BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

Positive and Negative Cognate Amino Acid Bias Affects

Compositions of Aminoacyl-tRNA Synthetases and Reflects

Functional Constraints on Protein Structure

Hervé Seligmann

1The National Collections of Natural History, The Hebrew University of Jerusalem, 91404, Israel.

herves@bio.uio.no; podarcissicula@gmail.com: esTel and Fax: +972 2 6585876. Email address

doi:10.5618/bio.2012.v2.n1.2 || Received: 30-11-2011, Accepted: 24-02-2012, Available online: 08-03-2012

Abstract

By comparing phylogenetically related tRNA syn-

thetases (enzymes that specifically aminoacylate

tRNAs), a controlled natural experiment can reveal

synthetases’ coevolution with their cognate amino

acid substrate. Analyses of metabolic cost minimi-

zation confirm the existence of cognate avoidance

in tRNA synthetase compositions. It is found that

cognate avoidance increases and decreases,

respectively, with proteomic amino acid usage in

Escherichia coli and Bacillus subtilis.

In E. coli, cognate avoidance did not decrease

with cognate abundance in the colon, but decrease

with tRNA synthetase editing sites and cognate

impact on protein structure, revealing that

hydrophobic interactions and beta-sheet formation

constrain the folding of tRNA synthetase classes I

and II, respectively. Analyses of cognate bias yield

information on how proteins function in E. coli,

because function constrains cognate bias. In B.

subtilis, cognate avoidance occurred for rare and

abundant amino acids in the soil, positive bias

existed for cognates with intermediate abundances.

Presumably, life history strategies (endosymbiont

versus free living) and environmental compositions

modulate cost minimization of amino acid usages.

Avoidance of ‘expensive’ residues in tRNA

synthetases is inversely proportional to cognate

avoidance and protein size in E. coli, but not B.

subtilis. In relation to cost minimization, E. coli’s

predictable environment perhaps enabled to reach

evolutionary (balancing) equilibrium between

various factors affecting protein synthesis costs,

where decreasing costs through one factor requires

increasing other costs. Phylogenetically controlled

comparisons can detect more statistically

significant biases than similar analyses using

randomly selected control proteins, stressing the

power of carefully designed natural experiments.

This work confirms the importance of biosynthetic

cost minimization, and challenges neutralistic

approaches of biomolecular evolution.

Keywords: aminoacylation; phylogenetic contrast;

homology; catalysis; neutral evolution; tRNA

synthetase editing; secondary structure; antiparallel

betasheet.

1. Introduction

The molecular weight of amino acids associates

negatively with their abundances in protein

compositions [1]. This observation led to the

hypothesis that amino acid compositions of proteins

are in part determined by cost minimization principles

[2]. The aim of this study is to examine whether cost

minimization of protein synthesis is integrated with

functional constraints on proteins, and thus revealing

new information on protein structure and function.

The hypothesis of cost minimization of protein

synthesis resembles the hypothesis of synonymous

codon usage optimization, which minimizes transla-

tion delays by using synonymous codons that are

matched by the anticodons of the most common tRNA

for that cognate amino acid [3-5]. Similarly to

synonymous codon optimization, the extent of cost

minimization of amino acid usage increases with

protein expression levels [6-10]. Other observations

also follow predictions from the cost minimization

hypothesis. For instance, whole organism growth and

developmental rates increase with cost minimization

at molecular levels [7, 11]. Free-living organisms

minimize more costs of amino acid usages than

parasitic ones [7, 12], while some parasitic organisms

also minimize costs of amino acid usages [13].

Reduction in protein length, as a means to decrease

their biosynthetic costs, seems to follow similar

principles, at least in comparisons between free-living

and endocellular species [14]. Further observations

that protein size is inversely proportional to its

expression level are also in line with that principle

[15-16]. Cost minimization principles also explain

differences in amino acid compositions between

proteins expressed under different endogenic [17-18]

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

as well as environmental conditions [19]. Metabolic

costs also increase evolutionary conservation of amino

acids [7, 20]. Following the same principle at atomic

level, a cyanobacterium specifically expresses sulfur-

depleted versions of some highly expressed proteins

under sulfur-starvation [21]. This principle of

minimizing metabolic costs at the level of atomic

compositions also exists for nitrogen- and carbon-

assimilatory enzymes that are relatively depleted in

nitrogen and carbon, respectively [22-26]. This

introduces an additional prediction to the metabolic

cost avoidance hypothesis, that of substrate avoidance

in compositions of enzymes. This prediction was

confirmed specifically in the context of amino acids in

three species: Escherichia coli, its close relative which

shares its endosymbiontic habitat in the human colon,

Salmonella typhimurium and the more distantly

related Bacillus subtilis that inhabits the soil [27]. This

principle also affects an amino acid's secondary

metabolism [11]. In yeast, levels of cognate avoidance

in amino acid compositions of enzymes involved in an

amino acid's biosynthetic path can even predict a

following decrease in experimental growth after

depletion in that amino acid [11].

1.1. tRNA synthetases and cognate avoidance. In

this work, I explored the issue of cognate amino acid

bias in the realm of tRNA synthetases, a group of

enzymes that esterify amino acids to their cognate

tRNAs. The loading of each of the 20 tRNA

functional groups by its cognate amino acid is done by

a specific tRNA synthetase. This functional group of

20 enzymes has two further characteristics that

constitute an actual controlled natural experiment in

which cognate bias can be investigated. It is

subdivided into two equal sized classes of proteins,

class I and class II tRNA synthetases. Note that it is

possible that class I and II tRNA synthetases are

matched in terms of each being originally coded on

the antisense (complementary) strand of the other [28-

33]. This is a controversial hypothesis (see digression

by Williams et al. [34]) that would result from cost

minimization of genome lengths in the ancestral

genome(s) [35]. Minimizing genome lengths is also

frequently redeveloped in modern parasitic organisms

[36-39]. Independently of these, all enzymes in class I

and class II tRNA synthetases are evolved from a

common ancestor [40]. Hence comparisons between

phylogenetically matched sister tRNA synthetases

with different cognate amino acids can be used to

quantify extents of cognate amino acid avoidance in

their respective compositions.

In order to compare the relative power of

compositional tests on proteins based on evolutionary

homology to a randomly chosen control group as done

in [27], I tested cognate bias at the compositional level

of tRNA synthetases in the organisms examined by

Alves and Savageau [27]. These organisms are also

particularly adequate for a test of the „cognate amino

acid bias‟ because of the availability of the relative

abundances of amino acids in their respective natural

environments. This can enable tests on whether

cognate bias is regulated by environmental factors.

1.2. Substrate avoidance hypothesis. Increased

charging of some tRNAs, as a result of negative

feedbacks in metabolism, can be presumed to occur

specifically under conditions where the loaded tRNA

is relatively rare [41]. This means that tRNA

synthetase synthesis systematically occurs under

circumstances of lack of the reaction product it

catalyzes, i.e., the tRNA loaded with its cognate. This

rationale suggests that the composition of tRNA

synthetases includes their respective cognate amino

acid only at locations where that amino acid is

functionally important, or even unavoidable. By

comparing amino acid compositions of tRNA

synthetases that are „phylogenetic‟ sisters [40], I

expected that the cognate amino acid would be less

frequent in the composition of its esterifying tRNA

synthetase than in the composition of its phylo-

genetically closest tRNA synthetase, which esterifies a

different amino acid. Exploring differences in

cognate bias between tRNA synthetases with different

cognates but from the same tRNA synthetase class

could also reveal which properties of the cognate

make the cognate amino acid unavoidable for protein

function. For example, one could expect that amino

acids with great impact on protein structure

(quantified for example by their Chou-Fasman

conformational indices) can less be avoided (weaker

negative bias) than those with lower impact. Because

class I and II tRNA synthetases belong to different

types of proteins, it is possible that different cognate

properties associate with cognate bias in each group,

reflecting the functional importance of each property

on the different protein types. Such analyses of

differences in cognate bias would, beyond confirming

the existence of cognate bias, have the potential to

reveal functional properties of amino acids and

proteins through the study of cognate bias in particular

and cost minimization principles in general.

1.3. Why tRNA synthetases. In addition to the

adequacy of tRNA synthetases for non-manipulative

natural experiments, there are other reasons regarding

processes that influence the evolution of these

proteins. Indeed, the various tRNA synthetases did

not only evolve from a common ancestor, but

surprisingly other proteins apparently evolved from

tRNA synthetases. For example, the vertebrate

mitochondrial gamma polymerase, the sole replication

polymerase for animal mitochondrial genomes [42],

originates from horizontal transfer of prokaryotic

glycyl tRNA synthetase [43], which is also predicted

to interact with DNA templating for tRNAs [44]. The

evolutionary versatility of tRNA synthetases [45] and

their functionally crucial sites build a bridge between

the RNA and the protein „worlds‟: they have sites that

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

specifically recognize RNA regions, and the

properties of these sites are determined by those of

RNA, which affect any protein that interacts with

those RNAs [to some extent, also DNA stretches (as

suggested by the homology with DNA gamma

polymerase reported above)]; they have a site that

interacts with their specific cognate amino acid,

whose properties are determined by those of that

amino acid. This could bear insights on how proteins

interact with other proteins.

1.4. Comparative experiments: the power of

adequate control. In principle, the cognate amino

acid bias hypothesis was confirmed in [27], but this

was by comparing the amino acid contents of specific

proteins with a pool of proteins at large in that

organism. The results presented below on tRNA

synthetases are based on better controlled

comparisons (between pairs of phylogenetically

“sister” tRNA synthetases). These analyzes yield

results that are far more statistically significant than

those obtained in [27], thereby stressing the power of

the tRNA synthetase system for testing protein

evolution hypotheses. The study shows that careful

comparative evolutionary analysis is a powerful tool

for scientific research. In addition, the analysis can

reveal functional properties that modulate cognate

bias in relation to the constraints of each tRNA

synthetase class.

2. Materials and Methods

2.1. Overall compositional bias. I counted amino

acids in each of the tRNA synthetases of Escherichia

coli and Bacteria subtilis and calculated the

percentage that each amino acid represents in the

composition of each tRNA synthetase. Using the

phylogenetic relationships among tRNA synthetases

of class I and class II described in Figure 9 from [40],

I matched each tRNA synthetase with its

phylogenetically closest tRNA synthetase.

Considering the composition percentage of the

cognate amino acid for each tRNA synthetase, I

subtracted it from the composition percentage of that

amino acid in the phylogenetically matched tRNA

synthetase. Negative values reflect the level of

avoidance of the cognate amino acid in the tRNA

synthetase‟s primary structure. In some cases, the

phylogenetic topology indicated that a pair of closely

related tRNA synthetases form a group that is the

sister group of another tRNA synthetase. For

example, threonyl synthetase and prolyl synthetase

form a sister group for serinyl tRNA synthetase. In

such cases, I averaged the % that serine represents in

the compositions of threonyl and prolyl synthetases,

and subtracted this average from the % serine in

serinyl synthetase.

2.2. Testing the hypothesis at three different levels.

I tested the statistical significances of specific cognate

biases at three different levels using sign tests. In the

sign tests, cognate bias fits the „cognate avoidance‟

hypothesis if the index is negative. a) across tRNA

synthetases and cognate amino acids, I checked the

number of times that the cognate amino acid bias is

negative; b) within single tRNA synthetases, across

amino acids, I checked the number of times that the

subtraction of the % composition of all 19 non-

cognate amino acids in the sister tRNA synthetase

from the % in that specific tRNA synthetase was less

negative than the subtraction for the cognate amino

acid; c) across all tRNA synthetases, but for the same

amino acid, I checked the number of times that the

subtraction for a specific amino acid in the other

tRNA synthetases was less negative than for the tRNA

synthetase for which this amino acid is the cognate

amino acid. For the data used for tests (b) and (c), I

also calculated the Z standardized cognate bias by

dividing it by the standard deviation of all biases.

3. Results

3.1. General overview. Table 1 presents the „sister‟

tRNA pairs used to calculate cognate bias, and the

cognate bias calculated for each tRNA synthetase of

Escherichia coli, as well as similar reductive statistics

for Bacillus subtilis. The index is negative in the

majority of cases (75% in E. coli, 60% in B. subtilis, P

= 0.01 and P = 0.126, respectively, one sided sign

tests), indicating an overall avoidance of substrates in

the constitution of their catalysts. This result, which

follows the approach of test level (a) as indicated in

„Materials and Methods‟, was significant by sign tests

at P < 0.05 in E. coli for class I synthetases (90%, P =

0.005, one sided sign test), but not significant for class

II synthetases for B. subtilis (70%, P = 0.086, one

sided sign test). This tendency is confirmed, although

not statistically significantly, in class II synthetases of

E. coli (60%, P = 0.189, one sided sign test) and class

I synthetases of B. subtilis (55%, P = 0.25, one sided

sign test). Fisher‟s method for combining P values

uses the -2*sum of ln (P) as a chi square statistic with

2*k degrees of freedom (where k is the number of

independent tests being combined, two in this case,

for each tRNA synthetase class). Using this method,

the combined P‟s are P = 0.0075 for E. coli and P =

0.104 for B. subtilis. These results confirm in a more

rigorously controlled manner (because amino acid

compositions are compared between „matched‟

proteins, and not between specific proteins and all

proteins at large) the principle that substrates are

avoided in the composition of their catalysts.

Cognate bias in E. coli and B. subtilis are positively

correlated for class I tRNA synthetases (r = 0.63, P =

0.034, one tailed test). This result is qualitatively

similar for class II synthetases (r = 0.16, P = 0.33) and

for all synthetases (r = 0.29, P = 0.116, one tailed

test), though is not statistically significant. Hence, a

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

wide variation exists in the extents and directions of

the amino acid cognate bias index, between bacteria,

synthetase classes, and specific synthetases within the

same species and class. These results suggest that

additional phenomena affect cognate bias.

3.2. Tests for E. coli

3.2.1. E. coli: Cognate bias tests within single tRNA

synthetases. For tests within single tRNA synthetases,

but across amino acids (rows in Table 1, test level (b)

in „Materials and Methods‟), the bias was more

frequently more negative for the cognate amino acid

than the biases calculated for non-cognate amino acids

in a non-significant majority of the tRNA synthetases

(65%). This was significant at P <0.05 according to a

sign test within single tRNA synthetases for 11 cases

(7 cases in class I: Gln, Ile and Met, P = 0.000122 for

each; Leu, P = 0.00065; Trp, P = 0.0087; Tyr and Val,

P = 0.00002; 4 cases in class II: Gly, large subunit, P

= 0.000001; His, P = 0.009; Phe, large subunit, P =

0.00002; and Ser, P = 0.00012; all one sided sign

tests). There were also two tRNA synthetases with

significant positive cognate bias, those with cognates

Phe (a different subunit of the tRNA synthetase than

in the previous group) and Arg (P = 0.004 and P =

0.0013, respectively, two-sided sign tests were used

here because the null hypothesis does not expect

positive cognate bias). Note that Phe is one of the

largest amino acids. Because at P < 0.05, when 23

tests are done, one can expect 0.05*23= 1.15 false

positive results, meaning one expects on average one

false detection of a bias. Because there are 13

significant cases in total, one can be sure that overall

the phenomenon of cognate bias exists, but one cannot

be certain that bias exists in any tRNA synthetase. An

extreme method to handle this problem is

Bonferroni‟s adjustment [46]. It divides 0.05 by the

number of tests done. P values that meet that criterion

(P = 0.05/23= 0.00214) are significant at P < 0.05

considering the number of tests done. P values in bold

remain significant according to this overconservative

criterion (in total 11 among 13 cases). According to a

slightly less conservative adjustment, called the

Benjamini-Hochberg adjustment [47] of the

Bonferroni method, all 13 cases remained significant.

The working hypothesis can also be tested by

combining the P values across the various tRNA

synthetases, using Fisher‟s method for combining P

values. This yields chi-square statistics of 129.64 and

103.38 with 10 and 13 degrees of freedom, and P =

4.6*10-18

and P = 3.5*10-11

for class I and II tRNA

synthetases, respectively. The working hypothesis is

hence confirmed by very significant results for each

tRNA synthetase class.

3.2.2. E. coli: Cognate bias tests across tRNA

synthetases. For tests keeping the amino acid

constant, but across tRNA synthetases (columns in

Table 1, test level (c) in „Materials and Methods‟),

two tests can be done. One considers only

comparisons within the class of the specific tRNA

synthetase, and the other considers all tRNA

synthetases. Keeping the class constant, the bias was

more frequently negative for the cognate of the tRNA

synthetase which specifically recognizes this amino

acid than for other tRNA synthetases. The bias is

65% (P = 0.065, one sided sign test) in all the amino

acids, 80% in amino acids with tRNA synthetases

from class I (P = 0.0275, one sided sign test), and 30%

from class II). The bias is significant at P < 0.05 for

11 amino acids, with 8 amino acids from class I tRNA

synthetases (Cys, Leu, Val, P = 0.027 for each; Gln,

Ile and Met, P = 0.0054 each; Trp and Tyr, P =

0.0005, each) and 3 amino acids from class II (Gly

and Thr, P = 0.027, each; Lys, P = 0.0005). There was

one amino acid for which the bias was significantly (P

< 0.05) less negative when it was calculated for tRNA

synthetases specific to those cognate amino acids than

for other non-specific synthetases (Pro from class II, P

= 0.011, two sided sign test because the working

hypothesis does not expect positive cognate bias). The

tests that remain significant at P < 0.05 after

Bonferroni‟s correction are indicated in bold, while

the tests remaining significant according to the

Benjamini-Hochberg adjustment of Bonferroni‟s

correction are underlined. According to the latter

Table 1 (pages 15 and 16). Contrasts in amino acid compositions between sister tRNA synthetases in E. coli.

Rows are subtractions (contrasts) of frequencies (multiplied by 10000) of amino acids in the composition of

evolutionary sister tRNA synthetase from frequencies in compositions of specified tRNA synthetases. Contrasts

for cognate amino acids of the specified tRNA synthetase are in bold. Z indicates the normalized cognate bias

(in bold) divided by the standard deviation of the contrasts in that tRNA synthetase across all amino acids

(column Zb, corresponds to test level (b) in Materials and Methods and in Results), and for that amino acid

across tRNA synthetases (row Zc, corresponds to test level (c) in Materials and Methods and in Results). Sign

tests (see text) use numbers of contrasts that are more negative than the cognate bias. These numbers are

indicated by ± (column (±b) and row (±

c) after those corresponding for Z; for test level c (last rows), the first

number is for cases within the same tRNA synthetase class, and the second number for all tRNA synthetases).

These statistics are also indicated for B. subtilis (Zbb, Zb

c and ±b

b, ±b

c, respectively), N indicates tRNA

synthetase size in E. coli, followed by its size in B. subtilis. Bold numbers indicate statistical significance of

cognate bias at P < 0.05 according to a sign test. Note that in B. subtilis, data are missing for glutamynil tRNA

synthetase which is absent in that genome (complete genome sequence, NC_000964).

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

AA composition Arg Cys Gln Glu Ile Leu Met Trp Tyr Val

Synthetase NCBI entry N

I Arg - Cys AAC74946 577-556 96 39 -121 -94 5 -242 74 65 -52 -82

Cys – Arg AAC73628 461-466 -96 -39 121 94 -5 242 -74 -65 52 82

Gln – Glu AAC73774 554 43 32 -111 2 80 -85 -38 -25 15 -5

Glu – Gln AAC75457 471-483 -43 -32 111 -2 -80 85 38 25 -15 5

Ile - Val AAC73137 941-921 -153 65 -49 -225 -174 -12 -81 -8 77 145

Leu – Met AAC73743 860-804 -11 10 -65 1 -125 -86 85 82 64 217

Met - Leu AAC75175 677-664 11 -10 65 -1 125 86 -85 -82 -64 -217

Trp – Tyr AAC76409 334-330 -23 -45 -68 -102 37 -110 134 -82 159 206

Tyr - Trp AAC74709 424-419 23 45 68 102 -37 110 -134 82 -159 -206

Val – Ile AAC77215 951-880 153 -65 49 225 174 12 81 8 -77 -145

II Ala -His AAA03208 876-878 -104 -49 -83 -71 126 -288 28 -50 -137 -14

Asn - Asp AAC74026 466-430 18 1 -43 116 -127 -211 -141 82 106 331

Asp - Asn AAC74936 590-430 -18 -1 43 -116 127 211 141 -82 -106 -331

Gly - Phe AAC76584 303-295 -206 134 263 58 -93 78 -42 137 380 -18

AAC76583 689-679 16 -120 61 -19 -165 46 56 0.1 13 -254

His - Ala AAC75567 424-424 104 49 83 71 -126 288 -28 50 137 14

Lys – Asn, Asp AAC77090 505-499 -1533 -105 -716 -1647 -1042 -1769 -642 -149 -587 -1398

AAC75928 505 -1473 -105 -657 -1647 -1042 -1828 -642 -149 -568 -1398

Phe – Gly AAC74783 795-804 -16 120 -61 19 165 -46 -56 -0.1 -13 254

AAC74784 327 206 -134 -263 -58 93 -78 42 -137 -380 18

Pro – Thr AAC73305 562-564 753 20 356 911 614 951 356 40 277 555

Ser – Pro, Thr AAC73979 430-426 121 6 -24 36 -101 221 -5 -60 -68 -58

Thr - Pro AAC74789 642-643 -753 -20 -356 -911 -614 -951 -356 -40 -277 -555

Z

Zb

-1.11

-0.24

0.87

0.90

-1.28

0.44

-0.02 -1.50

0.13

-15.41

-1.39

-0.90

0.14

-1.29

-0.81

-1.66

-0.57

-0.86

-0.41

±

±b

8, 15

3, 7

2, 4

6, 14

1, 3

6, 11

3, 5

1, 3

4, 8

2, 5

1, 3

1, 4

5, 10

0, 3

2, 4

0, 3

1, 5

2, 6

2, 4

do no nob eb oT

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

AA

composition

Ala Asn Asp Gly His Lys Phe Pro Ser Thr

Synthetase Zb Zb

b ±

b | ±b

b

I Arg - Cys 81 -61 -200 4 -104 73 -83 4 -135 112 -0.90, -0.16 19| 11

Cys – Arg -81 61 200 -4 104 -73 83 -4 135 -112 0.37, 0.40 10| 11

Gln – Glu -167 63 -54 1 -115 68 115 81 102 -0.1 -1.92, 0.24 3 | 12

Glu – Gln 167 -63 54 -1 115 -68 -115 -81 -102 0.1 -0.04 9 | -

Ile - Val 158 -197 103 103 87 91 -39 5 163 -58 -1.46, 0.12 3 | 9

Leu – Met 47 25 10 100 -50 37 -233 4 -327 214 -0.86, -0.84 4 | 2

Met - Leu -47 -25 -10 -100 50 -37 233 -4 327 -214 -0.85, 0.12 3 | 11

Trp – Tyr 49 -83 63 -203 98 58 -230 143 74 -76 -0.70, -1.10 6 | 3

Tyr - Trp -49 83 -63 203 -98 -58 230 -143 -74 76 -1.37, -0.82 2 | 5

Val – Ile -158 197 -103 -103 -87 -91 39 -5 -163 58 -1.22, -1.07 2 | 4

II Ala -His 72 -24 166 -54 62 159 128 -79 193 19 0.74, 0.11 14| 11

Asn - Asp -86 39 -114 -42 19 -135 109 -50 176 -49 -1.15, -0.24 13| 6

Asp - Asn 86 -39 114 42 -19 135 -109 50 -176 49 -0.40, 1.61 9 | 18

Gly - Phe -196 -96 -53 23 -171 27 -183 41 -68 -13 0.25, -0.05 12| 4

353 -42 86 -228 -69 155 196 -30 -98 41 -1.36 1 | -

His - Ala -72 24 -166 54 -62 -159 -128 79 -193 -19 -0.65, -2.21 6 | 1

Lys – Asn, Asp -1735 -806 -1435 -1361 -380 -871 -1160 -809 -764 -1092 -2.20, -0.58 10| 4

-1715 -825 -1475 -1401 -400 -950 -1120 -809 -744 -1053 -2.42 9 | -

Phe – Gly -353 42 -86 228 69 -155 -196 30 98 -41 -1.17, -1.05 2 | 2

196 96 53 -23 171 -27 183 -41 68 13 2.02 18| -

Pro – Thr 792 416 753 693 238 475 574 376 317 535 2.01, -0.47 8 | 2

Ser – Pro, Thr 41 32 37 -2 -27 -72 -94 14 -82 85 -1.37, -0.42 3 | 6

Thr - Pro -792 -416 -753 -693 -238 -475 -574 -376 -317 -535 -2.86, 0.07 9 | 9

Zc

Zbc

0.10

0.02

0.11

-0.64

0.18

2.03

-0.38

-0.05

-0.33

-1.12

-2.34

-0.47

-0.39

-1.43

1.10

-1.18

-0.25

-0.58

-1.16

0.11

±c

±bc

5, 12

6, 11

6, 11

2, 5

7, 16

9, 18

2, 3

4, 7

4, 8

1, 1

0, 0

2, 4

6, 11

0, 1

9, 19

0, 2

5, 9

4, 8

2, 3

4, 10

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

method, 3 among the 5 tests (significant at P = 0.027)

remain significant, but because the decision of which

three among the five is arbitrary, none can be

specifically assigned as significant. Nevertheless,

according to the latter approach, 10 among 20 tests are

significant at P < 0.05 after adjusting P values.

Analyses can be repeated by considering all tRNA

synthetases across classes. Following this approach,

there were 12 cognates with P < 0.05 (8 in class I: Cys

and Met, P = 0.003, each; Gln, Trp and Tyr, P =

0.0006, each; Glu and Leu, P = 0.01, each; and Val, P

= 0.029; and 3 in class II: Gly and Thr, P = 0.0006,

each; and Lys, P = 0.0000005). There were 3 cognates

for which there was significant (P <0.05) positive

cognate bias, one from class I (Arg, P = 0.021) and

two from class II (Asp and Pro, P = 0.006 and P =

0.00002, respectively). These results also confirm the

working hypothesis at the level of test 3. Combining

the P values according to Fisher‟s method, this yields

P = 5.6*10-14

and P = 3.2*10-7

for class I and II,

respectively.

Results on positive cognate biases, together with

parallel ones from the previous section, suggest that

positive cognate bias exists in some few cases

(nevertheless 10%). Consequently, cognate bias is

bimodal, with a majority of cases in the realm of the

negative bias expected by the working hypothesis, but

positive bias apparently occurs too. The working

hypothesis expects negative bias as a result of the

principle of minimizing costs of protein synthesis by

avoiding the cognate/substrate of the protein in the

protein‟s own composition. This hypothesis is

incompatible with positive bias. It accommodates

cases where no systematic tendency for positive bias

exist (hence weak positive bias), but not cases where

the positive cognate bias is strong and statistically

detectable. This implies that a factor other than cost

minimization of cognate usage operates here, at least

in some cases, so as to create these rarer positive

biases.

3.2.3. E. coli: Effects of intrinsic usage and

environmental abundance. The bias against the

cognate amino acid decreases in E. coli as a function

of the usage frequency of that amino acid in the

species‟ proteome (table 1 in [27]). I found this

tendency, yet it was not statistically significant (r =

0.31 and P = 0.093 for all synthetases; r = 0.41 and P

=0.118 for class I; r = 0.20 and P = 0.29 for class II,

all one tailed tests). There was also no statistically

significant trend between cognate amino acid bias and

the abundance of the amino acid in the colon, the

habitat of E. coli (classes pooled, r = -0.36, P = 0.147;

class I, r = -0.23; class II, r = 0.23, P = 0.51; two

tailed tests). This suggests that substrate avoidance in

E. coli affects mainly the composition of

rare/expensive amino acids, but not of highly

expressed synthetases. If one remains within the issue

of cost minimization of protein synthesis metabolism,

these data suggest that E. coli protein compositions

are relatively independent of the composition of its

host environment, but are more affected by the

composition of its own intrinsic cellular environment,

because it seems to minimize costs according to its

own usage frequency of an amino acid. This result

could be interpreted as resulting from its

endosymbiontic life habit.

3.3. Tests for B. subtilis

3.3.1. B. subtilis: Cognate bias tests within single

tRNA synthetases. In Bacteria subtilis, for tests

within single tRNA synthetases (rows in Table 1, test

level (b) as indicated in „Materials and Methods‟), but

across amino acids, in 68% of the tRNA synthetases

(P = 0.042, one sided sign test), bias was more

frequently negative for the cognate amino acid than

the biases calculated for the 19 non-cognate amino

acids (56% for class I (P = 0.25), 80% for class II (P =

0.027), one sided sign tests). The negative bias is

significant at P 0.05 (one sided sign test) in 58% (11

among 19 cases) of all tRNA synthetases, 40% for

class I (Leu, P = 0.00018; Trp, P = 0.0011; Tyr, P =

0.016 and Val, P = 0.0048, one sided sign tests), and

70% for class II (Asn and Ser, P = 0.042 in each; Gly

and Lys, P = 0.0048 in each; His, P = 0.00002; Phe

and Pro, P = 0.00018 in each). There was one class II

tRNA synthetase for which cognate bias was

significantly less negative than for the other non-

cognate amino acids (aspartyl synthetases, P =

0.00004). Hence these results independently confirm

the existence of positive cognate bias, in addition of

negative cognate bias. Combining P values using

Fisher‟s method yields chi-square = 44.7, and chi-

square = 66.9 for class I and II, respectively, which

yields P = 6.98*10-7

and P = 1.76*10-10

, respectively.

This P value is greater by 11 orders of magnitude than

that for class I tRNA synthetases in E. coli, and

greater by one magnitude than for class II tRNA

synthetases in E. coli. Overall, the phenomenon of

negative cognate bias exists and is clearly detectable

in both species at the level of single tRNA

synthetases, but it is even more obvious in E. coli than

in B. subtilis.

3.3.2. B. subtilis: Cognate bias tests across tRNA

synthetases. For tests keeping the amino acid

constant, but across tRNA synthetases, test level (c) as

indicated in „Materials and Methods‟, tests can be

done considering only cognates from tRNA

synthetases of the same class, and considering all

cognates. Considering only those from tRNA

synthetases within a given class, bias was more

frequently more negative for the tRNA synthetase for

which this amino acid is its cognate amino acid than

for the other non-specific tRNA synthetases from the

same class in 74% of the cases (P = 0.016, one sided

sign test),66% for tRNA synthetases from class I and

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

80% from class II (P = 0.127 and P = 0.027, respectively, one

sided sign tests). This was significant at P < 0.05 for 9 (47%)

of the cases, 4 (44%) from class I (Leu and Tyr, P = 0.0098, in

each; Trp and Val, P = 0.045, in each) and 5 (50%) from class

II (Asn and Lys, P = 0.027 in each; His, Phe and Pro, P =

0.0054 in each).

These analyses were repeated pooling all tRNA synthetases,

independently of their class. According to this approach, there

was negative bias for cognates from 13 (68%), and this bias

was significant for 9 (47%) cases, 4 from class I ( Leu, P =

0.0011; Trp and Val, P = 0.0048, in each; and Tyr, P =

0.0159) and 5 from class II (Asn, P = 0.0159; His and Phe, P =

0.000019 in each; Lys, P = 0.0048; and Pro, P = 0.00018).

There was statistically significant positive cognate bias for one

cognate, Asp (P = 0.000038). Combining P values to test for

the working hypothesis considering all tRNA synthetases

together yields chi square statistics 53.95 and 90.28 with P =

1.87*10-5

and 6.62*10-11

for class I and II respectively. This is

9 orders of magnitude less than for class I in E. coli, and 4

orders of magnitude more than for class II in E. coli.

3.3.3. B. subtilis: Effects of environmental abundance and

intrinsic usage. In B. subtilis, avoidance of cognate amino

acids decreases (albeit not statistically significantly) with

usage frequencies of those amino acids in the proteome (r = -

0.01, P = 0.96, 2 tailed test, for all synthetases; class I

synthetases, r = -0.17, P = 0.67 and r = 0.51, P = 0.13 for class

II synthetases). Although these results are not per se

statistically significant, the tendencies are systematically

opposite of those from E. coli, where negative cognate bias

seems to decrease with usage. Figure 1 plots cognate amino

acid bias as a function of the abundance of the amino acid in

the soil, the environment of B. subtilis. It shows that cognate

amino acid bias, in terms of avoidance, occurs for synthetases

Phe

Thr

Ser

Gly

Ala

Asp

Asn

Leu

Arg

HisPro

TrpLys,Val,Tyr

Met

Cys, Glu

Ile

-0.03

0

0.03

0 1 2 3 4

Abundance category in Soil

Co

gn

ate

av

oid

an

ce

Figure 1. Cognate bias in B. subtilis tRNA synthetases as a

function of amino acid abundance class in the soil.

that esterify amino acids that are rare in the

soil, and those that are abundant in the soil.

Positive cognate bias exists for amino acids

with intermediate abundances. These results

contrast with those in E. coli, where substrate

avoidance seems unrelated to environmental

concentrations, and decreases, rather than

increases, with amino acid usage.

4. Discussion

4.1. General overview. The results described

here confirm the main result in [27] that

cognate amino acid avoidance exists. Results

also confirm that cost minimization exists.

The cost minimization might be even slightly

stronger in the bacterium with symbiontic

lifestyle than in the free living B. subtilis, as

suggested by lower numbers of specific tRNA

synthetases with significant negative bias

detected through test levels b and c in B.

subtilis (tests b and c, 11 and 9, respectively)

than in E. coli (11 and 12). Regarding the

hypothesis that symbionts are less likely to

minimize costs, it is strengthened by the fact

that positive cognate bias is more common in

E. coli than in B. subtilis: tests b and c

detected 5 statistically significant positive

biases in total in E. coli, but only two in B.

subtilis. This observation could be interpreted

as a confirmation of the effects of life style on

cost minimization by cognate bias.

4.2. Statistical considerations. Results

discovered by using a controlled (natural)

experimental setup here are very significant

for specific tRNA synthetases and amino

acids, in spite of the use of the conservative

(and robust) sign test. Results stress the

increased power of this system at uncovering

patterns, and at detecting more subtle and

finely tuned phenomena than the approach

used in [27].

Although analyses presented in this work

are of confirmative nature of principles

already described earlier, I used statistical

adjustment methods to control the multiplicity

of tests only as a precaution, but not to

confirm the cases with significant value at P <

0.05 without adjustments [46]. Considering

that such methods are overconservative and

can artificially inflate negative results [47],

the methods of correcting test multiplicity can

be valuable if the main aim is to be certain

that a phenomenon exists. Therefore they are

important for the fewer cases where positive

cognate bias was detected. However previous

studies already showed that the phenomenon

of negative cognate bias occurs, hence

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

asserting the existence of negative bias is not the

primary aim here.

The results suggest that the comparative system

formed by the two classes of phylogenetically related

tRNA synthetases could prove more useful when

trying to uncover the functional aspects of amino acid

conservation and replacements at active sites.

However, one has to bear in mind that the increased

statistical power comes at the costs of the generality of

the results: observations within tRNA synthetases may

be affected by the specifics of tRNA synthetase

biology (including differences between classes I and

II). This situation is not exceptional for this specific

system. The more precise approach used here is

considered to complete, rather than replace, the more

general results based on less „carefully‟ designed

comparisons [27]. The fact that both analyses confirm

the same principles increases the confidence that the

hypothesized principle is indeed valid at biological

level.

4.3. E. coli versus B. subtilis

It makes sense that the absolute rarity in the medium

leads to cost minimization for an amino acid.

However, it is less straightforward to understand why

cognate avoidance occurs for amino acids that are

frequent in the medium. Note that proteome-wide

amino acid usage in B. subtilis correlates positively

with their abundances in the soil (r = 0.42, P = 0.031,

one sided test). However, while cognate avoidance

exists in E. coli and B. subtilis at similar levels

(though slightly weaker in B. subtilis), patterns clearly

contrast between them. It is not clear whether the

differences in associations between cognate avoidance

and environmental versus proteomic abundances are

related to the endosymbiontic versus „free living‟

lifestyles of these two bacteria. It makes sense that the

endosymbiont is less dependent on variation in amino

acid environmental abundances because these are all

kept high by the host (or with predictable daily

fluctuations), and hence cognate avoidance will

mainly reflect the own metabolic needs of E. coli. In

contrast, results in B. subtilis indicate two factors

accounting for the differences: one related to

environmental amino acid abundances (its

environmental abundances seem to affect proteomic

amino acid abundances, as suggested by the positive

correlation between proteome-wide amino acid

abundances and those in the soil), and the other

unknown factor (for amino acids which are abundant

in the soil) that might be due to intrinsic metabolic

needs (perhaps usage frequency, as in E. coli). These

results would be in line with general principles

deduced from observations that usually, parasites and

symbionts minimize metabolic costs less than

comparable free living organisms [7-8, 48]. There is

here an apparent conundrum: parasites minimize less

costs at the level of protein synthesis, including by

having larger proteins and hence longer protein coding

genes. Since they minimize costs of DNA replication

by reducing genome sizes, they are not (or less)

minimizing costs according to proteins, but are more

than free-living bacteria according to genome size.

4.4. Negative cognate bias and cost minimization

strategy by avoiding large residues

The data analyzed here enable to develop the theory

about cost minimization of protein metabolism

beyond confirmation of previous results [27]. Cost

minimization can be related to various properties of

proteins, and be estimated using different methods

[49]. I previously used the (negative) correlation

between amino acid molecular weight and its

proteomic usage to estimate an overall extent of cost

minimization of protein metabolism based on the

usage of all amino acids [7]. In that approach, the

more negative the correlation coefficient, the rarer

large amino acids are in the composition of proteins.

The analyses presented above are related specifically

to avoiding usage of cognate amino acids. This is a

different type of cost avoidance. Figure 2 shows that

in E. coli, extents of negative cognate avoidance are

negatively associated (r = -0.41, P = 0.049, two tailed

test) with extents of cost minimization in relation to

amino acid size. This is estimated by the negative

correlations between amino acid usages and their

molecular weights. Each datapoint in Figure 2 is for

a given tRNA synthetase (note that for Gly, Lys and

Phe, the tRNA synthetases consist of two subunits).

This correlation is also qualitatively confirmed for B.

subtilis (r = -0.07) but is clearly much weaker and not

statistically significant in this species. This result for

E. coli suggests a balancing situation between

minimizing metabolic energetic costs (at the level of

favoring usage of small amino acids) and biosynthetic

time (negative cognate bias mainly avoids long

durations until a tRNA loaded with the cognate is

encountered during protein synthesis). This inverse

proportionality between the two phenomena suggests

that overall cost minimization reached an equilibrium

where increasing cost minimization according to one

factor (time) will decrease cost minimization

according to the other factor (energy). This implies the

possibility that protein function can be affected by

replacing one amino acid by another one due to cost

considerations. Hence decreases in protein

functionality due to avoiding cognates can be

compensated by using more expensive non-cognates.

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

12

http://ccaasmag.org/BIO

Figure 2. Cognate bias (Z standardized) in E. coli tRNA

synthetases as a function of cost minimization of amino acid

abundances in these proteins, estimated by the Pearson

correlation coefficient r between amino acid abundances and

the molecular weight of the amino acids for each protein.

For both x and y axes, the more negative values indicate

greater cost minimization. Note that cognate bias can differ

widely for different subunits of the same tRNA synthetase,

such as for those acylating Phe, as well as be very similar, as

found for those acylating Lys.

The fact that this phenomenon is not observed in B.

subtilis suggests that in this species the system did not reach

equilibrium in this respect. This makes sense because an

evolutionary equilibrium requires stable and especially

predictable environmental conditions. The colon might be

more stable than the soil.

4.5. Negative cognate bias and protein size. The

interesting result about a compensating equilibrium between

amino acid energetic cost minimization and cognate usage in

tRNA synthetases in the previous section can be

independently confirmed by looking at a further factor

affecting costs of protein synthesis, its length. As expected,

tRNA synthetases are larger in the endosymbiont E. coli

than in the free living B. subtilis (in 63 percent of the cases,

which is however not statistically significant in this small

sample according to a sign test). However, if one considers

that three proteins forming tRNA synthetases are missing in

B. subtilis (for example, the smaller subunits for tRNA

synthetase Lys and Phe), it is clear that the freeliving

bacterium reduces protein size as compared to E. coli. When

one compares cost minimization between these two species

for homologous tRNA synthetases, by calculating the

difference between correlation coefficients r between amino

acid usages and their molecular weights, one finds that the

correlations are more negative (hence more cost

minimization) in E. coli than in B. subtilis in 16 out of 20

cases (P = 0.012 according to a two sided sign test). This

situation is opposite to the one found for protein

length as cost criterion, according to which E.

coli seems to minimize costs less than B.

subtilis. This suggests compensatory effects

between different types of cost minimization, as

indicated in Figure 2. Indeed, the increase in

tRNA synthetase length observed in E. coli is

inversely proportional to the increase in cost

minimization in that species, as compared to B.

subtilis, and as estimated by the correlation

between amino acid usage and their molecular

weights (r = -0.39, P = 0.044, one tailed test,

data not shown; a one tailed test is used here

because a negative correlation is expected,

considering the result in Figure 2 that suggests

that compensation is expected). Besides

confirming the principle of reaching

equilibrium in a stable/predictable rather than

an unpredictable environment, and that of

optimization according to several criteria at

once, these results suggest that a

unidimensional approach to proteins is

inadequate, that these are complex constructs

reacting to different pressures, and that their

structure is optimized in relation to several

factors, even if those reflect (more or less) the

same simplistic economic principle.

4.6. Cognate bias and functional properties:

secondary structure formation. A wide

variation exists among amino acids in terms of

cognate bias, from negative to positive bias, and

this varies also between E. coli and B. subtilis.

One could assume that this extent depends on

how much avoidable the cognate is in the

structure of the protein. For example,

membrane proteins are very hydrophobic and

with low polarity, hence these could probably

be limited in the extents they could avoid using

hydrophobic, non-polar amino acids in their

structure, even if these were not optimal from

the point of view of metabolic/energetic

considerations. The same principle functions

also for tRNA synthetases, which are not

membrane bound but have other function-

associated constraints. Class I tRNA

synthetases have the Rossmann fold, both

classes are made up of anti-parallel beta strands

[50]. Amino acid polarity is also an important

property in protein folding, low polarity

favoring hydrophobic interactions, while polar

regions with opposite charges tend to attract

each other and those with similar charges repel

each other, both principles that contribute to

protein folding. Hence one expects that cognate

bias associates with the betasheet-formation

index, which estimates the tendency of an

amino acid to be more (or less) frequent than

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

expected in protein regions forming

betasheet secondary structures. Indeed,

cognate bias in E. coli correlates negatively

with the beta index (class I, r = - 0.79, P =

0.0033; class II, r = -0.45, P = 0.096; all, r =

-0.69, P = 0.00038, one tailed tests). This

phenomenon was much weaker for B.

subtilis (class I, r = - 0.20, P = 0.30; class II,

r = -0.51, P = 0.066; all, r = -0.14, P = 0.28,

one tailed tests). Because the structure of

both classes of tRNA synthetases is based

on antiparallel betasheets, it is not surprising

to see that correlations of cognate bias with

antiparallel betasheet indices (E. coli: class

I, r = -0.75, P = 0.0062; class II, r = -0.57, P

= 0.043; all, r = -0.73, P = 0.00013; B.

subtilis: class I, r = -0.72, P = 0.014; class II,

r = -0.46, P = 0.09; all, r = -0.38, P = 0.054,

see Figure 3) are generally stronger than

those indicated above, for indices calculated

pooling parallel and antiparallel betasheets.

It is unlikely that this difference between the

two species indicates that secondary

structure formation during protein folding is

not related to the beta index in B. subtilis,

but rather that the latter optimizes cognate

bias according to additional criteria. Similar

analyses for another important amino acid

property, polarity, suggest for E. coli (but

not B. subtilis) differences between class I

and II tRNA synthetases in relation to

cognate bias optimization: in class I, cognate

bias decreases with cognate polarity (r = -

0.66, P = 0.038, two tailed test), while the

tendency was opposite for class II (r = 0.31,

P = 0.38, two tailed test). Figure 4 shows

that this difference between classes is

mainly due to the tRNA synthetases with

cognates Phe and Pro. Other class II tRNA

synthetases overall fit into the trend

observed for class I tRNA synthetases. This

trend suggests that polar cognates can more

be avoided in the structure of its tRNA

synthetase than non-polar ones. This

suggests that polar interactions are not

preponderant in tRNA synthetase folding,

besides the exceptions for the tRNA

synthetases of Phe and Pro indicated above.

For B. subtilis, this tendency for greater

negative cognate bias in polar cognates

exists but is weaker (class I, r = -0.44, P =

0.118; class II, r = -0.40, P = 0.126, one

tailed tests).

Further analyses reveal again differences

between class I and II tRNA synthetases in

E. coli, in relation to cognate bias. Betasheet

formation indices above unity indicate that a

specific amino acid is more frequently part

of betasheets than of other structural domains. Amino acids that

are avoided in beta sheets get values below unity. Such residues

are also important for betasheet formation, because this involves

(relatively short) regions that should not form a beta sheet.

Hence one could assume that also the absolute value of the

subtraction of the betasheet index from unity, as an estimate of

structural specificity of an amino acid in relation to betasheet

Figure 3. Cognate bias in tRNA synthetases as a function of the

betasheet formation index of the amino acids for E. coli (filled

symbols) and B. subtilis (circles). Negative cognate bias is

stronger for amino acids with low tendencies to form beta

sheets.

Figure 4. Cognate bias in E. coli tRNA synthetases as a

function of amino acid polarity.

formation, might associate with cognate bias: those with high

specificity could not be avoided, whether this specificity

indicates specificity promoting, or preventing betasheet

formation. This is the case for class II tRNA synthetases in E.

coli (r = 0.66, P = 0.038, one tailed test) but the exact opposite

is observed for class I (r =-0.79, P = 0.0065, two tailed

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

test; Figure 5). Tendencies were similar but weaker for B.

subtilis (class I, r =-0.24, P = 0.27; class II, r = 0.27, P = 0.23,

one tailed tests). The results indicate that cognate avoidance is

prevented for class II tRNA synthetases if that amino acid has

a high impact on betasheet formation, whether by preventing

or promoting it. Interpreting the result for class I is less

straightforward, as the more an amino acid impacts secondary

structure formation, the more it is avoided. However, results

from Figure 4 indicated that nonpolar (hydrophobic)

interactions, a physicochemical property, rather than a

structural one (beta index), is optimized in relation to cognate

bias in class I tRNA synthetases of E. coli. This seems to be

the explanation for the negative correlation with cognate bias

observed in class I tRNA synthetases: the structural impact on

betasheet formation used in Figure 5 increases with amino

acid polarity for class I tRNA synthetases (class I, r = 0.78, P

= 0.0077; class II, r = 0.27, P = 0.45, two tailed tests). These

analyses focused on cognate bias confirm the antiparallel

betasheet structure of tRNA synthetases, but further analyses

show that additional properties are important for protein

structure, and that there are differences between the tRNA

synthetase classes. The differences in this respect between the

two bacteria suggest that much can be learned from such

analyses. It is probable that comparing larger numbers of

species, as similar data on tRNA synthetase sequences are

available from genomic sequences of numerous species, could

enable to study the details of interactions between principles

of protein folding and the species‟ life histories, based on

relatively simple analyses of cognate bias.

Figure 5. Cognate bias in E. coli tRNA synthetases as a

function of amino acid structural impact on betasheet

formation. The x axis is the absolute value of the subtraction

of the betasheet index from unity. Hence the more an amino

acid affects betasheet formation, whether by promoting or

preventing it, the larger the value on the x axis. In class II

tRNA synthetases, cognate avoidance decreases with

structural impact of the cognate on betasheet

formation, the opposite occurs for class I

tRNA synthetases.

4.7.Cognate bias and functional properties:

tRNA editing. Many tRNA synthetases

possess catalytic sites that are not involved in

aminoacylation of the tRNA, but in its pre-

and post-transfer editing, thus preventing

misacylations of tRNAs by non-cognate

amino acids [51]. The existence of specific

sites with editing activities has been indicated

for several tRNA synthetases (class I: Ile [52],

Leu [53], Met [54], Val [55]; class II: Ala

[56], Asn [57], Phe [58], Pro [59], Thr [60])

but evidence indicates the lack of specific

editing sites for some class I tRNA

synthetases (Cys [61], Gln [62] and Tyr [63])

and apparently inefficient pre-transfer editing

for the class II tRNA synthetase

aminoacylating Lys [64]. The nature of cost

minimization of amino acid composition is

that an „expensive‟ amino acid is used only

where it is unavoidable. Obviously, catalytic

and other functionally important sites (such as

editing sites) are regions where cost

minimization (by cognate avoidance) is

unlikely because of their functional

importance. Functional sites are formed by

relatively few residues, but tRNA synthetases

with editing sites should on average possess

more functionally important regions than

those lacking one. Hence one can expect a

slightly less negative cognate bias for tRNA

synthetases that possess editing sites. No such

effect could be detected for cognate biases of

E. coli and B. subtilis. However, after

controlling for the effect of cognate polarity,

using the regression in Figure 4 to calculate

residual cognate bias, a positive association

between cognate bias and editing sites can be

detected, considering that the confirmed lack

of editing site is „-1‟ for the tRNA synthetases

of Cys, Gln and Tyr, „1‟ for those with

confirmed editing sites, and „0‟ for the others

(r = 0.44, P = 0.027; Spearman rank

correlation rs = 0.52, P = 0.0112, one tailed

tests, see Figure 6). Hence after controlling

for the effects of a functional constraint on

cognate bias that relates to the overall

structure of the protein, the effect of the

number of specific functional sites on cognate

bias becomes detectable. It is notable that no

such effect is observed if a similar residual

analysis is done, but using the regression lines

in Figure 5, where cognate bias is shown

constrained by the formation of betasheets.

This could mean that the function of editing

sites is unrelated to betasheet formation, but

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

that polarity affects interactions with loaded tRNAs. This

interesting hypothesis seems plausible, but should not be

overstressed at this stage. It is nevertheless a potential

example of how the analysis of cognate bias, in relation to

other properties, can reveal how proteins function.

Figure 6. Residual cognate bias in E. coli tRNA synthetases

as a function of the presence of editing sites in the tRNA

synthetase. The y axis is the residual cognate bias calculated

from the regression line in Figure 4. The status of editing sites

on the x axis is: -1, tRNA synthetases for which evidence

exists for a lack of editing site; 1, tRNA synthetases with

confirmed editing site(s); 0, tRNA synthetases with no

reported information.

4.8. Cost minimization hypotheses. I have already expressed

my disappointment that so much biology can be understood

via such a simple economy-minded principle as amino acid

cost minimization [7]. Similar principles also predict high

densities of off frame codons in protein coding genes,

reducing costs of synthesis of dysfunctional proteins after

ribosomal slippages [65-67]. It also seems that the adaptation

for translational activity of antisense tRNAs, tRNAs templated

by DNA complementing regular, sense tRNA genes, is also

governed by cost minimization principles [68-70], of reducing

genome size. Some results even confirm that selection favors

the compensatory combination of codon-anticodon

mismatches with tRNA misacylations by non-cognates, which

result in correct amino acid insertion because the mismatched

codon codes for the misacylated amino acid [71-73]. The latter

effects predicted by cost minimization principles (in these

cases in relation to inaccuracies in protein synthesis) could

have been expected to be weak, these nevertheless seem to

have substantial effects on genomes [73]. This means that cost

minimization principles are a powerful approach for the

analysis of biochemical systems. These results present serious

challenges to those promoting neutralistic views of evolution

[74].

5. Conclusions

The principle of cost minimization of protein

synthesis predicts that cognates are avoided in

tRNA synthetases, as compared to closely

related tRNA synthetases with a different

cognate. This negative cognate bias is more

easily detected in a phylogenetically matched

setup than when comparing cognate abund-

ances in specific proteins with abundances in

randomly selected proteins. The bias is also

found less negative in the free living B.

subtilis than in the endosymbiont E. coli.

However, cost minimization is a multidimen-

sional phenomenon, and has to be observed

simultaneously at different levels. They are,

for instances, avoidance of costly residues

(independently of cognate status), protein

size, and negative cognate bias. In E. coli,

these various components of protein meta-

bolic costs seem to be at evolutionary

equilibrium, where one component is at the

expense of the other. This is probably due to

constraints related to the protein‟s function.

Analyses show that cognate avoidance is

prevented for amino acids that promote

betasheets, the major secondary structure of

tRNA synthetases. They also reveal that non-

polar interactions are important for class I

tRNA synthetases, but not class II. Cognate

avoidance is found to be greater in tRNA

synthetases lacking editing sites, which is

probably due to the fact that these proteins

possess more functionally critical sites/

regions that limit cost minimization. These

phenomena are not detected in B. subtilis.

This is possibly because its environment is

less stable, and proteins do not evolve toward

a stable steady state structure that integrates

functional and cost minimization constraints.

Acknowledgments

I thank Neeraja M. Krishnan for comments on

a previous version of this manuscript.

References

[1] Barrai, I.; Volinia, S.; Scapoli, C. The

usage of oligopeptides in proteins

correlates negatively with molecular

weight. Int. J. Peptide Prot. Res. 1995, 45,

326-331. http://dx.doi.org/10.1111/j.1399-3011.1995.tb01045.x

[2] Dufton, M.J. Genetic code synonym

quotas and amino acid complexity: Cutting

the cost of proteins? J. Theor. Biol. 1997,

187, 165-173. http://dx.doi.org/10.1006/jtbi.1997.0443

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

[3] Ikemura, T. Correlation between the abundance of

Escherichia coli transfer RNAs and the occurrence

of the respective codons in its protein genes. J. Mol.

Biol. 1981, 146, 1-21. http://dx.doi.org/10.1016/0022-2836(81)90363-6

[4] Bennetzen, J.L.; Hall, B.D. Codon selection in

Saccharomyces cerevisiae. J. Biol. Chem. 1982,

257, 3026-3031.

[5] Gouy, M.; Gautier, C. Codon-usage in bacteria:

correlation with gene expressivity. Nucl. Acids Res.

1982, 10, 7055-7074. http://dx.doi.org/10.1093/nar/10.22.7055

[6] Akashi, H.; Gojobori, T. Metabolic efficiency and

amino acid composition in the proteomes of

Escherichia coli and Bacillus subtilis. PNAS 2002,

99, 3695-3700. http://dx.doi.org/10.1073/pnas.062526999

[7] Seligmann, H. Cost minimization of amino acid

usage. J. Mol. Evol. 2003, 56, 151-161.

http://dx.doi.org/10.1007/s00239-002-2388-z

[8] Das, S.; Ghosh, S.; Pan, A.; Dutta, C.

Compositional variation in bacterial genes and

proteins with potential expression level. FEBS

Letters 2005, 579, 5205-5210.

http://dx.doi.org/10.1016/j.febslet.2005.08.042

[9] Swire, J. Selection on synthesis cost affects

interprotein amino acid usage in all three domains

of life. J. Mol. Evol. 2007, 64, 558-571.

http://dx.doi.org/10.1007/s00239-006-0206-8

[10] Raiford, D.W.; Heizer, E.M.; Miller, R.V.;

Akashi, H.; Raymer, M.L.; Krane, D.E. Do amino

acid biosynthetic costs constrain protein evolution

in Saccharomyces cerevisiae? J. Mol. Evol. 2008,

67, 621-630. http://dx.doi.org/10.1007/s00239-008-9162-9

[11] Perlstein, E.O.; de Bivort, B.L.; Kunes, S.;

Schreiber, S.L. Evolutionary conserved

optimization of amino acid biosynthesis. J. Mol.

Evol. 2007, 65, 186-196. http://dx.doi.org/10.1007/s00239-

007-0013-x

[12] Das, S.; Pan, A.; Paul, S; Dutta, C. Comparative

analyses of codon and amino acid usage in

symbiotic island and core genome in nitrogen-

fixing symbiotic bacterium Bradyrhizobium

japonicum. J. Biomol. Struct. & Dynamics 2005,

23, 221-232.

[13] Chanda, I.; Pan ,A.; Saha, S.K.; Dutta, C.

Comparative codon and amino acid composition

analysis in Tritryps-conspicuous features of

Leishmania major. FEBS Letters 2007, 581, 5751-

5758. http://dx.doi.org/10.1016/j.febslet.2007.11.041

[14] Brocchieri, L.; Karlin, S. Protein length in

eukaryotic and prokaryotic proteomes. Nuc. Acids

Res. 2005, 33, 3390-3400.

http://dx.doi.org/10.1093/nar/gki615

[15] Warringer, J.; Blomberg, A. Evolutionary

constraints on yeast protein size. BMC Evol. Biol.

2006, 6, 61. http://dx.doi.org/10.1186/1471-2148-6-61

[16] Gong, X.; Fan, S.; Bilderbeck, A.; Li, M.; Pang,

H.; Tao, S. Comparative analysis of essential genes

and nonessential genes in Escherichia coli K12.

Mol. Genet. Genomics 2008, 279, 87-94.

http://dx.doi.org/10.1007/s00438-007-0298-x

[17] de Bivort, B.L. ; Perlstein, E.O. ; Kunes, S. ;

Schreiber, S.L.Amino acid metabolic origin as an

evolutionary influence on protein sequence in

yeast. J. Mol. Evol. 2009, 68, 490-497.

http://dx.doi.org/10.1007/s00239-009-9218-5

[18] Sajitz-Hermstein, M.; Nikoloski, Z. A novel

approach for determining environment-specific

protein costs: the case of Arabidopsis thaliana.

Bioinformatics 2010, 26, i582-i588.

http://dx.doi.org/10.1093/bioinformatics/btq390

[19] Smitch, D.R.; Chapman, M.R. Economical

evolution: Microbes reduce the synthetic cost of

extracellular proteins. mBio 2011, 1, e00131-10.

[20] Heizer, E.M.; Raymer, M.L.; Krane, D.E. Amino

acid biosynthetic cost and protein conservation. J.

Mol. Evol. 2011, 72, 466-473.

http://dx.doi.org/10.1007/s00239-011-9445-4

[21] Mazel, D.; Marlière, P. Adaptive eradication of

methionine and cysteine from cyanobacterial light-

harvesting proteins. Nature 1989, 341, 245-248.

http://dx.doi.org/10.1038/341245a0

[22] Baudouin-Cornu, P.; Surdin-Kerjan, Y.; Marlière,

P.; Thomas, D. Molecular evolution of protein

atomic composition. Science 2001, 293, 297-300.

http://dx.doi.org/10.1126/science.1061052

[23] Barton, M.D.; Papp, B.; Delneri, D.; Oliver, S.;

Rattray, M.; Bergman, C. Systems biology of

energetic and atomic costs in the yeast

transcriptome, proteome, and metabolome. Nature

Preceedings 2008,

http://hdl.handle.net/10101/npre.2008.1841.1

[24] Jankovic, B.; Seoighe, C.; Alqurashi, M.;

Gehring, C. Is there evidence of optimisation for

carbon efficiency in plant proteomes? Plant Biol.

2011, 13, 831-834. http://dx.doi.org/10.1111/j.1438-

8677.2011.00494.x

[25] Grzymski, J.J.; Dussaq, M.A.The significance of

nitrogen cost minimization in proteomes of marine

microorganisms. ISME J., 2011.

http://dx.doi.org/10.1038/ismej.2011.72

[26] Bragg, J.G.; Wagner, A. Protein material costs:

single atoms can make an evolutionary difference.

Trends Genetics 2009, 25, 5-8.

http://dx.doi.org/10.1016/j.tig.2008.10.007

[27] Alves, R.; Savageau, M.A. Evidence of selection

for low cognate amino acid bias in amino acid

biosynthetic enzymes. Mol. Microbiol. 2005, 56,

1017-1034. http://dx.doi.org/10.1111/j.1365-2958.2005.04566.x

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

[28] LéJohn, H.B.; Cameron, L.E.; Rennie, S.L.

Molecular characterization of an nad-specific

glutamate-dehydrogenase gene inducible by L-

glutamine-antisense gene pair arrangement with L-

glutamine-inducible heat-shock 70-like protein

gene. J. Biol. Chem. 1994, 269, 4523-4531.

[29] Carter, C.W., Duax, W.L. Did tRNA synthetase

classes arise on opposite strands of the same gene?

Mol. Cell 2002, 10, 705-708.

http://dx.doi.org/10.1016/S1097-2765(02)00688-3

[30] Rodin, A.S.; Rodin, S.N. Translation of both

complementary strands might govern early

evolution of the genetic code. In Silico Biol. 2007,

7, 309-318.

[31] Rodin, S.N.; Rodin, A.S. On the origin of the

genetic code: signatures of its primordial

complementarity in tRNAs and aminoacyl-tRNA

synthetases. Heredity 2008, 100, 341-355.

http://dx.doi.org/10.1038/sj.hdy.6801086

[32] Rodin, A.S.; Rodin, S.N.; Carter, C.W. On

primordial sense-antisense coding. J. Mol. Evol.

2009, 69, 555-567.http://dx.doi.org/10.1007/s00239-009-9288-4

[33] Rodin, A.S.; Szathmary, E.; Rodin, S.N. One

ancestor for two codes viewed from the perspective

of two complementary modes of tRNA

aminoacylation. Biol. Direct 2009, 4, 4.

http://dx.doi.org/10.1186/1745-6150-4-4

[34] Williams, T.A.; Wolfe, K.H.; Fares, M.A. No

Rosetta Stone for a sense–antisense origin of

aminoacyl tRNA synthetase classes. Mol. Biol.

Evol. 2009, 26, 445-450.

http://dx.doi.org/10.1093/molbev/msn267

[35] Trifonov, E.N. Tracing life back to elements.

Physics Life Rev. 2008, 5, 121-132.

http://dx.doi.org/10.1016/j.plrev.2008.03.001

[36] Delmotte, F. ; Rispe, C. ; Schaber, J. ; Silva, F.J. ;

Moya, A. Tempo and mode of early gene loss in

endosymbiotic bacteria from insects. BMC Evol.

Biol. 2006, 6, 56. http://dx.doi.org/10.1186/1471-2148-6-56

[37] Marais, G.A.B. ; Calteau, A. ; Tenaillon,

O.Mutation rate and genome reduction in

endosymbiotic and free-living bacteria. Genetica

2008, 134, 205-210. http://dx.doi.org/10.1007/s10709-007-

9226-6

[38] McCutcheon, J.P.; McDonald, B.R.; Moran,

N.A. Origin of an alternative genetic code in the

extremely small and gc-rich genome of a bacterial

symbiont. PLOS Genetics 2009, 5, e1000565.

http://dx.doi.org/10.1371/journal.pgen.1000565

[39] McCutcheon, J.P. The bacterial essence of tiny

symbiont genomes. Curr. Op. Microbiol. 2010, 13,

73-78. http://dx.doi.org/10.1016/j.mib.2009.12.002

[40] Nagel, G.M.; Doolittle, R.F. Phylogenetic

analysis of the aminoacyl-tRNA synthetases. J.

Mol. Evol. 1995, 40, 487-498.

http://dx.doi.org/10.1007/BF00166617

[41] Pelchat, M.; Lapointe, J. Aminoacyl-tRNA

synthetase genes of Bacillus subtilis: organization

and regulation. Biochem. & Cell Biol.-Biochimie et

Biol. Cell. 1999, 77, 343-347. http://dx.doi.org/10.1139/bcb-

77-4-343

[42] Kaguni, L.S. DNA polymerase gamma, the

mitochondrial replicase. Annu. Rev. Biochem. 2004,

73, 293–320.

http://dx.doi.org/10.1146/annurev.biochem.72.121801.161455

[43] Wolf, Y.I.; Koonin, E.V. Origin of an animal

mitochondrial DNA polymerase subunit via

lineage-specific acquisition of a glycyl-tRNA

synthetase from bacteria of the Thermus-

Deinococcus group. Trends in Genetics 2001, 17,

431-433. http://dx.doi.org/10.1016/S0168-9525(01)02370-8

[44] Seligmann, H. Mitochondrial tRNAs as light

strand replication origins: Similarity between

anticodon loops and the loop of the light strand

replication origin predicts initiation of DNA

replication. Biosystems 2010a, 99, 85-93.

http://dx.doi.org/10.1016/j.biosystems.2009.09.003

[45] Schimmel, P., De Pouplana, L.R.Footprints of

aminoacyl-tRNA synthetases are everywhere.

Trends Biochem. Sci. 2000, 25, 207-209.

http://dx.doi.org/10.1016/S0968-0004(00)01553-X

[46] Perneger, T.V. What is wrong with Bonferroni

adjustments. Brit. Med. J. 1998, 136, 1236-1238.

http://dx.doi.org/10.1136/bmj.316.7139.1236

[47] Benjamini, Y.; Hochberg, Y. Controlling the

false discovery rate: a practical and powerful

approach to multiple testing. J. Roy. Stat. Soc. B

1995, 57, 289-300.

[48] Chanda, I.; Pan, A.; Dutta, C. Proteome

composition in Plasmodium falciparum: Higher

usage of GC-rich nonsynonymous codons in highly

expressed genes. J.Mol. Evol. 2005, 61, 513-523.

http://dx.doi.org/10.1007/s00239-005-0023-5c

[49] Barton, M.D.; Delneri, D.; Oliver, S.G. ; Rattray,

M.; Bergman, C.M. Evolutionary systems biology

of amino acid biosynthetic cost in yeast. Plos One

2010, 5, e11935. http://dx.doi.org/10.1371/journal.pone.0011935

[50] O‟Donoghue, P.; Luthey-Schulten, Z. On the

evolution of structure in aminoacyl-tRNA

synthetases. Microbiol. Mol. Biol. Rev. 2003, 67,

550-573. http://dx.doi.org/10.1128/MMBR.67.4.550-573.2003

[51] Martinis, S.A.; Boniecki, M.T. The balance

between pre- and post-transfer editing in tRNA

synthetases. FEBS Letters 2010, 584, 455-459.

http://dx.doi.org/10.1016/j.febslet.2009.11.071

BIO © [2012], Copyright CCAAS Hervé Seligmann, BIO 2012, 2, 11-26

11

http://ccaasmag.org/BIO

[52] Baldwin, A.N.; Berg, P. Transfer ribonucleic

acid-induced hydrolysis of valyladenylate bound to

isoleucyl ribonucleic acid synthetase. J. Biol.

Chem. 1966, 241, 839-845.

[53] Englisch, S.; Englisch, U.; von der Haar, F.;

Cramer, F. The proofreading of hydroxyl analogues

of leucine and isoleucine by leucyl-tRNA

synthetases from E. coli and yeast. Nuc. Acids Res.

1986, 14, 7529-7538. http://dx.doi.org/10.1093/nar/14.19.7529

[54] Fersht, A.E.; Dingwall, C. An editing mechanism

for the methionyl-tRNA synthetase in the selection

of amino acids in protein synthesis. Biochemistry

1976, 18, 1250-1256.

[55] Fersht, A.E.; Kaethner, M.M. Enzyme

hyperspecificity. Rejection of threonine by the

valyl-tRNA synthetase by misacylation and

hydrolytic editing. Biochemistry 1976, 15, 3342-

3346. tp://dx.doi.org/10.1021/bi00660a026

[56] Tsui, W.-C.; Fersht, A.R. Probing the principles

of amino acid selection using the alanyl-tRNA

synthetase from Escherichia coli. Nuc. Acids Res.

1981, 9, 4627-4637. http://dx.doi.org/10.1093/nar/9.18.4627

[57] Danel, F.; Caspers, P.; Nuoffer, C.; Hartlein, M.;

Kron, M.A.; Page, M. G.P. Asparaginyl-tRNA

syntetase pre-transfer editing assay. Curr. Drug

Discov. Technol. 2011, 8, 66-75.

http://dx.doi.org/10.2174/157016311794519947

[58] Roy, H.; Ling, J.; Irnov, M.; Ibba, M. Post-

transfer editing in vitro and in vivo by the beta

subunit of phenylalanyl-tRNA synthase. EMBO J.

2003, 23, 4639-4648.

http://dx.doi.org/10.1038/sj.emboj.7600474

[59] Beuning, P.J.; Musier-Forsyth, K. Hydrolytic

editing by a class II aminoacyl-tRNA synthetase.

PNAS 2000, 97, 8916-8920.

http://dx.doi.org/10.1073/pnas.97.16.8916

[60] Dock-Bregeon, A.-C.; Sankaranarayanan, R.;

Romby, P.; Caillet, J.; Springer, M.; Rees, B.;

Francklyn, C.S.; Ehresmann, C.; Moras, D.

Transfer RNA-mediated editng in threonyl-tRNA

synthase. Cell 2000, 103, 877-884.

http://dx.doi.org/10.1016/S0092-8674(00)00191-4

[61] Fersht, A.E.; Dingwall, C. Cysteinyl-tRNA

synthetase from Escherichia coli does not need an

editing mechanism to reject serine and alanine.

High bindning energy of small groups in specific

molecular interactions. Biochemistry 1979, 18,

1245-1249. http://dx.doi.org/10.1021/bi00574a020

[62] Gruic-Sovulj, I.; Uter, N.; Bullock, T. ; Perona,

J.J. tRNA-dependent aminoacyl-adenylate

hydrolysis by a nonediting class I aminoacyl-tRNA

synthetase. J. Biol. Chem. 2005, 280, 23978-23986.

http://dx.doi.org/10.1074/jbc.M414260200

[63] Fersht, A.E.; Shindler, J.S.; Tsui, W.-C. Probing

the limits of protein-amino acid side chain

recognition with the aminoacul-tRNA synthetases.

Discrimination against phenylalanine by tyrosyl-

synthetases. Biochemistry 1980, 19, 5520-5524.

http://dx.doi.org/10.1021/bi00565a009

[64] Jakubowski, H. Misacylation of tRNAlys with

noncognate amino acids by lysyl-tRNA synthetase.

Biochemistry 1999, 38, 8088-8093.

http://dx.doi.org/10.1021/bi990629i

[65] Seligmann H.; Pollock, D.D. The ambush

hypothesis: hidden stop codons prevent off frame

gene reading. DNA & Cell Biol. 2004, 23, 701-705.

http://dx.doi.org/10.1089/dna.2004.23.701

[66] Seligmann, H. Cost minimization of ribosomal

frameshifts. J. Theor. Biol. 2007, 249, 162-167.

http://dx.doi.org/10.1016/j.jtbi.2007.07.007

[67] Seligmann, H. The ambush hypothesis at the

whole-organism level: Off frame, 'hidden' stops in

vertebrate mitochondrial genes increase

developmental stability. Comput. Biol. & Chem.

2010, 34, 80-85.

http://dx.doi.org/10.1016/j.compbiolchem.2010.03.001

[68] Seligmann, H. Avoidance of antisense,

antiterminator tRNA anticodons in vertebrate

mitochondria. Biosystems 2010, 101, 42-50.

http://dx.doi.org/10.1016/j.biosystems.2010.04.004

[69] Seligmann, H. Undetected antisense tRNAs in

mitochondrial genomes? Biol. Direct 2010, 16, 39.

http://dx.doi.org/10.1186/1745-6150-5-39

[70] Seligmann, H. Pathogenic mutations in antisense

mitochondrial tRNAs. J. Theor. Biol. 2010, 269,

287-296. http://dx.doi.org/10.1016/j.jtbi.2010.11.007

[71] Seligmann, H. Do anticodons of mysacylated

tRNAs preferentially mismatch codons coding for

the misloaded amino acid? BMC Molecular Biology

2010, 11, 41. http://dx.doi.org/10.1186/1471-2199-11-41

[72] Seligmann, H. Error compensation of tRNA

misacylation by codon-anticodon mismatch

prevents translational amino acid misinsertion.

Comp. Biol. Chem. 2011, 35, 81-95.

http://dx.doi.org/10.1016/j.compbiolchem.2011.03.001

[73] Seligmann, H. Coding constraints modulate

chemically spontaneous mutational replication

gradients in mitochondrial genomes. Cur.

Genomics, in press.

[74] Seligmann, H. Positive correlations between

molecular and morphological rates of evolution. J.

Theor. Biol. 2010, 264, 799-807.

http://dx.doi.org/10.1016/j.jtbi.2010.03.019