We had reported earlier the construction of a whole-genome PCR-product based S. coelicolor microarray with probes for more than 95% of the predicted ORFs 25. Comparative genomic hybridizations between S. coelicolor M145 and S. lividans TK21 using this microarray revealed extensive homology between the two chromosomes. This is apparent from Figure 1a showing the majority of ORFs with relatively similar signal intensity levels from both S. coelicolor and S. lividans gDNA channels (i.e. log₂ of hybridization signal ratio close to zero). The observation concurs well with earlier reports of genome-scale conservation between the two species based on restriction fragment based linkage maps 15. Genes with log₂ hybridization ratios less than the overall mean (of all genes) minus one standard deviation were designated as potentially absent or divergent in S. lividans. Of 7579 S. coelicolor genes probed in our array, we found that only ~8% fell in this category. Most often, these genes are clustered in localized regions of the chromosome. The clusters themselves, however, are scattered extensively across the entire length of the chromosome, albeit with a definite bias toward the right half. The absence of these clusters in S. lividans were confirmed by PCR with primers flanking the missing region (refer to supplementary material for details). Based on the size of these clusters and gene order in S. coelicolor M145, we classified the resident genes as belonging to either one of 5 large genomic islands (GI) (≥25 kb) or 18 smaller islets (Gi) (< 25 kb). These are in addition to the two chromosome ends and about 70 other genes scattered across the chromosome which, despite having low S. lividans gDNA signal, did not satisfy the three-adjacent-gene criteria used here to define a genomic island (see Methods section). The chromosome ends were not termed as genomic islands because large-scale DNA rearrangements are known to occur in the terminal regions leading to difficulties in ascertaining the exact boundaries of islands. Also PCR verification using flanking primers was impossible in such cases.

To assess whether these genomic deletions/insertions are consequences of horizontal gene transfer from distant species, we examined each island for characteristic features that define laterally acquired elements. Hallmarks of laterally acquired element include some or all of the following features – (1) unusual GC bias (2) anomalous codon usage (3) presence in the 3' end of tRNA genes (4) tendency to be flanked by direct repeats and (5) inclusion of mobile genetic elements 2627. We therefore, sought to evaluate the GC bias and codon usage pattern of genes across the entire genome of S. coelicolor. We adopted a windowless cumulative GC plot based on z' curve (see Methods section) to evaluate genome-wide GC variation. This approach is more robust, gives higher resolution and has been shown to be useful when the traditional sliding-window based GC plots fail 28. A sharp rise in the z' value at a localized region in the chromosome implies an unusually AT-rich segment. For assessing codon usage patterns, we calculated the

C

A

I

A total of 625 genes reside within the S. coelicolor genomic islands/islets identified above. They were categorized based on their functional assignments (Table 2). Unsurprisingly, more that two-thirds of laterally acquired elements in S. coelicolor are absent in S. lividans. Mobile genetic elements like integrative plasmids, bacteriophages and transposon related genes constitute this class. Other genes missing in S. lividans are comprised largely of hypothetical and periplasmic/exported proteins although a significant fraction belongs to a diverse set of functional categories shown in Table 2.

GI-1 (22 ORFs; ~26 kb) consists of a variety of genes including those coding for acetate uptake, amino acid biosynthesis and possible iron uptake systems. Superoxide dismutase – SCO0999 (sodF2), an Fe-dependent antioxidant was previously hypothesized to have been acquired by horizontal gene transfer 30. SCO0981-83 constitutes an acetate uptake system with a DNA-binding regulator (aceR), isocitrate lyase (aceA) and malate synthase (aceB2) genes. Isocitrate lyase and malate synthase are required for a functional glyoxylate pathway which promotes acetate utilization; yet, expression of these genes in S. cinnamonensis did not restore growth of acetate-uptake-deficient mutants on acetate as sole carbon source giving rise to the possibility that some, as yet unknown, factors are missing 31. SCO0997 and SCO0998 are annotated as iron-uptake system proteins – ftrD and ftrE respectively. However, the presence of an upstream gene (SCO0996) homologous to a lipoprotein found in daptomycin biosynthesis cluster has raised questions about their involvement in iron uptake 32. This is further corroborated by the presence of chromosomally linked genes homologous to SCO0991/92 (a hypothetical protein and a putative cysteine synthase) near the daptomycin cluster of S. roseosporus 33. If true, the presence of these antibiotic biosynthesis linked genes in a genomic island of S. coelicolor would be intriguing in view of evidence for horizontal transfer of antibiotic clusters among streptomycetes isolated from soil 34.

GI-2 (70 ORFs; ~75 kb) includes ORFs encoding a number of transposases and insertion elements. Interspersed within this region are a putative extracytoplasmic function sigma factor (SCO3450), a methyltransferase (SCO3452), a set of putative ABC-type transporters related to spermidine/putrescine transport family (Pot) (SCO3453-56) and an assortment of consecutive genes with various putative metabolic activities (SCO3473-3506). It is noteworthy here that the only extracellular agarase (dagA, SCO3471) of S. coelicolor maps to this genomic segment. This explains the lack of agarase activity in cultures of S. lividans noted elsewhere 35.

GI-3 (53 ORFs; ~57 kb) contains elements of three putative two-component systems (SCO6353/54, SCO6362/63/64 and an orphan kinase SCO6369). The region also encompasses ORFs with bacitracin transport permease domains (SCO6356, 6360, 6378). The last 16 ORFs in this island comprise mainly of mobile genetic elements like transposases and recombinases.

GI-4 (18 ORFs; ~30 kb) contains a phage growth limitation (Pgl) system. This is a defense mechanism against φC31 bacteriophage infections in which an infecting phage undergoes a single burst of attack and the resulting progeny are severely attenuated in subsequent infectious cycles. Two loci, pglWX and pglYZ – both of which map to this genomic locus – are necessary for the Pgl⁺ phenotype. Absence of these genes in S. lividans leads to unstable φC31 lysogens while complementing them with these elements give rise to Pgl⁺ colonies 36.

GI-5 (148 ORFs; ~153 kb) is the largest genomic island identified in this study. The size of this cluster implies that it hosts a multitude of genes with varied functions. Importantly, this cluster hosts genes that are involved in various modes of microbial defense. SCO6826/27 encodes a pair of type I modular polyketide synthase genes. In close proximity are two genes (SCO6832/33) coding for subunits of methlymalonyl-CoA mutase – an enzyme that catalyzes conversion of succinyl-CoA to methylmalonyl-CoA, which serves as building blocks for certain polyketide antibiotic synthesis 37. Two other genes, SCO6929/30 are similar to those involved in lantibiotic biosynthesis. In addition, SCO6835/36/37 belong to the arsenic resistance family. SCO6809/10 are similar to several multi-drug efflux transporters. Also, a probable acyl-CoA dehydrogenase (SCO6938) belonging to this island has been implicated in control of actinorhodin production and timing of sporulation 38. This observation is particularly interesting considering the extremely low-levels of actinorhodin production 39 and absence (or silencing) of the ram-independent developmental pathway in S. lividans 40. Furthermore, two putative DNA methylases (SCO6844 and SCO6885) also map to this locus.

The smaller genomic islets (Gi-1 to Gi-18) also contain several potentially interesting genes. Foremost among them is Gi-15. This islet includes the bldB locus (SCO5723) which has been attributed to pleiotropically regulate both antibiotic synthesis and morphological differentiation 41. Two other genes found in this islet (SCO5722 and SCO5731) encode putative secreted serine proteases. This islet is also unusual in being flanked on one side by a highly expressed rpsO gene rather than a tRNA. Among other smaller islets, Gi-12 contains an adenine-specific DNA methylase (SCO5331, pglS). This gene has been confirmed to interfere with the Pgl system of GI-4 and might extend resistance to phages other than φC31 42. Gi-10 and Gi-11 each encode a methyl specific restriction endonuclease – SCO4213 (mrr-like) and SCO4631 (mcrA-like) respectively. The rest of the genes in the genomic islands/islets are largely comprised of hypothetical proteins and other genes whose functions are yet to be elucidated.

Assuming that these genomic islands have evolved under selective pressure, we asked which, if any, of these genes are actively expressed under normal (laboratory) culture conditions. We chose to analyze mRNA and protein expression data derived from two distinct liquid media – one, an R5-based rich medium with yeast extract and the other, a minimal medium supplemented with casaminoacids (SMM). Due to significant differences in their composition, we expected that they will elicit markedly distinctive gene expression patterns. Data for the R5-based cultures were generated using our microarray and LC-MS experiments (Jayapal et al., in preparation) while those for SMM cultures were downloaded from a public repository http://streptobase.org/ or from previously published 2-D gel proteomic analyses 4344. Since the microarray hybridizations were performed with S. coelicolor gDNA reference, the resulting log₂ expression ratios (= log₂ [mRNA_Scoe/gDNA_Scoe]) are an indication of the expression level for each gene 25. The highest log₂ expression ratio observed for a given gene over the range of growth phases and conditions analyzed can be taken as its potential for transcription under laboratory conditions. Figure 2 shows this transcriptome expression data plotted against comparative genomics hybridization data. Genes in the top-left region of the plot (marked in Figure 2) are, in general, designated as absent in S. lividans, yet highly expressed in S. coelicolor at least in some conditions. Note that this region also includes a few genes with quite low log₂ [gDNA_Sliv/gDNA_Scoe] value but were not designated as absent/divergent in S. lividans simply because they did not co-localize as three- or more-gene clusters and hence did not satisfy out absence criteria (see Methods section). Reciprocally, some genes with high log₂ [gDNA_Sliv/gDNA_Scoe] were assigned to certain genomic islands because of overwhelming evidence of absence from neighboring genes. Many such genes with considerable homology are transposon/plasmid related elements and are probably present elsewhere in the S. lividans chromosome. 42 genes from various genomic islands exceeded a threshold of max(log₂ [mRNA_Scoe/gDNA_Scoe]) value greater than 2.0 indicating appreciably high mRNA expression. 20 of these genes also had an average(log₂ [mRNA_Scoe/gDNA_Scoe]) > 0 indicating probable constitutive basal expression (data not shown in Figure). Currently, more than half of those 42 genes (23/42) do not have a reasonably descriptive functional annotation.

The most striking observation from the transcriptome analysis was the high expression of ~30 contiguous genes (SCO3473-3506) within GI-2 exclusively in SMM cultures. These genes encode a variety of probable carbohydrate metabolism enzymes including putative aldolase, isomerase, dehydrogenase, beta-galactosidase, sugar kinase and sugar permease; the precise context of these functions remain to be elucidated. Nevertheless, such coordinated but conditional transcription of genes within genomic islands leads one to speculate that, in addition to collective horizontal transfer as an island, they are also regulated in an independent and rather concerted fashion in response to varying environmental stimuli. In addition, the extracellular agarase precursor (SCO3471) in GI-2 is also expressed highly only in SMM cultures.

Several genes in GI-1 are also highly expressed. The superoxide dismutase (SCO0999, sodF2) has an average(log₂ [mRNA_Scoe/gDNA_Scoe]) of over 2.0 in cultures of R5-based medium and was also identified in LC-MS proteomic studies with multiple peptide hits (Jayapal et al., in preparation). However, this gene is not as highly expressed in SMM cultures. On the contrary, adjacent genes SCO0985/0993/0995 were identified in the 2-D gel experiments conducted on minimal medium derived samples 43. Genes in bldB locus (Gi-15) are also expressed at moderate to high levels from microarray data (in both rich and minimal medium). This is corroborated by the identification of some of these proteins (SCO5723/24 and SCO5729) in proteomic experiments. Interestingly, SCO5724/25 possibly codes for a novel secretion system 45. Such transporters are widely associated with pathogenicity islands in many bacteria 26. Furthermore, phage resistance proteins of the Pgl system (SCO6627/28/37/38; GI-4) were identified in several proteomic experiments. Notably, one of them – SCO6627 (PglX) was identified in multiple forms in the membrane fraction of mycelium undergoing programmed cell death 44. Likewise, polyketide precursor synthesis enzymes (SCO6832/33, methylmalonyl-CoA mutase subunits in GI-5) were identified in LC-MS experiments of R5 medium samples (Jayapal et al., in preparation).

Our results indicate that over 90% of S. coelicolor genes are conserved remarkably well in S. lividans. This observation is not surprising considering that even bacteria from other genera, namely Mycobacterium and Thermofibida, share a significant homology with genes in Streptomyces 46. Punctuating this pattern of widespread conservation are 23 genomic islands comprising > 600 genes that are conspicuously absent in S. lividans. Many of these islands exhibit hallmarks of horizontal gene transfer. Although it is generally believed that chromosome extremities are more prone to massive gene flux compared to the central "core" 2, we did not observe a higher occurrence of genomic islands near the chromosome ends (10 genomic islands in the "core" compared to eight in the chromosome "arms"). This could partly be explained for certain islands containing phage-related elements or integrative plasmids that target specific sequences rather than rely on genome instability for chromosome integration. There is at least one report of DNA amplifications in the "core" region of S. coelicolor indicating the chromosomal instabilities are not strictly limited to the extremities 47. Many of these regions (eg. GI-1, GI-4, GI-5, Gi-15) harbor elements that could potentially benefit the host, asserting that they truly are "fitness islands". Such elements include antibiotic biosynthesis genes, drug resistance determinants, phage defense systems, and numerous metabolic functions.

It is notable that 14 of the 23 islands detected in this study were reported as putative laterally acquired regions through in silico analysis during sequencing of the S. coelicolor genome 5. Half of those (7/14) were reported precisely as described here; for the rest, the exact boundaries of the islands do not match our results. This does not, however, imply that our results contradict previous observations. One likely possibility is that genes in any extended genomic island predicted by Bentley et al., although recently acquired, are well-conserved in S. lividans (probably acquired before speciation). Another possibility is that the apparent discrepancies are merely consequences of our strict criteria for absence that eliminated certain marginal cases. On the other hand, all boundary extensions (and new islands) reported in this work provide conclusive experimental evidence of recent gene transfer in those regions.

Over the years, identification of genetic determinants that suppress actinorhodin production in S. lividans has aroused considerable interest among researchers. Although numerous mutational and overexpression studies have led to activation of antibiotic synthesis in S. lividans 4849505152, to our knowledge, none have succeeded in identifying the evolutionary transformation that led to modulation of actinorhodin synthesis. Our results provide two important clues in this regard. One is the absence of a probable acyl-CoA dehydrogenase (SCO6938) in S. lividans, whose mutation in S. coelicolor caused a marked delay in onset of actinorhodin synthesis as well as aerial mycelium formation 38. Acyl-CoA dehydrogenases are enzymes associated with breakdown of long-chain fatty acids and might potentially generate precursors for polyketide biosynthesis 53. Of particular note is a mutant of an acyl-CoA synthetase (another enzyme involved in fatty acid catabolism) in S. coelicolor that was severely impaired in actinorhodin synthesis particularly in glucose containing medium 54 – an observation curiously reminiscent of the conditional suppression of actinorhodin biosynthesis in S. lividans in glucose containing medium. The other clue comes from the apparent absence of bldB (SCO5723) locus in S. lividans. At least two other paralogs of bldB (SCO3424 and SCO4542) are also known to exist in S. coelicolor chromosome, both of which are conserved in S. lividans. However, these paralogs do not substitute for bldB – S. coelicolor mutants of bldB fail to erect aerial hyphae and lack actinorhodin production; yet, S. lividans appears to be quite capable of forming aerial mycelia and spores. Despite the apparent absence of SCO5723 locus in S. lividans, we also noted that a S. lividans gene annotated as bldB (GenBank accession AF071232) has been cloned and expressed in E. coli 55. A closer look revealed that this S. lividans bldB shares only ~82% homology with SCO5723 and hence gave a weak hybridization signal – an observation further corroborated by the failure of stringent Southern hybridizations to reveal the presence of bldB in S. lividans 56. Perhaps, this level of similarity is sufficient for most of the functional activity of BldB. Notably, all the conserved residues reported by Eccleston et al. 57 are present in both versions of the protein. Nonetheless, our data suggests that this ortholog in S. lividans is present at an alternate locus since an entire 17-gene S. coelicolor segment (Gi-15) including SCO5723 is absent in S. lividans. We note that the other bldB paralogs of S. coelicolor have a much lesser degree of homology with S. lividans bldB.

Another question of particular interest in this comparative study is what genetic factors make S. lividans a preferred host for heterologous protein expression? As mentioned earlier, the two contributing factors in this regard are the absence of a strong restriction-modification system and diminished extracellular protease activity. The phage growth limitation (Pgl) system found in GI-4 contains elements required for DNA restriction. Although this explains the unstable φC31 lysogens of S. lividans, introduction of this system into S. lividans did not impair its ability to uptake methylated DNA 36. Two other potential candidates – SCO4213 and SCO4631 (both annotated as hypothetical proteins) were identified in a scan for methyl-sensing restriction endonucleases. SCO4213 contains a signature Mrr_cat type II restriction enzyme domain while SCO4631 is similar to an E. coli methylcytosine-specific restriction enzyme with an HNH endonuclease domain. In addition, SCO2863 – a putative helicase, also contains an hsdR-like type I restriction endonuclease domain. Moreover, the absence of two DNA-methylases, SCO6844 and SCO6885 might explain why DNA obtained from S. lividans is readily transformable into other streptomycetes. The attenuated extracellular protease activity in S. lividans is, perhaps, explained by the absence of two secreted serine proteases – SCO5722 and SCO5731.

An interesting observation that we noted in our analysis is the unusually high frequency of the rare TTA codon containing genes designated as absent in S. lividans. About 30% (43/145) of such genes are absent in S. lividans – statistically a much higher frequency compared to only 8% of all genes absent. Considering that AT-rich segments frequently occur in bacterial phages and plasmids 58 and also the fact that occurrences of TTA codons are quite rare in a GC-rich organism like S. coelicolor, it is likely that many of these genes are of foreign origin. In fact, a recent report by Chater and Chandra postulated that over 80% of TTA-containing S. coelicolor genes were acquired through horizontal gene transfer 46. They speculated that bldA (the sole tRNA that can efficiently translate TTA) might itself have been laterally acquired, and that TTA-modulated protein expression is a very recent evolutionary adaptation. Our experimental results are consistent with their hypothesis.

Spores for S. coelicolor M145 and S. lividans TK21 were generated on Mannitol-Soy flour or R5 agar 4. Cultures for genomic DNA preparation were performed in YEME medium with 0.5% glycine supplement at 30°C until early stationary phase.

Genomic DNA (gDNA) extraction was carried out using Kirby mix procedure as described elsewhere 4. About 500 μl of 20 μg/μl gDNA was sonicated briefly for 30–40 sec for shearing them to ~500 bp average size (confirmed by gel electrophoresis). The DNA was then labeled with Label IT^® Cy3 or Cy5 Labeling Kit (Mirus Bio Corp., Madison, WI) according to suppliers instructions.

Samples containing ~500 ng each of Cy3 and Cy5 labeled gDNA from S. coelicolor M145 and S. lividans TK21, respectively were hybridized to a whole-genome S. coelicolor microarray as described previously 25. Hybridizations were carried out in triplicate for ~16 hr at 50°C; arrays were washed and scanned using ScanArray5000 (Perkin Elmer, Wellesley, MA). Images were analyzed using GenePix (Axon Instruments, Union City, CA) to obtain raw intensity data for each spot. The median fluorescence intensity from each spot was used for all subsequent analysis.

Raw relative intensity values for each spot were first normalized by scaling one of the gDNA channel signal intensities by a normalization factor to set the total intensity from both channels as equal. Log₂ hybridization signal ratios were then calculated from normalized intensities as log₂ [gDNA_Sliv/gDNA_Scoe] and values were averaged using the median from triplicate experiments. Presence of at least three consecutive (in chromosome order) log₂ [gDNA_Sliv/gDNA_Scoe] values less than its global mean minus one standard deviation was taken as evidence for absence (or divergence) of a contiguous genomic segment in S. lividans. Certain marginal cases were thereafter manually reassigned as present/absent in S. lividans based on overwhelming trends of neighboring genes. For visualization purposes, median intensity ratios from successive 19-gene sliding windows were plotted as a function of genes in chromosome order.

For transcriptome analysis, cDNA-gDNA based microarray data was normalized using quantile normalization 25.

All microarray data discussed here are available at Gene Expression Omnibus (GEO): Accession – GSE7167 (comparative genomics hybridizations) and GSE7172 (transcriptome data).

Regional GC variations in S. coelicolor genome were calculated using z' curve method proposed by Zhang and Zhang 59. Briefly, z_n was calculated at genomic location n as:

z_n = (A_n + T_n) - (G_n + C_n), n = 0, 1, 2...N

where A_n, T_n, G_n and C_n are the cumulative number of bases A, T, G and C occurring in a sequence from 1^st to n^th base. To amplify the deviations of z_n from its average trend, a linear least square fit: z_n = kn is performed and z_n' is calculated as

z_n' = z_n - k_n

Codon Adaptation Index (CAI), a measure of utilization of "optimal codons" was calculated as described previously 29. First, a relative adaptiveness factor for every codon i coding for amino acid j is estimated as:

w i , j = f i , j ( G ) f max ⁡ , j ( G ) ,  for all codons  i  and amino acids  j MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@714B@

where f_i,j(G) is the frequency of occurrence of codon i among the set for amino acid j across the entire genome G and f_max,j(G) is the corresponding value for the most frequently used codon for amino acid j. The CAI for a gene g is then calculated as

C A I g = ∏ i = 1 N w i 1 / N MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaaieaacqWFdbWqcqWFbbqqcqWFjbqsdaWgaaWcbaGaem4zaCgabeaakiabg2da9maarahabaGaem4DaC3aa0baaSqaaiabdMgaPbqaaiabigdaXiabc+caViabd6eaobaaaeaacqWGPbqAcqGH9aqpcqaIXaqmaeaacqWGobGta0Gaey4dIunaaaa@3F07@

where N is the number of amino acids in gene g. The algorithms for both cumulative GC plot and CAI plot were implemented using Matlab 7.0 with bioinformatics toolbox.

The absence of genomic islands in S. lividans were verified by PCR using primers flanking each island. Primers were chosen so as to fall in or outside of the DNA probe segment used in microarray to improve chances of amplification. The PCRs were conducted using GC-rich™ PCR system or Expand™ long template PCR system (Roche Applied Science, Indianapolis, IN) with 2% DMSO. Successful amplification of a relatively small product from S. lividans indicates absence of the intervening genomic island in each case.