transcripts suggest that the frataxin mRNA terminates within the coatomer complex protein gene, but also that frataxin sequence forms part of the 3′ UTR of the upstream gene transcript. Results from neither method are consistent with contaminating DNA. Although these examples illustrate that A. locustae transcription patterns are unusual, transcripts of the most common single gene ESTs have a strong tendency to end at a single particular polyadenylation site (Table 2, which is published as supporting information on the PNAS web site), indicating that transcription termination/polyadenylation can be consistent.

The diversity and complexity in structure of A. locustae multigene transcripts raise questions about the large proportion of cDNAs that encode only the antisense of identifiable genes (Fig. 1 A). Given the structure of many multigene transcripts and the fact that some cDNAs encode fragments of more than one antisense gene, it is possible that antisense transcripts may represent truncated cDNAs of multigene transcripts. These data suggest that the overall proportion of multigene transcripts may be higher than currently recognized: multigene plus antisense transcripts together represent approximately a quarter of A. locustae cDNAs with homologues in National Center for Biotechnology Information databases.

Microsporidian genomes are atypical, conspicuously in their highly reduced and compacted nature. The content and characteristics of these genomes have been studied in some detail (6, 8, 9), but whether this has any effects on genome function is not clear. The unusual nature of transcription in A. locustae indicates that compaction may indeed have an impact on expression. One of the obvious characteristics of compacted microsporidian genomes is the short intergenic regions: in A. locustae intergenic spaces average only 211 bp (8). If the pressure to shrink intergenic regions is strong enough, they could be reduced beyond the minimal size needed to encode the essential control regions for expression. The severe reduction of a 3′ intergenic space could, for example, eliminate existing transcription termination signals and force transcription termination fortuitously within the next gene or, if it is in the same strand, using the downstream gene's existing termination signals. Consistent with this idea, the mean intergenic space between genes encoded on multigene transcript mRNAs is only 119 bp, just over half the average for this genome.

Transcript Structure in Nucleomorphs. If multigene transcripts in A. locustae are the result of its compacted genome, then we might expect transcription in other gene-dense nuclear genomes to share similar characteristics; therefore, we carried out EST surveys of G. theta and B. natans to examine transcription of nucleomorph genes. Not only are these nucleomorph genomes the smallest and most compact of any nuclear genomes, but the G. theta and B. natans nucleomorphs have evolved in parallel from a red and green alga, respectively (4), so they give us two independent points of comparison. Two multigene transcripts from the B. natans nucleomorph have been characterized previously (7); however, with only two examples it is not clear whether these messages are representative or not. Because in one case the coding regions characterized were in the sense strand, they could be interpreted as polycistronic or processed messages (7). A larger sample from both nucleomorphs would discern these possibilities from multigene transcripts as observed in microsporidia.

A total of 2,125 and 3,448 ESTs were sequenced from G. theta and B. natans, respectively, and transcripts derived from nucleomorph genes were identified by comparison with genomic sequence resulting in 52 and 38 nucleomorph loci, respectively (the vast majority of transcripts being from host nuclear genes because nucleomorph transcription does not form a large proportion of expression in the cell). Once completely sequenced, 19 and 3 of these loci, respectively, appeared to represent spurious products that do not clearly correspond to mRNAs of any particular gene or are truncated at A-tracks in the genome, and these were not analyzed further. The structure of the cDNAs for the remaining loci were examined for all genes and gene fragments encoded within them (Tables 3 and 4, which are published as supporting information on the PNAS web site). The overall picture from these data are that multigene transcripts are more common in nucleomorphs than in microsporidia (Fig. 4A), but they are simpler in some respects. Identifying the likely expressed gene was not difficult, because virtually every cDNA began with a sense-strand gene. Three G. theta cDNAs and 28 B. natans cDNAs corresponded to intron-containing genes, one G. theta gene a formerly unannotated rpl30. All three G. theta introns were spliced in the cDNA, and all three were part of multigene transcripts. In B. natans not all introns were spliced, but all cDNAs contained at least one spliced intron, confirming beyond doubt that they correspond to mRNA. Only three G. theta transcripts appear to initiate within an upstream coding sequence, two in protein-coding genes and one within a tRNA gene (nearly all B. natans cDNAs were truncated near the 5′ end of the first gene, suggesting that 5′ leaders are likely not long). However, at the 3′ end, many cDNAs overlap with at least one additional gene, and sometimes several (Fig. 4A). In B. natans, cDNAs from 27 of 35 loci included sequence of more than one gene, and one of these included sequence from three genes. In G. theta, the proportions were higher, with cDNAs from 31 of 32 loci including sequence from at least one additional gene, and five cases included two or three additional genes or gene fragments. In G. theta, tRNA genes were found on five multigene transcripts, whereas none were found in B. natans, although one ended at a tRNA gene. In all cases in both nucleomorphs, multigene transcripts ended within a downstream coding region rather than beyond it, as was sometimes found in microsporidia, which suggests that processing signals are encoded within many genes. Indeed, in multiple cases from both nucleomorph genomes, transcripts from two adjacent genes were identified that show that mRNA processing can occur within otherwise actively transcribed genes. In some cases these genes were convergent (Fig. 4C), and in others they are transcribed from the same strand (Fig. 4D).

Compaction and Multigene Transcripts. Considering microsporidian and nucleomorph genomes together, there is a correlation between compaction and the presence of multigene transcripts, particularly in light of the further correlation between multigene transcription and short intergenic size within the A. locustae genome. This finding has several interesting implications for the process of genome compaction and for the process of transcription.

Why genomes compact is not obvious, but several reasons have been proposed. One of the underlying themes to explain compaction is the economy of reducing the amount of DNA to replicate (21, 22). This may be true in part, but this economy also appears to spawn waste at a different level, because many transcripts have exceedingly long untranslated regions, perhaps because the short intergenic regions have lost the capacity to encode control elements and these are replaced fortuitously upstream or downstream with whatever happens to work. Whatever the cause of compaction, another effect appears to be the stagnation of genomic rearrangements. This stagnation has been attributed to various constraints (23), including the reduction in intergenic spaces making it difficult to reorganize genes without introducing deleterious breakpoints (8, 23). The presence of multigenic transcripts adds another possible cause. The movement of promoter and terminator elements into adjacent genes will make it more difficult to reorder the genome, because removing a gene from one region of a genome might also remove control elements important for the expression of adjacent genes. It is likely that complex traits like this are caused by the sum of a variety of forces.

Multigene transcripts also raise some interesting questions about the process of transcription in nuclear genomes in general. Instances of overlapping transcription are certainly known from bacteria (24) and other eukaryotes, including well studied examples in other fungi, some of which have no detectable effect on gene expression (17, 18). However, on balance, overlapping transcription appears to be rare in systems where it has been examined globally (16). One of the reasons for this rarity is that transcriptional overlap can interfere with the expression of genes where it is normally prevented by control elements. This transcriptional interference can take several forms. Transcriptional collision between two convergent genes has been shown in an artificially convergent construct in Saccharomyces cerevisiae (25) or when the promoter of one yeast gene is enhanced to the point that transcripts read through the normal termination sites into a downstream gene (19). In both cases, the expression of one or both genes was disrupted. Similarly, enhancing the yeast ARO4 promoter so that transcription reads into the downstream HIS7 gene, which is on the same strand, blocks expression of HIS7 (26). Interestingly, this effect has also been used by yeast to control expression at the SER3 locus. Here, a small upstream noncoding transcript that extends into the SER3 promoter acts as an inhibitor of SER3 expression (27). Examples of both of these situations can be seen in compacted genomes (e.g., Fig. 4 B and C). As the SER3 locus shows, overlapping expression has to be considered in time as well as space, because temporal separation of expression could mitigate any deleterious effects of overlap. Indeed, it has been shown that coexpressed genes tend to be divergent when they are adjacent, perhaps to limit the possibility of promoter occlusion (28). However, it seems unlikely that this temporal separation of expression could explain the massive amounts of overlap observed in microsporidia and particularly nucleomorphs. These systems have more likely adapted in different ways to cope with high levels of overlapping expression.

One of the interesting aspects of this coping mechanism must be found in adjacent genes in the same strand where the upstream gene transcript terminates within the downstream gene (e.g., Fig. 4D). If there are termination signals within the downstream gene, how are they recognized by one transcription complex and not another? It is possible that transcription of the downstream gene is terminated at some frequency at these positions, but we have not observed this. It is also possible that relatively weak termination signals may be strong enough to terminate transcripts that have proceeded some distance from the promoter but not others that are still relatively short. Unfortunately, the process of transcription termination is complex and still only partially understood. It is clearly not as simple as a set of signals recognized by a complex but instead is tightly coupled to other processes, such as elongation (25, 29).

Implications for Other Systems. Transcription in nuclear genomes is not as simple as once thought. Numerous studies have found high levels of noncoding transcripts (sometimes called “orphan” or “sterile”), which sometimes overlap with protein-coding genes but apparently do not direct the translation of those genes (30–32). These important complexities are potentially universal to nuclear genomes but different from that described here. The major distinction is that multigene transcripts in microsporidian and nucleomorph genomes are not a class of transcripts apart from those that direct the expression of protein-coding genes. Instead, multigene transcripts are canonical mRNAs altered by extreme conditions in the genome. This effect is most obvious in the nucleomorphs, where multigene transcripts represent nearly 100% of mRNAs. In the microsporidia the situation is harder to interpret, but some of the very highly represented cDNAs (e.g., DNA polymerase α) are multigene transcripts. It is possible that many microsporidian cDNAs are noncoding transcripts like those found in other nuclear systems, but there is also another layer of complexity apparently introduced by the density of the genome, and this complexity is specific to such genomes.

Although microsporidia and nucleomorphs represent the most compact nuclear genomes presently known, there are also other genomes to which these observations may be extended. The marine picoalga Ostreococcus has a small genome (33), and it is reasonable to assume that it is likely compact as well. Similarly, complete genome sequence of the apicomplexan parasite Cryptosporidium has shown it to be reduced and somewhat compacted (34). We examined a small number of short Cryptosporidium EST sequences from public databases (www.cryptodb.org) and found at least 10 of 576 instances where cDNAs encoded fragments of more than one gene. Whether these all correspond to mRNA is not verifiable, and, because the sequences are short, these data are not appropriate to estimate the frequency of multigene transcripts in this organism. Nevertheless, observations from microsporidia and nucleomorphs suggest that this process should be sought in Cryptosporidium. Last, there is no reason to expect that multigene transcripts are restricted to cases where whole genomes are compacted. Transcription of genes within compacted regions of otherwise normal genomes may follow similar patterns, as indeed has been shown in one region of the yeast genome (35).