ProSite is a catalogue of patterns identified by sequence or profile (weight matrix). This paper is based on the analysis of the May 2000 release (release 38), which contains more than 1300 patterns. Since profiles allow for the detection of signatures in sequences with a high degree of divergence, we only worked with patterns. ProSite patterns are sequences of brackets, each containing the list of the possible amino acids that can be found in a given position (e.g., [NF]), separated by a number d of gaps, each characterized by lower and upper boundary values (e.g., a⩽d⩽b).

The princoms of a ProSite pattern were obtained by replacing the amino acids within each bracket by the amino acids encoded by the i.c. of all their possible codons (see Table 1), and the order of the brackets and the gaps, inverted.

We asked how many of the ProSite patterns have princoms in the proteins registered in SwissProt. The probability of a bracket equals the sum of the frequencies of its letters (e.g., p_[KEIV]=p_K+p_E+p_I+p_V). By this way, we take into account the fact that a given protein generally has more than one princoms.

The probability p of the occurrence of a sequence of brackets separated by gaps is the product of the probabilities of the brackets in the sequence, multiplied by the factor b−a+1 for each gap of length d with a⩽d⩽b. Taking into account the number N of amino acids in SwissProt (80 000 entries with N=29 085 265 amino acids), the expected number E of the observed number of occurrences, O, for this sequence is equal to the product Np. The standard deviation s.d. of O was calculated through a Poisson approximation. The statistical signification of the observed number of occurrences, O, was based on the statistic (O−E)/s.d. [2].

To avoid very large files, we selected the patterns for which the expected number E of princoms occurrences does not exceed 100, the cut-off number 100 being arbitrary. This selection generated a set of 594 sequences. Out of these 594 motifs, the mathematical expectation of the number of occurrences greater than 1.96 s.d. is less than 30, and we observed 273 (Group A, p<10⁻¹⁰). The mathematical expectation of the number of occurrences greater than 5 s.d. is less than 2×10⁻⁴, and we observed 93 (Group B, p<10⁻¹⁰). From these results, we conclude that the princoms that we reported earlier are a subset of a much larger set. This strongly suggests that princoms are a common feature in the proteome.

The amino acid composition of the Group-B patterns is significantly different from that of all ProSite entries (Table 2). Cysteine is almost 2.5 times more frequent in Group B than in the whole ProSite; arginine, glycine, lysine, alanine, and tryptophan almost 1.25 times. The over-representation of these amino acids could be explained by the fact that all of them can be generated by a single site mutation of the cysteine codon, while lysine is a habitual replacement of arginine. A similar behaviour could be expected from serine, whose codons can also be derived from those of cysteine by a one-letter change. The codons for cysteine, tryptophan, arginine and glycine, as well as one of the stop codons, all have a central guanine, and the frequency of central guanine in the codon usage of Group B (40.60%) is significantly higher than in the whole ProSite (p<10⁻¹⁸). In the case of serine, we estimated its frequency as 1/3 (Table 3).

Protein data banks have certain inherent characteristics that can lead to sampling biases – e.g., the arbitrary selection of proteins studied and reported, the inconsistencies of the annotation systems and the existence of numerous entries corresponding to the same protein in different species (say ‘redundancy’). Furthermore, the criteria for identification of the patterns registered in ProSite are not exhaustive.

We addressed the problem of redundancy by comparing the statistical signification of princoms data obtained from the whole SwissProt and that derived from four specialized data banks – Human proteins in SwissProt (5913 proteins), Yeast Protein Database (4531 proteins), an Enzyme sub-bank (1519 proteins), and a PDB sub-bank (2139). The statistical signification was comparable except in the case of the PDbank (Table 4) [3].

3.1 The biological signification of princoms

3.2 Apoptotic proteins and the hemoglobins

Duplications and inversions are characteristic genomic features, and play a central role in the evolution of chromosomal architecture. Large size, low-copy repeats with high-sequence identity (several kb to Mb duplicons) lead to deletions, duplications, inversions, and inverted duplications. Contemporary mosaic proteins are often the result of the iteration of small-size genetic domains (duplicated segments of up to 1 kb). Many of these iterated domains preserve their characteristic sequence patterns motifs and signatures as well as their 3D structure. A substantial amount of motifs and signatures are cysteine C-rich, and the constancy of the positions of cysteine allows the classification of proteins in families and superfamilies, and to identify new members belonging to them [10,11].

While studying the patters of cysteine signatures in several families of autacoid peptides, we became aware of the fact that in the precursor polypeptides cysteine-rich regions alternate with threonine and/or alanine-rich regions. This clustered distribution of cysteine, threonines, and alanines is also found in vertebrate and invertebrate membrane glycoproteins, mucins, metalloenzymes of the extracellular matrix, proteoglycans, DNA-binding proteins, nuclear membrane proteins, and viral capsides. Since threonine and alanine are encoded by the inverse complementary codons of cysteine, we asked whether the threonine/ alanine-rich regions were in fact the result of inversions of duplicated cysteine-rich domains [1]. To answer this question, we applied Hidden Markov Models (HMM) in the statistical tool R'HOM [12,13] to study the DNA encoding the threonine/alanine-rich regions flanking the three cysteine-rich trefoil patterns present in two small proteins MUA1-XENLA [sw P10667] and MUC1-XENLA [sw Q05049]. Our results showed that the cysteines and these threonine/alanine-rich regions actually are princoms pairs. These trefoil peptides can be described, therefore, as mosaics made up by the linear combination of direct and inverse gene segments. The analysis of the amino acid sequences of other peptides containing cysteine signatures revealed that the also have princoms pairs, e.g., the prepropeptides of six endothelins, and the Zn²⁺ finger proteins of the classes 1, 2, 4, 4^∗ (half of the type 4 signature), and 5 knots. In other cases, the princoms pairs are found in different polypeptides: the i.c. sequence of the cysteine signature of the somatomedins is present in 39 different proteins, but not in the somatomedin prepropetides.

In this paper, we provide evidence that the princoms pairs reported in our previous paper are just a small subset of a larger universe set of primcoms inserted in contemporary host-proteins registered in SwissProt. Our results are not due to biases introduced by the inherent characteristics of the protein data banks (problems of annotation and redundancy), because essentially the same results are obtained with different protein data banks. From this date, we conclude that many proteins are mosaics composed by direct and i.c. sequences (princoms pairs). Our present results allow us to generalize these findings and to postulate that many proteins contain sequences that are the princoms of known ProSite entries. Since we only analysed entries in protein data bank, we do not know if there are traces of princoms in intergenic regions. However, there is evidence that inverted segments are also translocated to non-coding, intergenic regions in the fish Tetraodon nigrotiridis (J. Weissenbach, personal communication). Furthermore, our data show that the role of genetic inversions in the determination of protein structure extends beyond the case of RAG2 and RAG1-mediated V/C recombinations to create antibody diversity. We have found examples in which the ancestral gene that has given rise to a multidomain protein by n-plication is in fact the princoms of a sequence found in a different, totally unrelated kinds of proteins.

Furthermore, we show that princoms change significantly the biochemical and physiological characteristics of the grafted proteins by providing new opportunities for intra-molecular and inter-molecular bonding, thus conferring distinct biochemical and physiological functions to their host-proteins. In the case of PS01113, one of the princoms of C1q provides the hinge of IgG heavy chain, another gives vWF its CTCK, and still other makes the heme-binding residues of Hp. In the case of Hp, the molecule itself is the result of the tetraplication of a primordial princoms of C1q. In fact, the princoms of C1q is the characteristic ProSite motif of Hp. The grafting of a princoms opens the possibility of substantial structural modifications of the host-protein, e.g., polypeptide length, stability, catalytic specificity, folding, and associativity. Princoms inserted in phase and devoid of non-sense codons do not disrupt the reading of the host-protein, but if they contain a non-sense codon they will cause premature end of translation. When inserted out of phase, princoms will shift the open reading frame of the protein, and the new reading frame will replaces previous stop signals and introduce a new stop. It is plausible that the grafting of a princoms could cause radical 3D changes or confer new catalytic profiles to the host-protein. They may contribute hinge regions and divide a single domain in two domains, and the hinge may contain target sequences for proteolytic processing. Finally, princoms may create new interactive surfaces leading to non-covalent homodimerization or heterodimerization. We do not know the actual size of the set of princoms. Since we limited our search to the i.c. sequences of the ProSite patterns, we do not know yet the real length of the duplicated and inverted sequences. Work is in progress to devise mathematical and computational tools to find the real princoms length.

Our findings give raise to several structural questions. We have shown that a princoms pair, one present in the hemoglobin β-chain and the other in domain BH-3 of apoptotic proteins, have essentially the same secondary (α-helical) structure. Is this a general phenomenon? Do all the highly similar princoms pairs have the same secondary structure in solution? If they do, will they conserve it when grafted in their host proteins, or do they adopt a new secondary structure as a function of the context provided by the host protein? In fact, the existence of highly similar princoms pairs in different types of proteins is an experiment of nature for testing the generality of the findings of Milnor and Kim [14] concerning the importance of context-dependent effects in protein folding.

Chromosomes are mosaics of ancestral and horizontally transmitted sequences [15]. Genomes evolve by acquiring new sequences through duplications, inversions, horizontal genetic transfers, transposition events, and rearrangements (duplications and inversions), and ulterior divergence [15–19]. The widespread occurrence of palindromes and their biological relevance in genetic regulation and RNA structure and function indicates that inversions played a crucial role in the diversification of the portfolio of biological opportunities at the polynucleotide level throughout evolution [20,21]. Princoms offer a new way for detecting generalized lateral transfer among kingdoms, taxa, and species. Lateral gene transfer is a significant mechanism in the evolution (diversification and speciation) of bacterial genomes, introducing traits of antibiotic resistance, virulence attributes, and metabolic properties, and accounts for the ability of bacteria to exploit new environments [22]. So far, the detection and identification of cases of lateral gene transfer in bacteria relays on the finding of unusually high degrees of similarity between the donor and the recipient strains, atypical base compositions, and patterns of codon usage bias, as well as the detection of vestiges of genetic elements involved in their transference and integration. However, this approach is restricted to cases in which the putative recipient and the donor species (or taxons) are known, and is prone to underestimate the actual number of transferred genes [15]. On the other hand, princoms allow the detection of small lateral transfers. Since the statistical methods used so far to draw protein phylogenetic trees do not take into consideration the horizontal transfer of dincoms and their corresponding princoms, new approaches are needed for inferring the historical patterns of protein evolution, including their estimated time of grafting. This would provide a more refined appraisal of the tempo of evolution and reflect the multiple sources of genetic material from which a given protein family derives its contemporary structure.

The existence of princoms pairs implies the existence of DNA inverse complementary sequences (dincoms pairs), and their corresponding RNA inverse complementary sequences (rincoms pairs). The existence of families of short, non-coding RNAs having the characteristics of rincoms has been recently reported [23–27]. Rincoms can generate self-folding sequences that can change alternative splicing targets, give raise to anti-sense RNA and post-transcriptional gene silencing (PTGS) structures, form regulatory hairpins in primary transcripts and messenger RNAs and thus affect the expression of genes and the rate of protein synthesis.