Exercise PSI-BLAST

Written by: Carolina Barra Quaglia

Overview

In this exercise you will learn how to

• Critically assess when BLAST fails (e.g., no significant hits) and explore alternative strategies.
• Use PSI-BLAST to search for remote homologues of a given protein sequence (an orphan gene).
• Interpret iterative PSI-BLAST output (number of hits, coverage, E-value, identity/positives) to assess significance.
• Save and reuse a PSSM (profile) to search specialized databases (e.g., PDB, RefSeq) for structural or functional insights.
• Make a reasoned functional hypothesis for a gene of unknown function (the orphan gene) based on remote homology, domain architecture, structural clues, conserved residues, etc.

Introduction: What are orphan genes?

In genomics and evolutionary biology, an orphan gene (also called a taxonomically-restricted gene, TRG) is a gene for which no detectable homologue exists outside a given species or lineage.

In today's exercise, you will use the iterative BLAST program (PSI-BLAST) to calculate sequence profiles and to discover what is the function of a real human orphan gene called C22orf45. We will aim to do a research‐style annotation of a “dark” gene that is not well annotated.

Interestingly this gene (C22orf45) may have once originated from 'Junk DNA' and it is supposed to have gained function through mutations that allowed it to start producing proteins. (You can find more known information of the gene here: C22orf45 Publications)

When BLAST fails

Here you have the protein‐coding sequence with unknown function from the human gene named "C22orf45". This gene is currently poorly annotated in the human genome, and initial BLAST searches show no obvious homologues. Your task is to use PSI-BLAST to search for remote homologues, explore whether this gene might belong to a known protein family, gain insight into its possible function and structure, and reflect on its status as a potential orphan gene.

>C22orf45
MEQDWQPGEEVTPGPEPCSKGQAPLYPIVHVTELKHTDPNFPSNSNAVGTSSGWNRIGTG
CSHTWDWRFSCTQQALLPLLGAWEWSIDTEAGGGRREQSQKPCSNGGPAAAGEGRVLPSP
CFPWSTCQAAIHKVCRWQGCTRPALLAPSLATLKEHSYP

First we are going to check that BLAST does not find any homologous sequence. Go to the BLAST web-site at NCBI. Select blastp as the algorithm. Paste in the query sequence. Change the database from nr to Protein Data Bank (pdb), and press BLAST (Figure 1).

Figure 1. Partial screenshot of the Blast interface. The red arrow shows the settings change to the database to pdb

Note: If BLAST collapses you can check pre-run results using this ID: GPHA6F6K016 in here [Lookup BLAST Job]

• QUESTION 1: How many significant hits does BLAST find (E-value < 0.005)?

Trying another approach

Now go back to the search web-site of BLASTP. Paste in the query sequence again. This time, set the database to Non-redundant protein sequences (nr) and select PSI-BLAST (Position-Specific Iterated BLAST) as the algorithm (Figure 2).

IMPORTANT: To allow for more remote homologues we will increase the E-value of our search to 100. Note that this will riks finding non-homologous proteins in our results.

Figure 2. Partial screenshot of the PSI-BLAST interface. The red arrow shows the settings change to PSI-BLAST.

Note: If BLAST collapses you can check pre-run results using this ID: GPJM9RYM014 in here [Lookup BLAST Job]

• QUESTION 2: How many hits do you obtain (E-value < 10)? (Tip: you can see the number by selecting all hits (clicking All under Sequences producing significant alignments with E-value BETTER than threshold) and then looking at the number of selected hits)

• QUESTION 3: Excluding the identical match, what is the highest sequence identity and coverage among the hits? Are the hits only human, or do they include other mammals/vertebrates?

• QUESTION 4: Based on the first result, is there a clear homologue in non‐human species? What does that suggest about the gene’s taxonomic distribution?

Constructing the PSSM

Now run a second BLAST iteration in order to construct a PSSM (Position-Specific Scoring Matrix). Press the Run button at Run PSI-Blast iteration 2 (you can find it at both the bottom and top of the results table).

• QUESTION 5: How many significant hits does BLAST find (E-value < 0.005)?
• QUESTION 6: How large a fraction of the query sequence do the 20 most significant hits match (do not include the first hit since this is identical to the query)?
• QUESTION 7: Why does BLAST come up with more significant hits in the second iteration? Make sure you answer this question and understand what is going on!

Figure 3. Partial screenshot of the PSI-BLAST interface before running Iteration 2. The red square shows how to change the settings for the run.

Saving and reusing the PSSM

This time, we will not ask you to look for PDB identifiers manually among the significant hits. Instead, you should save the PSSM that PSI-BLAST has created and use it for searching PDB directly.

Go to the top of the PSI-BLAST output page and click Download All, then click PSSM. Save the file to a place on your computer where you can find it again! You can take a look at this file using Geany, but it is really not meant to be human-readable.

Then, open a new BLAST window (this is important—you need your first BLAST window again later) where you again select PSI-BLAST as the algorithm. Select pdb as the database. Do not limit your search to Archaea this time. Click on Algorithm parameters to show the extended settings. Click the button next to Upload PSSM and select the file you just saved. Note: You don't have to paste the query sequence again, it is stored in the PSSM!

• QUESTION 8: Do you find any significant PDB hits now? If yes, how many?
• QUESTION 9: What are the PDB identifiers and the E-values for the two best PDB hits?
• QUESTION 10: What are the values for Query coverage, sequence identity, and sequence similarity (Positives) for the two best PDB hits? (Tip: click on the description to get to the actual alignment between the query sequence and the PDB hit)?
• QUESTION 11: What is the function of these proteins?

One more round

Let's try one more iteration of PSI-BLAST:

• Go back to your first BLAST window (the one with the results from the nr database limited to Archaea) and press the Run button at Run PSI-Blast iteration 3.
• Save the resulting PSSM file (make sure you give it a different name!).
• Launch a new PSI-BLAST search against pdb in all organisms using this PSSM (you may have to click on Clear to erase your first PSSM file from the server).
• QUESTION 12: Answer questions 8-10 again for the new search.

Finding a remote homolog (on your own)

PSI-Blast is not only useful for finding a remote homolog in a specific database such as PDB — now it is time to search the broader database "Reference proteins" (refseq_protein). (Note: we would have liked to do this exercise in the broadest database nr, but that search runs into technical problems). PSI-BLAST can be used in the same way for finding a remote homolog in a specific organism or taxonomic group. Your task in this round is to find out whether the protein with the UniProt ID GPAA1_HUMAN has a homolog in the genus Trypanosoma (unicellular parasites which cause diseases like sleeping sickness or Chaga's disease).

• First, try a standard BlastP (where you set Organism to Trypanosoma, Database to refseq_protein (not refseq_select), switch the Low complexity regions filter off, and set the E-value threshold to 10).
• QUESTION 13: Do you find any significant (E<0.005) hits? What is the E-value of the best hit?
• Then, try PSI-BLAST. Hint: You need to search in all organisms (still using refseq_protein) to build a PSSM, then save your PSSM and use that to search in Trypanosoma.
• QUESTION 14: How many significant (E<0.005) hits do you find now? What is the E-value of the best hit?

Concluding remarks

Now you have seen the power of sequence profiles in general and the PSI-BLAST program in particular. Using sequence profiles you have been able to identify a relationship between protein sequences far below 30% sequence similarity. Further, you have made qualified predictions on the protein function and selected a set of essential amino acids suitable for experimental validation of the structural and functional predictions.