Protein length in eukaryotic and prokaryotic proteomes.

We analyzed length differences of eukaryotic, bacterial and archaeal proteins in relation to function, conservation and environmental factors. Comparing Eukaryotes and Prokaryotes, we found that the greater length of eukaryotic proteins is pervasive over all functional categories and involves the vast majority of protein families. The magnitude of these differences suggests that the evolution of eukaryotic proteins was influenced by processes of fusion of single-function proteins into extended multi-functional and multi-domain proteins. Comparing Bacteria and Archaea, we determined that the small but significant length difference observed between their proteins results from a combination of three factors: (i) bacterial proteomes include a greater proportion than archaeal proteomes of longer proteins involved in metabolism or cellular processes, (ii) within most functional classes, protein families unique to Bacteria are generally longer than protein families unique to Archaea and (iii) within the same protein family, homologs from Bacteria tend to be longer than the corresponding homologs from Archaea. These differences are interpreted with respect to evolutionary trends and prevailing environmental conditions within the two prokaryotic groups.

INTRODUCTION

Several studies reported that eukaryotic proteins are, on average, significantly longer than prokaryotic proteins and that, among Prokaryotes, bacterial proteins tend to be longer than archaeal proteins (1–6). As expected, these global relations do not always apply to individual protein families [see e.g. (7) for contrasting length relations of homologous families of membrane transport proteins]. It was also observed that the most conserved, functionally essential and/or highly expressed proteins tend to be longer (8–11). Biological events of diverse significance may be at the origin of the observed overall differences. A greater fraction of conserved, functionally important proteins in some proteomes could, for example, fully explain observed differences in protein lengths. Also, because proteins from different families and different functional classes have different lengths, an overall difference in protein length may reflect the variable proteome composition of different organisms (see Results). Finally, different criteria of genome annotation may affect the overall average length of the proteins in different organisms, as suggested for the overall small length difference observed between bacterial and archaeal proteins (4).

To investigate the role of conservation, functional specificities, annotation criteria and other factors in determining the average protein size in eukaryotic and prokaryotic species, we have analyzed the length of proteins from different classes of function and conservation. We were guided by the following questions: (i) are there differences in protein length among well-characterized proteins? (ii) do protein length differences appear in most or only in special functional classes of proteins? (iii) are differences in length due to proteins unique to each phylum or do they appear among proteins conserved between different phyla? (iv) do protein lengths correlate with environmental conditions and life styles? Our analyses confirmed the broad difference in length between eukaryotic and prokaryotic proteins. We were also able to conclude that the small overall length difference observed between bacterial and archaeal proteins is biologically significant and results from different evolutionary events and ecological conditions.

MATERIALS AND METHODS

Proteomes

We examined proteomes from 5 eukaryotic species, 16 archaeal species and 67 bacterial species (Table 1). For human sequences, we used the Ensembl genome database (12) Release 29.35b collection of ‘known and new’ proteins based on the NCBI 35 assembly of the human genome. These proteomes contained 104 394 eukaryotic proteins, 37 141 archaeal proteins and 191 518 bacterial proteins. Protein lengths were compared with respect to taxonomic, functional and ecological classes.

Table 1

Proteomic collections

	Species	Abbreviation
Eukaryota	Homo sapiensa	HUMAN
	Drosophila melanogaster	DROME
	Caenorhabditis elegansa	CAEEL
	Saccharomyces cerevisiae	YEAST
	Arabidopsis thalianaa	ARATH
Euryarchaeota	Pyrococcus abyssi	PYRAB
	Pyrococcus horikoshii	PYRHO
	Pyrococcus furiosus	PYRFU
	Archaeoglobus fulgidus	ARCFU
	Thermoplasma acidophilum	THEAC
	Thermoplasma volcanium	THEVO
	Methanothermobacter thermoautotrophicus	METTH
	Methanococcus jannaschii	METJA
	Methanosarcina acetivoransa	METAC
	Methanosarcina mazeia	METMA
	Methanopyrus kandleria	METKA
	Halobacterium sp. NRC-1	HALN1
Crenarchaeota	Aeropyrum pernix	AERPE
	Pyrobaculum aerophiluma	PYRAE
	Sulfolobus solfataricus	SULSO
	Sulfolobus tokodaii	SULTO
γ-Proteobacteria	Escherichia coli K12	ECOLI
	Salmonella typhimurium	SALTY
	Salmonella enterica	SALTI
	Yersinia pestis CO92	YERPE
	Shigella flexneria	SHIFL
	Wigglesworthia brevipalpisa	WIGBR
	Buchnera sp. APS	BUCAI
	Buchnera aphidicolaa	BUCAP
	Haemophilus influenzae	HAEIN
	Pasteurella multocida	PASMU
	Xanthomonas campestrisa	XANCP
	Xanthomonas axonopodisa	XANAC
	Xylella fastidiosa	XYLFA
	Pseudomonas aeruginosa	PSEAE
	Vibrio cholerae	VIBCH
	Shewanella oneidensisa	SHEON
β-Proteobacteria	Neisseria meningitides MC58	NEIMB
	Ralstonia solanacearuma	RALSO
α-Proteobacteria	Agrobacterium tumefaciens C58 Cereon	AGRT5
	Mesorhizobium loti	RHILO
	Sinorhizobium meliloti	RHIME
	Brucella melitensisa	BRUME
	Brucella suisa	BRUSU
	Caulobacter crescentus	CAUCR
	Rickettsia prowazekii	RICPR
	Rickettsia conorii	RICCN
ɛ-Proteobacteria	Helicobacter pylori 26 695	HELPY
	Campylobacter jejuni	CAMJE
Actinobacteria	Mycobacterium tuberculosis H37 Rv	MYCTU
	Mycobacterium leprae	MYCLE
	Streptomyces coelicolora	STRCO
	Corynebacterium glutamicum	CORGL
	Corynebacterium efficiensa	COREF
	Bifidobacterium longuma	BIFLO
Firmicutes	Oceanobacillus iheyensisa	OCEIH
	Bacillus subtilis	BACSU
	Bacillus halodurans	BACHD
	Staphylococcus aureus N315	STAAN
	Listeria innocua	LISIN
	Listeria monocytogenes	LISMO
	Lactococcus lactis	LACLA
	Streptococcus agalactiae 2603 V/Ra	STRA5
	Streptococcus mutansa	STRMU
	Streptococcus pneumoniae TIGR4	STRPN
	Streptococcus pyogenes M1	STRPY
	Clostridium acetobutylicum	CLOAB
	Clostridium perfringensa	CLOPE
	Thermoanaerobacter tengcongensisa	THETN
	Ureaplasma urealiticum	UREPA
	Mycoplasma genitalium	MYCGE
	Mycoplasma pneumoniae	MYCPN
	Mycoplasma pulmonis	MYCPU
	Mycoplasma penetransa	MYCPE
	Fusobacterium nucleatuma	FUSNN
Cyanobacteria	Synechocystis sp. PCC 6803	SYNY3
	Nostoc sp. PCC 7120b	ANASP
	Thermosynechococcus elongatusa	SYNEL
Chlamydiae	Chlamydia trachomatis	CHLTR
	Chlamydia muridarum	CHLMU
	Chlamydophila pneumoniae CWL029	CHLPN
Spyrochaetes	Borrelia burgdorferi	BORBU
	Treponema pallidum	TREPA
	Leptospira interrogansa	LEPIN
Others	Deinococcus radiodurans	DEIRA
	Chlorobium tepiduma	CHLTE
	Thermotoga maritima	THEMA
	Aquifex aeolicus	AQUAE

aProteins from these species are not classified in the COG database and are excluded from the functional group analyses.

bThe COG classification of proteins from this species does not follow the standard coding and has been excluded from the COG analyses.

Datasets of selected proteins

We evaluated results using the set of proteins classified in the COG (Clusters of Orthologous Groups of proteins) database (13,14) and the set of genomic proteins included in the Pfam (15–17) database of functional/structural domain alignments verified by human intervention (Pfam-A). The COG database classifies orthologous proteins in functional groups. At the onset of this study, classification of proteins into COGs was available for 2 eukaryotic proteomes (yeast and Drosophila melanogaster), 12 archaeal proteomes and 44 bacterial proteomes (Tables 1 and 2). COG data were obtained for prokaryotic organisms and yeast from the corresponding tables (*.ptt) available from the NCBI genomes ftp site (). Data for D.melanogaster were obtained from the classification table available at the COG website (). All proteomes included in our analysis are also represented in the Pfam-A database.

Table 2

Median protein lengths in eukaryotic, bacterial and archaeal organisms

Speciesa	All species		Classified in COG		Classified in Pfam-A
	Numberb	Medianc	Numberb	Medianc	Numberb	Medianc
Eukarya	104 394	361	5177	471	71 584	419
HUMAN	33 869	375	–	–	21 686	416
DROME	14 226	373	3092	492	13 091	475
CAEEL	21 124	344	–	–	13 316	391
YEAST	6315	379	2085	438	3953	448
ARATH	28 860	356	–	–	19 538	407
Bacteria	191 541	267	83 513	304	131 915	306
ECOLI	4289	278	3289	309	3483	303
SALTY	4553	271	3408	303	3527	300
SALTI	4767	253	3258	300	3118	300
YERPE	4083	268	2991	299	3003	304
SHIFL	4180	261	–	–	2613	304
WIGBR	654	268	–	–	571	291
BUCAI	574	282	558	284	544	285
BUCAP	545	279	–	–	536	285
HAEIN	1709	262	1470	286	750	314
PASMU	2014	286	1740	302	780	289
XANCP	4181	291	–	–	2976	325
XANAC	4312	286	–	–	3056	326
XYLFA	2832	201	1549	305	1544	305
PSEAE	5565	291	4355	310	4309	309
VIBCH	3828	259	2794	315	2731	312
Chromosome 1	2736	273	2133	314
Chromosome 2	1092	225	661	316
SHEON	4778	245	–	–	2913	308
NEIMB	2025	239	1448	291	1310	305
RALSO	5116	276	–	–	3518	310
Chromosome 1	3440	271
Chromosome 2	1676	296
AGRT5	5402	280	3984	316	4062	307
Circular chr.	2785	258	2098	305
Linear chr.	1876	302	1424	329
Plasmids	741	273	462	316
RHILO	7275	269	5184	305	5107	303
Chromosome	6746	270	4888	304
Plasmids	529	243	296	327
RHIME	6205	281	4614	312	4669	308
Chromosome	3341	276	2602	302
Plasmid A	1294	265	890	310
Plasmid B	1570	303	1122	330
BRUME	3198	263	–	–	2322	300
Chromosome 1	2059	252
Chromosome 2	1139	279
BRUSU	3264	254	–	–	2351	301
Chromosome 1	2116	239
Chromosome 2	1148	278
CAUCR	3737	275	2551	317	2686	312
RICPR	834	283	687	295	672	299
RICCN	1374	173	861	247	769	264
HELPY	1566	266	1083	303	1052	315
CAMJE	1634	268	1309	294	1197	298
MYCTU	3918	287	2554	322	2213	326
MYCLE	1605	282	1145	326	1138	324
STRCO	7897	278	–	–	5330	317
CORGL	2993	275	1954	307	1985	314
COREF	2950	287	–	–	1961	323
BIFLO	1729	321	–	–	1286	341
OCEIH	3496	261	–	–	2583	295
BACSU	4100	256	2818	298	2974	297
BACHD	4066	261	2838	303	2916	300
STAAN	2625	257	1801	300	1542	293
LISIN	3043	255	2176	289	2234	291
LISMO	2846	267	2206	289	2211	292
LACLA	2266	251	1602	288	1690	281
STRA5	2124	254	–	–	1480	290
STRMU	1960	250	–	–	1445	282
STRPN	2043	243	1465	287	1409	291
STRPY	1696	263	1178	294	1159	299
CLOAB	3848	262	2487	298	2634	299
CLOPE	2723	268	–	–	1997	303
THETN	2588	269	–	–	1903	306
UREPA	614	286	409	298	395	303
MYCGE	484	292	384	292	375	304
MYCPN	677	286	407	299	507	299
MYCPU	782	297	489	302	498	320
MYCPE	1037	304	–	–	664	315
FUSNN	2067	261	–	–	1432	303
SYNY3	3169	274	2141	318	2344	306
ANASP	6129	256	–	–	3600	320
SYNEL	2475	272	–	–	1759	315
CHLTR	894	289	615	316	639	327
CHLMU	916	290	644	321	641	330
CHLPN	1052	289	646	324	716	333
BORBU	1637	220	635	318	981	265
Chromosome	850	286	–	–
Plasmids	787	179	–	–
TREPA	1031	293	708	331	691	337
LEPIN	4727	207	–	–	2243	309
Chromosome 1	4360	206
Chromosome 2	367	223
DEIRA	3182	264	2249	303	2050	307
Chromosome 1	2629	257	1873	294
Chromosome 2	368	304	265	347
Plasmids	185	303	111	336
CHLTE	2252	239	–	–	1431	311
THEMA	1846	284	1509	303	1459	304
AQUAE	1560	272	1321	291	1231	297
Archaea	37 141	247	18 219	283	24 067	288
PYRAB	1765	265	1450	281	1407	282
PYRHO	1801	257	1398	283	1312	285
PYRFU	2065	253	1627	273	1477	281
ARCFU	2420	243	1887	270	1720	276
THEAC	1482	269	1233	293	1083	301
THEVO	1499	259	1247	287	1074	304
METTH	1869	242	1382	273	1325	277
METJA	1770	241	1298	266	1260	272
METAC	4540	256	–	–	2677	306
METMA	3371	255	–	–	2141	294
METKA	1691	257	–	–	1067	272
HALN1	2622	242	1746	297	1471	303
AERPE	1840	239	1191	293	1067	301
PYRAE	2603	208	–	–	1411	267
SULSO	2977	251	1983	294	1917	293
SULTO	2826	226	1777	284	1658	279

aSee Table 1 for abbreviations.

bNumber of proteins in each set.

cMedian length.

Statistical significance evaluations

We compared median protein lengths (the midpoint of all lengths arranged in order of magnitude) rather than average lengths between two sets of proteins to reduce the effect of outliers. The statistical significance of the difference in median length between the proteins of two sets was evaluated estimating the distribution of median length differences between samples created by randomly redistributing all sequences from the two sets into two new sets of the original sizes. For each determination, from 200 to 1000 independent data shufflings were implemented. We highlight the differences in median length observed in <1% of all shuffled samples (P < 0.01, boldfaced in the tables) and observed in the range 1–10% of all shuffled samples (0.01 ≤ P < 0.1, underlined in the tables). The significance of asymmetric counts of longer or shorter protein families comparing two evolutionary groups was evaluated on the basis of exact binomial probabilities or of their normal approximation.

RESULTS

We compared medians from the protein length distributions of eukaryotic and prokaryotic proteomes (Table 1) from the sets of proteins characterized in COGs (Clusters of Orthologous Groups) (13,14) or included in Pfam-A (15–17) alignments. We then used the COG database classification to investigate median protein length relations separately for different functional classes of proteins, among protein groups unique to Eukarya, Bacteria and Archaea and among protein groups shared by Bacteria and Archaea or Eukaryotes and Prokaryotes. We then compared proteins included in the Pfam database to evaluate the influence of domain structure in determining protein length differences between domains. Finally, we evaluated the influence of growth temperature on protein length.

Protein lengths across proteomes

Median protein lengths in all species and for the collections of 5 eukaryotic species, 16 archaeal species and 67 bacterial species are shown in Table 2. For prokaryotic organisms, the median lengths of individual chromosomes and of collections of plasmids are also indicated. The median length of the proteins annotated among Eukaryotes (361 amino acids) is much higher than in Bacteria (267 amino acids) and this in turn is higher than in Archaea (247 amino acids). These differences are significant (P < 0.001) and are in agreement with the previously published results (1,2,5). The median difference in length between Prokaryotes and Eukaryotes is unambiguous, whereas among Prokaryotes the distributions of median lengths of individual bacterial and archaeal species overlap. The median lengths of bacterial species that are higher than the median archaeal length are underlined in Table 2.

In evaluating the distribution of protein lengths in eukaryotic and prokaryotic proteomes, a major concern is the reliability of the dataset. In fact, eukaryotic and prokaryotic proteomes contain a large fraction of proteins of uncertain determination (14). Proteins described as putative, hypothetical, predicted, poorly characterized, uncharacterized or unknown in genome annotations of Eukaryotes, Archaea and Bacteria constitute 56.1, 51.5 and 51.7%, respectively, of all annotated proteins. A more reliable set of proteins is provided by the COG database (13,14), which identifies proteins conserved in different organisms and characterizes them in functional classes, and by the Pfam database (15–17) of well-curated alignments (Pfam-A), which characterizes proteins by the presence of conserved domains. The COG database provides a reasonable compromise between reliability of annotation and size of dataset. Although based on automated procedures, it has the advantage of using consistent criteria with no obvious biases over a large set of organisms. The Pfam-A database provides the advantage of human supervision. COG orthologs comprise 25.0% of all annotated proteins from DROME and YEAST (see Table 1 for abbreviations), 72.5% of all proteins from bacterial proteomes and 73.1% of all proteins from archaeal proteomes, whereas proteins in Pfam including domains verified by human intervention (Pfam-A) comprise 68.6% of all eukaryotic proteins, 68.9% of all bacterial proteins and 64.8% of all archaeal proteins. The median lengths of proteins that are classified in COG or in Pfam-A (Table 2) confirm the significant difference between the three domains observed for complete proteomes, yielding Eukaryotes ≫ Bacteria > Archaea.

Protein length differences in functional classes

We analyzed protein lengths within functional classes of proteins defined in the COG database (Table 3). We noticed (Figure 1) that in all three phylogenetic domains (Eukarya, Bacteria and Archaea) proteins involved in the broad functional classes of cellular processes and metabolism are the longest in median value, followed by the sets of poorly characterized proteins, by the group of proteins involved in information storage and processing and finally by the non-conserved proteins that are not classified in the COG database. The representation of these broad functional classes in Eukarya, Bacteria or Archaea is shown in Figure 2. Among eukaryotic proteins, only those with prokaryotic homologs are classified in the COG database. These comprise only 26% of the eukaryotic proteomes, whereas 74% of the eukaryotic proteomes (yeast and Drosophila) are composed of relatively shorter proteins not found in Prokaryotes. Among Prokaryotes, the long proteins functioning in cellular processes are present in higher proportions in Bacteria than in Archaea, whereas Archaea have a higher proportion of the shorter, poorly characterized proteins (Figure 2).

Table 3

COG functional classification

Information storage and processing (Isp)
J	Translation, ribosomal structure and biogenesis
K	Transcription
L	DNA replication, recombination and repair
Cellular processes (Cp)
D	Cell division and chromosome partitioning
O	Posttranslational modification, protein turnover, chaperones
M	Cell envelope biogenesis, outer membrane
N	Cell motility and secretion
P	Inorganic ion transport and metabolism
T	Signal transduction mechanisms
Metabolism (Me)
C	Energy production and conversion
G	Carbohydrate transport and metabolism
E	Amino acid transport and metabolism
F	Nucleotide transport and metabolism
H	Coenzyme metabolism
I	Lipid metabolism
Q	Secondary metabolites biosynthesis, transport and catabolism
Poorly characterized (Pc)
R	General function prediction only
S	Function unknown

Figure 1

Relative median length of proteins within major functional classes in Eukaryotes (Euk), Bacteria (Bac) and Archaea (Arc). Lengths are normalized by the global median length within each phylum. Major functional classes follow the definition in COG (see also Table 3): Isp, information storage and processes; Cp, cellular processes; Me, metabolism; Pc, poorly characterized. N.C. signifies proteins not classified in the COG database.

Figure 2

Representation in eukaryotic and prokaryotic proteomes of proteins belonging to the major functional classes. Isp, information storage and processes; Cp, cellular processes; Me, metabolism; Pc, poorly characterized. N.C. signifies proteins not classified in the COG database.

Median protein lengths within all COG functional classes are shown in Table 4. Eukaryotic proteins feature a median length greater than prokaryotic proteins for every functional class. The greatest length difference, of ∼400 amino acids, is observed among proteins functioning in DNA replication, recombination and repair. For most other functional classes, differences in length are in the approximate range of 100–200 amino acids. The shortest eukaryotic proteins are those not classified in the COG database. Even these, however, are >200 amino acids longer than the corresponding prokaryotic proteins and are also longer than prokaryotic proteins from each major functional class (with overall median length ∼300 amino acids).

Table 4

Median protein lengths in COG functional classes

COG classa	Tot grpb	All COGs
		EUK			BAC			ARC			p(ΔBA)e
		Grpb	Seqc	Medd	Grpb	Seqc	Medd	Grpb	Seqc	Medd
Isp	541	258	1157	399	434	17 621	252	345	4034	218	0.000
J	220	175	623	296	155	6110	208	155	1736	205	0.231
K	139	31	295	444	114	6122	240	75	1027	156	0.000
L	188	55	318	723	167	5389	315	116	1271	321	0.309
Cp	689	161	1355	507	667	19 275	325	279	2720	297	0.000
D	35	8	33	439	34	936	346	12	181	282	0.000
O	116	47	421	370	110	3167	270	62	584	246	0.012
M	166	24	82	449	166	4713	355	49	550	341	0.011
N	131	13	43	508	121	3194	320	38	368	292	0.077
P	167	50	338	538	164	4014	314	91	732	294	0.003
T	89	21	439	605	87	3251	323	28	305	253	0.001
Me	1005	438	2034	494	970	31 258	338	640	6625	331	0.000
C	228	85	377	480	210	5149	366	160	1620	346	0.000
G	178	61	456	519	175	6478	371	84	929	372	0.506
E	240	117	577	515	227	8268	356	163	1632	353	0.240
F	89	61	186	376	82	2396	274	65	586	244	0.002
H	147	73	184	393	137	3434	307	106	911	283	0.000
I	80	47	240	518	78	2658	313	41	515	360	0.000
Q	68	15	225	505	63	2875	294	22	432	261	0.000
Pc	1372	207	1057	444	1167	15 359	262	645	4840	246	0.000
R	501	146	962	459	423	9003	291	282	2806	274	0.000
S	897	64	95	318	764	6355	210	368	2035	202	0.018
Chr	2201	845	4791	481	2051	68 154	315	1261	13 379	299	0.000
All	3482	1027	5177	471	3162	83 513	304	1894	18 219	283	0.000
N.C.	n.a.	n.a.	15 492	365	n.a.	31 739	158	n.a.	6717	157	0.342

aChr = Isp + Cp + Me; All = Chr + Pc; N.C. = Not classified in COGs. See Table 3 for other class symbols.

bNumber of COG groups within each class.

cNumber of sequences.

dMedian length.

eProbability of the median length difference observed between bacterial and archaeal sequences.

Table 4 also shows that the overall relation between bacterial and archaeal proteins (bacterial longer than archaeal) is also valid within most individual functional classes. In Table 4, significant (P ≤ 0.01) length differences between bacterial and archaeal medians are shown in boldface and less significant differences (0.01 < P ≤ 0.10) are underlined. The most pronounced differences occurred for proteins functioning in transcription (240 amino acids median in Bacteria versus 156 amino acids median in Archaea), in signal transduction (323 amino acids versus 253 amino acids), and in cell division and chromosome partitioning (346 amino acids versus 282 amino acids). Smaller differences occurred for proteins in nucleotide transport and metabolism (274 amino acids versus 244 amino acids) or in secondary metabolite biosynthesis, transport and catabolism (294 amino acids versus 261 amino acids). Proteins longer in Bacteria than in Archaea also stand out in most of the other functional classes, with median length differences in the range of 5–20 amino acids. Equivalent lengths were only observed in the functional classes of translation, ribosomal structure and biogenesis, DNA replication, recombination and repair, amino acid transport and carbohydrate transport and metabolism. A reverse relation, with archaeal proteins longer than bacterial proteins, was only found for proteins involved in lipid metabolism (313 amino acids in Bacteria versus 360 amino acids in Archaea).

The comparison of well-characterized proteins from Bacteria and Archaea confirms the biological significance of the overall length differences between bacterial and archaeal proteins. In this respect, it is interesting to emphasize that no length differences were observed for proteins that are not classified in COGs (Table 4, 157 amino acids in Archaea and 158 amino acids in Bacteria). This result, together with the observation that non-classified proteins are equally frequent in Bacteria and Archaea (Figure 2), indicates that the length difference between bacterial and archaeal proteins cannot be ascribed to over-annotation of short open reading frames in Archaea as previously suggested (4).

Length differences of proteins unique to Eukarya, Bacteria or Archaea

A relevant question is whether the significant protein length differences observed in most functional classes are due to proteins unique to each class of organisms or are also present among homologs conserved between Eukaryotes and Prokaryotes or between Archaea and Bacteria. Here, we compared the lengths of proteins unique to each phylogenetic group. As mentioned before, in Eukaryotes proteins without prokaryotic orthologs were not classified in the COG database. These were shorter than those classified in COGs (medians 365 amino acids versus 471 amino acids) but still longer than the median bacterial (267 amino acids) or archaeal proteins (247 amino acids). Among Prokaryotes, 1588 families of protein orthologs have representatives only in Bacteria and 320 families are unique to Archaea (Table 5). Unique families of proteins comprise 23.9 and 10.9% of the bacterial and archaeal proteomes, respectively. Table 5 shows a substantial length difference (73 amino acids) between proteins unique to Bacteria (median 275 amino acids) and proteins unique to Archaea (median 202 amino acids). This difference was emphasized among well-characterized proteins involved in information storage and processing (Isp), metabolism (Me) or cellular processes (Cp), which were 58.6% longer in Bacteria than in Archaea (295 amino acids versus 186 amino acids). The overall length difference between unique bacterial and archaeal proteins is determined by two factors: (i) within all major functional classes of characterized proteins, unique bacterial proteins are ∼40% longer than unique archaeal proteins and (ii) unique bacterial proteins are mostly (73%) represented by metabolic proteins (Me) or proteins involved in cellular processes (Cp), which tend to be longer in all organisms (Figure 1), whereas unique archaeal proteins are mostly (79%) represented by generally shorter proteins involved in information storage and processing (Isp).

Table 5

Median lengths of proteins unique to Bacteria or Archaea among Prokaryotesa

COG class	BAC			ARC			p(Δ)
	Grp	Seq	Med	Grp	Seq	Med
Isp	192	5771	240	103	1161	172	0.000
J	61	2178	166	61	653	145	0.009
K	62	1990	256	23	281	136	0.000
L	71	1607	276	20	227	369	0.000
Cp	406	8432	312	18	165	219	0.000
D	23	464	384	1	2	253	0.516
O	53	984	234	5	39	368	0.008
M	117	2492	348	0	0	n.a.	n.a.
N	92	1990	269	9	99	197	0.000
P	74	1061	386	1	8	376	0.611
T	61	1447	254	2	17	287	0.402
Me	359	6757	331	29	145	237	0.000
C	64	1009	383	14	58	312	0.122
G	91	2010	339	0	0	n.a.	n.a.
E	70	1062	389	6	25	273	0.000
F	20	440	214	3	29	189	0.053
H	36	862	286	5	27	181	0.000
I	37	927	300	0	0	n.a.	n.a.
Q	42	447	374	1	6	263	0.120
Pc	698	6744	228	176	1246	224	0.221
R	193	2520	284	52	471	273	0.105
S	524	4247	196	128	776	191	0.215
Chr	940	20 945	295	150	1471	186	0.000
All	1588	27 570	275	320	2709	202	0.000

aSee Table 3 and footnotes of Table 4 for abbreviations.

Length differences of conserved proteins

We distinguished eukaryotic proteins with prokaryotic orthologs between those with only archaeal orthologs, those with only bacterial orthologs and those conserved in all three domains (see Supplementary Table S1). About 79% of the eukaryotic sequences with orthologs in Archaea but absent from Bacteria are involved in information storage and processing (Isp) and among these 60% function in translation. In contrast, sequences with orthologs present in Bacteria but absent from Archaea, or present in all three domains, are rather evenly distributed among functional classes, with a higher frequency (∼41%) of metabolic proteins (Me). In virtually all classes the median length of conserved eukaryotic proteins is significantly greater than the median length of the corresponding prokaryotic orthologs. Notably, eukaryotic proteins with bacterial orthologs are considerably longer (∼79%) than those with archaeal orthologs. This asymmetry applies to all four major functional categories (information storage and processing, cellular processes, metabolism, poorly characterized) and to most individual classes. The only exceptions are proteins functioning in post-translational modification, turnover and chaperoning and proteins functioning in transport and metabolism of inorganic ions.

We have shown that the overall length differences observed between bacterial and archaeal proteins are amplified among proteins unique to each of the two domains. It remains to be established whether length differences are also present among proteins conserved between Bacteria and Archaea. According to the COG database, 1574 families of orthologs are conserved between Bacteria and Archaea. These comprise 48.5 and 62.2% of the average bacterial and archaeal proteome, respectively. The results shown in Table 6 indicate that orthologs conserved between Bacteria and Archaea are also longer in Bacteria, with an overall significant length difference of 17 amino acids (compared with an overall length difference of 73 amino acids between unique proteins). Qualitatively similar length differences prevail within most functional groups of proteins.

Table 6

Median lengths of orthologs shared by Bacteria and Archaeaa

COG class	# Grp	BAC		ARC		p(Δ)
		Seq	Med	Seq	Med
Isp	242	11 850	260	2873	246	0.001
J	94	3932	245	1083	255	0.105
K	52	4132	221	746	157	0.000
L	96	3782	336	1044	306	0.004
Cp	261	10 843	330	2555	302	0.000
D	11	472	318	179	283	0.021
O	57	2183	290	545	246	0.002
M	49	2221	359	550	341	0.000
N	29	1204	388	269	357	0.023
P	90	2953	302	724	293	0.058
T	26	1804	358	288	252	0.000
Me	611	24 501	340	6480	332	0.000
C	146	4140	364	1562	346	0.011
G	84	4468	390	929	372	0.020
E	157	7206	351	1607	354	0.312
F	62	1956	310	557	257	0.000
H	101	2572	312	884	285	0.000
I	41	1731	322	515	360	0.019
Q	21	2428	281	426	261	0.000
Pc	469	8614	281	3595	253	0.000
R	230	6483	294	2335	274	0.000
S	240	2108	234	1259	206	0.000
Chr	1111	47 209	321	11 908	311	0.000
All	1574	55 943	315	15 510	298	0.000

aSee Table 3 and footnotes of Table 4 for abbreviations.

To determine whether these overall differences represent a trend common to most families or reflect strong asymmetries present in few families, we counted how many of the 1574 shared families of orthologs have a greater median length in Bacteria and how many are longer in Archaea. Table 7 shows that among all shared protein families almost twice as many (63.3%) involved longer sequences in Bacteria than in Archaea (boldface or underlined in Table 7 when the differences in counts have probability P ≤ 0.01 or 0.01 < P ≤ 0.10, respectively, see Materials and Methods). For comparison, 89.9% of 1027 protein families shared by Eukaryotes and Prokaryotes were longer in Eukaryotes (data not shown). Similar proportions (2:1) of longer bacterial proteins were observed within most functional categories. We also partitioned shared protein families by absolute length difference, i.e. distinguishing among all protein families those with an absolute length difference between bacterial and archaeal orthologs of <20 amino acids, in the range of 21–100 amino acids or >100 amino acids (Table 7). Remarkably, within each of these classes we observed similar proportions (2:1) of families with longer bacterial sequences. Similar results were obtained for a wide variety of other length intervals (data not shown). These results suggest that there are evolutionary pressures for proteins of different length in Archaea and Bacteria, acting both on protein families unique to each lineage and on proteins conserved between the two lineages. Among the latter, length differences seem both compatible with addition/deletion in different families of orthologs of small sequence element (e.g. <20 amino acids) or of entire structural domains (e.g. >100 amino acids).

Table 7

Median length relations of orthologs conserved between Bacteria and Archaeaa

COG class	Bacteria versus Archaea
	# Grp	Bacteria > Archaea		Archaea > Bacteria
		#	%	#	%
Isp	242	145	59.9	92	38.0
J	94	46	48.9	47	50.0
K	52	35	67.3	16	30.8
L	96	64	66.7	29	30.2
Cp	261	163	62.5	96	36.8
D	11	11	100.0	0	0.0
O	57	34	59.6	23	40.4
M	49	33	67.3	15	30.6
N	29	16	55.2	12	41.4
P	90	50	55.6	40	44.4
T	26	20	76.9	6	23.1
Met	611	409	66.9	195	31.9
C	146	95	65.1	50	34.2
G	84	60	71.4	24	28.6
E	157	107	68.2	46	29.3
F	62	39	62.9	21	33.9
H	101	67	66.3	34	33.7
I	41	27	65.9	14	34.1
Q	21	15	71.4	6	28.6
Pc	469	284	60.6	178	38.0
R	230	145	63.0	81	35.2
S	240	140	58.3	97	40.4
Chr	1111	717	64.5	380	34.2
All	1574	997	63.3	556	35.3
1–20 amino acids	750	469	62.5	281	37.5
21–100 amino acids	567	366	64.6	201	35.4
>100 amino acids	246	162	65.9	74	34.1

a# is the number of COG groups within each collection; % is the corresponding percent of COG groups compared to the total within each class (# Grp). See text, Table 3 and footnotes of Table 4 for other abbreviations.

Domain composition of eukaryotic, bacterial and archaeal proteins

To substantiate our speculations that the protein length relations between Eukarya, Archaea and Bacteria may be influenced by their composition in structural domains, we analyzed the domain composition of genomic proteins included in the Pfam-A database. In particular, we calculated the average number of domains per eukaryotic, bacterial or archaeal protein, and their median length. We repeated the analysis for domains classified in Pfam-A and for all domains classified in Pfam-A or Pfam-B (automatically generated). Table 8 shows that eukaryotic proteins tend to include substantially more domains per protein than bacterial proteins (2.16 versus 1.46 Pfam-A domains and 3.48 versus 2.09 Pfam-A+B domains, respectively). They also show that bacterial proteins tend to include marginally more domains than archaeal proteins (1.46 versus 1.39 Pfam-A and 2.09 versus 2.00 Pfam-A+B). The median length of each Pfam-A domain was substantially similar in the three phylogenetic groups (179–188 amino acids) but greater differences were observed including also Pfam-B domains, with domain lengths following the same overall ordering observed between proteins (eukaryotic domain 257 amino acids > bacterial domain 217 amino acids > archaeal domain 205 amino acids).

Table 8

Structural/functional domains in eukaryotic, bacterial and archaeal proteomes

Database	Domains	EUK	BAC	ARC
Pfam-A	Total number	154 979	192 680	33 372
	Mean number/seq.	2.16	1.46	1.39
	Median length/amino acids	185	188	179
Pfam-A + Pfam-B	Total number	249 163	275 630	48 060
	Mean number/seq.	3.48	2.09	2.00
	Median length/amino acids	257	217	205

Protein length relations of mesophilic versus thermophilic Prokaryotes

Only 5 of the 67 bacterial proteomes in our collection are from thermophilic organisms: CHLTE [optimal growth temperature (OGT) 48°C], SYNEL (55°), THETN (75°C), THEMA (80°C) and AQUAE (96°C), of which only proteins from THEMA and AQUAE were classified in COGs. In contrast, only 3 of the 16 archaeal proteomes were from mesophilic organisms (HALN1, METAC and METMA), of which only proteins from HALN1 were classified in COGs. The archaeal thermophiles live under OGTs ranging from 60°C (THEAC and THEVO) to >100°C (PYRAB, PYRHO and PYRAE). Considering the disproportionate number of mesophilic organisms among Bacteria and the disproportionate number of thermophilic organisms among Archaea, the median protein length differences observed between Bacteria and Archaea may reflect differences between mesophiles and thermophiles. In the following analyses, we evaluated the relation of OGT with protein length. We then examined median protein lengths partitioning prokaryotic species into separate sets of bacterial mesophiles (BM, 62 species, 42 species for COG comparisons), archaeal mesophiles (AM, 3, 1), bacterial thermophiles (BT, 5, 2) and archaeal thermophiles (AT, 13, 11).

We compared the median protein length in a variety of datasets with the OGT of bacterial and archaeal organisms. In all comparisons we found a negative correlation of the protein length with OGT. The correlation was not significant for comparisons of all proteomic proteins (P = 0.239) and marginally significant for proteins classified in the COG database (P = 0.069). However, we found a highly significant correlation (P = 0.0018) for proteins classified in Pfam-A (Figure 3). Figure 3 also shows an overall tendency for proteins of bacterial thermophiles to be longer than proteins from archaeal thermophiles for all OGTs.

Figure 3

Relation of median length of genomic proteins included in the Pfam-A database of curated alignments and OGT of the corresponding organism. Each point represents the median protein length within each bacterial (red) or archaeal (green) species.

In Table 9, we compared the overall median protein lengths among thermophiles or among mesophiles from groups of orthologs conserved between Bacteria and Archaea (Shared) and of orthologs unique to each of the two lineages (Unique). In Table 10, we counted the number of shared families that are longer in bacterial or archaeal species of similar thermophilicity (BM versus AM and BT versus AT) and, within the same domain, the number of protein families that are longer in thermophiles or mesophiles (BT versus BM and AT versus AM). Among conserved proteins, there is only a small, non-significant length difference (0.01 < P ≤ 0.10, underlined in Table 9) between proteins from bacterial and archaeal species of similar temperature preferences (Table 9). Consistently, among mesophiles or thermophiles, the number of shared orthologous groups that are longer in Bacteria or Archaea is equally distributed between the two groups (Table 10). In contrast, comparisons of mesophilic versus thermophilic species within Bacteria or within Archaea show in both domains significantly more orthologous groups longer in mesophiles compared with thermophiles (P ≤ 0.01, boldface in Table 10).

Table 9

Median length of orthologous groups from mesophilic or thermophilic Prokaryotesa

Type of ortholog	# of groups	Set I			Set II			P(Δ)
		Bacteria	Seq	Med	Archaea	Seq	Med
Shared	977	BT	2036	306	AT	10 664	298	0.071
	860	BM	36 891	317	AM	1481	309	0.064
Unique	465 831	BT	794	267	AT	5810	244	0.002
	2 250 187	BM	43 791	294	AM	265	206	0.000

a# of groups is the number of COG groups in each comparison (the two different numbers shown for comparisons of Unique orthologs correspond to the unique groups found in Bacteria and Archaea, respectively. Pairwise comparisons between bacterial thermophiles (BT), archaeal thermophiles (AT), bacterial mesophiles (BM) and archaeal mesophiles (AM). See text and footnote of Table 4 for other abbreviations.

Table 10

Number of conserved orthologous groups longer in bacterial or archaeal thermophiles and mesophilesa

Set I	Set II	# Grp	Set I > Set II		Set II > Set I
			#	%	#	%
BT	AT	977	484	49.5	478	48.9
BM	AM	860	411	47.8	435	50.6
BM	BT	1390	947	68.1	422	30.4
AM	AT	961	585	60.9	361	37.6

aSee text and footnote of Table 7 for abbreviations.

A different picture emerges comparing the median lengths of orthologous groups of proteins that are present only in Bacteria or only in Archaea of similar temperature preferences (Table 9, Unique). In contrast to shared proteins, proteins unique to Bacteria tend to be significantly longer than proteins unique to Archaea (P ≤ 0.01, boldface in Table 9) independently of temperature, following the ordering: bacterial mesophiles (294 amino acid median) > bacterial thermophiles (267 amino acid median) > archaeal thermophiles (244 amino acid median) > archaeal mesophile (206 amino acid median).

DISCUSSION

It has been consistently reported (1–7) that eukaryotic proteins are generally longer than bacterial proteins and these in turn are marginally longer than archaeal proteins. We have concluded that the overall protein length differences observed between Eukarya, Bacteria and Archaea represent a genuine trend among the three Domains involving most functional groups of proteins. Well-characterized eukaryotic proteins are ∼55% longer (median values) than bacterial proteins. COG groups showing the greatest length difference between eukaryote and prokaryote proteins are listed in Supplementary Table S2. These proteins span all functional groups, including membrane-associated, cytoplasmic and nuclear proteins. What can account for the difference in length between eukaryotic and prokaryotic proteins? A greater length of eukaryotic proteins may reflect upon the greater complexity of the eukaryotic cell compared with the prokaryotic cell. Our analyses confirm previous suggestions (5,8) that eukaryotic proteins have a strong tendency to fuse into multi-domain and multi-functional units. It has been proposed that eukaryotic proteins are expanded by the addition of sequence motifs or structural domains that act as functional regulators (2). Compared with prokaryotic proteins, it may also be more difficult for eukaryotic proteins to associate in a crowded cytoplasmic space partitioned by a complex array of compartments. Fusion of interacting single-function proteins into multi-domain units may facilitate the interaction between functional units and diminish the need to produce proteins in greater amounts to achieve appropriate concentrations of their complexes. It seems plausible that the acquisition of longer, multi-functional proteins in eukaryote organisms may have evolved concomitant with the acquisition of multi-exon proteins. However, we do not find that yeast genes, with relatively few introns (of which many are in ribosomal proteins), code for proteins that are shorter than in more intron-rich genomes.

In contrast to the substantial length difference between eukaryotic and prokaryotic proteins, the length difference between median bacterial and archaeal proteins is relatively small and has been described as an artifact of genome annotation (4). In fact, predicted uncharacterized proteins tend to be shorter than conserved, functionally characterized proteins (8–11) and the inclusion in proteome annotations of a high proportion of putative proteins could bias the overall protein length (4). However, we found that putative proteins in Bacteria and Archaea have similar lengths and comprise similar proportions of the bacterial and archaeal proteomes. In contrast, we found that well-characterized proteins from most functional categories are significantly longer in Bacteria than in Archaea.

Our analyses suggest that differences in environmental temperature may govern the length biases observed between bacterial (mostly mesophilic) and archaeal (mostly thermophilic) orthologs. The small reduction in the median length of archaeal versus bacterial orthologs (median 17 amino acids) is consistent with the length reduction of disordered loops or N-terminal and C-terminal tails that presumably confers extra stability to proteins subject to high temperatures (18–23).

Although the differences in median length between bacterial and archaeal homologs are relatively small, length differences within specific orthologous groups suggest that a substantial proportion of bacterial orthologs differ from archaeal orthologs by the addition of entire structural domains. Groups of orthologs with the greatest difference between bacterial and archaeal median protein lengths are listed in Supplementary Table S3. Insertion or deletion of domains is common among bacterial or archaeal orthologs, as attested to by the wide range of protein lengths observed within many groups of orthologous sequences. In fact, in ∼70% of all prokaryotic groups of orthologs, bacterial or archaeal sequences span length differences of >100 amino acids. Our results suggest that bacterial proteins are more prone than archaeal proteins to domain-fusion.

We speculate that proteins tend to be longer among Bacteria in relation to the protected environment of many bacterial species. High temperatures and harsh environmental conditions may have instead favored the evolution of shorter, less complex and more stable proteins in archaeal species. In contrast to the modest length difference between shared proteins, proteins unique to Archaea are substantially (∼59%) shorter than proteins unique to Bacteria. Among the shortest proteins unique to Archaea are many ribosomal proteins, subunits of RNA polymerase, transcriptional regulators and RNA-binding proteins, whereas the longest archaeal-unique proteins function in DNA replication and protein turnover. Among Bacteria, the shortest unique proteins also include ribosomal proteins, but the majority are proteins contributing in cell envelope biogenesis, carbohydrate transport, cell motility and secretion, signal transduction mechanisms, amino acid transport and metabolism and ion transport and metabolism.

We cannot account for the length difference between proteins unique to Bacteria or Archaea on the basis of environmental temperature. In fact, unique sequences tend to be longer in Bacteria also when comparing only thermophiles or only mesophiles. Functional and structural constraints limit protein adaptation to different environments. In particular, the divergence of homologous proteins with respect to bacterial and archaeal lineages must have been constrained by the functionality achieved in their common ancestor. The modest length differences that we observe among most protein groups shared by Bacteria and Archaea may reflect such constraints. In contrast, proteins that are present only in Bacteria or in Archaea have evolved their functionality (or have survived) in only one of the two evolutionary groups. The length difference that we observe between proteins unique to Bacteria or Archaea may relate to unconstrained adaptations to different bacterial or archaeal conditions.

Among the prokaryotic species in our collection, archaeal species are free-living, mostly in extreme environments, whereas bacterial species are adapted to a wide variety of ecological conditions and include endocellular parasites, obligate and facultative parasites of animals and plants, and species that spend different proportions of their life-cycle in free-environments (aqueous or terrestrial) where they are subject to stresses of a variable nature and amplitude. Small proteomes of parasitic organisms, such as Mycoplasma and Chlamydia, that live in a protected environment seem to have longer median proteins than most other species. Also, proteins tend to be longer in the obligate intracellular parasite Rickettsia prowazekii, but not in its close relative Rickettsia conorii. Long proteins also characterize Gram-positive Actinobacteria (high G+C Gram-positives), whereas short proteins are predominant in all Firmicutes (low G + C Gram-positives) except Mycoplasmas (Mollicutes). Different groups of Proteobacteria exhibit great protein length variability.

Factors other than growth temperature are likely to influence the median protein length among Prokaryotes and may underlie its great variability particularly among bacterial species. Less complex and more stable proteins can be expected among free-living species exposed to more intense stresses and environmental fluctuations. It has also been suggested that minimization of costs related to amino acid usages is a significant force in protein evolution (24,25). Minimization of the length of proteins can be an effective mechanism to reduce their cost. In this respect, a selective pressure for shorter, less expensive proteins would be more intense among those species that are likely to encounter starving conditions (e.g. free-living species versus parasites). Finally, a propensity toward longer proteins in Bacteria may be a consequence of the phenomenon of genome reduction among bacterial obligate parasites. Genes of low expression and poorly conserved are likely to be eliminated from the genomes of obligate parasites when under no or weak selection (26). Consequently, the proteomes of these species would be enriched in longer conserved proteins of fundamental function (9).

Our findings indicate that the differences in protein length between Eukaryotes, Bacteria and Archaea are biologically meaningful. They suggest that proteins present in Prokaryotes as single units have often fused in Eukaryotes into multi-domain units. Among Prokaryotes, extreme environmental conditions may account for the shorter length of archaeal proteins compared with their bacterial orthologs, through loop and terminal element deletions and a tendency among Archaea to evolve less complex single-domain proteins. There is a greater variability in median protein length between sequenced bacterial species than between archaeal species. This variability may reflect the diversity of the bacterial ecological adaptations, ranging from free-living in extreme environments to endoparasitic life-styles. The precise relation of protein length with environmental conditions is likely to be complex and remains to be explored.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at NAR Online.