Arguments in favour of a phylogenetic analysis of the corresponding protein rather than the DNA

Codon bias

• Amino-acid codons have been degenerated with wobble in the third position. Yeasts, protozoa, and animals have different codon preferences, which would result in differences in DNA sequence that are related to codon bias and not to evolution. Compare the codon usage tables, for human and yeast. (Click here to access the Japanese CUTG database with codon usage information for many different species)
• Also, the protozoa use the codons TAA and TGA to encode glutamine, rather than STOP. and in mitochondria the codon TGA encodes tryptophane, rather than STOP. The inclusion of unique codons in a subset of the sequences will tend to make that subset appear more divergent than they really are. (For more information about organelles visit the Organelle Database.)
• Therefore, it may be advantageous to first translate a coding sequence or open reading frame of a gene into its corresponding protein sequence. The results will be a peptide sequence in either the one- or three-letter code.
• Working with proteins rather than with DNA means that you have to know the three- and especially the one-letter code for amino acids. It is important to familiarise you with the one-letter code as quickly as possible. Here you may find them
  • The One- and Three-Letter Amino-Acid Codes
  • The Universal Genetic Code
  • The Mitochondrial Genetic code

Back to the Table of Contents

Long Time Horizon

• Homologous sequences that diverge with time tend to incorporate mutations more ore less at random. This will make the two sequences more different from each other when time evolves. The chance that a certain position in the DNA incorporates a second mutation, so obscuring the first mutational event or even a back mutation resulting in no obversable difference, will increase with the total number of mutations that have been incorpotated and thus will also increase with time. In protein-coding sequences the first and second position of each codon are less prone to the incorporation of mutations, because this will almost always lead to a change in amino acid in the corresponding position of the protein. The third position, also called the wobble position, in most cases, may be mutated without directly affecting the protein. Check this by analysing the universal genetic code.
• When comparing protein coding sequences that have diverged for possibly a billion years or more, it is very likely that the wobble bases in the codons will have become randomized. By excluding the wobble bases by removing every third nucleotide from the potein coding sequence (a general technique in phylogenetic analyses), one is actually looking at amino-acid sequences. Sometimes it is cumbersome to remove the wobble bases from your DNA sequences, while it is much easier to simply translate the open reading frame into its corresponding protein sequence.
• There are several more reasons why it may be advantageous to translate an open reading frame into protein before carrying out phylogenetic analyses (see below)

Conclusion: due to wobble-base degeneracies the third bases in the codons of a protein-coding gene is of little value in the analysis of distantly related proteins!

Back to the Table of Contents

Advantages of the translation of DNA into protein

• DNA is composed of only four kinds of unit: A, G, C and T
• If gaps are not allowed, on average 25% of residues in two randomly chosen aligned sequences would be identical.
• If gaps are allowed, as much as 50 % of residues in two randomly chosen aligned sequences can be identical. Such a situation may obscure any genuine relationship that may exist. Especially when comparing distantly related or rapidly evolving gene sequences.
• Moreover, it is easier to translate a gene sequence into its corresponding protein sequence than to remove the third wobble base from each of the codons in the gene.
• Translation of DNA into 21 different types of codon (20 amino acids and a terminator) allows the information to sharpen up considerably. Wrong frame information is set aside.
• Third-base degeneracies are consolidated.
• After insertion of gaps to align two random protein sequences it can be expected that they are between 10-20% identical.
• As a result of the translation procedure the protein sequences with their 20 amino acids are much more easy to align than the corresponding DNA sequences with only 4 nucleotides.
• If, after this, you still want to align distantly related gene sequences, you better prepare first a protein alignment and then base your self on this alignment for the alignment of the corresponding gene sequences and the precise placement of indels in the aligned sequences.

Conclusion: The signal to noise ratio is greatly improved when
using protein sequences over DNA sequences!

Back to the Table of Contents

Nature of Sequence Divergence in Proteins (the PAM)

• The observed sequence difference of two sequences that diverge with time takes the course of a negative exponential. This is the result not only of the fact that each position is subject to the incorporation of mutations, but also to reverse changes ("back mutations") and multiple hits. Such events increase in number as the evolutionary distance between two homologous proteins increases.
• This leads to an underestimation of evolutionary distances between two homologous proteins and as a consequence the observed percentage of difference between two protein sequences is not proportional to the actual evolutionary difference.
• A measure that is proportional to the true evolutionary distance between proteins is the PAM value. PAM (Dayhoff and Eck, 1968) is number of Accepted Point Mutations per 100 amino acids.

Example of some PAM values and their corresponding observed distances are given in the following Table.

Back to the Table of Contents

Proteins Evolve at Highly Different Rates.

• Pseudogenes and proteins with functions that are less essential to the organism rapidly evolve. As a result evolutionary information is quickly erased.
• House-keeping proteins, such as histones, enzymes of essential metabolic pathways and proteins of the cytoskeleton, evolve slowly and incorporate between 1 to 10 mutations per 100 residues and per 100 million years (inspect the Table on protein-evolution rates). Therefore, it takes a considerable time before they have incorporated sufficient (250-350 substitutions per 100 residues) before all evolutionary information has been erased. Because of this slow rate of evolution house-keeping proteins are excellent tools to trace evolutionary relationships over long periods of time. For instance the slow mutation rate of glycolytic enzyme glyceraldehyde-3-phosphate dehydrogenase provides us with a look-back window of some 9 billion years, twice the age of our solar system (see the Table on important dates).

Conclusion: proteins are excellent tool to study the evolutionary relationships of both closely as well as distantly related taxa!

Back to the Table of Contents

Introns and Non-Coding DNA

• Many proteins are encoded on each piece of DNA and so, when confronted with a DNA sequence, a biologist needs to figure out where the code for a protein starts and stops. This problem is even more difficult because a eukaryotic genome contains much more DNA than is needed to encode proteins; the sequence of a random piece of DNA is likely to encode no protein whatsoever.
• The DNA which encodes proteins is often not continuous, but rather is frequently scattered in separate blocks called exons. Many of these problems can be reduced by sequencing of RNA (via cDNA) rather than DNA itself, because the cDNA contains much less extraneous material, and because the separate exons have been joined in one continuous stretch in the RNA (cDNA). There are situations, however, where analysis of RNA is not possible and the DNA itself needs to be analyzed
• Eukaryotic genes in general have been fragmented into exons and interspering introns. Due to differences in evolutionary pressure on exons and introns the rate of incorporation of base substitutions in these two elements of eukaryotic genes may be dramatically different. Therefore, a study of the evolution of a protein using its DNA sequence should only include coding sequences. This requires that in every DNA sequence all the introns are being edited out. This may be cumbersome and time consuming. Therefore, it may be easier to translate a cDNA into its corresponding protein, rather than using with the genomic DNA sequences.
• Although a much greater fraction of RNA encodes protein than does DNA, it is certainly not the case that all RNA encodes protein. In the first case, there can be RNA up- and down-stream of the coding region. These non-coding regions can be quite large, in some cases dwarfing the coding region. Further, not all RNAs encode proteins. Ribosomal RNA (rRNA), transfer RNA (tRNA), and the structural RNA of small nuclear ribonucleoproteins (snRNA) are all examples of non-coding RNA.
• By and large, global, complete solutions are not available for determining an encoded protein sequence from a DNA sequence. However, by combining a variety of computational approaches with some laboratory biology, scientists have been fairly successful at accomplishing this in many specific cases. Nonetheless, this problem is currently considered one of the most important in computational biology.

Back to the Table of Contents

Multigene Families

• Organisms may contain many highly similar genes, while only one peptide sequence can be identified (e.g. histones and GAPDH in humans). Using these DNA sequences, it would be difficult to decide which genes are expressed and which are not and thus to decide which genes to include in the analysis. Moreover, if all the genes that are expressed encode the same protein, then DNA differences are not significant.

Back to the Table of Contents

Protein is the Unit of Seletion

• The fundamental building blocks of life are proteins. Enzymes, which are the molecular machines responsible for virtually all of the chemical transformations that cells are capable of, are proteins. In addition, much of the structure of a cell is made up of proteins. That part of the structure which is not made up of proteins is produced by enzymes which are proteins. Mycoplasma genitalium is the smallest known genome that is not a virus. It codes for 468 proteins, that have been called the minimal set for life. A recent estimate of the smallest number of proteins required to make a cell is 256 (click here for more details on how many proteins are essential to make a cell). E. coli contains 4288 genes and a yeast cell contains about 6000 genes. A human contains on the order of 100,000 different proteins. It is the properties of and the interactions between these 100,000 proteins that make us what we are.
• Proteins are variable length linear, mixed polymers of 20 different amino acids. Other terms used more or less interchangably for amino-acid polymers are peptides and polypeptides. These topologically linear polymers fold upon themselves to generate a shape characteristic of each different protein, and this shape along with the different chemical properties of the 20 amino acids determine the function of the protein. One of the most important concepts in modern biology is that the functional properties of proteins is determined largely by the sequence of the 20 amino acids in the linear polypeptide chain; that in many cases proteins are largely self-folding. Thus, in theory, knowing the sequence of a protein (the order with which the amino acids occurred) one could infer its function.
• For protein-encoding genes, the object on which natural selection acts is the protein itself and not the DNA. The underlying DNA sequence reflects this process in combination with species-specific pressures on DNA sequence (like the need for thermophiles to have DNA that is resistant to melting or a very high or low GC content). Thus if function demands that a protein maintains a specific sequence, there still is sufficient room for the DNA sequence to change.

Conclusion: where possible use a translated cDNA sequence for your protein analysis!

Back to the Table of Contents

RNA Editing

The DNA sequence doesn't always translate into amino-acid sequence. The pre-mRNA may require alteration of its coding sequence before it can be translated into a funtional protein. This is called post-transcriptional editing. In post-transcriptional editing several different mechanisms are known. These are:

• RNA editing in the Kinetoplastida.
  This involved the insertion or deletion of one or more Us in the pre-mRNA, using guide RNAs as templates. This way non-coded initiation codons or amino acids are added or coded amino acids are removed during the editing process. This could lead to major differences between DNA and mature mRNA sequence. In some extreme cases (in Trypanosoma brucei sometimes more than 50% of a genes is edited. Such genes are called pan-edited genes). DNA and mature mRNA do not hibridise anymore to each other. Nevertheless this leads to roughly the same protein sequence after final editing. (Some details about editing in Trypanosomatidae can be found here or on the RNA editing site of Larry Simpson).
• Post-transcriptional base modification in some gene products.
  Examples of these are:
  • Modification of rRNAs
  • Modification of tRNAs
  • Modification of the apo-lipoprotein B mRNA creating an additional termination codon

Conclusion: Peptide sequences are not always identical to what is predicted by the corresponding genes!

Back to the Table of Contents

Some good advice

It is recommended to analyse your data set both ways (DNA and protein).

Keep in mind that:

• For a group of species or taxa that are relatively close in time or that are closely related (like viral proteins or vertebrate enzymes) DNA-based analysis is probably a good way to go, since you avoid such problems as differences in codon bias or saturation of the third position of codons. It is nevertheless strongly recommended to carry out an analysis on the protein data as well.
• Multigene families (for instance genes coding for different, but similar, isoenzymes) may cause you problems and be careful when you decide to exclude or include such sequences (this may result in paralogous sequences in your data set and peculiarly looking phylogenetic trees).

Back to the Table of Contents

Construction of Phylogenetic Trees

Alignment of two protein sequences

In order to study molecular evolution of proteins one has to compare the sequence of a protein with that of other homologous proteins (click here for an explanation of the term homologous and the difference between homology and identity). However, such a comparison is not easy to make because the sequences that need to be compared are usually not identical, but only similar. In addition to having substitutions, there will be insertions and deletions in one sequence relative to the other. Phylogenetic analyses should be carried out on homologous residues only (i.e. those residues in each of the two sequences that originates from a common ancestral residue). Thus it is essential that two or more sequences be properly aligned relative to each other. Several excellent programs have been developed for the alignment of multiple DNA or protein sequences. Some like DNA*, GeneWorks or MacVector are commercial and expensive but run on the PC or Macintosh sitting on your desktop. You may have access to a computer center where various programs, both commercial and freeware, have been installed, such as the Pileup program of the Wisconsin GCG (Genetics Computer Group) package that is available through the Belgian EMBL node computer (BEN) to which you can connect for a reasonable annual fee. You might go out on the network and download the source code for a program such as ClustalW, which is available for many different platforms and which you compile and run on your own computer. Or finally, there are now places on the network where such programs such as ClustalW and Darwin run that you can connect to and use. This latter possibility will be covered in my Demo chapter.

Back to the Table of Contents

The visual method

How do programs such as ClustalW, Pileup, Multalign, etc. work ? The easiest way to understand this is to have a look at a visual type of alignment of two sequences such as is carried out by the "Dot-matrix" method. In this method the two sequences to be aligned are written out as column and row headings of a matrix. Dots are put in the matrix wherever the residues in the two sequences are identical. If the two sequences are identical there will be dots in all the diagonal elements of the matrix. If the two sequences are different, but can be aligned without gaps, there will be dots in most of the diagonal elements. If a gap occurred in one of the sequences, the alignment diagonal will be shifted vertically or horizontally.

The figure here shows a dot-matrix for two highly homologous trypanosome phosphoglycerate kinases and two more distantly related phosphoglycerate kinase sequences from Trypanosoma and Euglena . It is obvious that there is no problem in aligning two sequences as long as they are of similar length and have more than 50 % identity.

Back to the Table of Contents

Computer algorithms

Computer algorithms that have been developed for the alingment of two homologous sequences in principle use the same procedure as the dot matrix method. Each residue of one sequence is compared with each residue of the other and when there is an identity a certain value is given to that position in the matrix.

	C	D	E	G	L	D
C	x
D		x				x
E			x
G				x
L					x
D		x				x

Identity score for wo identical sequences

Then a diagonal line is drawn connecting the points with the highest score. Horizontal or vertical shifts from the diagonal due to the presence of gaps are given a penalty.

	C	D	E	L	D
C	x
D		x			x
E			x
G
L				x
D		x			x

Identity score for two sequences with one indel

The choice of the value for each positive score relative to that of the gap penalty will of course stronly influence the quality of the resulting alignment. Too high a gap penalty value will lead to a situation where dissimilar regions will not be aligned with each other at all, but with gap regions, while a too low gap penalty will lead to alignment of non-homologous residues or regions. Most programs allow the user to select an appropriate weigth matrix for the scoring of either identities or similarities of amino acids and for adjustment of the gap penalty value.

Many different weight matrices have been developed for the use with sequence alignment programs. Some of these are:

• Identity matrix
• Mutation-cost matrix
• Hydrophobicity matrix
• PAM1 matrix
• PAM250 matrix
• BLOSUM (Block Substitution) matrix
• Log odds matrix

More detailed information about these matrices can be obtained from the document: "Weight Matrices for Similarity Scoring" compiled by David Wheeler.

It is up to the scientific judgement of the user, and depending on the dataset that is being analysed, what kind of weight matrix will be chosen.

Several algorithms for the alignment of protein sequences have been developed. Some of the better known are the Pearson-Lipman algorithm, used in Pearson's well known FASTA program, the Needleman Wunsch (check, ref), the Smith-Waterman (ref) and the BLAST algorithms. They are used in sequence-comparison programs for the search of homologous sequences in large sequence databases and in programs used for the creation of multiple sequence alignment.

Back to the Table of Contents

Multiple sequence alignment

In order to create a multiple alignment of homologous protein sequences you have to collect all related sequences from a database. Several databases are available for this purpose. First of all the SwissProt database would be the most reliable source for your sequences. The advantage of this database is that each sequence is checked and extensively annotated. Moreover, homologous sequences from different organisms have highly similar locus names, which tremendously facilitates the recognition and retrieval of related sequences. The Enzyme or EC database is a useful tool to retrieve all homologous sequences of one specific enzyme by FTP. Other databases for the retrieval of related sequences are the homologous protein domain databases Prodom and the WIT (What Is There (PUMA)) databases.

For the construction of reliable phylogenetic trees the quality of a multiple alignment of the protein sequences is of the utmost importance.

There are many programs available for the multiple alignment of proteins. Click here for an extensive upto date list of multiple alignment software complied by Georg Fuellen

Good programs in the public domain are:

• on the World-Wide Web or available as freeware:
  • ClustalW
  • Multalign
  • Darwin
  • ProteinPredict
• Pileup of the Wisconsin GCG (Genetics Computer Group) package
  Click here for an example of an alignment in Pileup MSF format.

They quickly align pairs of sequences and roughly determine the degrees of identity between each pair. Then the sequences are aligned more precisely in a progressive way, starting with the two most related sequences.

Darwin is a large suite of programs that allows you to carry out many different types of analysis on a collection of proteins, including the construction of a phylogenetic tree.

ProteinPredict is a program for the prediction of the secondary structure of a protein. It makes use of the information on the mutatability of each residue in a multiple sequence alignment. Therefore it first aligns the query sequence with all homologous sequences available in the SwissProt database. So this is an easy way to create a multiple sequence alignment that includes your query sequence.

The algorithm of the Pileup program of GCG resembles that of Clustal and both programs give more or less the same result. The Wisconsin GCG (Genetics Computer Group) package is a commercial package for DNA and protein analysis. It can be accessed via the BEN (Belgium EMBNet Node) computer, provided that you or your laboratory have an account there.

NB: Most multiple sequence alignment programs work best when the sequences have similar length.

For an extensive list of multiple alignment resources, both for the creation and the analysis of alignments, compiled by Georg Fuellen and available on the net: click here.

Back to the Table of Contents

Prodom and Blocks databases

There is another way to obtain multiple sequence alignments to which your sequence can be aligned. There exist databases of prealigned sequences that share domain structures or homologous blocks of sequences. These databases are called Prodom, Pfam and Blocks, and are accessible via the Internet. They have been compiled by comparing and aligning all homologous sequences of the SwissProt database of protein sequences. The differences between the Prodom and Blocks databases is that in Prodom gaps are allowed, whereas in the Blocks database, that has been compiled using the Blast algorithm, no gaps are allowed. Therefore, in general the compiled sequences in Blocks are shorter than in Prodom. Prodom and Blocks alignments serve as an excellent basis to start a multiple sequence alignment and/or phylogeny project.

Back to the Table of Contents

Manual adjustment of a protein alignment

An automatically produced multiple sequence alignment often needs manual adjustment to improve the quality of the alignment. Such improvement can be obtained by using all the knowledge that is available about a protein.

Some rules of thumb:

• The rules for mutation of amino acids are dependent on their physicochemical properties.
• Surface residues (DRENK) are preferably mutated to residues of similar properties. Since they are polar or charched they are mainly on the outside of the folded protein. Since they are not, or less, involved in protein folding they mutate rather easily.
• Hydrophobic residues (FAMILYVW) are preferentially replaced by other hydrophobic ones. These residues are mainly internal and determine the folding of the protein. They thus mutate rather slowly.
• The residues CHQST are generally indifferent and may be replaced with any other type of residue.
• The residues (DRENKCHQST), when conserved throughout the alignment, are very likely residues that are involved in the active site. So the multiple alignment should be adjusted in such a way as to maintain them aligned.
• Periodicity of charged residues may provide information as to the presence of elements of secondary structure such as alpha-helices and beta-strands. Alpha helices have a repetition of 3.6 residues per turn. Stretches of more than 12 amino acids with a charged amino acid every 3rd or 4th residue may be indicative of the presence of an alpha helix. Repetition of charged amino acids every 2nd residue may be indicative of a beta strand structure
• Indels (insertions/deletions) are almost never found in elements of secondary structure but only in loops. Moreover P and G interfere with secondary structure elements and thus have a preference for loops. Since loops easily acquire or loose residues you should always try to align indels with P and G residues.
• Hydrophobicity (or hydrophathy) profiles according to Kyte and Doolittle of two homologous proteins are in general strikingly similar (see figure below).

.

Hydrophobicity profile according to Kyte and Doolittle

Click here for a comparison of of the hydropathy profiles of the phosphoglycerate kinases from Trypanosoma congolense, T. brucei and Euglena gracilis.

Click here to see a typical alignment of homologous PGK sequences

Another useful tool for the manual alignment of proteins, in the case there exists a crystal structure of at least one of the enzymes of the alignment, is the NLR-3D database. This database contains protein sequences plus secondary structure information. It tells you which residues of a protein sequence belong to conserved areas of secondary structure such as alpha helices and beta strands. Such areas almost never contain indels.

Back to the Table of Contents

Continue with Protein Phylogeny