A survey of HK, HPt, and RR domains and their organization in two-component systems and phosphorelay proteins of organisms with fully sequenced genomes

Identification of proteins containing TCS/PR domains

The fully annotated proteomes of 9,961 organisms were downloaded from NCBI’s genome database (January 2014 version). 2,352 of these proteomes were eliminated because they belonged to phages, virus, satellite DNA sequences, or organisms whose taxonomic classification was still not fully resolved. The remaining 7,609 proteomes belonging to 35 phyla from Bacteria, 6 phyla from Archaea and 11 eukaryotic phyla (Table S1) were further analyzed in search for proteins containing domains of types HK, RR and HPt. These domains are associated with IST in all TCS/PR cascades. Other protein domains (such as the CHEW adaptor domain or the P2 protein from NRI/NRII, among many others) were not included in the analysis because they are specific of certain TCS/PR cascades. The sensor domain of TCS/PR cascade proteins was also not included due to its sequence variability. Thus, the results from our study are general for all TCS/PR cascades.

We used PROSITE (http://prosite.expasy.org) to obtain a set of well curated sequences that can be used as a seed to identify TCS/PR proteins in the relevant proteomes. We downloaded a multiple alignment of all relevant ortholog sequences for each protein domain (HK—PS50109 PROSITE Domain, RR—PS50110 PROSITE Domain and HPt—PS50894 PROSITE Domain) from PROSITE. We then used these three multiple alignments as a set of query sequences for two independent searches. One was done using HMMER (Johnson, Eddy & Portugaly, 2010). For each multiple alignment downloaded from PROSITE, we built a profile HMM using hmmbuild, and performed the search of the profile HMM against all proteomes selected from the NCBI database using jackhmmer. The second search was done in parallel using PSI-BLAST (Altschul et al., 1990) and the three multiple alignments downloaded from PROSITE as a query. HMMER finds homologues that are more distantly related than those found by BLAST.

We simultaneously use BLAST and HMMer because they have different sensitivities in detecting sequence similarities. BLAST generates a higher number of false negatives, while HMMer generates a higher number of false positives. By using both and filtering the results, we hope to obtain a more precise picture of the conserved domains. In each search we queried the 7,609 proteomes in order to identify proteins with domains that are homologous to those used as queries.

In addition, the consensus sequence was calculated for each domain (HK, RR and HPt) independently. Using an in-house PERL script, the most common residue in each position was identified for each of the three multiple alignments. This residue was taken as the consensus value for that position in the corresponding protein domain. Subsequently the three consensus sequences were used to search each proteome using PSI-BLAST (Altschul et al., 1990). In all three searches, the hits selected were the ones with an e-value lower than 10⁻⁶ and with a domain coverage of at least 80%.

After performing these three searches, a PERL script was also used to perform a fourth text-mining search and identify the proteins that were annotated in each proteome as being histidine kinases, sensory kinases, hybrid kinases, response regulators or histidine phosphotransferases.

The results of the four searches were merged into a non-redundant set. A total amount of 469,421 proteins containing HK, HPt and/or RR domains were identified. This set was curated in the following way:

1. First we manually looked at the annotation of the proteins to identify functions that are not involved in TCS/PR cascades (e.g., serine kinase).

2. Then, we build a PERL script that automatically eliminates proteins annotated with those functions from the list.

3. We finish by automatically comparing the proteins in the list and the number of proteins containing terms related to TCS/PR cascades.

4. We repeat steps 1–3 until the number of proteins in the list and the number of proteins containing only terms related to TCS/PR cascades are the same.

In this way, we semi manually identified 36,169 proteins that were annotated as being something other than a TCS/PR protein. These proteins were eliminated. Frequent protein types found in the discarded set of proteins are serine/threonine kinases and several types of regulatory transcription factors.

The remaining 433,255 proteins were then reanalyzed and an additional set of 17,727 proteins were found to be annotated as being hypothetical or partial proteins. For each of the three domains, the set of 17,727 hypothetical and partial proteins were aligned using Clustal X in order to identify the conserved histidine motif in the HK and HPt domains, and the conserved aspartate residue in RR domains. Those sequences without a conserved histidine or aspartate residue were eliminated from the data, leaving a grand total of 415,525 annotated proteins and 17,724 partial/hypothetical proteins containing HK, RR and/or HPt domains.

A PERL script was developed to filter the curated data sets and determine both, the domain composition of each protein and, when they belonged to the same organism, the relative position of their corresponding genes with respect to each other in the genome.

Once we had identified all proteins containing HK, RR or HPt domains, and the relative genomic position of their corresponding genes, we looked for all type of operons of TCS/PR coding genes that occur in the organisms with fully sequenced genomes. For this purpose, we performed a search of all genes coding HK, RR or HPt protein domains that are located in consecutive positions on prokaryotic genomes. We assumed that they constitute a transcription unit, although this may introduce a small error, as consecutive operons coding for independent TCS/PR exceptionally exist. In our search, we allow the presence of a gap in the operon, that is, a gene which does not encode any HK, RR or HPt domain, because this could be a gene with regulatory functions in the operon.

The statistical treatment of data was carried out independently with and without taking into account the hypothetical and partial proteins found. Both results are qualitatively the same. In the Results section of this paper we give the results from the analysis of the set of proteins without the hypothetical and partial proteins. The results of the whole set of proteins, including hypothetical and partial proteins, are given as Supplemental Information. The sequence files for all domains are also available at http://web.udl.es/usuaris/pg193845/Salvadoretal.html.

Numerical and statistical data analysis

To estimate how the clustering of the various TCS and PR proteins in a genome differed from what one would expect by chance in the context of that genome, we took the following approach. First, we calculated how frequently one would expect proteins containing TCS/PR domains to be coded by neighboring genes in a genome if the order of genes was fully random, given the total number of proteins in that genome, and the number of proteins involved in TCS/PR cascades. The expected neighboring frequencies under this assumption can be computed by Eqs. (1)–(6). In these F(P1↔P2) represents the expected frequency of the neighboring events in a genome for genes coding proteins of types P1 and P2, n_RR represents the number of proteins containing one RR domain in the proteome, n_HK represents the number of proteins containing one HK domain in the proteome, and P represents the total number of proteins annotated to the proteome. (1)${\displaystyle F\left(\mathrm{HK}\leftrightarrow \mathrm{RR}\right)=\frac{n_{\mathrm{RR}}}{P-1}+\frac{n_{\mathrm{RR}}}{P-2}.}$ Equation (1) represents the probability that a gene localized in position j of the genome is located next to a gene coding for a protein that contains an RR domain, either in positions j − 1 or j + 1, if gene order is random in a genome. The first term of the sum represents the probability of the presence of an RR gene in one of the two possible locations irrespective of its presence also in the other genome location, and the second term is the probability of the presence of the RR gene in one of the two genome locations when it is not found in the other one. We note that we are not calculating the probability of having a consecutive gene pair containing HK and RR domains. Rather, for any genomic position j, we ask what the probability of its neighboring a gene containing an RR domain is. Equation (1) gives a good estimation of this random probability, given that the total number of protein coding genes is tens to hundreds of times larger than the number of IST domain coding genes, and assuming that position j represents neither the first nor the last genomic position. This expected RR neighboring frequency will be compared with the actual fraction of HK genes that are found next to RR genes in order to study their genomic distribution. (2)${\displaystyle F\left(\mathrm{HK}\leftrightarrow \mathrm{RR}\leftrightarrow {\mathrm{HK}}_2\right)=6\times \frac{n_{\mathrm{HK}}}{P-1}\times \frac{n_{\mathrm{RR}}}{P-2}.}$ Equation (2) computes the probability of finding an RR gene and a second HK gene in the genomic neighborhood of a given HK gene. Because these three consecutive genes can be sorted in 6 different ways, we must multiply by 6 the probability of an individual neighboring event. Again, note that we assume having an HK domain containing gene, and ask what the probability of its neighboring genes containing additional HK and RR domains is. (3)${\displaystyle F\left(\mathrm{HK}\leftrightarrow \mathrm{RR}\leftrightarrow {\mathrm{HK}}_2\leftrightarrow {\mathrm{RR}}_2\right)=12\times \frac{n_{\mathrm{RR}}}{P-1}\times \frac{n_{\mathrm{HK}}-1}{P-2}\times \frac{n_{\mathrm{RR}}-1}{P-3}.}$ Similarly, in Eq. (3) we compute the probability that, considering that we have found a gene containing an HK domain in a given place in the genome, we also find in consecutive genomic positions around that HK gene location another HK gene and two RR genes, if gene organization is random. These four genes can be sorted in 24 different ways, but we don’t differentiate between the two RR genes and therefore there are only 12 possible spatial arrangements of these series of four genes. (4)${\displaystyle F\left(\mathrm{HKRR}\leftrightarrow \mathrm{HK}\leftrightarrow \mathrm{RR}\right)=6\times \frac{n_{\mathrm{HK}}}{P-1}\times \frac{n_{\mathrm{RR}}}{P-2}.}$ In Eq. (4), the probability of the event is computed in exactly the same way as in Eq. (2). (5)${\displaystyle F\left(\mathrm{HKRRHPt}\leftrightarrow \mathrm{RR}\right)=\frac{n_{\mathrm{RR}}}{P-1}+\frac{n_{\mathrm{RR}}}{P-2}}$ (6)${\displaystyle F\left(\mathrm{HKRRHK}\leftrightarrow \mathrm{RR}\right)=\frac{n_{\mathrm{RR}}}{P-1}+\frac{n_{\mathrm{RR}}}{P-2}.}$ Equations (5) and (6) compute the probability of finding an RR gene placed in the genome next to an HKRRHPt or an HKRRHK gene respectively, exactly in the same way as described above for Eq. (1).

Once these expected frequencies were computed using Eqs. (1)–(6), we calculated the odds ratios of the observed neighboring events with respect to the expected neighboring event. All numerical and statistical calculations were done using Mathematica (Wolfram, 1999).

Statistical models

To analyze the relationship between the number of TCS/PR gene fusion events and the proteome size, we built a linear model that would better fit our data for % of fused HK (RR, HPt) domains vs. total number of HK (respectively, RR, HPt). We also built linear models of total number of IST domains in an organism vs. total number of proteins in the proteome and phylogeny (prokaryote, eukaryote). In other words, we fit the data to Eq. (7): (7)${\displaystyle \text{Number of IST domains}={\alpha }_1\left(\text{Total number of proteins in proteome}\right)+{\alpha }_2\text{ Phylogeny}+\epsilon .}$ In Eq. (7), the variable phylogeny can assume two values. If the organism is a prokaryote, the variable has value 1; otherwise it has value 2. An ANOVA analysis was used to determine whether the coefficients for each control variable of the linear model are significantly different from zero. If so, this implies that the variable is relevant in explaining the variation observed in the dependent variable.

When fitting the data to the linear models we also calculated the R² and adjusted R² of the models. R² shows how well terms (data points) fit a curve or line; adjusted R² also indicates how well terms fit a curve or line, but adjusts for the number of terms in a model.

Survey of proteomes containing proteins with domains involved in internal signal transduction (IST) in TCS/PR cascades

Percentage of proteins with HK, RR or HPt domains in the surveyed proteomes

For simplicity, hereafter we shall refer to proteins containing IST domains typical from TCS/PR cascades as TCS/PR proteins. On average, between 1 and 2% of a prokaryotic proteome is composed of TCS/PR proteins (mean = 1.37%). In contrast, when an eukaryotic proteome contains TCS/PR proteins, they account for between 0.05% and 0.2% of the entire proteome (mean = 0.11%). In bacteria, Deltaproteobacteria is the group with the highest average percentage of TCS/PR proteins (Fig. 3). In contrast Tenericutes and Chlamydiae almost tie with the lowest average percentage of TCS/PR proteins (Fig. 3).

It has been observed in previous analyses that the number of proteins containing IST domains associated with TCS/PR cascades increases almost quadratically with the number of total proteins in a proteome (Ulrich, Koonin & Zhulin, 2005; Galperin, Higdon & Kolker, 2010). We further wanted to assess if this dependency is significantly different between eukaryotes and prokaryotes. To do so we fit the data to the linear model described by Eq. (7). An ANOVA analysis shows that phylogeny is important in explaining the variation in total number of IST domains found in a proteome (p < 10⁻²⁵). Because of this we divided the dataset in prokaryotes and eukaryotes, and fit each dataset to the linear model shown in Fig. 4. We find that the fraction of variability in number of IST domains explained by proteome size in eukaryotes doubles that of prokaryotes. This suggests that the number of IST domains could evolve differently in prokaryotic and eukaryotic organisms.

Survey of TCS/PR protein types

We find fifty unique types of TCS/PR proteins, when it comes to IST domain organization within a single polypeptide chain. These unique types of TCS/PR proteins are shown in Table 2, sorted by abundance. In that table, the protein identifier describes the type of IST domain (HK, HPt, or RR) and the number describes how many domains of a given IST type are found in each protein. Hereafter we shall refer to proteins containing only one HK IST domain as HK protein type, proteins containing one HK domain and one RR domain as HKRR protein type, and so on and so forth.

Overall, all phyla where IST domains associated with TCS/PR cascades were identified have RR and HK protein types, with the exception of Monocots, which lack RR domains. HKRR protein type (also known as hybrid HK) is present in all phyla where TCS/PR proteins were identified, except in Aquificae, Tenericutes, Chlamydiae, and Crenarchaeota (Table S2). Together, HK, RR, and HKRR proteins represent 94% of all TCS/PR proteins that were identified.

In prokaryotes, RR or HK protein types are the most abundant. Together, they represent more than 90% of all TCS/PR proteins found in the genomes of many organisms (Table S2). HKRR represent the third most abundant type of TCS/PR protein, oscillating between less than 1% (Firmicutes) and more than 10% (Cyanobacteria) of all TCS/PR proteins (Table S2). The remaining protein types (HPt, HKRRHPt, HK₁RRHK₂, HKRR₁HPtRR₂, HK₁RR₁HK₂RR₂, …) range from less than 1% to 5% of all TCS/PR proteins identified in a phylum. Of these less abundant protein types, the three-domain HKRRHPt protein is more abundant than HK₁RRHK₂. The HPt domain is more frequently found in combination with other IST TCS/PR protein domains than alone in a protein, with the exception of Firmicutes, Tenericutes, Actinobacteria, Bacteroidetes and Spirochaetes. We also observe that HKRR₁HPtRR₂ is more abundant than HK₁RR₁HK₂RR₂ (Table S2).

The relative abundances of proteins containing IST domains associated with TCS/PR cascades in eukaryotes are different from those of prokaryotes. In broad terms, HK and RR protein types tend to make for a smaller fraction of TCS/PR proteins in eukaryotes than in prokaryotes, while the opposite is observed for HKRR proteins. Another clear distinction between prokaryotes and eukaryotes refers to HPt-containing proteins: HPt protein type represents more than 10% of all TCS/PR proteins in eukaryotes. In prokaryotes, except in Tenericutes, HPt proteins typically account for less than 1% of TCS/PR proteins. Moreover, no HKRRHPt or HKRR₁HPtRR₂ protein types were found in eukaryotes (Table S2).

Among protists, Euglenozoa proteomes contain mostly HK protein type, although HKRR type is the most abundant in D. discoideum (Amoeboflagellate). There are cases of inactive HK domains that have lost their histidine. When identified, these proteins were eliminated from the analysis as described in Methods. However, there is always the possibility that some such proteins have passed our filters. To control for that possibility we created a multiple alignment of the Euglenozoa HK proteins. We found that the HK domains contained the conserved histidine motif that is needed for HK signal transduction. Hence, these proteins could be active HK proteins. Furthermore, if we lower our e-value for cut-off to 10⁻⁴, many of these proteins will also be flagged as containing RR domains with conserved aspartate residues, suggesting that such proteins could be HKRR types with a high degree of sequence divergence from other HKRR proteins we identified. Thus, the HK proteins in this clade might either be hybrid HKs or be active in a context that does not involve a TCS/PR cascade. TCS/PR proteins are almost absent in Alveolates. In the fungi phyla (Table 1), HKRR is the most abundant protein type in Basidiomycetes, making up for almost 50% of total TCS/PR proteins. In contrast, RR, HK and HPt protein types are relatively more abundant than HKRR protein type in Ascomycetes. A remarkable result in fungi is the relative abundance of HK₁RRHK₂ and HK₁RR₁HK₂RR₂, which are much more frequent in eukaryotes (above 10%) than in prokaryotes. In plants, RR is the most abundant protein type in Eudicots, making for about 60% of all TCS/PR proteins.

Distribution of genes coding for TCS/PR protein types in the genomes

Previous surveys found that many of the TCS/PR proteins are mostly organized in operons and/or regulons in prokaryotes (Alm, Huang & Arkin, 2006; Williams & Whitworth, 2010; Galperin, 2010; Galperin, Higdon & Kolker, 2010). Consistent with this, we find that between 60% and 90% of genes containing HK domains are neighbors to genes containing RR domains. Exact percentages depend on the phylum, but below 20% of the total prokaryotic HK coding genes are orphan, that is, they are not neighboring any other gene coding for a protein that contains at least one IST domain. We also have found some clusters of genes coding HK, RR or HPt domains in eukaryotes, but all of them are a succession of genes with identical domain composition. Although the existence of operons has been reported in the eukaryote C. elegans (Blumenthal, Davis & Garrido-Lecca, 2015), the gene clusters identified in our search have independent promoters.

Altogether, we found 530 different types of gene clusters coding for TCS/PR proteins. We now briefly describe these results, shown in Table S9.

Neighborhood analysis for HK and RR protein types

In most prokaryotes neighboring genes coding for HK and RR protein types are between 50 and 100 times more frequent than one might expect by chance alone. In some species, this frequency is even higher (Fig. S1 and Table S3). Several phyla have a small percentage of species containing only orphan HK and RR protein types in their genomes (20 out of 2066 species in Firmicutes, 2 out of 635 in Actinobacteria, 6 out of 235 in Bacteroidetes, 11 out of 2,246 in Gammaproteobacteria, 48 out of 451 in Alphaproteobacteria, 7 out of 366 in Betaproteobacteria, 4 out of 108 in Chlamydiae, 3 out of 118 in Cyanobacteria and 9 out of 179 in Euryarchaeota). Most of these species have a number of TCS/PR proteins below the average of their phylum.

Gene fusion of TCS/PR proteins

Scope, caveats, and limitations of our analysis

In this work we analyze the distribution and prevalence of different types of TCS/PR proteins in 7609 organisms belonging to 52 phyla. These proteins are responsible for sensing and adequately regulating the cellular responses to environmental cues. To date, this is the largest survey of TCS/PR proteins we are aware of. We confirm that these proteins are predominantly prokaryotic, although they are also present in many eukaryotic phyla. However, functional TCS/PR cascades appear to be absent in animals. This is also consistent with previous findings (Attwood, 2013).

An important feature in this study is that we include all organisms with fully sequenced and annotated genomes in our analysis. For example, on the order of one thousand Escherichia coli strains are included in our analysis. This would clearly bias any deletion/duplication or horizontal gene transfer study of TCS/PR proteins that one might make in the full dataset. However, considering all strains and subspecies in our analysis is fundamental for identifying extremely low-frequency unique IST domain and operon organization types.

Identifying unique types of IST domain organization in TCS/PR cascades

The main goal of this analysis is to identify the unique types of organization for IST domains in proteins of TCS/PR cascades. In addition we also perform a less thorough identification of operon organization for TCS/PR proteins. This study was independently made in two ways: first, we eliminate all proteins annotated as hypothetical or partial. Subsequently we include such proteins in the analysis. The results for the analysis that include the hypothetical and partial proteins can be found as a ZIP supplementary file (Appendix S1). Results are qualitatively similar in both cases, and the raw sequences in FASTA format can be downloaded from http://web.udl.es/usuaris/pg193845/Salvadoretal.html.

Our analysis identifies 50 unique types of TCS/PR proteins, when it comes to intra protein IST domain organization. The most frequent types of proteins with fused IST domains are the hybrid histidine kinases, a design with one HK and one RR protein domains fused in a single protein. This organization has been observed in most of the eukaryotic PRs that have been well characterized genetically and biochemically (for example the Sln1p-Ypd1p-Ssk1p pathway in S. cerevisiae (Maeda, Wurgler-Murphy & Saito, 1994) or the ETR1 system in A. thaliana (Chang et al., 1993)). It is also present in some prokaryotic systems (for example, the RcsC/YojN/RcsB pathway, involved in the regulation of capsular polysaccharide synthesis in E. coli (Takeda et al., 2001), and the Lux pathway regulating bioluminescence in V. harveyi (Freeman & Bassler, 1999)). Another relatively frequent type of IST domain organization is when one HK, one HPt, and one RR domain are found within a single protein. Such proteins are called unorthodox histidine kinase or tripartite HK. Some examples of systems with this design are: BvgS-BvgA (Uhl & Miller, 1996), EvgS/EvgA (Bock & Gross, 2002), ArcB/ArcA (Georgellis, Lynch & Lin, 1997), TorS/TorR (Bordi et al., 2003), BarA/UvrY (Sahu et al., 2003), TodS/TodT (Silva-Jiménez, Ramos & Krell, 2012) and GacS/GacA (Sahu et al., 2003).

We also identify 530 unique types of possible operons in prokaryotes and some eukaryotes, such as ascomycetes and eudicots (Table S9). This variety will be used in subsequent works to infer naturally occurring variations in the pattern of regulatory interactions between the proteins involved in TCS/PR networks. For example, if we find a gene cluster formed by one HK and two RR coding genes, we can infer that the signaling pathway has a branching point in which the HK phosphorylates both RR. This alternative circuitry is important because it has been proved that network architecture affects network dynamics and can define the operational limits of the system in a way that is independent of the specific biological processes being regulated (Alves & Savageau, 2003; Igoshin, Alves & Savageau, 2008; Cağatay et al., 2009; Tiwari et al., 2010; Salvado et al., 2012).

We have no way of identifying TCS/PR cascades at the regulon level using only sequence data. Many examples for this type of organization exist, such as the Kin-SpoO pathway (Burbulys, Trach & Hoch, 1991).

Why do we focus only on the IST domains of TCS/PR cascades, rather than also including also other protein domain that are involved in TCS/PR signal transduction? By focusing on these domains and their organization, our results set the stage for an analysis of general dynamics organization principles in the internal transmission of signals within TCS/PR cascades. The organization of IST domains, either within a protein or within an operon, plays an important role in determining the dynamics of the signal transmission in a cascade (Alves & Savageau, 2003; Igoshin, Alves & Savageau, 2008; Ray & Igoshin, 2010; Narula et al., 2012). Hence, that organization is likely to be the subject of natural selection. Had we included other types of domains, we would be also analyzing aspects of the input and output of the cascades that are case specific and not general to all cascades of a given type.

Some physiological, phylogenetic and evolutionary considerations

In prokaryotes, approximately 90% of all PR proteins have only one HK domain or one RR domain (Table 2 and Table S2), and most of the genes encoding these proteins are located in the chromosome next to other PR/TCS genes, forming operons. In contrast to this, in eukaryotes proteins of types HK and RR are less common, and genes encoding these proteins are never located next to other TCS/PR genes in the species surveyed. On the other hand, in eukaryotes there is a higher fraction of TCS/PR proteins containing a combination of the HK and RR domains (the HPt domain was not found in these eukaryotic multi domain TCS/PR proteins), such as HKRR, HK₁RRHK₂ and HK₁RR₁HK₂RR₂. This implies that TCS/PR signal transduction in eukaryotes is in principle less prone to cross-talk and noise, as the signal is internally transmitted within the same peptide chain (Tiwari et al., 2010; Tiwari & Igoshin, 2012).

Our analysis confirms the ad hoc observation that coordinated expression of IST domains and/or TCS/PR proteins involved in the same cascade is frequent. We also quantify how much more frequent this coordinated expression is with respect to what one would expect by chance alone. Although this is not unexpected (Price, Arkin & Alm, 2006; Alm, Huang & Arkin, 2006; Ray & Igoshin, 2012; Tiwari & Igoshin, 2012), to our knowledge, such quantification had not been done before on such a large dataset.

This suggests that alternative IST regimes might be favored by evolution in prokaryotic or eukaryotic TCS/PR cascades. This can be inferred from the fact that the three types of gene expression coordination (regulon, operon, or gene fusion) imply different characteristics when it comes to internal signal transmission within the cascade. In general, fused genes will have a lower level of noise in signal transduction, followed by genes coded in the same operon, and with genes coded in the same regulon permitting the highest level of noise to enter the signal transduction process (Ray & Igoshin, 2012).

Why is this so? TCS/PR proteins whose expression is coordinated either at the regulon or operon levels are potentially translated in different amounts. RR proteins typically are orders of magnitude more abundant than HK proteins (Igoshin, Alves & Savageau, 2008). This leads to a type of signal transduction where amplification of the signal can be high, as many RR molecules can be modified by a single HK protein. In contrast, in hybrid kinases where the HK and RR domains are fused in the same protein, the ratio of HK/RR domains is one to one. This means that each HK domain will likely only phosphorylate one RR domain. Moreover, independent HK protein types might also be leakier, phosphorylating non-cognate RRs. Similarly, independent RR protein types can be more prone to phosphorylation by non-cognate sources. Such non-cognate phosphorylation events are physically harder to achieve in HKRR protein types. Thus, proteins with fused TCS/PR domains represent a design that will on average transduce signals with smaller amplification, but higher fidelity than TCS/PR cascades composed only of proteins with individual TCS/PR domains.

Taking these considerations into account, one might think that maximization of internal signal amplification is likely to be an important selective pressure for the evolution of TCS/PR cascades in prokaryotes, while fidelity of internal signal transmission appears to be a more important selective pressure for the evolution of TCS/PR cascades in eukaryotes. These two functional requirements for IST in TCS/PR cascades are generic and independent of more specific pressures, such as the type of signal they transduce, whether the organism is uni- or multi-cellular, or other similar considerations (Alm, Huang & Arkin, 2006; Laub & Goulian, 2007; Williams & Whitworth, 2010; Galperin, Higdon & Kolker, 2010; Capra & Laub, 2012; Podgornaia & Laub, 2013). If and why amplification and fidelity of internal signal transmission are indeed shaping the general organization of TCS/PR cascades is a matter to be investigated further in the future. This will be done in a forthcoming study by creating mathematical models for the TCS/PR cascade architectures identified in this study and comparing the dynamic behavior of each of the alternatives.