BIO Assignment 5 2011

Note! This assignment is currently inactive. Major and minor unannounced changes may be made at any time.

 
 

   

Assignment 5 - Phylogenetic Analysis

Introduction  

Nothing in Biology makes sense except in the light of evolution.

Theodosius Dobzhansky

... but does evolution make sense in the light of biology?

As we have seen in the previous assignments, the Mbp1 transcription factor has homologues in all other fungi, yet - looking at orthologues - this is not always a clear one-to-one mapping of related genes to each other. It appears that various systems of APSES domain transcription factors have evolved independently. Of course this bears directly on our notion of function - what it means to say that two genes in different organisms have the "same" function. In case two organisms both have an orthologous gene for the same, distinct function, this may be warranted. But what if that gene has duplicated in one of them, and the two paralogues now perform different, related functions in one organism? In order to be able to even ask such questions, we need to understand how we can make the evolutionary history of gene families explicit. This is the domain of phylogenetic analysis. We can ask questions like: how many paralogues did the cenancestor of a clade possess? Which of these underwent additional duplications in the phylogenesis of the organism I am studying? Did any genes get lost? And - adding additional biological insight to the picture - did the observed duplications lead to the "invention" of new biological systems? When was that? And how did the species benefit from this event?

We will develop some of this kind of analysis in this assignment. In the previous assignment you have established which genes are the reciprocally most closely related orthologues to Mbp1 in yeast. In this assignment, we will analyse their evolutionary relationship and compare it to the evolutionary relationship of all fungal APSES domains. The goal is to define families of related transcription factors and their evolutionary history.

A number of good tools for phylogenetic analysis exist; general purpose packages include the (free) PHYLIP package and the (commercial) PAUP package. Specialized tools for tree-building include Treepuzzle or Mr. Bayes. This assignment is conctructed around programs that are availble in PHYLIP, however you are welcome to use other tools that fulfil a similar purpose if you wish. In this field, researchers consider trees that have been built with ML (maximum likelihood) methods to be more reliable than trees that are built with parsimony methods, or distance methods such as NJ (Neighbor Joining). However ML methods are also much more compute-intensive. Just like with multiple sequence alignments, some algorithms will come closer to guessing the truth and others will not and usually it is hard to tell, which is the more trustworthy of two diverging results. The prudent researcher tries out alternatives and forms her own opinion. Specifically, we may usually assume results that converge, independent of the algorithm, to be more reliable than those that depend strongly on a particular algorithm or details of input data.

But regarding algorithm and rersources: we will take two shortcuts in this assignment (and both shortcuts are things you should not do in real life):

One: we will use an efficient tree-building algorithm, not the best-available one. This is an algorithm which is available on the Web, without the need for you to install software on your own machine. In real life you would of course use the most accurate algortihm you can lay your hands on, regardless of the resources this requires, since it makes no sense to waste your time on a careful analysis of inaccurate trees. Your supervisor would want it so as well. And if not she, the reviewers of your manuscript.

Two: we will assume the tree the algorithm constructs is correct. In real life you would establish its reliability with a bootstrap procedure: repeat the tree-building a hundred times with partial data and see which branches and groupings are robust and which depend on the details of the data. But we should still acknowledge that bifurcations that are very close to each other have not been" resolved". Any conscientious reviewer would flag such leniency and send your results back to you for a bootstrapping exercise at the computer. In phylogenetic analysis, not all lines that the program draws are equally trustworthy. Dont take the trees as a given fact just because a program suggests this. Look at the evidence, use your reasoning, and analyse them critically.

In case you want to review concept of trees, clades, LCAs OTUs and the like, I have linked an excellent and very understandable introduction-level article on phylogenetic analysis to the resource section at the bottom of this page.

 

Preparation, submission and due date

Read carefully.

Be sure you have understood all parts of the assignment and cover all questions in your answers! Sadly, we always get assignments back in which important aspects have simply overlooked marks unnecessarily. If you did not notice that the above did not make sense, you are reading what you expect, not what is written.

Review the guidelines for preparation and submission of BCH441 assignments.

The due date for the assignment is Friday, December 7 at 15:00 in the afternoon.

   

   

(1) Preparations

   

(1.1) Preparing Input Files (2 marks)

 

Introduction: Task

For this assignment, we start from the multiple sequence alignments we have constructed in the last assignment. We will edit alignments to make them suitable for phylogenetic analysis. We will construct a tree and we will discuss tree analysis in the end.

The phylogenetic tree we will construct will contain all APSES domains we have found. In order to interpret such a tree it is crucial to have some sense of what these domains are, i.e. to cluster them according to their orthologues. Only then can we analyse the tree by asking which subclades mirror the accepted phylogeny of fungi and which ones differ. In the third assignment, you have defined the true orthologues for most of the domains we had previously found with our PSI-BLAST search. (I have filled in the rest.) From this information, I have revised the gene names in the MUSCLE alignment of all APSES domains. When we calculate a phylogenetic tree with these sequences, we should expect orthologues to cluster into the same subclade. Of course, not all fungi have the same number of APSES domain homologues, but from the data we have compiled it should be possible to define their evolutionary history with reference to the other species.

Introduction: Principle

In order to use molecular sequences for the construction of phylogenetic trees, you have to build a multiple alignment first, then edit it. This is important: all rows of sequences have to contain the exact same number of characters and to hold aligned characters in corresponding positions. Phylogeny programs are not meant to revise an alignment but to analyse evolutionary relationships, given the alignment. Their inferences are made on a column-wise basis and if your columns contain data from unrelated positions, the inferences are going to be questionable.

The result of the tree construction is a decision about the most likely evolutionary relationships. Fundamentally, tree-construction programs decide which sequences had common ancestors.

Distance based phylogeny programs start by using sequence comparisons to estimate evolutionary distances:

• they apply a model of evolution such as a mutation data matrix, to calculate a score for each pair of sequences,
• this score is stored in a "distance matrix" ...
• ... and used to estimate a tree that goups sequences with close relationships together. (e.g. by using an NJ, Neigbor Joining, algorithm).

They are fast, can work on large numbers of sequences, but are less accurate if genes evolve at different rates.

Parsimony based phylogeny programs build a tree that minimizes the number of mutation events that are required to get from a common ancestral sequence to all observed sequences. They take all columns into account, not just a single number per sequence pair, as the Distance methods do. For closely related sequences they work very well, but they construct inaccurate trees when they can't make good estimates for the required number of sequence changes.

ML, or Maximum Lieklihood methods attempt to find the tree for which the observed sequences would be the most likely under a particular evolutionary model. They are based on a rigorous statistical framework and yield the most robust results. But they ar also VERY compute intensive and a tree of the size that we are building in this assignment is already almost beyond the resources of common workstations (runs about a day on my computer). However, one may split a large problem into smaller, obvious subtrees (e.g. analysing orthologues as a group, only including a few paralogues for comparison) and then reassemble the smaller trees and in this way even very large problems can become tractable. They also suffer less from "long-branch attraction" - the phenomenon that weakly similar sequences can be grouped inappropriately close together in a tree due to spurious shared diferences distinguishing them from more highly conserved subfamilies.

Clearly, in order for tree-estimation to work, one must not include fragments of sequence which have evolved under a totally different evolutionary model as all others, such as domain fusion, or large stretches of indels. Thus it is appropriate to edit the sequences and pare them down to a most characteristic subset. The goal is not to be as comprehensive and complete as possible but to input those columns of aligned residues that will best represent the true phylogenetic relationships between the sequences.

Introduction: Problems

Gaps are a real problem here, as usual. Strictly speaking, the similarity score of an alignment program as well as the distance score of a phylogeny program are not calculated for an ordered sequence, but for a sum of independent values, one for each aligned columns of characters. The order of the columns does not change the score. Hoever in an optimal sequence alignment with gaps, this is no longer strictly true since a one-character gap creation has a different penalty score than a one-character gap extension! Most alignment programs use a model with a constant gap insertion penalty and a linear gap extension penalty. This is not rigourously justified from biology, but parametrized (or you could say "tweaked") to correspond to our observations. However, most phylogeny programs, (such as the programs in PHYLIP) do not work in this way. PHYLIP strictly operates on columns of characters and treats a gap character just like a residue with the one letter code "-". Thus gap insertion- and extension- characters get the samescore. For short indels, this underestimates the distance between pairs of sequences, since any evolutionary model should reflect the fact that gaps are much less likely than point mutations. If the gap is very long though, all events are counted individually as many single substitutions (rather than one lengthy one) and this overestimates the distance. And it gets worse: long stretches of gaps can make sequences appear similar in a way that is not justified, just because they are identical in the "-" character. It is therefore common to edit gaps in the alignment to one or two character, or to remove them.

Introduction: Practice

In practice, follow the fundamental principle that all characters in a column should be related by homology. This implies the following rules of thumb:

• Remove all stretches of residues in which the alignment appears ambiguous (not just highly varible, but ambiguous regarding the aligned positions).
• Remove all frayed N- and C- termini, especially regions in which not all sequences that are being compared appear homologous and that may stem from unrelated domains.
• Remove all but approximately one column from gapped regions, and all residues N- and C- terminal of the gap in which the alignment appears questionable. ( I would keep one gapped column as a placeholder for a rare and very distinct evolutionary event, rather than simply deleting them all, some researchers remove all gaps).
• Also, consider that neither residues that are completely different between all species, nor residues that are completely conserved are informative for relationship distances.
• If your sequences are too long, you may run out of memory. 60-80 aligned residues should be plenty and if the sequences fit on a single line you will save yourself potential trouble with block-wise vs. interleaved input.

(A very useful trick with Microsoft Word is that you can select blocks of text and entire columns in the document with your mouse: hold the "ALT" key depressed while you click and drag your mouse to select. This will greatly facilitate the preparation of sequences. You can treat that selection as any other selected text, color characters, or delete them. Importantly, you can also cut and paste entire columns! Of course, this will only work as expected if you use a fixed-width font such as Courier. )

The preparation of the input file of aligned residues, used by the PHYLIP package is straightforward in principle; just carefully follow the instructions in PHYLIP's well written documentation. If you plan to use an outgroup for your tree, it is a good idea to move that to the first line of your alignment, since this is where PHYLIP will look for it by default.

Some notes on how to avoid common editing troubles. Copy the sequences frrom the link provided below. Paste them into a document, using the Word "Edit -> Paste special -> Unformatted text". Set the page-setup to "landscape", the font-size to something small, then you can put every sequence into one line. You can replace all paragraph marks ("^p") with (nothing) to remove them, then replace the FASTA header line character ">" with paragraphs ("^p") to separate them by line again. Take special note that your files must not include tab characters. You can use Word to globally replace all tabs (specified as "^t") with a blank, to make sure. Spaces count, so display your alignment in a fixed-width font, such as Courier ("Courier New" on Windows), not a proportional-width font such as Times, Arial, or Helvetica, and ensure all characters in your alignments align as they should. As always, make sure you save your input files as "Text Only".

A note if you are working on a Mac and saving input on disk, to run with a locally installe PHYLIP version: here MS Word will play one of its usual shenanigans on you since it writes text files with the old-style OS 9 Carriage Return characters (\r; ASCII 13; hex 0D; CR). Just by looking at the file, this is quite invisible but such "Carriage returns" are not going to be recognized by PHYLIP and most other self-respecting UNIX based programs. It may not make a difference when you paste your sequences to a Web server; but if you compute things locally it will appear to the program as though everything were in one line). And this can (and did) lead to head-banging rounds of frustration. You need to replace them with Linefeed resp. Newline characters (\n; ASCII 10; hex 0A; LF) and you can't even do that within Word(!). Open a UNIX terminal window and navigate to the directory where your files reside. Then type:

tr "\r" "\n" < infile > outfile

... where outfile is different from infile (careful: if a file by the name of outfile already exists, tr will cheerfully overwrite it.) Alternatively you could type the following perl one-liner :

perl -e 'while(<>){tr/\r/\n/;print}' < infile > outfile

In your assignment submission, clearly identify the source sequences you are using, as well as the alignment method you have used. Paste your unaltered source alignment into your document, clearly highlight or otherwise color the columns that you have selected, annotate why you have selected them and paste your result as well. Here is an example of what this might look like:

(Possible) steps in editing a multiple sequence alignment towards a PHYLIP input file. a: raw alignment (CLUSTAL format); b: sequences assembled into single lines; c: columns to be deleted highlighted in red - 1, 3 and 4: large gaps; 2: uncertain alignment and 5: frayed C-terminus: both would put non-homologous characters into the same column; d: input data for PHYLIP: names for sequences must not be longer than 10 characters, the first line must contain the number of sequences and the sequence length. PHYLIP is very picky about incorrectly formatted input, read the PHYLIP sequence format guide.

Introduction: Web Service and data

You have two choices for completing the assignment: either to use one of the PHYLIP on-line Webservices that generously provide public computing resources, or to download and install the PHYLIP program package on your own computer at home. If you choose the former, one of your options is the PHYLIP service at the Institut Pasteur in France.

I have tried the Pasteur service many times, and it works - however not always entirely without problems. Uninformative errors will occur when your input is too large for the system's memory (like: "sequences not aligned" ... "out of memories" and such) and once, after submitting a number of jobs, the system locked me out, asking me to what an unspecified time until results would be sent by e-mail (which then hasn't happened). Regrettably, this is not documented. However the integration of their services in a logical sequence of steps is good and some of their services are a bit more advanced than plain out of the box PHYLIP. If you rather decide to install PHYLIP, good for you. That is easy to do, well documented, there are much less limitations on memory - but if you don't read and understand the instructions carefully, you may be in for a spell of frustration.

Either way, I have posted typical input files and result files on the fallback data page, to allow you to bail out in case technical problems become overwhelming. If you use the data posted here instead of your own, you must document that fact and explain what you have tried, and why that has failed. The posted data is a fallback, not a shortcut.

For this assignment, we will use a simple distance based tree construction method. This represents a reasonable compromise between accurracy and speed, especially when applied to moderately dissimilar sequences. In genereal, distance methods include two steps: (1) calculate a pairwise-distance matrix between sequences, (2) construct a tree, based on the matrix. Thus all the information in the alignment bewtween two pairs of sequences is collapsed into a single number: their pairwise distance. Alternative approaches, parsimony as well as ML based algorithms, take individual columns into account. Parsimony based methods construct inaccurate trees when they can't make good estimates for the required number of sequence changes, if the sequences become too dissimilar. ML based methods are considered the most accurate for dissimilar sequences, however they are also very compute intensive and the full-length APSES domain alignment for 70 sequences run about half a day on my workstation. Thus we will use distance based methods here, specifically the UPGMA variant of the neighbor joining algorithm.

 

Prepare an input file that is representative of the APSES domains.

• Access the revised MSA for all APSES domains, linked here (and from the resources section at the bottom of the page). Prepare a PHYLIP formatted input file from this MSA, restricting the number of sequence characters to no more than 70. Read the PHYLIP format documentation and follow the considerations dicussed above. (See the fallback data in case you get stuck, but you must prepare (and document) an input file according to the instructions, even if you end up using the fallback data for whatever reason.) Do not forget to document how you have prepared your input file: define where your source-sequences came from, define which columns you have deleted by highlighing the deleted residues in one sequence, and include your input file in the assignment. (2 marks)

 
 

(1.2) Calculating a Tree (3 marks)

 
 

• Using the protdist program of PHYLIP, calculate a distance matrix for the input file you have prepared. (See the fallback data in case you get stuck) (1 mark)

• If you use the PHYLIP Webserver, select the neighbor joining algorithm from the menu options (neighbor on the PHYLIP server) and click the button "run the selected program on outfile" ; on the next form, click the button to the "advanced neighbor form", choose the option "UPGMA" and click on the button "run neighbor". When the program is done, select the option drawgram and click Run the selected program on outtree. Choose a cladogram tree-style and a suitable output format (e.g. postscript). Paste the trees into your assignment.

• If you use a locally installed version of PHYLIP use neighbor with the UPGMA method to construct a tree for both input files. Open the file outfile in a text-editor, copy and paste the trees into your assignment.

In both cases, the process is: protdist → neighbor → drawgram

(1 mark for constructing a tree, 1 mark for displaying it).

 
 

(2) Analysis

I have constructed a cladogram for the species we are analysing, based on data published for 1551 fungal ribosomal sequences. Such reference tres from rRNA data are a standard method of phylogenetic analysis, supported by the assumption that rRNA sequences are monophyletic and have evolved under comparable selective pressure in all species.

Cladogram of fungi studied in the assignments. This cladogram is based on small subunit ribosomal rRNA sequences, and largely follows Tehler et al. (2003) Mycol Res. 107:901-916. Even though many details of fungal phylogeny remain unresolved, the branches shown here individually appear to have strong support. In a cladogram such as this, the branch lengths are not drawn to any scale of similarity. I have labeled all speciation events so you can refer to these labels in your assignment.

In order to study the evolutionary history of the entire gene family you can use the tree you have computed or access the APSES domains reference tree here.

This is a complicated tree, and it can look impenetrably confusing at first. Here are two principles that will help you make sense of the tree.

A: A gene that is present in an ancestral species, is inherited in all descendent species. The gene has to be observed in all OTUs, unless its has been lost (which is a rare event). This means, if a gene is present in two widely divergent species, but in none other of the descendants of the LCA, it is possible that there is some problem with the tree (long branch attraction maybe), or the sequence has been acquired through horizontal gene transfer.

B: Paralogous genes in an ancestral species should give rise to monophyletic subtrees for each of the genes, in all descendants; this means: if the LCA of a branch has e.g. three genes, we would expect three copies of the species cladogram below this branchpoint, one for each of these genes. Each of these subtrees should recapitulate the reference phylogenetic tree of the OTUs, up to the branchpoint of their LCA.

With these two simple principles (you should draw them out on a piece of paper if they do not seem obvious to you), you can probably pry the reference tree of all APSES domains apart quite nicely. A few colored pencils and a printout of the tree will help.

   

(2.1) The Cenancestor's APSES Domains (2 marks)

 

 

• Identify at least two (more or less) monophyletic subtrees of the tree that recapitulate (more or less) the species tree. Write down their branch-point numbers. In one of these subtrees, define the correspondence between the letters in the species tree and the branchpoints in the gene tree as far as possible. (The correspondence is not going to be exact. ) (1 mark)

• Discuss briefly if there are features of the gene tree that are systematically inconsistent with the reference cladogram: choose one species from the gene tree that violates the expectation from the species tree and check what it's close neighbors are for the other parts of the tree. Do these sequences always appear more closely or more distantly related than they should be? (1 mark)

 
 

(2.3) Unraveling the Mbp1 clade (2 marks)

 

If we consider the Mbp1 clade (the clade containing all 16 of our Mbp1 orthologues), descending from branchpoint 115, we note that it contains 16 additional genes, one from each of the species we have studied, however both subtrees are polyphyletic within the clade. (In what follows I will refer to subgroups of the tree as Clades and label them with their branch number in this reference tree.) It is thus tempting to speculate that this subtree, Clade 115 is actually a mix-up of two ancestral genes, one of them an orthologue to Mbp1, the other to a different gene - perhaps Swi4. If this were true, there would be three possible explanations for the polyphyletic tree: (i) either we have mislabeled the orthologues, or (ii) we have constructed an incorrect tree, or (iii) our model of how these genes have evolved is wrong. Actually, it is rather obvious when we compare the clades that branch off at 117 and 128, why this is not a simple case of mislabeling the genes ...

 

• Explain briefly why it is unlikely that the clades at branchpoints 117 and 128 are each descendants of a single cenancestral gene. (1 mark)

 

Could this point to a problem with the columns we have selected for the tree building? Have we biased our input data and then constructed an incorrect tree? To test this, a revised tree was constructed, taking the genes from Clade 115, realigning them and constructing a new tree. You can have a look at the revised Mbp1 APSES domain tree, and the procedure - or you can just take my word for it, the trees are substantially similar. Apparently the precise choice of columns has a minor effect on the results, as it should have. We should note in general however that branch points that are close together in time may be poorly resolvable, while branch points that are separated by longer evolutionary distances are more reliable. In summary, save for some inaccuracies, we have no reason to believe the tree is fundamentally incorrect.

So what about or model of evolution? What do we actually expect?

As we had written above, we expect to see anumber of more or less faithful copies of the species tree, for each of gene in the cenancestor (or LCA for later duplications).

And as you see, this is not exactly what we find in Clade 115. What we do find however is that Clade 115 reflects our expectations in part. Clade 123 pretty well recapitulates the Saccharomycotina, as does Clade 118, and when we consider Clade 128, apart from some "noise" we see both the Dikaryomycota (in Clade 142) reasonably grouped together, as well as the Euascomycotina (in Clade 129") - in fact the most striking departure from our expected tree for Mbp1 is the placement of Clade 123, which we would expect attached below branch 128, not above it where we have found it.

This may be due to insufficient resolution of the trees. Whenever we have long, evolutionary distances with a few OTUs clustered at the end, we can be reasonbly certain of the branch. When branching events occur at similar times, the order is uncertain and any specific ordering would probably not be well supported in bootstrapping.

What is striking however is the parallelism between Clade 123 and Clade 118. Apparently branchpoint 117 marks a duplication event of the ancestral Mbp1 gene in the Saccharomycotina. Here our expectation appears to be borne out. And given the uncertainty in the precise ordering between branchpoints 117 and 128, the Clade overall is thus actually quite consistent with the species tree.

 

• Finally, find one example of a gene duplication, below branchpoint 128.

   

(3) Summary of Resources

 

Links

• Review (PDF, restricted) Sandra Baldauf: Phylogeny for the Faint of Heart
• Assigned Organisms
• PHYLIP home page
• PHYLIP Web Service at the Institut Pasteur
• Fallback data

Sequences

• All APSES domains (Do not use ... old sequence names!)

APSES domain alignment

• All APSES domains - MUSCLE aligned and sequence names revised

Trees

• APSES domains reference tree

   

[End of assignment]

If you have any questions at all, don't hesitate to mail me at boris.steipe@utoronto.ca or post your question to the Course Mailing List