Shotgun Metaproteomics of the Human Distal Gut Microbiota VerBerkmoes et al. Supplementary Online Information



Download 87.99 Kb.
Date conversion16.05.2016
Size87.99 Kb.
Shotgun Metaproteomics of the Human Distal Gut Microbiota

VerBerkmoes et al. Supplementary Online Information
All datasets, databases and supplementary data files (spreadsheets in .xls format) can be found at http://compbio.ornl.gov/human_gut_microbial_metaproteome


Proteome informatics

All MS/MS spectra were searched with the SEQUEST algorithm (Eng et al., 1994 [(enzyme type, trypsin; Parent Mass Tolerance, 3.0; Fragment Ion Tolerance, 0.5; up to 4 missed cleavages allowed (internal lysine and arginine residues), and fully tryptic peptides only (both ends of the peptide must have arisen from a trypsin specific cut, except N and C-termini of proteins)] and filtered with DTASelect/Contrast (Tabb et al., 2002) at the peptide level [Xcorrs of at least 1.8 (+1), 2.5 (+2) 3.5 (+3)]. Only proteins identified with two fully tryptic peptides from a 22 h run were considered for further biological study. Monoisotopic theoretical masses for all peptides identified by SEQUEST were generated and compared to observed masses. Observed high resolution masses were extracted from .raw files from the full scan preceding best identified spectra; parts per million (ppm) calculations were made comparing each identified peptides observed and theoretical mass. When quality MS/MS spectra didn’t have an observed mass (low intensity) a mass of 0 was reported and ppm was calculated as infinity.


Four database searches were performed with the above settings. The databases are outlined in Supplemental Table S1. The first database (db1) contained two human subject’s metagenomes (Gill et al., 2006) a human database, and common contaminants such as trypsin, human keratins, etc. The existing metagenome databases were deficient in Bacteroides sequences and since Bacteroides are known to be common and abundant in the human intestine (Eckburg et al., 2005) we also included Bacteroides genome sequences in a second database (metadb), plus other sequences from representatives of the normal gut microbiota deposited and available at the Joint Genome Institute (JGI) IMG database (http//img.jgi.doe.gov/), including representatives of Bacteroides, Bifidobacteria, Clostridia and Lactobacilli. In addition, we included representative human pathogens and included distracters that one would not commonly expect in the gut; i.e. environmental bacterial isolates, plus the rice (Oryza Sativa) genome (to help identify food-related proteins). Distracters were uniquely numbered so they could be easily extracted and compared with identifications from proteins thought to be associated with the human microbiome. While this is not actually a false positive rate or false discovery rate as properly defined it gives an indication of how well the method uniquely identifies gut related proteins vs. other protein databases. See discussion below on false discovery rates.
The third and fourth databases were used for estimating false discovery rates on the db1 search as previously described (Lo et al., 2007; Peng et al., 2003). For the third database (db3), we took the db1 database and precisely reversed each protein entry (i.e., n-terminus became c-terminus in each case) and then appended these reversed sequences onto the original database. The same was done for the fourth database (db4), but in this case proteins were not reversed but were randomized (Elias et al., 2007). Proteins with the reversed or randomized orientations were given a unique identifier for easy extraction. All databases, peptide and protein results, MS/MS spectra and supplementary tables for all database searches are archived and made available as open access via the following link: http://compbio.ornl.gov/human_gut_microbial_metaproteome/

Raw files are available on request.


False positives

Currently, there are many ways of estimating error associated with peptide identifications. Until the field of proteomics comes to a conclusion on the proper way of reporting proteomic data, different versions will exist (Tabb et al., 2008). Even the semantics of calling it false discovery rates or false positive rates are under debate. Below we refer to them as false positive rates based on the publications the formulas were derived from. For this large scale study, false-positive rates were used in order to differentiate between true and false peptide identifications rather than false discovery rates (FDR). The overall false-positive rate (FPR) was estimated using the formula: false-positive rate = 2[nrev/(nrev + nreal)]*100 where nrev is the number of peptides identified from the reverse database and nreal is the number of peptides identified from the real database (Peng et al., 2003). A false-positive rate (FPR) was calculated using three different database searches. First, a composite target-decoy database was created with the db1 database. Data analysis consisted of only the forward peptide identification except in the calculation of the FPR where both forward (correct) and reverse (false) identifications were required (Peng et al., 2003) (Elias et al., 2007). The data was separated based on ppm values that were between +10 and -10 ppm (<±10ppm), values that were able to define a charge state for the peptide but were not between + 10 and -10 ppm (>±10ppm), and values that were unable to properly identify the charge state of the full scan peptide mass spectra (unresolved values). This +10 and -10 ppm division is based on earlier research (Lefsrud et al., 2007), with this metaproteome data summarized in Supplementary Figures S1 and S2. The majority of identified peptides have <±10 ppm values, accounting for an average of 83.0% of the total peptides (Table S4). Peptides with >±10 ppm accounted for an average of 8.8% over the data set with 8.2% resulting in unresolved values (Table S4).


The average ppm for the total identified peptides from the human microbial metaproteome data was 33.7 ppm, however when the data was filtered for <±10 ppm this value dropped to -4.1 ppm (Table S5). Also, the average delta amu (atomic mass units) for the total identified peptides for the human microbial metaproteome data was 0.054 amu, but this value dropped to -0.006 amu when filtered for <±10 ppm (Table S5). The FPR for the total identified peptides from the human microbial metaproteome data set was between 3.17% and 1.18% for both samples and duplicate runs (Table S6) when all peptides were considered. However, when the data was filtered for only those peptides with <±10 ppm these values dropped significantly, the highest FPR for filtered data <±10 ppm was 0.21% and the lowest was 0.05%.
A second approach to estimate the error associated with peptide identifications also involved using a composite target-decoy database with db1, except in this case each protein sequences was randomly shuffled creating a “decoy” database. The purpose of this decoy database was to create a more randomized database by shuffling the amino acids of each protein rather than simply reversing the n-terminus and c-terminus. A shuffled database creates more nonsense, thus, reducing the overall chances of making false identifications. Any proteins that are identified with the decoy database indicate that the forward peptide is in fact illegitimate. A FPR can be estimated using a similar formula as previously described. The number of peptides identified from the shuffled database is multiplied by 2 and divided by the sum of all shuffled peptides plus forward peptides identified from the target database. A FPR was estimated for both samples and runs (Table S7) and as described (Elias et al., 2007) was similar to the rate determined by the reverse database method.
We also estimated the false discovery rate in the metadb database search by different method. Here we were interested in seeing the number of unique and total peptides identified to known gut isolates, metagenomes, human proteins and rice proteins vs. distracter sequences including the genomes of Leptospirillum ferrooxidans, Shewanella oneidensis MR-1, Rhodopseudomonas palustris and others. For the entire list of database entries, please visit the website url: http://compbio.ornl.gov/human_gut_microbial_metaproteome/databases/ .
The majority of peptides that matched to the distracter database were in fact non-unique peptides. These shouldn’t be counted as false peptides since they overlap with peptides and proteins from isolates, metagenomes etc that could be in the gut. Thus we only counted unique peptides matching the distracter sequences. A FPR was estimated for both samples per run by comparing the number of total unique peptides from the distracter database to both the total unique peptides from the rest of the database and the total peptides from the rest of the database by the same equation given above for the other two methods of determining FPR. When only unique peptides were considered a false positive rate of 3-5% was found. When all peptides were considered then a false positive rate of ~1% was found. These results are very similar to the false positive rates determined for db1 with the reverse and shuffled methods.
Assigning proteins to COG groups

To create Figure 2 in the manuscript the JGI IMG/M database was used (http://imgweb.jgi-psf.org/cgi-bin/m/main.cgi). This database contains COG information for all proteins which we used for COG assignment for bacterial isolates and the human gut metagenome sequences. We found that 37.9% of the proteins could not be assigned to COGs when screening the metagenome databases, similar to 34% estimated for the existing metagenome data deposited at JGI. Most of these proteins were hypothetical or conserved hypotheticals, therefore, they were assigned "S" which is function unknown. On the other hand, several known proteins, such as DNA-directed RNA polymerase, did not have an assigned COG function. In these cases, we assigned them based on our own knowledge to a COG category and if we did not know, we assigned them to "R" which is general function prediction only. For example, we assigned the DNA-directed RNA polymerase protein to "K", for transcription. Supplemental Table S2 and S3 have COG entries for all detected microbial proteins.


Proteins found in all replicates and runs

We extracted the list of all proteins found in each technical replicate and in both biological samples, i.e. the “conserved proteins”. The list for db1 and associated NSAF values for each run can be found in supplemental table 9 first tab (474 proteins total), the list for metadb and associated NSAF values for each run can be found in supplemental table 10 first tab (749 proteins total).


Label free Quantitation methods.

The label free methods rely on intrinsic values obtained in the course of the experiment such as peak intensities or areas of peptides (Old et al., 2005), spectral counts (Liu et al., 2004) and normalized spectral abundance factors (Florens et al., 2006; Zybailov et al., 2006) to quantify peptides and thus proteins. They have grown in popularity due to simplicity, cost considerations and the fact they can be used on any sample assuming proper experimental design is implemented. There is strong effort in the proteome informatics community to develop better tools and statistics for label free methods (Zhang et al., 2006), (Lu et al., 2007). The absolute protein expression (APEX) method recently developed by Lu et al. (2007) may allow for a better statistical comparison of two data sets but was derived specifically for isolate data such as E. coli and yeast. The APEX method will not be applicable to complex microbial communities because it requires an estimate on the number of proteins being expressed in the system. This is not possible with a complex microbial community from the gut where it’s impossible to estimate the number of different cell types, species or total proteins. Thus, we applied simpler methods for protein quantitation based on spectral counts and normalized spectral abundance factors (NSAFs). Unlike spectral counting, NSAF is based on spectral counts but takes into account protein size and the total number of spectra from a run, thus normalizing the relative protein abundance between samples. A spectral abundance factor (SAF) is first calculated by dividing the number total number of spectral counts for each protein by its mass or length. The NSAFs are then calculated by normalizing each SAF to one by dividing by the sum of all SAFs for all protein (Florens et al., 2006; Zybailov et al., 2006). We first compared the NSAF results from all proteins found in Samples 7 and 8 using db1 and metadb, but limited ourselves to those that were only found in both technical replicates. As can be seen in Figures S3 and S4 the reproducibility of technical runs, based on NSAFs is high with an R2 of 0.77 for Sample 7 and 0.85 for Sample 8 with the metadb (similar results with db1 not shown). We then averaged the NSAF values for Samples 7 and 8, but left all proteins in the graph to determine what was found uniquely in one sample and not the other. The results from this comparison are found in Figure S5 and suggest some proteins that differed significantly in expression between the two samples. The figure indicates four major clusters; two clusters are located on each extreme where proteins were found only in one individual but not the other. The other two intermediate clusters were found where proteins were present, but expressed in different amounts in comparison with the other individual. We found the cutoff of these clusters to be around a log ratio difference between 1.1 and 2.4. Thus, we created a sub-table of those proteins showing large differences in expression between the two samples in db1 via those boundaries. We further manually curated the data to only include only those proteins represented by identification in the both runs were they were considered “higher protein abundance” as well as 2x increase in average spectral counts over the other sample (Supplementary Table S9, second tab). In total 225 proteins were found differentially expressed between Sample 7 and Sample 8. This same process was repeated for the metadb dataset. Again there was similar reproducibility in the technical replicates for the two samples (data not shown) and a similar trend in the comparison of Samples 7 and 8 (Figure 4 manuscript). In total, 308 proteins were found differentially expressed between Sample 7 and Sample 8 (for metadb Supplementary Table S10, second tab).
Hypothetical Protein Prediction

Sequences of all hypothetical proteins identified above (116 from the Gill metagenomes and 89 from bacterial isolate genomes), were submitted to the distant homology recognition server FFAS03 (Jaroszewski et al., 2005). This server automatically builds a sequence profile for the submitted sequences and compares it against a curated library of sequence profiles, encompassing several sets of annotated proteins (COG, PDB, PFAM and structure determination targets from the JCSG structural genomics center). In independent tests, FFAS03 was shown to consistently outperform PSI-BLAST and other distant recognition algorithms. In Supplementary Table S11 we have summarized results of the analysis. For 80% of the hypothetical proteins a statistically significant match (Z-score below 9.5) to one of the proteins in the reference databases can be obtained. Functions of the matching proteins were used to assign a provisional function for the hypothetical proteins identified in this study. It is important to such analysis can narrow down the possible function of the analyzed protein but, because of the distant homology, detailed function may have diverged from that of the homolog identified in this analysis. More detailed analysis of active site residue conservation and other features is necessary for more detailed function assignment. All the FFAS03 results are available from the FFAS03 server at http://ffas.burnham.org/ffas-cgi/cgi/login.pl (Login: Janet_new, password: Janet_new). Links provided on the site can be followed to obtain detailed alignments, three dimensional models and other information.


Figure S4 (manuscript). Supplementary information:
Genome neighborhood analysis was performed and figures were prepared using the SEED environment for genome annotations as implemented at the National Microbial Pathogen Data Resource project website (http://www.nmpdr.org/cur/FIG/wiki/view.cgi/Main/WebHome). Sequences of BF3046 and BT2437 genes, representing the two families discussed in the text, were compared against the Bacteroides fragilis ATCC genome (NCBI Taxonomy Id: 272559).
B) BF3046 conserved genomic neighborhood

(1) red CAC2564 family, as discussed in the text

(2) green new family of hypothetical proteins, as discussed in the text

(3) light brown LysR family transcriptional regulator

(4) blue N-succinyltransferase

(5) yellow DNA damage inducible protein

(6) aquamarine new family of hypothetical proteins

(7) violet telluride resistance protein

(8) dark green multiple antibiotic resistance protein

(9) dark brown new family of hypothetical proteins

(10) light blue Glycerophosphoryl phosphodiesterase
C) BT2437 conserved genomic neighborhood

(1) red BT2437 family, as discussed in the text

(2) green new family of hypothetical proteins, as discussed in the text

(3) brown tripeptidyl aminopeptidase

(4) blue MarR family transcriptional regulator

(5) yellow Aspartate decarboxylase

(6) aquamarine Coenzyme A disulphate reductase
Supplementary Tables
Supplementary Table S1. Description of databases that were screened, see website for complete breakdown.


Database

Sequences included

References

db1

Metagenome, Individual 7

Metagenome, Individual 8

Human proteins

Gill et al. 2006

Gill et al. 2006




metadb

db1

Human commensals and pathogens

Bacteroides

Bifidobacterium

Etc.


Environmental isolates

Leptospirillum

Etc.


Rice (Oryza Sativa)

JGI/IMG


db3 and db4

db1 in reverse (db3) or random (db4) orientation and appended to db1




Supplementary Table S2. Protein Identification with NSAF counting from all runs from db1. The results from each individual run can be found on the website.

Supplementary Table S3. Tab 1: Categorical breakdown of identifications to each database type. Tab 2: Protein Identification with NSAF counting from all runs from Metadb. The results from each individual run can be found on the website.


Supplementary Figure S1. PPM Variability Verse XCorr on Forward Peptide Distribution from Human Microbial Metaproteome Data (db1).



Supplementary Figure S2. Delta Atomic Mass Units Verse PPM on Forward Peptide Distribution from Human Microbial Metaproteome Data.




Supplementary Table S4. Identified Forward Peptides from db1 database search.







Forward Identified Peptides

% of Forward Identified Peptides

Sample ID

Values <±10ppm

Values >±10ppm

Unresolved values

Values <±10ppm

Values >±10ppm

Unresolved values

Sample 7

Run1


2316

279

195

83.0%

10.0%

7.0%

Sample 7

Run2


2592

275

330

81.1%

8.6%

10.3%

Sample 8

Run1


3664

357

309

84.6%

8.2%

7.1%

Sample 8

Run2


3397

355

350

82.8%

8.7%

8.5%

Supplementary Table S5. Average Delta AMU and Average ppm.






Average Delta AMU

Average ppm

Sample ID

Values excluding unresolved values

Values <±10ppm

Values excluding unresolved values

Values <±10ppm

Sample 7

Run1


0.065

-0.007

46.9

-4.4

Sample 7

Run2


0.041

-0.003

20.5

-2.2

Sample 8

Run1


0.068

-0.008

43.6

-5.2

Sample 8

Run2


0.040

-0.006

24.2

-4.2

Supplementary Table S6. Total Identified Peptides, Identified Reverse Peptides and False Positive Rate from db1 determined by reverse database method (db3).



Sample ID

Total Identified Forward Peptides

Identified Reverse Peptides

False Positive Rate

Values <±10ppm

Values >±10ppm

Unresolved values

Total False Positive

False Positive <±10ppm

Sample 7

Run1


2790

3

40

2

3.17%

0.21%

Sample 7

Run2


3197

1

14

4

1.18%

0.06%

Sample 8

Run1


4330

4

37

11

2.37%

0.18%

Sample 8

Run2


4102

1

32

7

1.93%

0.05%

Supplementary Table S7. Total Identified Peptides, Identified Shuffled Peptides and False Positive Rate from Human Microbial Metaproteome Data (db1) determined by random database method (db4)




Sample ID

Total Identified Forward Peptides

Identified Shuffled Peptides

False Positive Rate

Values <±10ppm

Values >±10ppm

Unresolved values

Total False Positive

False Positive <±10ppm

Sample 7

Run1


2789

3

31

1

2.48%

0.21%

Sample 7

Run2


3279

3

51

3

3.42%

0.18%

Sample 8

Run1


4324

0

42

6

2.20%

0.00%

Sample 8

Run2


4230

5

40

3

2.24%

0.23%

Supplementary Table S8. Total Identified Distracter Peptides, Identified Gut Peptides and False Positive Rate from Human Microbial Metaproteome Data (metadb)




Sample ID

Unique Distracter Peptides

Non-Unique Distracter Peptides

Unique Gut Peptides

Total Gut Peptides

FPR (%)

unique distracter & unique gut peptides

FPR (%)

unique distracter & total gut peptides

Sample 7 Run 1

30

272

1135

5080

5.15

1.17

Sample 7 Run 2

31

184

1436

6036

4.23

1.02

Sample 8 Run 1

31

205

1899

7115

3.21

0.87

Sample 8 Run 2

33

151

1808

6511

3.59

1.01

Supplementary Figure S3. Comparison of NSAF values. Sample 7, run 1 and run 2 NSAF values are plotted on a log scale. The dark blue squares represent all of the proteins that were identified in both runs from metadb.



Supplementary Figure S4. Comparison of NSAF values. Sample 8, run 1 and run 2 NSAF values are plotted on a log scale. The dark blue squares represent all of the proteins that were identified in both runs from metadb.


Supplementary Figure S5. Comparison of NSAF values for Samples 7 and 8. NSAF values were averaged amongst two individual technical runs pre sample and plotted on a log scale. The dark blue squares represent all of the proteins identified in each sample from db1. The straight diagonal line is for visualizing the location of all proteins that had approximately equal expression in both samples


Supplementary Table S9. Tab one Proteins found in both samples and replicates with db1. Tab two proteins showing abundance differences based on NSAF calculations Samples 7 and 8 with db1.


Supplementary Table S10. Tab one Proteins found in both samples and replicates with metadb. Tab two proteins showing abundance differences based on NSAF calculations Samples 7 and 8 with metadb.
Supplementary Table S11. All identified hypothetical proteins and predicted functions. Column B is original predicted function, column C is the new computational predicted function. More detailed listing can be found on website.
Supplementary References
Elias JE, Gygi SP (2007) Target-decoy search strategy for increased confidence in large-scale protein identifications by mass spectrometry. Nature Methods 4:207-214.
Lefsrud MG, et al., (2007) Proteogenomics reveals Key Insight into the Microbial Activities of Enhanced Biological Phosphorus removal in Activated Sludge. ASMS Conference Proceedings, Indianapolis, IN. June 3-7, 2007.
Liu H, Sadygov RG, Yates III JR, (2004) A model for random sampling and estimation of relative protein abundance in shotgun proteomics. Anal Chem 76:4193-4201.
Old WM, et al. (2005) Comparison of label-free methods for quantifying human proteins by shotgun proteomics. Mol Cell Proteomics 4:1487-1502.
Tabb DL (2008) What’s driving false discovery rates? J Proteome Res 7:45-46.
Zhang B. et al., (2006) Detecting differential and correlated protein expression in label-free shotgun proteomics. J Proteome Res 5: 2909-2918.


The database is protected by copyright ©essaydocs.org 2016
send message

    Main page