Characteristics of the T. oceanica genome
The genome of the centric diatom T. oceanica
CCMP1005 (Figure ) was de novo
assembled from 725 Mb of Roche 454 sequence read information, generated using nuclear genomic DNA (gDNA) of an axenic clonal culture as substrate [14
]. The current assembly version comprises 51,656 contigs of total size 92.15 Mb at N50 = 3,623 (that is, 50% of the genomic sequence information is present as contigs ≥3,623 bases). From a median 8.7-fold coverage of long contigs (≥10 kb) we estimated a true haploid nuclear genome size of 81.6 Mb, suggesting some redundancy in the current assembly. This estimate is in good agreement with the 159 Mb measured by van Dassow et al
] for the diploid G1 DNA content. The gene finder tool AUGUSTUS [16
] predicts 37,921 protein gene models that cluster into a non-redundant set of 29,306 models including pseudogenes and short ORFs; 10,109 models have BLAST hits to the National Center for Biotechnology Information (NCBI) nr protein database at a conservative E-value cutoff of 1.0E-30 and thus are more indicative of the expected true protein-coding gene number (that is, expressed genes excluding pseudogenes and short ORFs). Best BLAST hits are listed in Additional file 1
. In Table we present an overview of the most abundant Clusters of Orthologous Groups (COG) domains in T. oceanica
. The abundances of diverse groups of ATPases were overall very similar to those for other diatoms. A group of 19 chitinases is shared between the two centric Thalassiosira
Figure 1 T. oceanica CCMP1005 genome statistics. The sequenced strain T. oceanica CCMP1005 belongs to the Centrales group of radially symmetric diatoms and was first isolated from the oligotrophic Sargasso Sea by R Guillard. At 92.15 Mb, our genome assembly is (more ...)
Most abundant protein domains in diatom genomes
The chloroplast genome has been published previously [17
]. The mitochondrial genome encodes 31 protein genes and is represented by two contigs at a total of 35.3 kb (excluding the characteristic mitochondrial repeats). The current genome assembly, AUGUSTUS protein gene models, ESTs and proteomics peptides as well as updated versions thereof are publicly accessible with the Thalassiosira oceanica
Genome Browser [18
With an estimated haploid size of approximately 80 Mb, the genome of T. oceanica
is significantly larger than those of T. pseudonana
(approximately 34 Mb) or P. tricornutum
(approximately 28 Mb), and rather comparable to that of Fragilariopsis cylindrus
(approximately 80 Mb) [19
]. The genome expansion has occurred by DNA recruitment from both internal and external DNA sources.
A best BLAST hit analysis indicated a putative vertical inheritance for greater than 95% of the 10,109 predicted genes (that is, any genes that have not been acquired by a horizontal transfer event), with most of the genes (88%) having a match in the genome of T. pseudonana
, the most closely related species for which a sequenced genome is available (Figure ). However, a significant fraction (10%) of the genes mapped to P. tricornutum
instead. This could have resulted from frequent gene loss/replacement events in the genome of T. pseudonana
, thereby reflecting the overall high capacity for horizontal gene transfer in diatoms [5
]. Alternatively, the small genome size of T. pseudonana
may have arisen from reductional trends in this species.
Figure 2 Vertical versus horizontal inheritance of genes. For evaluation of the extent of laterally acquired genes we focused on the 10,109 AUGUSTUS gene models that have homologs in the NCBI nr protein database at a conservative E-value cutoff of 1.0E-30 (middle (more ...)
Further, the best BLAST screening revealed 530 genes whose best hits are assigned to taxa of diverse sources, indicating a putative lateral acquisition for these genes. The taxonomic distribution of the best BLAST hits at a conservative E-value cutoff of 1.0E-30 is presented in Figure . More refined phylogenetic analyses were obtained for 198 of these 530 genes (Additional file 2
). Of the 198 cases we examined, 180 had sister groups that contained no stramenopiles. The sister groups for the remaining 18 cases contained a heterogenous mix of taxa, suggesting frequent transfer between taxa for the respective genes (Figure S1 in Additional file 3
). Accordingly, a minimum of 1.8% of the 10,109 AUGUSTUS genes were confirmed to be from lateral gene transfer (LGT) based on the phylogenetic analyses. However, this may rise to 5% as more sequence information becomes available for the remaining >300 genes for which the limited number of homologous sequences did not permit construction of phylogenetic trees. The group of genes for which phylogenetic trees are available indicated that genes from LGT could be assigned as prokaryotic (35%) and eukaryotic (59%) with approximately 10% of questionable taxonomic assignments. Among the eukaryotic taxa are several expected to be present in the ecological niche of T. oceanica
, like the green algal genuses Micromonas
Genomic expansion originating from internal DNA sources may happen from genomic duplication events or transposon activity. In T. oceanica
we observe several paralogous gene pairs that could be the result of either mechanism (Figure S2 in Additional file 3
). Notably, several iron-regulated genes have either been duplicated (for example, the ISIP1
and the flavodoxin genes FLDA1
) or contain domain duplications (for example, CREGx2
) as discussed below.
For an in-depth analysis of the T. oceanica
low-iron response, we focused on approximately 300 genes that were identified from a log-likelihood ratio test statistic [22
] as significantly differentially regulated and that could be assigned a specific function (Figure ). Some additional genes for paralogous proteins were added. These were selected on the basis of their involvement in substitution between related proteins under iron-limited and replete conditions, or as members of a protein family exhibiting a differential response to low-iron conditions. In such cases, the response of a specific gene is better understood in the context of its respective group or family. The complement of organellar genes (encoded by the chloroplast and mitochondrial genome) was added as representative for the two well-defined and important pathways of photosynthetic and respiratory electron transport, or as proxy for organellar activity, respectively. A list of abbreviations for the genes discussed in this work can be found in Additional file 4
. All sequences of the selected proteins are provided in Additional file 5
, and the corresponding annotation is provided in Additional file 6
Figure 4 Basic cellular changes at low-iron. Differential gene expression of exponentially growing iron-limited versus iron-replete T. oceanica cells was assessed from global transcriptomics and proteomics approaches. (a) Transcriptomics data were screened with (more ...)
To determine the major metabolic differences found in iron-limited compared to iron-replete growth conditions, all annotated gene products together with their respective expression data were mapped on a cellular scheme. The major cellular trends that could be deduced are summarized in Figure . Identifier and detailed information on the discussed proteins (Additional file 7
) are given in Additional file 8
. In the following, proteins are referred to as exemplary (HSF1, p271) with HSF1 reflecting the gene name (or shortcut) and p271 being the identifier of its respective manually improved protein model (Additional file 5
Under stress conditions, maintaining cellular integrity is crucial to survival. During iron limitation, the electron flow through the impaired photosynthetic machinery leads to enhanced production of reactive oxygen species that damage biomolecules located near the thylakoid membranes [23
]. The need for protein repair and refolding induces an 'oxidative stress response' that is presumably coordinated by up-regulated heat shock factors (HSF1, p271; HSF2, p256). While all other chloroplast-encoded transcripts were down-regulated in the course of the general chlorosis response, the chloroplast chaperones dnaK and clpC were up-regulated. Additionally, an LHCSR (light harvesting complex stress responsive subunit) ortholog (LI818, p170), belonging to the FCP (fucoxanthin-chlorophyll a/c-binding protein) family of light-harvesting proteins and implicated in efficient non-photochemical quenching [24
], showed an increased transcript level.
The development of a chlorotic phenotype and the corresponding retrenchment of the chloroplast system is the most pronounced cellular response to low iron. Accordingly, we find substantial changes in organellar transcript levels, which suggests that major functions related to the cellular energy metabolism are adopted by the mitochondrial system instead ('metabolic shift'). Chloroplast transcript levels decreased (2,026 Fe(-) versus 14,931 Fe(+) total chloroplast reads), while mitochondrial transcripts showed a two-fold increase (31,261 Fe(-) versus 18,136 Fe(+) total mitochondrial reads). Much of this effect can be attributed to the organellar rRNA operons, whose transcription is indicative of organellar translational activity (Figure S3 in Additional file 3
). In parallel, diverse nuclear-encoded but chloroplast-targeted gene products were down-regulated. These included genes coding for enzymes involved in chlorophyll biosynthesis and the Calvin cycle, as well as components of the light reaction, such as photosystem (PS) subunits and several FCPs. Conversely, components of the mitochondrial respiratory chain, like cytochrome c oxidase, cytochrome b and several subunits of the NADH dehydrogenase, were up-regulated. This was also seen for a mitochondrial ATP/ADP-translocase (p242) involved in the transport of energy equivalents.
Cellular retrenchment (that is, the reduction of cellular biomass and activity) and decreased growth rates are general responses of nutrient-limited cells [13
]. While chloroplast reduction was readily observable in iron limitation due to the visual predominance of these organelles in the cells, we also saw indications of a general cellular retrenchment in the transcriptional response. The expression level of the 18S rRNA gene (represented by 1,154 Fe(-) versus 2,691 Fe(+) reads) suggests a lower translation rate under iron limitation. Though such inferences must be taken with care, this would be in agreement with the decreased growth rate and lower biomass, as cellular rRNA correlates with cellular biomass. The strong up-regulation of mitochondrial isocitrate lyase (ICL, p419) and glutamine synthetase (GS, p302) suggests biomass recycling strategies to avoid losing fixed carbon and nitrogen during the metabolite conversions associated with enhanced respiration. The isocitrate lyase bridges the two decarboxylation steps of the mitochondrial citric acid cycle (carried out by isocitrate dehydrogenase and α-ketoglutarate dehydrogenase), thereby preserving carbon as glyoxylate. The glutamine synthetase reincorporates free ammonium, preserving nitrogen as glutamine. Under low-iron conditions, utilization of ammonium is energetically advantageous due to the high iron requirements of the nitrate assimilation pathway [26
]. The concerted action of cellular retrenchment and biomass recycling allows for prolonged growth despite reduced carbon assimilation, thereby increasing the probability of cell survival.
Diverse genes, whose products are targeted to the secretory pathway, are up-regulated under iron limitation, suggesting extensive cell-surface remodeling as also observed for iron-limited P. tricornutum
]. Many of these genes are assigned adhesive or degradative functions. An enhanced capacity for adhesion favors recruitment of organic matter to the cell. As organic matter can be a rich and complex source for various nutrients, including iron, its recruitment to the cellular surface represents a required first step in iron uptake. Besides providing a source of iron, the bound organic matter could also serve as a source for other nutrients like nitrogen or phosphorus in the context of facultative mixotrophy. Example genes assignable to such a hypothetical scenario and highly responsive to low iron are given in Figure and include Adhesin 1
(Carbohydrate-binding 1, p230), CHIT
(chitinase, p88), M-Phosphoesterase
(p279), Redox 1
(p232). However, under the photoautotrophic experimental conditions, the cultures lacked any external organic carbon source except the essential vitamins.
Figure 5 Hypothetical categorization of low-iron-inducible cell surface proteins. In low-iron conditions we find an up-regulation of diverse genes, whose products are targeted to the secretory pathway, suggesting extensive cell surface remodeling. Many of these (more ...)
A straightforward strategy to survive in low-iron conditions is to lower cellular iron requirements by replacing components that are rich in iron with iron-free substitutes that are functionally equivalent, like the substitution of the chloroplast electron carrier ferredoxin with flavodoxin [10
]. The genome of T. oceanica
encodes two cytochrome c6
genes and one plastocyanin gene. While the cytochrome c6
are found to be weakly expressed, the plastocyanin gene PETE
shows high expression under high-iron conditions with a characteristic decrease in low-iron conditions as seen from many constitutively expressed chloroplast genes in the course of the chlorosis response. This suggests a constitutive use of plastocyanin (PETE, p175) instead of cytochrome c6
for photosynthetic electron transport and is consistent with prior findings [11
]. Constitutive expression of plastocyanin could certainly be regarded as a specific adaptation to low-iron regimes, although the retention of the cytochrome c6
genes suggests that these may play a role under specific environmental conditions. Fructose-bisphosphate aldolase (FBA) genes are redundant in some diatoms and have recently been described in more detail for P. tricornutum
]. The T. oceanica
genome, too, was found to encode several FBA enzymes, with the cytosol, the chloroplast stroma and the chloroplast pyrenoid harboring two FBA enzymes each (FBA1, p380, and FBA3, p153, in the chloroplast pyrenoid; FBA2, p381, and FBA5, AUG_g19407, in the chloroplast stroma; FBA6, AUG_g24977, and FBA4, p154, in the cytosol). As is the case in Phaeodactylum
, one of the T. oceanica
FBAs from each compartment (FBA1, FBA2, FBA6) appears to act through metal catalysis (class II) while the second (FBA3, FBA5, FBA4) is predicted to use Schiff-base catalysis (class I) instead. While the metal cofactor of different class II FBAs was found to be Mn2+
] or Cd2+
] in Escherichia coli
, the orthologous FBAs of T. oceanica
apparently are differentially regulated through the availability of iron, suggesting the involved metal in these enzymes might be Fe2+
, and implying a pairwise substitution by class I enzymes.
An essential part of iron-uptake systems are ferric reductases (FREs) and ferrous oxidases (MUCOX proteins) that act on the interconversion of the two ionic species Fe3+
. In the iron-limited transcriptome we find an up-regulated putative ferric reductase (FRE1, p157) and an up-regulated multicopper oxidase (MUCOX2, p67) that shows characteristics of a ferrous oxidase. Their differential regulation with respect to iron availability makes them candidates for iron-specific reductase and oxidase involved in iron uptake (Figure ). Iron uptake requires initial binding of iron and/or iron complexes. The involved receptors are presently unknown, though a number of genes, exclusively expressed under iron limitation, are targeted to the cell surface, making them candidates for iron-binding receptors. The low-iron-responsive gene ISIP1
(Iron-starvation induced protein) was first identified in P. tricornutum
, but has conserved orthologs in T. oceanica
, p159) and F. cylindrus
. We provide further evidence for a role of the ISIP1 protein as a putative receptor below. Additional members in this group are ISIP2 (p160) and ISIP3 (p161), both represented by orthologs in P. tricornutum
as well. Further, we list some proteins that contain duplicated domains known from P. tricornutum
low-iron-responsive genes, like an eight-fold duplicated ISIP2-like subdomain (ISIP2x8, p84) or a duplicated CREG-like domain (CREGx2, p90). Duplication of iron-binding domains would directly enhance the capacity for iron binding and enable increased uptake kinetics [26
Non-ribosomal peptide synthases (NRPSs) [31
] are responsible for the production of peptide antibiotics or - in some cases - siderophores that are capable of binding iron [32
]. In addition to a conserved fungal NRPS (NRPS1, p174) with orthologs in T. pseudonana
and P. tricornutum
, we find a putatively cytosolic NRPS of bacterial origin (NRPS2, p173) up-regulated in low-iron conditions. Co-regulated with this bacterial NRPS is a multidrug resistance-associated protein (MRP, p57) that might be involved in the export of the respective peptide products. The up-regulation of NRPSs likely indicates a defense mechanism in response to enhanced competition (either for iron or, under the premise of facultative mixotrophy, for organic matter).
We observe the induction of a reverse transcriptase (RT, p222) and a CRE-like recombinase (CRE, p321), potentially indicating an activation of mobile elements under iron limitation. These enzymes might also be involved in gene and/or domain duplication events through reverse transcription and genomic integration of cellular mRNA copies. Thereby, this molecular system may provide a link between environmental stresses and the structural dynamics of the diatom genome.
The transcriptomic data of T. oceanica
unveils extensive changes in cellular transcript levels in response to iron limitation. Although informative, transcript abundances do not necessarily reflect cellular protein levels [33
]. We therefore supplemented the transcriptomic data with proteomic data to determine the protein complement in action under the defined iron-replete and iron-limited growth conditions. Figure illustrates the dynamic range of differential abundances for all proteins detected by liquid chromatography-tandem mass spectrometry (LC-MS/MS) relative to equal amounts of total cellular protein for both conditions. The induction of flavodoxin is a hallmark of iron-deficiency responses in many diatoms and cyanobacteria (see above). In accordance with the transcriptome response, flavodoxin as well as ISIPs or class I FBAs could only be found under iron limitation. The extent of correlation between proteomics and transcriptomics data was assessed through plotting the relative abundance data from peptides (proteomics P) against the relative abundance data from their corresponding transcripts (transcriptomics T) (Figure S4 in Additional file 3
). A stretched cluster along the y-axis indicates a high dynamic range of the transcriptomics data, while the proteomics data is more uniform for this group.
Both transcriptomics and proteomics data are biased towards highly abundant transcripts/proteins. Especially the proteomics data, despite its relatively high number of signals, could resolve only a subset of the protein complement. Accordingly, we interpret the complement of differentially regulated genes and proteins recovered from both approaches as complementary in the information that they provide, and we do not expect them to show a complete overlap. However, the overlap in the response for the specifically induced proteins ISIP1 and class I FBAs shows that the data from both approaches are, in general, in good agreement with each other.
In the proteomics data it is of specific interest to have a closer look at proteins of the photosynthetic machinery. Chloroplast ribosomal proteins provide an appropriate internal reference for the regulation of chloroplast proteins and indicate a down-regulation of the ribosomes at a ratio of 0.8 relative to the iron-replete proteome. Protein subunits of PS I were reduced about two-fold under low iron conditions (0.45), except PsaL, which was only found under iron limitation. In cyanobacteria, PsaL, generally important for trimer formation, facilitates the formation of IsiA (iron stress induced protein A) rings around PS I monomers under iron-deprivation [34
]. We speculate that PsaL might be involved in the organization of PS I light-harvesting structures specifically formed under low-iron conditions and/or oligomerization of PS I in iron-limited T. oceanica
. Subunits of the iron-containing cytochrome b6
/f (cyt b6
/f) complex, were down-regulated, with ratios of 0.2 and 0.32. In contrast, PS II subunits PsbB, PsbC, PsbE, PsbH and PsbV remained almost constant, with ratios at about 1.1. While the PS II core complex seems to be retained to some extent, the labile D1 protein is down-regulated at 0.7, probably reflecting a proportional decrease in functional PS II. The differential regulation of the two photosystems (0.45 for PS I versus 0.7 for PS II D1 protein) supports an adaptive significance for the remodeling of the photosynthetic architecture under iron limitation, in contrast to earlier findings [12
While PS I and cyt b6/f complexes were down-regulated two- to threefold, it was still possible to detect the iron-rich mitochondrial complexes under iron limitation. Relative protein quantification was possible for subunits of complex III, complex IV and the ATPase with low-to-high iron ratios ranging from 0.95 for QOR2 (a NADPH-dependent quinone oxidoreductase) to 1.7 for the beta subunit of the mitochondrial cytochrome c oxidase (Figure ). This is in agreement with the transcriptomic data and supports the idea that mitochondrial electron transfer protein complexes are preserved under iron limitation relative to photosynthetic electron transfer protein complexes.
While the magnesium chelatase, involved in chlorophyll synthesis, is down-regulated at 0.35, the numerous FCP light-harvesting proteins showed very diverse responses to iron limitation (Figure S5 in Additional file 3
). Some FCPs showed down-regulation under low-iron whereas others were up-regulated. In particular, LHCSR-like FCPs, involved in photoprotection, were highly abundant under iron limitation, corroborating the transcriptome analysis. Notably, the xanthophyll cycle enzyme violaxanthin de-epoxidase showed significant up-regulation at 3.1, suggesting a possible linkage to the group of FCPs, which accumulate under iron limitation.
Comparative genomics reveals extensive genomic plasticity in T. oceanica
We used the genome information of T. oceanica, T. pseudonana, P. tricornutum and F. cylindrus to investigate central issues of the diatom low-iron response in a comparative genomics approach.
Taxonomic distribution of iron-regulated genes
We screened the four diatom genomes known to date (T. oceanica
, T. pseudonana
, P. tricornutum
and F. cylindrus
) for the highly conserved iron-regulated ISIP1
and class I and II FBA
genes (Table ; Additional file 9
). Phylogenetic trees for the important groups of flavodoxin [35
] and FBA proteins are provided in Figures S6 and S7 in Additional file 3
Presence and copy number of iron-regulated genes in the genomes of ecologically distinct diatoms
The short flavodoxin isoform, plastocyanin and the class I FBAs are known or assumed to replace iron-containing counterparts under low-iron conditions. The two oceanic diatoms T. oceanica and F. cylindrus, which have some of the highest tolerance to low-iron conditions, both contain all five of the respective genes while P. tricornutum lacks two of them. The typical coastal species T. pseudonana lacks all except the gene for the cytosolic class I FBA, while at the same time having the highest requirement for iron in the group of diatoms for which genome information is currently available. Further, we find multiple copies of the ISIP1 gene in T. oceanica and F. cylindrus, while this gene is absent in T. pseudonana. The presence or copy number of these genes in the tested diatom genomes suggests an adaptive significance with respect to the low-iron conditions found in oceanic waters.
Domain duplications of iron-regulated cell-surface proteins
While differentially regulated genes for cell-surface proteins, identified from the low-iron response of P. tricornutum
], like ISIP1
, represent single-copy genes encoding well-defined single-domain proteins, the situation in T. oceanica
is different (Figure S2 in Additional file 3
). Here, we find additional paralogous versions of several iron-regulated genes (ISIP1
), as well as diverse examples of domain duplications (CREGx2
). In the case of iron-binding proteins the duplication of domains might provide benefits under iron limitation through a higher density of exposed domains, thereby increasing the affinity for iron at the cell surface [26
With respect to the selective pressure encountered in the low-iron open ocean the duplication of complete genes may provide a possible mechanism for adaptation on the molecular level, in that it allows one of the two gene copies to vary, improve and optimize its iron-binding themes/motifs. This may potentially result in more efficient iron uptake. RT-qPCR allowed us to distinguish iron-regulated genes from their closely related paralogs (Figure S8 in Additional file 3
Iron uptake and the cell-surface protein ISIP1
Conservation between the predicted protein orthologs of ISIP1 in T. oceanica
, P. tricornutum
and F. cylindrus
was high, and the orthologs exhibited identical secondary structure predictions (Figure ). We found an amino-terminal signal peptide targeting the protein to the secretory pathway, while a carboxy-terminal transmembrane domain anchors the protein to a membrane. The major part of the protein is represented by a domain rich in β-strands that likely folds into a β-propeller-like structure. A clue to the structure and function of ISIP1 could be the low-density lipoprotein receptor LDLR, an important cell-surface receptor in humans [36
]. Although its extracellular domains differ from the single β-propeller domain of ISIP1, the remainder of the protein is strikingly similar with regard to amino acid composition and secondary structure prediction. Hence, we may transfer the respective LDLR annotation to the ISIP1 protein model.
Figure 6 The low-iron inducible receptor ISIP1. ISIP1 protein models and secondary structure from T. oceanica, P. tricornutum and F. cylindrus are compared. Conservation between the protein orthologs is high, with identical secondary structure predictions (center). (more ...)
Accordingly, the ISIP1 protein would represent a cell-surface receptor, anchored to the plasma membrane by a carboxy-terminal transmembrane helix. A small carboxy-terminal tail without well-defined secondary structure contains a conserved endocytosis motif responsible for endocytotic cycling. An α-helical region amino-terminal from the transmembrane helix is predicted to be O-glycosylated and would thereby serve to expose the large β-propeller as a putative receptor domain to the extracellular space.
An alignment of the ISIP1 proteins from T. oceanica
, P. tricornutum
and F. cylindrus
illustrates that the extracellular β-propeller domain contains a cysteine-rich center (Figure ) whose pattern is reminiscent of cysteines found in Fe-S cluster proteins and might be involved in binding Fe. The cysteine-rich center is not found in the orthologous p130B of Dunaliella salina
, which is thought to have undergone an evolutionary change in function and interacts with transferrin-like proteins [37
A conserved promoter motif associated with diverse iron-regulated genes
At the core of an organism's low-iron response are transcription factors (repressors or enhancers) and their respective binding sites (specific promoter motifs) that mediate the cellular response at the gene expression level. From pairwise promoter comparisons between the exclusively iron-regulated ISIP1
genes of T. oceanica
, P. tricornutum
and F. cylindrus
using dotlet [38
] and MEME [39
], we identified a conserved palindromic motif 'ACACGTGC' located around position -200 from the translation start. Upon genome-wide screening, a total of 45 gene models contained the complete motif (perfect match) at a position of 150 to 250 bases before the translation start. Functional assignments for genes with positive matches were rarely possible (mostly hypothetical genes of unknown function and without significant regulation). However, the accumulation of low-iron responsive genes (ISIP1
and three FBA
genes) in this group is remarkable. In Figure S9 in Additional file 3
we present only those genes whose orthologs in other species carry the motif in their promoters.
The complexity of the identified motif (A2T1C3G2) is high; its palindromic structure suggests binding of a homo- or heterodimeric protein factor. The remarkable conservation of this motif and its position (-200) relative to the translation start across three diatom species reinforces the suggestion that this motif plays a prominent role in iron-dependent gene regulation.