A repository of assays to quantify 10,000 human proteins by SWATH-MS

Abstract

Mass spectrometry is the method of choice for deep and reliable exploration of the (human) proteome. Targeted mass spectrometry reliably detects and quantifies pre-determined sets of proteins in a complex biological matrix and is used in studies that rely on the quantitatively accurate and reproducible measurement of proteins across multiple samples. It requires the one-time, a priori generation of a specific measurement assay for each targeted protein. SWATH-MS is a mass spectrometric method that combines data-independent acquisition (DIA) and targeted data analysis and vastly extends the throughput of proteins that can be targeted in a sample compared to selected reaction monitoring (SRM). Here we present a compendium of highly specific assays covering more than 10,000 human proteins and enabling their targeted analysis in SWATH-MS datasets acquired from research or clinical specimens. This resource supports the confident detection and quantification of 50.9% of all human proteins annotated by UniProtKB/Swiss-Prot and is therefore expected to find wide application in basic and clinical research. Data are available via ProteomeXchange (PXD000953-954) and SWATHAtlas (SAL00016-35).

www.nature.com/scientificdata

OPEN

SUBJECT CATEGORIES

Systems biology

Proteomics

Received: 11 June 2014

Accepted: 06 August 2014Published: 16 September 2014

A repository of assays to quantify 10,000 human proteins by SWATH-MS

George Rosenberger1,2, Ching Chiek Koh1,3, Tiannan Guo1, Hannes L. Rst1,2,Petri Kouvonen1, Ben C. Collins1, Moritz Heusel1,4, Yansheng Liu1, Etienne Caron1,Anton Vichalkovski1, Marco Faini1, Olga T. Schubert1,2, Pouya Faridi1,5, H. Alexander Ebhardt1, Mariette Matondo1, Henry Lam6, Samuel L. Bader7, David S. Campbell7, Eric W. Deutsch7, Robert L. Moritz7, Stephen Tate8 & Ruedi Aebersold1,9

Mass spectrometry is the method of choice for deep and reliable exploration of the (human) proteome. Targeted mass spectrometry reliably detects and quanties pre-determined sets of proteins in a complex biological matrix and is used in studies that rely on the quantitatively accurate and reproducible measurement of proteins across multiple samples. It requires the one-time, a priori generation of a specic measurement assay for each targeted protein. SWATH-MS is a mass spectrometric method that combines data-independent acquisition (DIA) and targeted data analysis and vastly extends the throughput of proteins that can be targeted in a sample compared to selected reaction monitoring (SRM). Here we present a compendium of highly specic assays covering more than 10,000 human proteins and enabling their targeted analysis in SWATH-MS datasets acquired from research or clinical specimens. This resource supports the condent detection and quantication of 50.9% of all human proteins annotated by UniProtKB/ Swiss-Prot and is therefore expected to nd wide application in basic and clinical research. Data are available via ProteomeXchange (PXD000953-954) and SWATHAtlas (SAL00016-35).

Design Type(s)

Measurement Type(s)

Technology Type(s)

Factor Type(s)

Homo sapiens monocyte neutrophil gut kidney lung muscle blood platelet blood plasma 293 cell THP-1 cell U2

OS cell HeLa cell NCI60 LNCAP cell CAL-51 cell

1Department of Biology, Institute of Molecular Systems Biology, ETH Zurich, CH-8093 Zurich, Switzerland. 2PhD Program in Systems Biology, University of Zurich and ETH Zurich, CH-8093 Zurich, Switzerland. 3Ruprecht Karls University of Heidelberg, DE-69117 Heidelberg, Germany. 4PhD Program in Molecular and Translational Biomedicine, Competence Centre for Systems Physiology and Metabolic Diseases (CC-SPMD), University of Zurich and ETH Zurich, CH-8093 Zurich, Switzerland. 5Department of Phytopharmaceuticals (Traditional Pharmacy), School of Pharmacy and Pharmaceutical Sciences Research Center, Shiraz University of Medical Sciences, 71345-1583 Shiraz, Iran. 6Division of Biomedical Engineering and Department of Chemical and Biomolecular Engineering, The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong, China. 7Institute for Systems Biology, Seattle, Washington 98109-5234, USA. 8AB SCIEX, Concord, Ontario L4K 4V8, Canada. 9Faculty of Science, University of Zurich, CH-8057 Zurich, Switzerland.

Correspondence and requests for materials should be addressed to R.A. ([email protected])

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 1

reference design replicate design protein expression proling quality control testing design

protein expression proling

mass spectrometry assay

purication proteolysis

Sample Characteristic(s)

www.nature.com/sdata/

Background & Summary

Much of science depends on reproducible and quantitatively accurate measurements. In the molecular life sciences, technological advances have moved the large-scale measurement of the molecules that constitute living cells to the forefront. For example, next generation sequencing (NGS) technology has made the routine quantitative analysis of complete genomes and transcriptomes a reality in many laboratories. In contrast, the analysis of proteins, the predominant class of functional effector molecules of the cell, has remained challenging and not generally accessible.

In most laboratories, proteins in complex samples are detected and quantied via immunoassays where specic reagents, frequently antibodies, are used to generate a signal that indicates the presence and quantity of a specic protein in a sample. Large-scale programs, exemplied by the Human Protein Atlas project1 and commercial efforts have attempted to generate specic afnity reagents for each human protein and to make them widely accessible. Undoubtedly, the availability of these reagents has the potential to signicantly impact life science research. At present, however, only a subset of the proteome is routinely measurable by afnity reagents, with the consequence that much of the literature knowledge about proteins is focused on a relatively small subset of the proteome, the fraction for which afnity reagents are readily available2. Furthermore, at least some of these reagents are of unknown and dubious quality3, limiting the utility of the results obtained. Therefore, life science research would greatly benet from the general availability of validated, high quality assays for the human proteome.

Mass spectrometry (MS) has become the method of choice for the deep and reliable exploration of the (human) proteome. In particular, liquid chromatography-coupled tandem mass spectrometry (LC-MS/ MS) operated in data-dependent acquisition mode (DDA), has achieved remarkable progress in the identication of proteins in complex samples. Proteome-wide identication and quantication have been achieved for human cell lines46 and efforts are being made to characterize at least one protein product of all 20,300 protein-coding genes. An example of such an effort is the HUPO Chromosome-centric Human Proteome Project7, which could detect at least one single peptide for ~ 14,000 proteins to date8. Recently, two independent studies from Kim et al.9 and Wilhelm et al.10 reported the cumulative analysis of more than 2,000 and 16,800 LC-MS/MS measurements, respectively, that yielded a map of identied peptides corresponding to 17,294 and 18,097 human protein-coding genes, respectively. However, the high degree of proteome coverage achieved in these studies depends on protein or peptide fractionation techniques like strong anion exchange (SAX) or off-gel electrophoresis (OGE) prior to MS analysis, to distribute the sample complexity among several instrument injections and the integration of the results of a high number of LC-MS/MS measurements. The high technical complexity and cost of generating and analyzing deep proteomic datasets and well understood technical tradeoffs11 have so far prohibited the distribution of this powerful technology to a large number of laboratories12 and limited the reproducibility of datasets generated within and across laboratories1315 thus limiting the breadth of its impact.

We and others have proposed that targeted mass spectrometry has the potential to democratize mass spectrometry-based proteomics, i.e., to make most or all proteins reliably detectable and quantiable in a large number of laboratories16. Under the umbrella of the HUPO Human Proteome Project17, we launched a program to make the targeting technology and associated measurement assays generally accessible. In targeted proteomics, exemplied by the prototypical quantitative MS technique selected reaction monitoring (SRM), also referred to as multiple reaction monitoring (MRM), predetermined sets of proteins are accurately quantied by means of specic mass spectrometric assays that have to be generated a priori once for each targeted protein. In support of SRM-based protein quantication, extensive, in some cases proteome-wide, assay libraries and empirical measurements of the same assays across multiple samples to judge performance of these assays have been created1821 and made freely

accessible (http://www.srmatlas.org, http://www.peptideatlas.org/passel/). While SRM and the recent implementations of the related method parallel reaction monitoring (PRM) on high performance mass spectrometers22 remain the best performing quantitative MS methods, they are limited by the relatively low number of proteins (50100) that can be quantied in a single injection and the fact that the targeted proteins need to be specied for each sample prior to data acquisition.

Recently, we introduced SWATH-MS, a new mass spectrometric technique that combines data-independent acquisition (DIA) with targeted data extraction on a high resolution mass spectrometer23. In

DIA mode, the instrument deterministically fragments all precursor ions within a predened mass-to-charge (m/z) range and acquires convoluted product ion spectra, containing the fragment ions of all concurrently fragmented precursors. By rapidly and recursively scanning through consecutive, adjacent precursor ion windows, termed swathes, the full precursor ion m/z range of trypsinized peptides is covered and consequently, fragment ion spectra of all precursors within a user dened retention time (RT) versus m/z window are recorded over time. This results in a data set that is continuous in both fragment ion intensity and retention time dimensions and essentially represents a digital recording of the protein sample analyzed. Within these data, specic peptides can be identied and quantied by applying a targeted data extraction strategy that results in signals analogous to those obtained by SRM, where sets of fragment ion signals uniquely associated with the targeted peptide are recorded over chromatographic time and the concluding peak groups are used as evidence for the conclusive identication and quantication of the targeted peptide in a sample. The data analysis depends on a priori assays, derived from fragment ion spectra of the targeted peptides that are best generated in the same high resolution instrument used for

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 2

www.nature.com/sdata/

SWATH-MS acquisition. In contrast to SRM where the targeted peptides need to be determined prior to data acquisition, SWATH-MS datasets are recorded independently and can then be perpetually re-mined using the targeted analysis strategy. Using freely or commercially available software (OpenSWATH24,

Skyline25, PeakView (AB SCIEX, Concord, Canada) or Spectronaut (Biognosys AG, Schlieren, Switzerland)) and a proteome-specic assay library, SWATH-MS can be used to carry out protein quantication at performance metrics that are comparable to SRM but at a much higher throughput23,24.

To date, most studies using SWATH-MS have relied on the generation of sample-specic assay libraries, acquired in fractionated or enriched samples, injected prior to SWATH-MS acquisition on the same instrument operated in DDA mode23,24,2629. Here we present a generic large-scale human assay

library to support protein quantication by SWATH-MS. It is optimized for targeted data analysis of SWATH-MS data sets acquired on AB SCIEX TripleTOF 5600+ Systems. It consists of 1,164,312 transitions identifying 139,449 proteotypic peptides and 10,316 proteins. It was generated by combining the results from 331 measurements of fractions from different cell lines, tissue and afnity enriched protein samples. The assays consist of precursor and fragment ion m/z, normalized RT and relative ion intensities, making this resource readily applicable for data analysis using state-of-the-art analysis software. We further demonstrate that the results and biological conclusions obtained with the combined assay library are comparable to those obtained with sample-specic assay libraries and applicable across laboratories. We expect that this resource will contribute signicantly to the simplied and reproducible analysis of human proteome samples across studies and laboratories.

Methods

Sample overview

To achieve broad representation of the human proteome we analyzed protein samples from a range of human cell and tissue types. The specic sample types analyzed are summarized in Table 1 (for complete annotation see Supplementary Table 1) and include human cell lines, tissues such as kidney, gut, monocytes, neutrophils and human blood. To increase the contents of the assay library of proteins of low abundance, we also added spectra obtained from afnity puried protein complexes. Figure 1 illustrates the experimental workow.

Cell culture, tissue sampling and protein-level separation

Cell culture. HEK293 cell samples were essentially generated as described before30.

HeLa and U2OS cells were obtained from ATCC and grown in DMEM with GlutaMAX-1 (Invitrogen, Carlsbad, CA) supplemented with 100 U/ml penicillin, 100 g/ml streptomycin and 10% fetal bovine serum at 37 C, 5% CO2, in a humidied incubator.

NCI60 and LNCaP cells were obtained as frozen, non-viable cell pellets from the Developmental Therapeutics Program (DTP), National Cancer Institute (NCI NIH).

CAL51 cells were grown in RPMI 1640 media depleted of arginine and lysine (Invitrogen) and supplemented with 10% Fetal Bovine Serum (Invitrogen, 26400-044) (FBS). The media was supplemented with 100 U/ml penicillin, 100 g/ml streptomycin, 2 mM L-glutamine (Gibco). THP1 cell line samples were generated as described before31.

Patient specimen. Kidney tissue samples (n = 18) were collected at the time of surgery and were provided by Dr Silke Gillessen, Dr Markus Joerger and Dr Wolfram Jochum (Kantonsspital St Gallen, Switzerland).

Gut tissue samples (n = 18) were provided by Dr Marko Kalliomaki (Turku University Hospital, Finland). The samples were collected during the diagnostic colonoscopy from nine patients.

Lung tissues samples (n = 12) were provided by Dr Wim Timens and colleagues at the University

Medical Center Groningen, Netherlands.

Muscle tissue samples (n = 12) were provided by Dr Carsten Jacobi (Novartis Pharma AG, Switzerland).

Blood plasma samples were obtained from 32 female healthy donors and mixed together before further processing. Plasma was depleted of the 14 most abundant plasma proteins with the multiple afnity removal system (MARS Hu-14 spin cartridge; Agilent Technologies) according to the manufacturers protocol. Depleted samples were exchanged with a 3,000 Da molecular weight cutoff lter (Pall Corporation) and denatured in 6 M urea and 0.1 M ammonium bicarbonate before digestion with trypsin and LC-MS analysis.

Monocytes & neutrophils samples were isolated from patients with active tuberculosis and were provided from Prof Dr Stefan Kaufmann (Max Planck Institute for Infection Biology, Berlin, Germany).

Puried platelets from a healthy donor were provided by Prof. Ohad Medalia (University of Zurich, Switzerland). The purication and protein digestion were performed essentially as described before32.

All clinical specimens were obtained under IRB approval and accepted protocols. Written informed consent was obtained from all patients from whom biopsy samples were taken.

Afnity purication. Previously published datasets from afnity purication samples of the 14-3-3 beta network were included27. In addition, pull-downs of human kinase baits according to the same protocol were generated and pooled for the purpose of spectral library generation.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 3

www.nature.com/sdata/

Sample type

HEK293 (CL) AP (Kinases) Trypsin None 12

HEK293 (CL) AP (14-3-3) Trypsin None 29

HEK293 (CL) SEC Trypsin None 81

HEK293 (CL) None Trypsin OGE 11

HEK293 (CL) None Trypsin None 1

U2OS (CL) None PCT None 13

HeLa (CL) None PCT None 9

U2OS and HeLa (CL) None Trypsin OGE 24

NCI60 (CL) None PCT None 13

NCI60 (CL) None Trypsin OGE 24

CAL51 (CL) None Trypsin None 5

CAL51 (CL) None Trypsin 1D GE 2

THP1 (CL) None Trypsin OGE 27

LNCaP (CL) None Trypsin SAX 6

LNCaP (CL) None Trypsin None 1

Kidney (T) None Trypsin 1D GE 15

Kidney (T) None PCT None 16

Large intestine (T) None Trypsin OGE 24

Muscle (T) None PCT None 3

Lung (T) None PCT None 2

Blood plasma (T) None Trypsin SAX 8

Monocytes (T) None Trypsin None 1

Neutrophils (T) None Trypsin None 1

Puried platelets (T) None Trypsin None 3

Total 331

Table 1. Overview of the contents of the combined assay library. CL refers to cell line and T refers to tissue, indicating the source of the specimen. The full sample annotation is provided in Supplementary Table 1.

Size-exclusion chromatography (SEC). Cycling HEK 293 wt cells were lysed essentially as described before27, except that the lysis buffer was not supplemented with avidin. Lysates were cleared by 15 min of ultracentrifugation (100,000 g, 4 C, Beckman Coulter Optima TLX ultracentrifuge) and lysis buffer was exchanged to SEC buffer (50 mM HEPES pH 7.5, 150 mM NaCl) over 30 kDa molecular weight cutoff membrane (Amicon Ultra-15, Millipore, MA, USA), at a ratio of 1:50 in three dilution and re-concentration steps of 1:2, 1:5 and 1:5. Proteins were concentrated to 2530 mg/ml as judged by OD280 and were then cleared from precipitates by 5 min of centrifugation at 16.9 krcf at 4 C (Eppendorf 5,418R) before protein level fractionation. SEC was performed on an Agilent 1,100 milliliter ow HPLC system (Agilent, CA, USA) utilizing a Yarra-SEC-4000 column (pore size 500 , dimensions 300 7.8 mm, Phenomenex, CA, USA) in 50 mM HEPES pH 7.5, 150 mM NaCl with temperature controlled at 4 C and at a ow rate of 500 ul/min. 1 g of concentrated lysate was injected for fractionation into 80 fractions collected from 1025 min post-injection. Two consecutive runs were pooled to yield the nal set of fractions for digestion and analysis via LC-MS/MS.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 4

Protein fractionation

Peptide fractionation

Proteolytic digestion

MS injections

www.nature.com/sdata/

Data acquisition Data analysis

SamplingCell culture Tissue

Protein fractionationAP SEC Depletion None

Proteolytic digestion Trypsin PCT

PeptideOGE 1D GE SAX None

LC-MS/MS (DDA)

Figure 1. Data acquisition and data analysis workows employed for the generation of assay libraries. (a) Data acquisition: Sampling of different cell lines and tissue types was followed by (optional) protein fractionation, proteolytic digestion (using trypsin or lys-c/trypsin using PCT), (optional) peptide fractionation and LC-MS/MS analysis in discovery proteomics mode. (b) Data analysis: Sequence database search was conducted using four different search engines and the results were statistically evaluated and combined using the Trans-Proteomic Pipeline. False discovery rate (FDR) control was conducted using MAYU. The identied peptides were used to generate a consensus, RT normalized spectral library using SpectraST. Assays were selected using spectrast2tsv. py and the OpenSWATH tool ConvertTSVToTraML.

Peptide sample preparation for MS

To maximize the proteome coverage of the individual specimen, the samples were fractionated using different physicochemical methods like off-gel electrophoresis or ion exchange chromatography. In this study, we included SEC and OGE fractionated samples from a HEK293 cell line, SAX fractionated samples from plasma and LNCaP cell lines and OGE fractionated samples from THP1 and NCI60 cell lines.

Proteolytic digestion. The protein samples were reduced with 5 mM TCEP, and alkylated with 10 mM iodoacetamide before overnight trypsinization. Some samples were trypsinized using the Pressure Cycling Technology (PCT) protocol as described below (indicated in Table 1). Protein from SEC fractions was denatured by incubation at 69 C for 10 min, reduced, alkylated and digested in the presence of 1% (v/v) Sodium-deoxycholate overnight. Trypsin was inactivated by lowering the pH to 2 and the peptides were immobilized onto C18 columns. After multiple washes, the peptides were eluted (50% acetonitrile/0.1% formic acid) and solvents were evaporated in a SpeedVac centrifuge. After re-suspension, the samples were briey sonicated before MS analysis.

PCT-assisted lysis and digestion. Pressure cycling technology (PCT)33 applies cycles of hydrostatic pressure between ambient and ultra-high levels to induce cell lysis and to enable precise thermodynamic control of biomolecular interactions. All PCT-processed samples were handled using Barocycler NEP2320 (PressureBioSciences, Inc, South Easton, MA). In brief, tissue or cell line samples were lysed in buffer containing 8 M urea, 100 mM ammonium bicarbonate supplemented with Complete protease and phosphatase inhibitor cocktail under Barocycler program (tissue samples: 60 cycles of 50 s 45 kpsi and 10 s 14.7 psi; cell line samples: 120 cycles of 20 s 45 kpsi and 10 s 14.7 psi) at 35 C. Whole cell/ tissue lysates were then sonicated for 25 s with 1 min interval on ice for 4 times. After removing tissue debris or unbroken cells, if any, by centrifugation, protein lysates were reduced and alkylated prior to proteolytic digestion. Lys-C (enzyme to substrate ratio: 1:50) and trypsin (1:30) were sequentially added to digest the proteins. Digestion was accelerated under a PCT scheme of 50 s 25 kpsi and 10 s 14.7 psi (cell line samples: 25 s 25 kpsi, 10 s 14.7 psi for 45 mins), under which both Lys-C and trypsin remain active. Lys-C digestion was performed in 6 M urea for 45 cycles, whereas trypsin digestion was performed in further diluted urea (1.6 M) for 90 cycles (cell line samples: 24 s 25 kpsi, 10 s 14.7 psi for 90 min). Subsequently, triuoroacetic acid (TFA) was added to a nal pH of around 2 before C18 desalting using SEP-PAK C18 cartridges (Waters Corp., Milford, MA, USA).

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 5

Shotgun database searchX!Tandem Myrimatch OMSSA Comet

Statistical evaluation PeptideProphet iProphet

FDR control

MAYU

Spectral library generation

SpectraST

fractionation

Assay library generation spectrast2tsv.py OpenSWATH/

ConvertTSVToTraML

www.nature.com/sdata/

Off-gel electrophoresis (OGE). After digestion and desalting steps, clean peptides were re-solubilised in OGE buffer, which contained 5% (v/v) glycerol, 0.7% ACN and 1% (v/v) carrier ampholytes mixture (IPG buffer pH 3.010.0, GE Healthcare). The peptides were separated on a 3100 OFFGEL (OGE) Fractionator (Agilent Technologies) using a 24 cm pH 310 IPG strip (GE Healthcare) at a maximum of 8,000 V, 50 A, and 200 mW until 50 kVhrs were reached. After all fractions were recovered, they were desalted on C18 reversed-phase MicroSpin columns (The Nest Group Inc.) and pooled according to the following schemes for MS injections:

HEK293pool 1 (fraction 12), pool 2 (fraction 3), pool 3 (fraction 4), pool 4 (fraction 5), pool 5 (fraction 67), pool 6 (fraction 89), pool 7 (fraction 1011), pool 8 (fraction 1216), pool 9 (fraction 1718), pool 10 (fraction 1921), pool 11 (fraction 2224).

NCI60 panelpool 1 (fraction 12), pool 2 (fraction 3), pool 3 (fraction 4), pool 4 (fraction 5), pool 5 (fraction 67), pool 6 (fraction 89), pool 7 (fraction 1011), pool 8 (fraction 1215), pool 9 (fraction 1619), pool 10 (fraction 2021), pool 11 (fraction 22), pool 12 (fraction 2324).

THP-1

No pooling was done. Each of the 24 fractions was injected once except for fraction 3, 4, 9, and 22, which were injected twice.

1D gel electrophoresis (1D GE). A pool of 18 kidney tissue samples was resolved into 15 gel fractions based on the molecular mass of proteins using SDS-PAGE34. These fractions were digested independently in-gel before mass spectrometric analysis using standard protocol35.

Strong anion exchange (SAX). A total of 50 g of peptides was separated on a pipet-based anion exchanger, which was assembled following the StageTip principle by stacking 6 layers of a 3 M Empore

Anion Exchange disk (Varian, 1214 5012) into a 200 l micropipet tip, as previously described36. Briey, the equilibration buffer was composed of 20 mM acetic acid, 20 mM phosphoric acid and 20 mM boric acid was titrated with NaOH to the desired pH. Peptides were loaded at pH 11 and fractions were subsequently eluted with buffer solutions of pH 8, 6, 5, 4, and 3, respectively by centrifugation at 7,000 g each time. The ow-through and the ve pH-eluted fractions were all captured on C18 StageTips.

RT normalization peptides. For the RT normalization and analysis, the peptides from the iRT Kit (Biognosys AG, Schlieren, Switzerland) were added to all samples prior to MS injection according to vendor instructions37.

DDA mass spectrometry for spectral library generation

For spectral library generation, an AB SCIEX TripleTOF 5600+ System mass spectrometer was used. It was operated essentially as described before23,24: All samples were analyzed on an Eksigent nanoLC

(AS-2/1Dplus or AS-2/2Dplus) system coupled with a SWATH-MS-enabled AB SCIEX TripleTOF 5600+ System. The HPLC solvent system consisted of buffer A (2% acetonitrile and 0.1% formic acid in water) and buffer B (2% water with 0.1% formic acid in acetonitrile). The samples were separated in a 75 m-diameter PicoTip emitter (New Objective) packed with 20 cm of Magic 3 m, 200 C18 AQ material (Bischoff Chromatography). The loaded material was eluted from the column at a ow rate of 300 nl/min with the following gradient: linear 235% B over 120 min, linear 3590% B for 1 min, isocratic 90% B for 4 min, linear 902% B for 1 min and isocratic 2% solvent B for 9 min. The mass spectrometer was operated in DDA top20 mode, with 500 and 150 ms acquisition time for the MS1 and MS2 scans respectively, and 20 s dynamic exclusion. Rolling collision energy with a collision energy spread of 15 eV was used for fragmentation.

Spectral and assay library generation

All raw instrument data (Data Citation 1) were centroided and processed as described previously24,27.

The assay library was generated according to the following protocol: The TPP38 (4.6.0) and SpectraST39(5.0) were used for the analysis of the shotgun proteomics runs. The datasets were searched individually using X!Tandem40 (2011.12.01.1) with k-score plugin41, Myrimatch42 (2.1.138), OMSSA43 (2.1.8) and Comet44 (2013.02r2) against the full non-redundant, canonical human genome as annotated by UniProtKB/Swiss-Prot45 (2014_02) with 20 270 ORFs and appended iRT peptide and decoy sequences. Carbamidomethyl (C) was used as a xed modication; oxidation (M) was the only variable modication. Parent mass error was set to 50 p.p.m., fragment mass error was set to 0.1 Da. The search identications were then combined and statistically scored using PeptideProphet46 and iProphet47 within the TPP38. MAYU48 (1.07) was used to select an iProphet cutoff of 0.999354, resulting in a protein FDR of 1.03%. SpectraST was used in library generation mode with CID-QTOF settings and iRT normalization at import against the iRT Kit peptide sequences (-c_IRTirtkit.txt -c_IRR) and a consensus library was consecutively generated49. The script spectrast2tsv.py (msproteomicstools 0.2.2; https://pypi. python.org/pypi/msproteomicstools) was then used to generate the asay library with suggested settings: -l 350,2000 -s b,y -x 1,2 -o 6 -n 6 -p 0.05 -d -e -w swath32.txt -k openswath. The OpenSWATH (OpenMS/ develop, revision: 03377b6) tool ConvertTSVToTraML converted the TSV le to TraML and decoys were

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 6

www.nature.com/sdata/

appended to the TraML assay library with the OpenSWATH tool OpenSwathDecoyGenerator as described before24 in reverse mode with a similarity threshold of 0.05 Da and an identity threshold of 1. The assay library (Data Citation 2) was further converted to table format compatible with OpenSWATH, PeakView, Skyline and Spectronaut.

DIA mass spectrometry (SWATH-MS)

For SWATH-MS data acquisition (Data Citation 3), the same mass spectrometer and LC-MS/MS setup was operated essentially as described before23,24, using 32 windows of 25 Da effective isolation width

(with an additional 1 Da overlap on the left side of the window) and with a dwell time of 100 ms to cover the mass range of 4001,200 m/z in 3.3 s. Before each cycle, an MS1 scan was acquired, and then the MS2 scan cycle started (400425 m/z precursor isolation window for the rst scan, 424450 m/z for the second... 1,1741,200 m/z for the last scan). The collision energy for each window was set using the collision energy of a 2+ ion centered in the middle of the window with a spread of 15 eV.

SWATH-MS data analysis

OpenSWATH. An improved development version of the OpenSWATH (OpenMS/develop, revision: 03377b6) analysis workow (OpenSwathWorkow) (http://www.openswath.org) was used for all data analyses. The parameters were selected analogously to the ones described before24: min_rsq: 0.95, min_coverage: 0.6, min_upper_edge_dist: 1, mz_extraction_window: 0.05, rt_extraction_window: 600, extra_rt_extraction_window: 100.pyprophet (0.9.2) (https://pypi.python.org/pypi/pyprophet) was run on the OpenSwathWorkow output adjusted to contain the previously described scores (xx_swath_prelim_score, bseries_score, elution_model_t_score, intensity_score, isotope_correlation_score, isotope_overlap_score, library_corr, library_rmsd, log_sn_score, massdev_score, massdev_score_weighted, norm_rt_score, xcorr_coelution, xcorr_coelution_weighted, xcorr_shape, xcorr_shape_weighted. yseries_score)24 and proteotypic peptides only with enabled MAYU export and 30-fold semi-supervised learning iterations. This generated an OpenSWATH peptide identication list, a FASTA library containing only the targeted peptides and proteins and the false target:decoy ratio (the ratio of targets which could not be detected and decoys) for direct analysis with MAYU.

MAYU (1.07) was used with a maximum mFDR of 0.1, 200 mFDR steps and the calculated false target: decoy ratio to compute assay-level q-value (m_score) cutoffs corresponding to the selected protein FDR. All further analyses were conducted on per run individually analyzed and ltered peptide and protein identications.

PeakView. A previously collected data set of AP-SWATH samples was reprocessed using PeakView (AB SCIEX) as described by Lambert et al.28 Essentially the raw data was processed using the sample-specic assay library or the combined assay library, extracting peak areas and scoring using the PeakView SWATH micro app. Peak areas were extracted and ltered to remove all peptides, which do not have a single measurement with an FDR less than 1% across all measurements.

The extracted peak areas were processed through most likely ratio normalization and fold change determination as described before28. The results for the fold change analysis from the sample-specic assay library were compared to the fold-change results from the combined assay library.

Data Records

Data Record 1

The mass spectrometry discovery proteomics data (instrument raw les, centroided mzXML and identied peptides in pepXML report) used to generate the combined assay library have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository51 with the dataset identier PXD000953 (Data Citation 1).

Data Record 2

The spectral libraries (SpectraST format) and assay libraries (CSV, TraML) are available for different SWATH-MS data analysis tools at the SWATHAtlas with the dataset identiers SAL00016-35 (Data Citation 2).

Data Record 3

The mass spectrometry SWATH-MS data (instrument raw les, mzXML and identied peptides in OpenSWATH report) used to validate the sample-specic and combined assay libraries have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository50 with the dataset identier PXD000954 (Data Citation 3).

Technical Validation

Assay library saturation analysis

Large-scale MS-based proteomics experiments are prone to accumulation of false identications, both at the peptide and protein level. It is thus crucial to lter these datasets restrictively, especially for the purpose of assay library generation. We applied the strategy implemented in MAYU48 to adjust the assay

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 7

www.nature.com/sdata/

library to an FDR of 1% at the protein level, resulting in an iProphet47 probability cutoff of 0.999354. At this cutoff, the number of true positive protein identications already reaches saturation (Figure 2a). This is in contrast to the number of true positive peptide identications, which could be further increased at the cost of accepting a higher number of false positive protein identications (Figure 2b). This result is in line with observations from other large-scale datasets, where the true detectable proteins generally have many associated peptides that match redundantly to the same protein. The false positive identications on the other hand do not show this redundancy and thus the error-rate needs to be controlled very strictly, resulting in a number of false negative identications51.

The number of proteins identied from a DDA dataset depends signicantly on the redundancy of the sequence database searched. Databases with a high degree of sequence redundancy can inate the protein identications because substantially similar or indistinguishable proteins are counted as separate species. Therefore, the application of redundant protein databases like UniprotKB/TrEMBL or the International Protein Index (IPI) is not recommended for the purpose of assay library generation because of their increased potential for generating random single hit identications48,52. For this study, we used

UniprotKB/Swiss-Prot as basis for protein annotation, which is considered to be the leading universal curated protein sequence database45,53 and which contains only non-redundant entries.

The combined assay library (CAL) contains injections from 16 different sample types and the relative contribution of each sample to the consensus spectral library varies from below 1 to 37%. In general, the

200000

13000

protein identifications

11000

12000

peptide identifications

140000

180000

160000

alltrue positive

alltrue positive

0.02 0.06 0.10 0.14

0.02 0.06 0.10 0.14

protein FDR

protein FDR

5506

040000

NCI60 (CL) 37%

HEK293 (CL) 18%

Gut (T) 17%

THP1 (CL) 9%

Kidney (T) 6%

U2OS & HeLa (CL) 5%

Blood plasma (T) 2%

U2OS (CL) 1%

Purified platelets (T) 1%

HeLa (CL) 1%

LNCaP (CL) 1%

CAL51 (CL) <1%

Monocytes (T) <1%

Neutrophiles (T) <1%

Muscle (T) <1%

Lung (T) <1%

4442 9514

802

CAL

protein-level evidence

Swiss-Prot

Figure 2. Statistics of the combined assay library and comparison to other human proteome mapping efforts. (a) True positive (red) and all protein identications (blue) as a function of protein FDR. The graph indicates that the number of true positive protein identications saturates at a protein FDR cutoff of 0.05. Additional identications at less strict FDR cutoffs are mainly false positive protein identications. (b) True positive (red) and all peptides identications (blue) as a function of protein FDR. The graph indicates that the number of true positive peptide identications correlates strongly with the total number of peptide identications and does not reach saturation within typical levels of protein FDR cutoffs. (c) The number of PSM per sample type contributed to the assay library. Multiple PSM can constitute a consensus spectrum and are individually counted per MS injection. The NCI60 cell line panel contributed most, and HEK293 cells, gut tissue and THP1 cells each contributed to more than 10% of all spectra. (d) Overlap of human proteins curated by UniProtKB/Swiss-Prot, a subset annotated with protein-level evidence and the presented combined assay library (CAL). On the protein level, the assay library provides 68.2% coverage of the proteins with evidence while providing assays for an additional 802 proteins. Compared to UniProtKB/Swiss-Prot, the assay library contains 50.9% of all 20,264 proteins.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 8

www.nature.com/sdata/

NCI60 cell line panel, the HEK293 and THP1 cell lines and gut and kidney tissue samples were the major contributors, collectively accounting for close to 90% of all consensus peptide spectrum matches (PSM) above the threshold (Figure 2c). This large coverage is mainly due to extensive fractionation on the protein and peptide level and the large number of MS injections per sample type.

Relation to present state of proteome discovery

In recent years, several studies and projects have aimed at mapping the complete human proteome, among them the HUPO Chromosome-centric Human Proteome Project (C-HPP)7,8, which attempts to characterize at least one protein product for each human protein-coding gene54. The proteomes of several human cell lines have been exhaustively identied46 and recently, draft maps of the human proteome have been published, accounting for 84%9 or 92%10 of the annotated human genome.

We compared the proteins contained in the combined assay library with the proteins annotated by UniProtKB/Swiss-Prot (version 2014_05) and the proteins annotated in there with evidence on protein-level55. We mapped the non-redundant, canonical list of UniProtKB/Swiss-Prot identiers to the proteins identied by proteotypic peptides contained in the combined assay library. Figure 2d indicates that on the protein level, our library reaches 68.2% coverage of the 13,956 proteins annotated with protein-level evidence, while providing assays for an additional 802 proteins. Compared to UniProtKB/Swiss-Prot, the combined assay library contains 50.9% of all 20,264 proteins. Table 2 provides an overview of the contents of the combined assay library.

Applicability of the combined assay library for SWATH-MS targeted data analysisAn analysis using whole cell digest samples from HeLa and U2OS cell lines was conducted to compare the performance of the combined (CAL) and sample-specic assay libraries (ss HeLa/ss U2OS). First, we generated sample-specic assay libraries from lysates of the respective cell lines by acquiring DDA datasets (which are also contained in the combined assay library) from three repeat injections of the unseparated peptide samples. For the HeLa cells the resulting sample-specic assay library contained 2,583 proteins, 16,096 peptides, 18,124 precursor ion sequences and 108,744 transitions. For the U2OS cells the library contained 2,610 proteins, 15,334 peptides, 17,360 precursors and 104,160 transitions. For both cell lines the data were ltered to a protein FDR of 1% and only proteotypic assays were considered for all further analyses. The overlap with the combined assay library was found to be over 99% on both peptide and protein level for both cell lines (Figure 3a). The overlap between the two sample-specic libraries is on peptide-level more than 70% and about 80% on protein-level. Both libraries were used to individually analyze the same sample acquired in DIA mode using OpenSWATH24. The q-value

threshold (m_score) on assay level was used to estimate the protein FDR as described above.

At a protein FDR of 1%, the number of true positive protein identications from a sample is very similar when the whole combined assay library or sample-specic assay libraries were used (Figure 3c,d). However, compared to the number of the non-single hits identied by the sample-specic assay libraries, the combined assay library provides an increased protein-level coverage of 4959% (Table 3). This apparent discrepancy can be resolved in context of the number of peptides that are identied as true positives using the combined assay library compared to the sample-specic assay libraries. Because the combined assay library enables detection of over 35% more peptides at a peptide FDR of 1% (Figure 3b), excluding single hits enables detection of more proteins. Overall, these data show that the combined assay library identies peptides at a higher level of sensitivity at typical levels of FDR control.

The reproducibility of the peptide identications among three technical replicates in dependency of the peptide FDR for the HeLa samples is depicted in Figure 3e. The number of peptides identied in all three samples is similar for both the combined and sample-specic libraries. However, the CAL detected a higher number of peptides in only one or two replicates. Further assessment of these peptides at 1% FDR for the CAL and sample-specic library indicates that they are mainly low-intensity peptides (CAL: 1/3 (detected in 1 out of 3 replicates) 33,433 38,083 (mean s.d. of summed fragment ion intensities per precursor), 2/3 (39,504 39,440), 3/3 (89,935 140,914); ss HeLa: 1/3 (35,865 38,467), 2/3

Proteotypic Proteotypic+Shared

Proteins 10,316 11,588

Peptides 139,449 146,576

Precursors 194,052 204,545

Transitions 1,164,312 1,227,270

Table 2. Assay statistics of the combined assay library. The number of proteins, peptides, precursors and transitions, ltered at protein FDR 1% is depicted. The combined assay library is provided with all target and decoy assays, but only proteotypic assays were considered for all downstream analysis.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 9

www.nature.com/sdata/

(39,440 52,346), 3/3 (97,226 152,470)). The majority (CAL: 77.4%; ss HeLa: 82.0%) of proteins mapped by these low-intensity peptides were also detected by different, higher-intensity peptides in all three replicates. This indicates that the assays are not resulting in false positive protein identications, but rather enable measuring of additional peptides of the same proteins and that the assays of the CAL and sample-specic assay libraries are very similar in terms of reproducibility of identication in targeted proteomics experiments. These assays are not present in the sample-specic assay libraries due the sample complexity and limitations of the DDA algorithms that only select the most intense precursors for fragmentation.

The coefcient of variation (CV) of the quantied signals on precursor level was found to correspond well with the expected technical variation between replicates of below 20%24 (Figure 3f). Further, the CV of the quantied signals using the combined and sample-specic libraries are very similar for the two cell lines, indicating conserved reliable quantication performance.

Portability of the combined assay library to different sample types and laboratoriesTo test the portability of the generated assay library we used a subset of assays for specic proteins from the combined assay library for reanalysis of the CDK4 AP-SWATH dataset of Lambert et al.28. This

dataset was generated on the same type of instrument used for the generation of the assay library presented here. However, the SWATH-MS data and the DDA data used to generate a sample-specic library were acquired in a different laboratory, at a different time point and using different chromatographic conditions. Using either the original sample-specic library or the corresponding assays contained in the combined library reported here we determined the fold change of the proteins between the wild type and the mutant CDK4 states (R23C, R23H). Figure 4 shows the comparison and overlap of the original analysis and reanalysis using the assays from the combined library. The protein fold change measurements between the different assay libraries are comparable. The data therefore indicate that the assays contained in the combined library can be used successfully to perform protein quantication even if the data were acquired at different times and in different laboratories. Investigation of the peptides within the combined in comparison to those in the sample-specic assay library created as part of the original publication showed that in most cases there was equivalent coverage of proteins between different libraries. In those cases where protein expression proles were different between the different assay libraries as in CD2A1 and CDN2C, the difference in the fold change can be attributed to the difference in the number of peptides present within the library. These results demonstrate that the assays contained in the combined assay library presented here are portable between different experimental setups.

Usage Notes

Application of the assay library to SWATH-MS data

There are two different ways to apply the assay library to search SWATH-MS datasets. The rst is a selective search for predetermined sets of proteins and the second is a comprehensive search of a SWATH-MS map with the whole library. In the rst case a selection of peptides or proteins of interest is available as prior information, e.g., from earlier proteomics or transcriptomics measurements or from the literature. The combined assay library can thus be ltered accordingly so that the query transition list only contains assays for these targeted proteins or peptides. To simplify this step, we provide querying of the combined assay library for specic proteins and peptides on the SWATHAtlas. These assays can be used in software like Skyline25 or PeakView for data analysis and visualization.

In the second case there is no pre-selection of target peptides or proteins and the whole assay library is used to search a SWATH-MS map by an automated software like OpenSWATH24. Since the whole combined library contains assays for more than 10,000 proteins and a typical short gradient single SWATH-MS map will typically lead to the identication of 2,0005,000 proteins, most proteins targeted by the whole assay library will either not be present in the sample or not be detectable. To avoid false positives due to the multiple comparisons problem, it is critical to appropriately set score cutoffs according to the peptide or protein FDR with tools like MAYU48. This approach is dependent on the proper application of the target-decoy approach56 and we have found that especially for very large assay libraries as the one presented here, it is crucial to generate decoy assays that both are guaranteed to be different from the target assays and that represent the full sample. To enable generation of decoy transitions for even highly repetitive or palindromic peptide sequence, we found that full reversion of the sequences fullls these requirements.

The effect of the multiple comparison problem is illustrated by the application of the whole combined assay library to the HeLa SWATH-MS datasets described above. In the analyses MAYU determined an assay FDR of approximately 0.0036% resulting in a protein FDR of 1%. In comparison, for a sample-specic library, the same protein FDR was reached with an assay FDR of about 0.6%. This discrepancy is partially related to the observation in shotgun proteomics database searching that searching very large databases, e.g., six-frame translations of genomic databases, increases the chances of random PSMs. However, the situation differs from sequence database searching in that the targeted approach attempts to detect specic signal groups in a variable number of experimentally observed ion chromatograms.

An updated version of OpenSWATH is provided (http://www.openswath.org) that directly enables protein FDR assessment using MAYU.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 10

www.nature.com/sdata/

Figure 3. Number of peptide and protein identications by SWATH-MS using different proteotypic assay libraries. (a) The proteotypic peptides contained in the combined assay library (CAL) and the samplespecic (ss) assay libraries and their overlap is depicted. The overlap on peptide-level between the sample-specic libraries is more than 70% and around 80% on protein-level. 239 peptides contained in the sample-specic libraries were not included in the CAL, since they did not meet the stricter quality cutoff of the CAL. (b) The number of true positive peptide identications in dependency of the peptide FDR is depicted. Using the combined library, the number of true positive peptide identications matches the sample-specic libraries at peptide FDR below 1% (dashed grey line). (c,d) The number of true positive protein identications of a HeLa (c) or U2OS (d) whole cell lysate in a single, unfractionated injection in dependency of the protein FDR is depicted. Protein FDR cutoffs are either reported for all identications or non-single hits (NS). The CAL provides similar sensitivity compared to the sample-specic libraries for HeLa and U2OS at typical levels of error-rate control. The non-single hit identications of the CAL generally provide a higher sensitivity at lower protein FDR cutoffs. The dashed grey line indicates the protein FDR cutoff at 1%. (e) Reproducibility of the peptide identications in dependency of the peptide FDR. The colors indicate reproducibility in 1 (green), 2 (blue) or 3 (red) of 3 technical replicates. Both ss HeLa (top) and CAL (bottom) enable detection of a similar number of assays among all replicates at the same peptide FDR. The CAL enables detection of more low intensity peptides in only one or two replicates. (f) Distribution of the coefcient of variation (CV) of summed transition intensities of precursors identied in all three replicates at 1% peptide FDR. The median CV of 5% (U2OS) to 10% (HeLa) corresponds well with the expected technical variation and is very similar between sample-specic and the combined assay library.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 11

www.nature.com/sdata/

Protein FDR CAL HeLa ss HeLa CAL U2OS ss U2OS

prot pep prot pep prot pep prot pep

1% 2,417 14,930 2,353 14,635 2,617 15,608 2,452 14,360

2% 2,730 17,294 2,467 15,416 2,989 18,321 2,541 14,982

5% 3,246 21,128 2,514 15,672 3,486 21,893 2,552 15,003

NS 1% 2,608 23,075 1,750 14,999 2,803 24,009 1,763 14,599

NS 2% 2,804 25,005 1,798 15,537 2,965 25,497 1,815 15,002

NS 5% 3,111 28,002 1,820 15,668 3,241 28,442 1,819 14,999

Table 3. Identication statistics of the combined and sample-specic assay libraries. The number of identied proteotypic peptides and proteins in SWATH-MS datasets of whole cell lysates of HeLa and U2OS cell lines at commonly used protein FDR cutoffs using combined (CAL) and sample-specic (ss) assay libraries is reported. Protein FDR cutoffs are either reported for all identications or non-single hits (NS). The true positive protein (prot) and peptide (pep) identications for the combined assay library and sample-specic assay libraries are reported as estimated by MAYU.

0

10

20

ss WT/R24C ss WT/R24H CAL WT/R24C CAL WT/R24H

Figure 4. Application of the combined assay library (CAL) to an independently acquired dataset (CDK4 APSWATH, Lambert et al.28) and comparison to the sample-specic assay library (ss). The fold changes of the comparison wild type (WT) and mutants (R24C or R24H) with whiskers for standard deviation are indicated. The assays contained in the combined library for CD2A1 and CDN2C covered fewer and different peptides than the sample-specic assay library and thus the fold change is smaller. The results indicate that comparable qualitative and quantitative results using the combined assay library can be retrieved from SWATH-MS experiments conducted using different experimental setups, data acquisition and data analysis strategies.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 12

fold change

CD2A1_HUMAN

CDN2C_HUMAN

CDK4_HUMAN

FKBP5_HUMAN

FKBP4_HUMAN

HS90B_HUMAN

HS90A_HUMAN

CDC37_HUMAN

www.nature.com/sdata/

The presented data was acquired on Eksigent nanoLC (AS-2/1Dplus or AS-2/2Dplus) systems coupled with an AB SCIEX TripleTOF 5600+ system and the combined assay library is therefore optimized for this type of instrument. However, the assay library could also be applied to DIA data acquired on other high-resolution instruments. In such a case, the expected fraction of detectable assays is depending on the similarity of the instrumentation in terms of fragmentation method and liquid chromatography. Particularly, when qTOF-CID spectra, as the ones presented here, are compared to ion trap HCD spectra, the conservation of the fragment pattern is high, indicating good portability of the assays57,58. Further, the

normalized retention time used here is a dimensionless value that can be transformed to different LC setups using spiked-in standards37. Finally, the semi-supervised learning approach employed by mProphet59 and related software like OpenSWATH, Spectronaut and Skyline adapts the inuence of potentially decreased fragmentation or retention time conservation on the discriminant scoring function to maintain accurate separation of true and false detected assays.

Generation of custom assay libraries from the presented data

Custom assay libraries can be optimized for specic sample types, proteoforms and proteomic background. For special applications such as the analysis of proteoforms, custom assay libraries can be generated by searching the spectral data additionally for post-translational modications such as phosphorylation or by using a different protein sequence database, e.g., one containing protein isoforms. It is recommended to apply an assay library generation workow that is scalable and enables control of the error rate. A manuscript providing detailed instructions for the generation of large-scale assay libraries is in preparation by the authors (Schubert, O. T., Gillet, L. C., Collins, B. C., Navarro, P., Rosenberger, G., Wolski, W. E., Lam, H., Amodei, D., MacLean, B., Mallick, P. & Aebersold, R.). Particularly for modications, the condence for correct site assignment needs to be assessed and accounted for ref. 60.

The transitions of the combined assay library have been selected according to a protocol that enables qualitative and quantitative comparable results as sample-specic assay libraries (Figures 3 and 4). Assays with many interfered transitions can be detected automatically by the software tools used in this study and rather affect the sensitivity than the selectivity and thus do not increase the number of false positives24. Because the combined assay library contains assays for more than one proteotypic peptide for86.5% of all proteins, a different peptide can be used for quantication in most such cases. However, for certain applications, especially when analysis of very complex human samples or differentially site-modied proteoforms is conducted, the transition selection could be altered according to the unique ion signature (UIS) concept61. Using tools like SRMCollider62, transitions could be selected for a given background proteome (e.g., based on previously identied proteins) to minimize potential interferences with other co-eluting peptides. Additionally, SWATH-MS enables iterative reanalysis using different assays for the same peptide and thus the combined assay library could be optimized for a particular sample type using empirical criteria.

Extension of the human assay library

This is a rst edition of the combined human SWATH-MS assay library and further extensions will be added. Analogous to the HUPO Human Proteome Project and the recent studies mapping the human proteome9,10, data fullling the requirements for SWATH-MS assay library generation can be collected in public repositories like ProteomeXchange63 and periodically, new builds of the assay library can be generated as new datasets covering extended parts of the human proteome become available. As demonstrated in this study, the extension will not compromise results derived from subsets of the assay library but enable a more complete and comparable targeted analysis of human SWATH-MS datasets.

References

1. Uhlen, M. et al. Towards a knowledge-based Human Protein Atlas. Nat. Biotechnol. 28, 12481250 (2010).2. Edwards, A. M. et al. Too many roads not taken. Nature 470, 163165 (2011).3. Marx, V. Finding the right antibody for the job. Nat. Methods 10, 703707 (2013).4. Beck, M. et al. The quantitative proteome of a human cell line. Mol. Syst. Biol. 7, 18 (2011).5. Geiger, T., Wehner, A., Schaab, C., Cox, J. & Mann, M. Comparative Proteomic analysis of eleven common cell lines reveals ubiquitous but varying expression of most proteins. Mol. Cell. Proteomics 11, M111.014050 (2012).

6. Moghaddas Gholami, A. et al. Global proteome analysis of the NCI-60 cell line panel. Cell Rep. 4, 609620 (2013).7. Omenn, G. S. The strategy, organization, and progress of the HUPO Human Proteome Project. J. Proteom. 100, 37 (2014).8. Farrah, T. et al. State of the human proteome in 2013 as viewed through PeptideAtlas: comparing the kidney, urine, and plasma proteomes for the biology- and disease-driven Human Proteome Project. J. Proteome Res. 13, 6075 (2014).

9. Kim, M.-S. et al. A draft map of the human proteome. Nature 509, 575581 (2014).10. Wilhelm, M. et al. Mass-spectrometry-based draft of the human proteome. Nature 509, 582587 (2014).11. Domon, B. & Aebersold, R. Options and considerations when selecting a quantitative proteomics strategy. Nat. Biotechnol. 28, 710721 (2010).

12. Bell, A. W. et al. A HUPO test sample study reveals common problems in mass spectrometry-based proteomics. Nat. Methods 6, 423430 (2009).

13. Tabb, D. L. et al. Repeatability and reproducibility in proteomic identications by liquid chromatography-tandem mass spec-trometry. J. Proteome Res. 9, 761776 (2010).

14. Paulovich, A. G. et al. Interlaboratory study characterizing a yeast performance standard for benchmarking LC-MS platform performance. Mol. Cell. Proteom. 9, 242254 (2010).

15. Rudnick, P. A. et al. Performance metrics for liquid chromatography-tandem mass spectrometry systems in proteomics analyses. Mol. Cell. Proteom. 9, 225241 (2010).

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 13

www.nature.com/sdata/

16. Picotti, P., Bodenmiller, B. & Aebersold, R. Proteomics meets the scientic method. Nat. Methods 10, 2427 (2012).17. Aebersold, R. et al. The biology/disease-driven human proteome project (B/D-HPP): enabling protein research for the life sciences community. J. Proteome Res. 12, 2327 (2013).

18. Picotti, P. et al. High-throughput generation of selected reaction-monitoring assays for proteins and proteomes. Nat. Methods 7, 4346 (2009).

19. Picotti, P. et al. A complete mass-spectrometric map of the yeast proteome applied to quantitative trait analysis. Nature 494,

266270 (2013).

20. Schubert, O. T. et al. The Mtb Proteome Library: A resource of assays to quantify the complete proteome of mycobacterium tuberculosis. Cell Host Microbe 13, 602612 (2013).

21. Karlsson, C., Malmstrm, L., Aebersold, R. & Malmstrom, J. Proteome-wide selected reaction monitoring assays for the human pathogen Streptococcus pyogenes. Nat. Commun. 3, 1301 (2012).

22. Peterson, A. C., Russell, J. D., Bailey, D. J., Westphall, M. S. & Coon, J. J. Parallel reaction monitoring for high resolution and high mass accuracy quantitative, targeted proteomics. Mol. Cell. Proteom. 11, 14751488 (2012).

23. Gillet, L. C. et al. Targeted data extraction of the MS/MS spectra generated by data-independent acquisition: a new concept for consistent and accurate proteome analysis. Mol. Cell. Proteom. 11, O111.016717 (2012).

24. Rst, H. L. et al. OpenSWATH enables automated, targeted analysis of data-independent acquisition MS data. Nat. Biotechnol. 32, 219223 (2014).

25. MacLean, B. et al. Skyline: an open source document editor for creating and analyzing targeted proteomics experiments. Bioinformatics 26, 966968 (2010).

26. Liu, Y. et al. Quantitative measurements of N-linked glycoproteins in human plasma by SWATH-MS. Proteom. 13, 12471256 (2013).

27. Collins, B. C. et al. Quantifying protein interaction dynamics by SWATH mass spectrometry: application to the 14-3-3 system. Nat. Methods 10, 1246 (2013).

28. Lambert, J.-P. et al. Mapping differential interactomes by afnity purication coupled with data-independent mass spectrometry acquisition. Nat. Methods 10, 12391245 (2013).

29. Liu, Y., Httenhain, R., Collins, B. & Aebersold, R. Mass spectrometric protein maps for biomarker discovery and clinical research. Expert Rev. Mol. Diagn. 13, 811825 (2013).

30. Glatter, T., Wepf, A., Aebersold, R. & Gstaiger, M. An integrated workow for charting the human interaction proteome: insights into the PP2A system. Mol. Syst. Biol. 5, 237 (2009).

31. Kristensen, A. R., Gsponer, J. & Foster, L. J. Protein synthesis rate is the predominant regulator of protein expression during differentiation. Mol. Syst. Biol. 9, 689689 (2013).

32. Burkhart, J. M. et al. The rst comprehensive and quantitative analysis of human platelet protein composition allows the comparative analysis of structural and functional pathways. Blood 120, e73e82 (2012).

33. Schumacher, R. T. et al. Automated solution for sample preparation: Nucleic acid and protein extraction from cells and tissues using pressure cycling technology (PCT). Am. Lab. 34, 3843 (2002).

34. Schgger, H. Tricine-SDS-PAGE. Nat. Protoc. 1, 1622 (2006).35. Shevchenko, A., Tomas, H., Havlis, J., Olsen, J. V. & Mann, M. In-gel digestion for mass spectrometric characterization of proteins and proteomes. Nat. Protoc. 1, 28562860 (2006).

36. Wisniewski, J. R., Zougman, A. & Mann, M. Combination of FASP and StageTip-based fractionation allows in-depth analysis of the hippocampal membrane proteome. J. Proteome Res. 8, 56745678 (2009).

37. Escher, C. et al. Using iRT, a normalized retention time for more targeted measurement of peptides. Proteom. 12, 11111121 (2012).

38. Keller, A., Eng, J., Zhang, N., Li, X.-J. & Aebersold, R. A uniform proteomics MS/MS analysis platform utilizing open XML le formats. Mol. Syst. Biol. 1, 2005.0017E8 (2005).

39. Lam, H. et al. Development and validation of a spectral library searching method for peptide identication from MS/MS. Proteom. 7, 655667 (2007).

40. Craig, R. R. & Beavis, R. C. R. A method for reducing the time required to match protein sequences with tandem mass spectra. Rapid Commun. Mass Spectrom. 17, 23102316 (2002).

41. MacLean, B., Eng, J. K., Beavis, R. C. & McIntosh, M. General framework for developing and evaluating database scoring algorithms using the TANDEM search engine. Bioinformatics 22, 28302832 (2006).

42. Tabb, D. L., Fernando, C. G. & Chambers, M. C. MyriMatch: Highly accurate tandem mass spectral peptide identication by multivariate hypergeometric analysis. J. Proteome Res. 6, 654661 (2007).

43. Geer, L. Y. et al. Open mass spectrometry search algorithm. J. Proteome Res. 3, 958964 (2004).44. Eng, J. K., Jahan, T. A. & Hoopmann, M. R. Comet: An open-source MS/MS sequence database search tool. Proteom. 13, 2224 (2013).

45. Magrane, M. & Consortium, U. UniProt Knowledgebase: a hub of integrated protein data. Database (Oxford), bar009bar009 (2011).

46. Keller, A., Nesvizhskii, A. I., Kolker, E. & Aebersold, R. Empirical statistical model to estimate the accuracy of peptide identi-cations made by MS/MS and database search. Anal. Chem. 74, 53835392 (2002).

47. Shteynberg, D. et al. iProphet: multi-level integrative analysis of shotgun proteomic data improves peptide and protein identi-cation rates and error estimates. Mol. Cell. Proteom. 10, M111.007690 (2011).

48. Reiter, L. et al. Protein identication false discovery rates for very large proteomics data sets generated by tandem mass spectrometry. Mol. Cell. Proteom. 8, 24052417 (2009).

49. Lam, H. et al. Building consensus spectral libraries for peptide identication in proteomics. Nat. Methods 5, 873875 (2008).50. Vizcano, J. A. et al. The Proteom. Identications (PRIDE) database and associated tools: status in 2013. Nucleic Acids Res. 41, D1063D1069 (2013).

51. Claassen, M. Inference and validation of protein identications. Mol. Cell. Proteom. 11, 10971104 (2012).52. Griss, J. et al. Consequences of the discontinuation of the International Protein Index (IPI) database and its substitution by the UniProtKB complete proteome sets. Proteom. 11, 44344438 (2011).

53. Apweiler, R., Bairoch, A. & Wu, C. H. Protein sequence databases. Curr. Opin. Chem. Biol. 8, 7680 (2004).54. Marko-Varga, G., Omenn, G. S., Paik, Y.-K. & Hancock, W. S. A rst step toward completion of a genome-wide characterization of the human proteome. J. Proteome Res. 12, 15 (2013).

55. Lane, L. et al. Metrics for the Human Proteome Project 2013-2014 and strategies for nding missing proteins. J. Proteome Res. 13, 1520 (2014).

56. Elias, J. E. & Gygi, S. P. Target-decoy search strategy for increased condence in large-scale protein identications by mass spectrometry. Nat. Methods 4, 207214 (2007).

57. Toprak, U. H. et al. Conserved peptide fragmentation as a benchmarking tool for mass spectrometers and a discriminating feature for targeted proteomics. Mol. Cell. Proteomics 13, 20562071 (2014).

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 14

www.nature.com/sdata/

58. de Graaf, E. L., Altelaar, A. F. M., van Breukelen, B., Mohammed, S. & Heck, A. J. R. Improving SRM assay development: a global comparison between triple quadrupole, ion trap, and higher energy CID peptide fragmentation spectra. J. Proteome Res. 10, 43344341 (2011).

59. Reiter, L. et al. mProphet: automated data processing and statistical validation for large-scale SRM experiments. Nat. Methods 8, 430435 (2011).

60. Chalkley, R. J. & Clauser, K. R. Modication site localization scoring: strategies and performance. Mol. Cell. Proteom. 11, 314 (2012).

61. Sherman, J., McKay, M. J., Ashman, K. & Molloy, M. P. Unique ion signature mass spectrometry, a deterministic method to assign peptide identity. Mol. Cell. Proteom. 8, 20512062 (2009).

62. Rst, H., Malmstrm, L. & Aebersold, R. A computational tool to detect and avoid redundancy in selected reaction monitoring. Mol. Cell. Proteom. 11, 540549 (2012).

63. Vizcano, J. A. et al. ProteomeXchange provides globally coordinated proteomics data submission and dissemination. Nat. Biotechnol. 32, 223226 (2014).

Data Citations

1. Rosenberger, G. et al. ProteomeXchange PXD000953 (2014).2. Rosenberger, G. et al. SWATHAtlas SAL00016-35 (2014).3. Rosenberger, G. et al. ProteomeXchange PXD000954 (2014).

Acknowledgements

G.R. was funded by the Swiss Federal Commission for Technology and Innovation CTI (13539.1 PFFLILS). H.L.R. was funded by ETH Zurich (ETH-30 11-2). P.K. was supported by the Finnish Cultural Foundation. E.C. was supported by a Marie Curie Intra-European Fellowship. M.F. was supported by a long-term fellowship from the European Molecular Biology Organization. M.M was funded by TRIREME. H.L. was funded by the General Research Fund (#602413) of the Research Grants Council of the Hong Kong Special Administrative Region Government. S.L.B was supported by a fellowship from the Swiss National Science Foundation (fellowship PBZHP3 143482). R.L.M., D.S.C. and E.W.D are supported in part by federal funds from the American Recovery and Reinvestment Act through Grant RC2 HG005805 from the National Human Genome Research Institute, the National Institutes of Health National Institute of General Medical Sciences under grant Nos. 2P50 GM076547/Center for Systems Biology, GM087221 and S10RR027584. R.A. was funded by the advanced European Research Council grant Proteomics v3.0 (ERC-2008-AdG_20080422), the PhosphonetX project of SystemsX.ch, and the Swiss National Science Foundation (3100A0-107679). We would like to thank Sharon Rashi-Elkeles for the generation of the CAL51 cells, the ITS Scientic IT Services of ETH Zurich for support and maintenance of the lab-internal computing infrastructure and the PRIDE Team of EBI for support of data deposition to the ProteomeXchange Consortium.

Author Contributions

G.R. conducted the study and computational analysis. C.C.K., T.G., P.K., B.C.C., M.H., Y.L., E.C., A.V., M.F., O.T.S., P.F., H.A.E. and M.M. contributed the datasets and conducted the sample preparation and mass spectrometric analysis of all discovery proteomics data. T.G. provided the SWATH-MS data. H.L.R. and G.R. developed and implemented the protein FDR estimation strategy using OpenSWATH and MAYU. H.L. implemented support for RT normalization of large spectral libraries in SpectraST. S.L.B., D.S.C., E.W.D. and R.L.M. designed and implemented the SWATHAtlas. S.T. analyzed the AP-SWATH data using the combined assay library. G.R. and R.A. wrote the manuscript with contributions from all authors. R.A. designed and supervised the study.

Additional information

Supplementary information accompanies this paper at http://www.nature.com/sdata

Competing nancial interests: S.T. is employee of AB SCIEX, which operates in the eld covered by the article. The research group of R.A. is supported in part by AB SCIEX by providing access to prototype instrumentation. R.A. holds shares of Biognosys AG, which operates in the eld covered by the article. The remaining authors declare no competing nancial interest.

How to cite this article: Rosenberger, G. et al. A repository of assays to quantify 10,000 human proteins by SWATH-MS. Sci. Data 1:140031 doi: 10.1038/sdata.2014.31 (2014).

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0

Metadata associated with this Data Descriptor is available at http://www.nature.com/sdata/ and is released under the CC0 waiver to maximize reuse.

SCIENTIFIC DATA | 1:140031 | DOI: 10.1038/sdata.2014.31 15