Combination of Edman degradation of peptides with liquid chromatography/mass spectrometry workflow for peptide identification in bottom-up proteomics

Article abstract of DOI:10.1002/rcm.6462

Rationale: High-throughput methods of proteomics are essential for identification of proteins in a cell or tissue under certain conditions. Most of these methods require tandem mass spectrometry (MS/MS). A multidimensional approach including predictive chromatography and partial chemical degradation could be a valuable alternative and/or addition to MS/MS. Methods: In the proposed strategy peptides are identified in a three-dimensional (3D) search space consisting of retention time (RT), mass, and reduced mass after one-step partial Edman degradation. The strategy was evaluated in silico for two databases: baker's yeast and human proteins. Rates of unambiguous identifications were estimated for mass accuracies from 0.001 to 0.05 Da and RT prediction accuracies from 0.1 to 5 min. Rates of Edman reactions were measured for test peptides. Results: A 3D description of proteolytic peptides allowing unambiguous identification without employing MS/MS of up to 95% and 80% of tryptic peptides from the yeast and human proteomes, respectively, was considered. Further extension of the search space to a four-dimensional one by incorporating the second N-terminal amino acid residue as the fourth dimension was also considered and was shown to result in up to 90% of human peptides being identified unambiguously. Conclusions: The proposed 3D search space can be a useful alternative to MS/MS-based peptide identification approach. Experimental implementations of the proposed method within the on-line liquid chromatography/mass spectrometry (LC/MS) and off-line matrix-assisted laser desorption/ionization (MALDI) workflows are in progress. Copyright

Full text of DOI:10.1002/rcm.6462

Research Article
Received: 27 August 2012
Revised: 1 November 2012
Accepted: 2 November 2012
Published online in Wiley Online Library
Rapid Commun. Mass Spectrom. 2013, 27, 391–400
(wileyonlinelibrary.com) DOI: 10.1002/rcm.6462
Combination of Edman degradation of peptides with liquid
chromatography/mass spectrometry workflow for peptide
identification in bottom-up proteomics
Anna A. Lobas1,2, Anatoly N. Verenchikov3, Anton A. Goloborodko1,2,4, Lev I. Levitsky1,2
and Mikhail V. Gorshkov1,2
*
1Institute for Energy Problems of Chemical Physics, Russian Academy of Sciences, Moscow, Russia
2Moscow Institute of Physics and Technology (State University), Dolgoprudny, Moscow region, Russia
3Mass Spectrometry Consulting Ltd., Bar, Montenegro
4Department of Physics, Massachusetts Institute of Technology, Boston, MA, USA
RATIONALE: High-throughput methods of proteomics are essential for identification of proteins in a cell or tissue under
certain conditions. Most of these methods require tandem mass spectrometry (MS/MS). A multidimensional approach
including predictive chromatography and partial chemical degradation could be a valuable alternative and/or addition
to MS/MS.
METHODS: In the proposed strategy peptides are identified in a three-dimensional (3D) search space consisting of
retention time (RT), mass, and reduced mass after one-step partial Edman degradation. The strategy was evaluated
in silico for two databases: baker’s yeast and human proteins. Rates of unambiguous identifications were estimated for
mass accuracies from 0.001 to 0.05 Da and RT prediction accuracies from 0.1 to 5 min. Rates of Edman reactions were
measured for test peptides.
RESULTS: A 3D description of proteolytic peptides allowing unambiguous identification without employing MS/MS of
up to 95% and 80% of tryptic peptides from the yeast and human proteomes, respectively, was considered. Further exten-
sion of the search space to a four-dimensional one by incorporating the second N-terminal amino acid residue as the
fourth dimension was also considered and was shown to result in up to 90% of human peptides being identified
unambiguously.
CONCLUSIONS: The proposed 3D search space can be a useful alternative to MS/MS-based peptide identification
approach. Experimental implementations of the proposed method within the on-line liquid chromatography/mass
spectrometry (LC/MS) and off-line matrix-assisted laser desorption/ionization (MALDI) workflows are in progress.
Copyright © 2012 John Wiley & Sons, Ltd.
Protein discovery is a rapidly evolving field of science in the
post-genome era in which proteomics is a technological
platform to effectively characterize proteins on a large scale.
To learn about the basic mechanisms associated with cell
functioning, protein expression has to be measured under con-
tinuously changing external conditions and/or in pathological
states of the organism. The dynamic nature of the proteome is
one of the main challenges in proteomics that stimulate the
current trend towards development of high-throughput
strategies. The most widely used strategy for protein identifica-
tion is bottom-up proteomics based on the identification of
proteolytic peptides, obtained from certain cell or tissue protein
mixtures by enzymatic digestion.[1–7] In this strategy mass
spectrometry (MS) provides data on the sequences of identi-
fied peptides. This sequence information is obtained from
fragmentation mass spectra of the peptides using a variety
of methods of tandem mass spectrometry (MS/MS). A two-
dimensional array of mass spectrometric data (peptide mass,
peptide fragment masses) is further used to identify peptides in
databases using search algorithms,[8–11] or for de novo sequen-
cing.[12] However, due to the sequential selection of precursor
ions in the first stage of the MS process and due to spreading
of signal intensity between multiple fragment peaks, the combi-
nation of speed and sensitivity of MS/MS experiments is limited,
and the majority of minor species (i.e. the species present in the
original mixtures in minute amounts) are lost from the
analysis.[13] In addition, due to well-known reasons, such as
incomplete fragmentation pattern, peptide ion losses during the
in-trap isolation for subsequent fragmentation, ’chimera’ spectra,
arising from simultaneous isolation and fragmentation of several
precursor ions,[13,14] etc., this strategy does not provide unambig-
uous peptide identifications and the search engines return a
number of possible sequence candidates resulting in wrong
sequence assignments. Because of the above reasons (and others
not mentioned) it is tempting to search for approaches to
unambiguous peptide identifications alternative to MS/MS.
* Correspondence to: M. V. Gorshkov, Institute for Energy
Problems of Chemical Physics, Russian Academy of
Sciences, Moscow, Russia.
E-mail: mike.gorshkov@gmail.com
Rapid Commun. Mass Spectrom. 2013, 27, 391–400
Copyright © 2012 John Wiley & Sons, Ltd.

A. A. Lobas et al.
When the mass of a peptide is measured with a sufficient
precision such that it is unique among all the peptides
predicted from a given genome sequence, it can be used as
a biomarker for the identification of its parent protein by
mass only.[15] Even at sub-ppm mass accuracy this one-
dimensional ’Accurate Mass Tag’ technique is useful only for
small protein databases, such as baker’s yeast (S. cerevisiae).
To improve further the efficiency of identification strategy
with lower requirements for the mass measurement accuracy
(MMA) any complementary information about peptide
sequences becomes a valuable addition to the MS data. One
possible strategy employs the fact that a typical bottom-up
proteomics workflow incorporates a separation of a mixture
of proteolytic peptides using liquid chromatography (LC)
prior to MS analysis.[16] Since the retention time (RT) of a
peptide depends on its primary structure, LC can provide
additional complementary information about peptide
sequence. This makes possible the strategies based on a
two-stage approach: at the first stage, standard LC/MS/MS
analyses are undertaken on fractionated protein digests to
yield tentative peptide identifications followed by the genera-
tion of a database containing the calculated masses based on
putative peptide sequences and their corresponding chroma-
tographic retention times. This database is subsequently
validated in an LC/MS experiment measuring the accurate
masses of the detected peptides at the normalized retention
times observed for their initial identification. At the second
stage, these accurate mass and time (AMT) tags are used in
a series of LC/MS measurements as biomarkers of the
presence of a given protein without resorting systematically
to MS/MS for identification. The above strategy has been
extensively explored in recent years and is known as the
AMT-tag approach.[17] In brief, instead of searching the
database using purely mass spectrometric data, peptides are
identified in a two-dimensional space of (mass, RT), in which
the chromatographic dimension provides sequence-related
information.
AMT database generation with LC/MS/MS requires
extensive experimental work using large sets of peptides.
Alternatively, these databases can be generated in silico by
employing peptide retention times predicted for the
specified HPLC conditions.[18] A number of retention time
prediction models have been developed recently allowing
a rather high level of prediction accuracy with the correlation
between predicted and experimental values of R2 ranging from
0.85 to 0.95. They include empirical models such as SSRCalc;[19]
additive models, accounting for amino acid composition of
peptides;[20,21] as well as physical models, BioLCCC (Liquid
Chromatography of Biomacromolecules under Critical
Conditions),[22–24] and QSRR (Quantitative Structure-Retention
Relationship).[25] Also there is a number of algorithms
for retention time prediction based on machine learning,
employing a kernel-based support vector machine approach[26]
or artificial neural networks.[27] While the AMT-tag approach is
a highly promising strategy for high-throughput identifica-
tion and quantification of peptides, it has been successfully
applied in proteome-wide scale analyses for rather small
genomes.[17] Indeed, the 2D search space (mass, RT) does
not provide unambiguous peptide identification for large
proteomes. To improve further the unambiguous identifica-
tion rate both mass measurement (MMA) and RT prediction
accuracies have to be elevated beyond the edge of current
technologies. Alternatively, peptide identification may be
enhanced by obtaining additional information on peptides with
partial sequencing.
The Edman degradation reaction (Fig. 1(A)) is a long-
known method for peptide sequencing, in which amino acid
residues are analyzed by chromatography.[28,29] It is used to
cleave the peptide bond at the N-terminal amino acid residue
thus producing analogues of yn–1 fragments (so-called reduced
peptides). The reaction rate is known to have little dependence
on peptide sequence except when proline is the first residue.
The combination of Edman degradation with mass spectrome-
try is known as ladder sequencing implemented previously
for de novo sequencing of peptides.[30,31] A decade ago the
single-step degradation was proposed to obtain additional
mass spectrometric information about peptide sequences for
subsequent database search.[32–34] In these earlier works a
combination of chemical degradation with mass spectrometry
was tested experimentally for single protein identifica-
tions,[32,33] and considered in silico for the whole proteomes of
E. coli, H. influenzae and C. elegans.[34] Mass spectrometry data
in resulting 2D search space (peptide mass, reduced peptide
mass) give higher identification rates, however, with lesser
requirements for the completeness of the fragmentation pat-
tern and/or the accuracy of peptide mass measurements. For
example, in silico studies have shown that up to 90% peptides
can be unambiguously identified for relatively small
proteomes using the partial degradation technique and peptide
masses only.[34]
The purpose of this work was to evaluate the feasibility of
unambiguous peptide identifications without employing the
MS/MS technique by combining three complementary
peptide descriptors obtained experimentally in an LC/MS
run. The proposed strategy can be considered as an extension
of the above described approaches and relies on a combina-
tion of predictive chromatography, mass spectrometry, and
partial chemical degradation. It can be called an ’extended
AMT-tag’ method for shotgun proteomics in which peptides
are identified in a 3D space (peptide mass, peptide RT,
reduced peptide mass). Using in silico simulations the rates
of unambiguous identifications of tryptic peptides from yeast
and human proteomes were evaluated. A number of possible
protocols for the Edman reaction were tested experimentally
to prove the feasibility of on-line implementation of the
proposed method.
EXPERIMENTAL
All calculations were performed using two databases:
S. cerevisiae (baker’s yeast) and human proteomes, down-
loaded from the UniProt Knowledge Base.[35] The yeast
proteome S. cerevisiae database contains 1877 entries (release-
2011_10) and is a common, widely used small-sized model
system for proteomic studies. The human proteome database
contains 56392 entries (release-2011_07) and was selected
to assess the capabilities of the proposed method for a
large database.
The software used for the calculations was written in
Python programming language and based on the annotated
library called ’Pyteomics’,[36] developed in our laboratory
and freely available for download.[37] This library allows the
processing of FASTA input files and generation of protein
wileyonlinelibrary.com/journal/rcm
Copyright © 2012 John Wiley & Sons, Ltd.
Rapid Commun. Mass Spectrom. 2013, 27, 391–400

Extended AMT-tag approach for proteomics
Figure 1. (A) Schematic of the Edman degradation reaction. First stage: phenyl
isothiocyanate (PITC) binding to the N-terminus of the peptide at high pH.
Second stage: cleavage of the N-terminal amino acid residue at low pH.
(B) Schematic of the possible experimental setup for on-line implementation of
the proposed strategy. The peptide mixture after trypsin digestion is injected into
the LC separation column. Then, the separated peptide fractions enter the Edman
degradation reactor, consisting of two chambers: the first one operated at high pH,
in which PITC is added to the peptide and attached to its N-terminus, and the
second one with low pH. PITC is added to the peptide fraction in the former
chamber. In the latter chamber, TFA is added to the mixture to provide the cleavage
of the N-terminal amino acid residue. After chemical degradation the mixture of
full-length and reduced peptides is transferred to the ESI source, followed by MS
measurements.
tryptic digests including peptide modifications in silico.
The peptides used in experiments were purchased from
Sigma-Aldrich (St. Louis, MO, USA). Edman degradation
consisting of two steps was performed as follows: for binding
PITC to the N-terminus of the peptide it was dissolved in the
mixture of water/ethanol/pyridine (1:1:1) containing 5% of
PITC at the concentration of 1 mg/mL. The reaction was held
at 45 or 50 ꢀC for 1 to 30 min. For cleavage of the N-terminal
residue an equal volume of TFA was added to the mixture,
which was subsequently incubated at 45 to 60 ꢀC for 1 to
10 min. For the chromatographic analysis the mixtures were
dissolved in four volumes of water and centrifuged for
10 min at 18 000 g.
LC separations were performed using a gradient HPLC
system (Rainin Dynamax SD-200; Mettler Toledo, Greifensee,
Switzerland) in binary water/acetonitrile solutions: solution
A (water, 0.1% formic acid (FA)), solution B (95% acetonitrile,
5% water, 0.1% FA). Two types of elution gradients were
used: linear 5–50% B in 30 min, as well as a step gradient,
including 10 min wash with 5% B and 20 min with 45% B.
Mass spectrometric analyses of the peptides were performed
using an in-house built hybrid 1.25 Tesla Fourier transform ion
cyclotron resonance (FT-ICR) mass spectrometer based on the
permanent magnet and equipped with a heated metal capillary
electrospray ionization (ESI) source and a linear quadrupole
Another in-house developed Python open-source library
’Pyteomics.biolccc’ was used for peptide retention time predic-
tion based on the BioLCCC peptide separation model.[22–24]
This library is available on-line.[38] Standard nano-LC separa-
tion conditions were used for RT calculations: column length of
150 mm, inner diameter of 0.075 mm, flow rate of 0.3 mL/min,
pore size of 100 Å and a 60 min linear gradient from 2 to 40%
ACN with 0.1% trifluoroacetic acid (TFA) as an ion-pairing
agent. For highly hydrophobic peptides which are not eluted
under these typical gradient conditions the upper ACN
concentration was automatically prolonged until the elution.
Monoisotopic masses were calculated for all peptides and
one possible missed cleavage was allowed when generating
the tryptic peptides list. The effects of post-translational
modifications of amino acids, including N-terminal acetylation,
methionine oxidation and phosphorylation of serine, threonine
and tyrosine residues, as well as phenyl isothiocyanate (PITC)
modification of the lysine side chain taking place during the
Edman reaction, were assessed in a separate set of simulations.
A peptide was considered as unambiguously identified if it had
no ’neighbors’ within certain mass and time accuracy windows
in the three-dimensional search space (peptide mass, reduced
peptide mass, peptide RT).
Rapid Commun. Mass Spectrom. 2013, 27, 391–400
Copyright © 2012 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/rcm

A. A. Lobas et al.
radio-frequency (rf) ion trap for ion accumulation and isola-
tion,[39] and a commercial rf ion trap ESI-LC/MS instrument
(Amazon SL; Bruker Daltonics, Bremen, Germany).
that the uniformity of this distribution (relative to the other
physicochemical properties of the residues) is an important
factor for the successful use of N-terminal residue informa-
tion as an additional search space dimension. Although, as
one can see in Fig. 2(C), this distribution is not uniform, the
improvements in the successful peptide identification rate
are expected since many of the residues are present at the
N-terminus of tryptic peptides with almost equal probabil-
ities. The most abundant N-terminal amino acid is leucine,
and, since it cannot be distinguished from isoleucine by MS,
their contributions were summed. About 15% of all tryptic
peptides under consideration start from either a leucine or
an isoleucine residue. For these peptides we expect minimal
improvement in unambiguous identification rate from
the additional information about the N-terminus. Note,
however, that leucine and isoleucine are chromatographi-
cally distinguishable residues.[24] Maximum improvement
is predicted for peptides starting with tryptophan (W) or
proline (P) residues. According to this distribution we
expect a shortening of the list of peptide candidates by a
factor of 3 to 30 compared to a standard two-dimensional
AMT-tag strategy.
RESULTS AND DISCUSSION
Database analyses
In the first round of in silico studies, the databases used for the
estimation of the efficiency of proposed approach were
analyzed. The yeast protein database, containing 1877 entries,
after in silico trypsin digestion with one possible missed
cleavage, produced 122 637 peptides (without repeats). Then,
the peptides with the sequence lengths exceeding 4 amino
acid residues and with masses less than 3000 Da were
selected. For the selected group of 99 900 peptides, the distri-
butions by masses, retention times and N-terminal amino
acid residues were calculated. The same procedure was
employed for the human proteins, for which the database
for the analysis consisted of 1 490 728 tryptic peptides. The
resulted distributions are shown in Figs. 2(A), 2(B), and 2(C)
and reveal similar patterns for both yeast and human
proteomes. Mass distributions (Fig. 2(A)) have a decaying
exponential form on a crude scale, and there are ca. 1 Da mass
clusters on a fine scale. This clustering was previously
described by Gay et al.[40] RT distributions are shown in Fig. 2
(B) and typically reveal two maxima at 35 min and in the
low RT region for the used gradient conditions. Finally, the
N-terminal residue distribution is shown in Fig. 2(C). Note
Two-dimensional (mass, RT) distributions of yeast tryptic
peptides with N-terminal leucine/isoleucine and tryptophan
residues are shown in Figs. 2(D) and 2(E), respectively.
These distributions have almost identical patterns that
further illustrate the absence of a strong correlation between
N-terminal amino acid residues and masses or retention times
of peptides. This figure also demonstrates the orthogonality
between mass and RT for the peptides under study.
Figure 2. Tryptic digest of model protein databases generated in silico. Black plots correspond to the baker’s
yeast proteome (99 900 peptides); gray plots represent the human proteome (1 490 728 peptides). (A) Normalized
mass distributions of peptides. (B) Normalized distributions of peptides by retention times. (C) Normalized
distributions of peptides by N-terminal amino acid residue. Analyses of tryptic digest of yeast proteins. Two-
dimensional (mass, RT) distributions of yeast tryptic peptides with N-terminal (D) leucine/isoleucine and
(E) tryptophan residue. The size of the cell is 25 Da 1 min. The bar with varying dark color intensity stands
for the number of peptides per cell.
wileyonlinelibrary.com/journal/rcm
Copyright © 2012 John Wiley & Sons, Ltd.
Rapid Commun. Mass Spectrom. 2013, 27, 391–400

Extended AMT-tag approach for proteomics
Evaluation of the rate of unambiguous peptide
identifications in 3D search space
range corresponds to more realistic performance of retention
prediction algorithms. For example, the statistically deter-
mined standard error of the retention time prediction using
the BioLCCC model for the gradient conditions used in this
study is ~2 min.[22,23]
The results of the evaluation of the unambiguous peptide
identification rate in the proposed 3D search space are shown
in Figs. 3(A) and 3(B). As shown in this figure one should
expect more than 60% of unambiguous peptide identifica-
tions using the proposed 3D search space for the human protein
database, and more than 85% for the yeast proteome, with
rather modest values of mass measurement and retention time
prediction accuracies of 0.01 Da and 2 min, respectively. This
demonstrates a significant improvement compared with the
2D search space (mass of peptide, mass of reduced peptide)
proposed previously.[33]
Figures 3(C) and 3(D) show the improvement in peptide
identifications provided by addition of information about
the N-terminal amino acid residue compared with the
standard AMT-tag approach (which is also two-dimensional).
In this figure the plots marked as ’3D’ correspond to the
proposed 3D search space. For comparison the figure also
contains the results from the standard AMT-tag approach for
mass measurement accuracy of 0.01 Da. The corresponding
At the next step of the analysis we assessed the utility of
combining the partial chemical degradation via Edman
reaction with LC and MS data for peptide identifications.
The rate of unambiguous identifications of tryptic peptides was
defined as the ratio between the number of peptides identi-
fied unambiguously and the number of all peptides under
consideration. This rate was calculated for a wide range of
mass measurement accuracies from 0.001 to 0.05 Da. These
values correspond to 1 to 50 ppm of relative mass measure-
ment accuracy for tryptic peptides with the average m/z of
1000. The lower end of this MMA range is attainable by
high-performance mass spectrometry such as LC/HRT,[41]
FTICR, or Orbitrap FTMS, while the upper end is typically
achieved when using low-resolution MS instruments based
on rf quadrupole ion traps. The range of retention time
accuracies was from 0.1 to 5 min. Note that the lower end
of this range corresponds to the experimental precision
of modern HPLC systems. None of the existing models
and/or algorithms for peptide retention time prediction are
capable of delivering this level of accuracy.[18,42] The upper
end of the considered retention time prediction accuracy
Figure 3. Percent ratio of unambiguous identifications for (A) yeast and (B) human
tryptic peptides in three-dimensional search space (mass of full-length peptide, mass
of reduced peptide, RT) as a function of the accuracies of mass measurements and
retention time predictions. Dependence of the rate (in %) of unambiguous identifications
of peptides on the values of time and mass accuracies. The values of mass accuracies in
ppm correspond to the average peptide mass of 1 kDa: (C) yeast peptides, (D) human
peptides database. ’2D’ designation stands for the standard two-dimensional (mass, RT)
AMT-tag strategy, ’3D’ designation stands for the proposed ’extended’ AMT-tag strategy
in which the information on the N-terminal amino acid residue is added into the
(mass, RT) search space. Finally, ’4D’ stands for the same proposed strategy, in which
two N-terminal residues are known.
Rapid Commun. Mass Spectrom. 2013, 27, 391–400
Copyright © 2012 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/rcm

A. A. Lobas et al.
plots are depicted as ’2D’ in Figs. 3(C) and 3(D) illustrating a
significant increase in the rate of unambiguous identifications
by incorporating the third search dimension. For example, for
the human peptide database (Fig. 3(D)) the increase in the
identification rate is 1.5-fold for the highest (and actually not
realistic) retention time accuracies, while it can be as high as
30-fold for the non-accurate retention time predictions
(10 min). The increase in successful identification rate can be
as large as 4-fold achieving 90% for the yeast proteome, as
shown in Fig. 3(C).
In addition to the 3D search space, the contribution of the
information about the second N-terminal residue as the
fourth dimension was also considered. The results of this
evaluation are shown in Fig. 3(D) by the plots marked as
’4D’: even with 100 times worse MMA of 100 ppm the rate
of unambiguous identifications is higher for the 4D search
space compared to the 3D one at the MMA of 1 ppm.
Moreover, obtaining the information about two N-terminal
residues, which is certainly an instrumental challenge, will
significantly reduce the requirements for the accuracy of
predicted chromatographic retention times. This will allow
the use of less accurate physical peptide retention models,
such as additive, or BioLCCC, rather then the empirical ones.
The latter provide better accuracy yet in the limited range of
separation conditions and require large sets of known
peptides for the model training.
In the next round of calculations, phosphorylation of ser-
ine, threonine and tyrosine, as well as methionine oxidation,
were taken into account. The lengths of peptides were in a
range from 4 to 30 amino acid residues and all possible com-
binations of modified and non-modified residues were con-
sidered. Accounting for the above modifications resulted in
a set of 1 499 546 peptides for the yeast database that is an
almost 20-fold expansion of the database compared to the
non-modified peptides (80 242 entries). The size of this data-
base is very close to the size of the human peptide database
without modifications (1 490 728 entries) which makes it easy
to compare the results for peptide identifications in both
cases. The results are shown in Fig. 4(D). The rate of unambig-
uous identifications of modified yeast peptides is rather low;
it has a weak dependence on mass accuracy and a strong
dependence on retention time prediction accuracy compared
with the human non-modified peptide database. This obser-
vation is expected and the explanation for this difference is
that there are many more isomers in the set of modified yeast
peptides when multiple modification sites are considered. For
instance, 55 881 of 80 242 peptides in the yeast database have
more than one possible phosphorylation site. This results in
the increase in the number of peptides which have the same
full-length peptide and reduced peptide masses, except the
case of N-terminal amino acid residue modification. Thus,
the only dimension in which these peptides can be distin-
guished is the retention time which is known to be sequence-
and modification-site specific.[43] Note also that additive
models for retention time prediction[21] are not applicable
because they are not sequence-specific. However, even for
sequence-specific models, such as BioLCCC and SSRCalc,
the required retention time prediction accuracy of 1 min or
less is beyond their current capabilities. Moreover, the modi-
fied yeast peptides have higher densities of mass and RT dis-
tributions, as shown in Figs. 4(A) and 4(B), which further
complicates their unambiguous identification. The large shift
in the mass distribution of the modified peptides to the higher
mass region is explained by the multiple modifications (Fig. 4
(C)). Note that the higher is the mass of a peptide, the more
sites of possible modification it has, and the more modified
peptides it produces. If the number of variable modification
sites in a peptide is n, it gives rise to 2n modified peptides.
Modified peptides
Proteins in eukaryotic organisms are usually subject to post-
translational modifications (PTMs). The multi-dimensional
search space strategy proposed in this study was tested for
the modified peptides as well. In the first step, N-terminal
acetylation of the proteins as a variable modification was
taken into account only. This modification is one of the easiest
for identification. Firstly, it can appear only at the N-terminus
of the protein, which means that no more than two acetylated
peptides correspond to each protein in the case of one possible
missed cleavage. Secondly, acetylated peptides are not subject
to the Edman degradation reaction as PITC can only bind to
an NH2- group of the peptide. Therefore, there will be no
reduced pairs for the acetylated peptides, which can be further
distinguished from the non-acetylated ones within the
proposed 3D search space. As a result, the addition of the
N-terminal acetylation as a variable modification does not
decrease the rate of unambiguous identifications as shown for
both the yeast and human protein databases. The lengths of
peptides under consideration were from 4 to 30 amino acid
residues. PITC modification of lysine residues was also taken
into account as a variable one. For the yeast protein database
the set of peptides under study consisted of 80 242 entries
without acetylation and 1591 acetylated peptides. Up to
69 910 of 80 242 (87.1%) non-acetylated peptides, 1508 of 1591
(94.8%) acetylated peptides when analyzed separately,
and 71 418 of 81 833 peptides in the whole set (87.3%)
were unambiguously identified in the 3D search space for mass
and time accuracies of 0.01 Da and 2 minutes, respectively. This
actually means that all of the acetylated peptides were distin-
guished from non-acetylated ones. In the case of human proteins,
702 962 of 1 152 560 entries without acetylation (61.0%), 20 455
of 25 973 acetylated (78.7%), and 723 417 of 1 178 533 (61.4%)
peptides in the whole set were unambiguously identified.
Experimental implementation
The ultimate goal of the present study is the MS/MS-free
peptide identification through a combination of a standard
AMT-tag approach for full-length peptides and partial chemi-
cal degradation followed by MS analysis of reduced-length
peptides. In previous reports,[32,33] the Edman degradation
reaction was performed separately, and the mixture of full-
length and reduced peptides was then analyzed by MS. The
major differences between the system used in the studies by
Van der Rest et al.[32] and the one proposed here are the addi-
tion of the LC separation stage in the workflow and chemical
cleavage of the N-terminal residue instead of in-trap ion frag-
mentation. For the off-line setup certain fractions of full-
length peptides eluting from the HPLC column are collected,
subjected to partial chemical degradation one after another,
and then transferred to a mass spectrometer. This approach
is easier for the practical implementation and similar to the
one described by Van der Rest et al.[32] The off-line workflow
wileyonlinelibrary.com/journal/rcm
Copyright © 2012 John Wiley & Sons, Ltd.
Rapid Commun. Mass Spectrom. 2013, 27, 391–400

Extended AMT-tag approach for proteomics
Figure 4. Mass and retention time distributions for PTM-modified yeast peptides and
unmodified human peptides. The distributions for modified peptides have sharper
peaks, providing lower unambiguous identification rate compared to the human peptide
database. (A) Mass distribution of PTM-modified yeast peptides is strongly shifted to
the higher mass region. (B) Retention time distribution has several sharp peaks. (C)
The distribution of modified yeast peptides by the number of modifications. Most of the
peptides under study are multiply modified. (D) Comparison of the efficiency of the
use of proposed three-dimensional search space for identification of yeast peptides with
PTMs and human peptides without modifications. Dependence of the rate (in %) of
unambiguous identifications of peptides on the value of time accuracy for modified yeast
peptides and for non-modified human peptides with various values of mass accuracy.
may also be automated using a fraction collector and a
reported side reactions related to Edman degradation affect
the released PTH-amino acids, which is important in the case
of a regular sequencing contrary to ladder sequencing. In the
latter case the reduced peptides are only analyzed. For this
reason the reaction protocols may be simplified by eliminating
the need for nitrogen blowing.[31] The main challenges are in
designing the two-chamber reactor for on-line implementation
of the Edman reaction, optimizing the speed of the reaction to
make it applicable for on-line chemical degradation, and
adjusting the mixture of reaction products to the conditions
suitable for the operation of the atmospheric pressure ioniza-
tion source of the mass spectrometer.
Here we present the results of the experiments with the off-
line implementation of the Edman degradation. Angiotensin I
(DRVYIHPFHL, monoisotopic mass 1295.7 Da) was used as a
model peptide. The reaction protocols were optimized to
make it compatible with the on-line setup. The products of
the reactions were analyzed using HPLC, which also helped
in removing the pyridine present in the reaction mixture.
Pyridine is one of the main components used in the Edman
robotic system to allow addition of the required chemical
reagents. Using peptide immobilizing membranes, several
steps of the Edman degradation may be performed in series
thus adding more sequence-specific information.[30] Besides,
using the volatile reagents, such as TFEITC instead of
PITC,[31] TFA and pyridine, would reduce the chemical
poisoning that sacrifices the ionization efficiency in the ESI
source. The off-line strategy requires rapid fraction analysis,
such as robotized rapid injections with the ESI source or
scanning using the MALDI source. The on-line workflow is
shown schematically in Fig. 1(B). In this setup, the peptides
separated by HPLC come into a reactor by fractions with their
subsequent chemical degradation, followed by the injection
of the degradation products into the ESI source of a mass
spectrometer. The Edman degradation reaction consists of
two steps: phenylisothiocyanate (PITC) binding to the NH2
group of the peptide, performed at high pH value, and the
cleavage of the PITC-modified N-terminal amino acid
residue, performed at low pH value. Note that most of the
Rapid Commun. Mass Spectrom. 2013, 27, 391–400
Copyright © 2012 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/rcm

A. A. Lobas et al.
degradation protocol and because of its high proton affinity it
dampens the peptide ionization in the ESI source, even if pre-
sent in minute amounts. The results of these analyses are
shown in Figs. 5(A)–5(C). Figures 5(A) and 5(B) show
chromatograms of the products of the first and second stages
of the reaction, respectively; and Fig. 5(C) contains mass
spectra of the fractions, in which triply charged species are
present. The first stage of the reaction is PITC binding to the
N-terminus of the peptide under elevated temperature condi-
tions. The duration of the incubation was varied from 0 to
30 min. As shown in Fig. 5(A), 10-min incubation with PITC
at 45 ꢀC is sufficient to completely modify angiotensin I
(fraction 3 contains modified peptide PTC-angiotensin I
(phenyl thiocarbamoyl)). To further simplify the reaction
Figure 5. Optimization of the incubation conditions for Edman degradation. LC/MS analyses of the reagents and
reaction products: (A) Chromatograms corresponding to the first stage, PITC binding to the N-terminus of angiotensin
I. (B) Chromatograms corresponding to the second stage, cleavage of the N-terminal residue. (C) Mass spectra of the
fractions collected after LC. LC/MS analyses of products of Edman degradation eluted in a single fraction in step
gradient. (D, E) Chromatogram and mass spectrum after long incubation (30 min with PITC, 10 min with TFA at
45 ꢀC). (F, G) Chromatogram and mass spectrum after short incubation (1 min with PITC, 1 min with TFA at 50 ꢀC).
wileyonlinelibrary.com/journal/rcm
Copyright © 2012 John Wiley & Sons, Ltd.
Rapid Commun. Mass Spectrom. 2013, 27, 391–400

Extended AMT-tag approach for proteomics
protocol, the incubation time was shortened to 1 min and the
temperature was raised to 50 C. Under these conditions, up
REFERENCES
ꢀ
[1] R. Aebersold, M. Mann. Mass spectrometry-based proteomics.
Nature 2003, 422, 198.
[2] W. P. Blackstock, M. P. Weir. Proteomics: quantitative and
physical mapping of cellular proteins. Trends Biotechnol.
1999, 17, 121.
[3] J. R. Yates III. Database searching using mass spectrometry
data. Electrophoresis 1998, 19, 893.
[4] J. R. Yates III. Mass spectrometry. From genomics to
proteomics. Trends Genet. 2000, 16, 5.
to 50% of the full-length peptides were modified with PITC.
The cleavage of the N-terminal residue was performed by
addition of an equal volume of TFA to the reaction mixture
with further incubation. The incubation for 30 min at 45 C
was insufficient for complete transformation of the peptide
ꢀ
into its reduced form. Further increase in the temperature to
ꢀ
60 C increased the yield of the reduced peptide (fraction 1
in Figs. 5(B) and 5(C)). However, the incubation at elevated
temperature caused the generation of the reaction byproducts
which are marked as fraction 4 in Fig. 5(B). These byproducts
have presumably low ionization efficiency in ESI since they
were not identified using MS. To check the possibility of rapid
degradation, the reaction mixture after 1-min incubation with
PITC was dissolved in an equal volume of TFA and incubated
for 1 min at 50 ꢀC. The yield of the reduced peꢀptide was
comparable to the longer incubation periods at 45 C.
The yields of the products of Edman degradation in the case
of long and short incubation periods were measured using
step-wise gradient HPLC, in which all the peptides were
collected in a single fraction shown by dashed lines in Figs. 5
(D) and 5(F). Expectedly, for longer incubations (Fig. 5(D) and
5(E)) the amount of unmodified angiotensin I was lower, while
for the shorter ones (Fig. 5(F) and 5(G)) it was the major compo-
nent in the collected LC fraction. However, in both cases the
peaks in the mass spectra corresponding to unmodified (full-
length) and reduced-length peptides have comparable heights.
Hence, we believe that prolonged incubation periods are not
necessary to measure unambiguously the masses of both full-
length and reduced-length peptides. Note also that for on-line
implementation it is essential that each peptide arrives at the
mass analyzer at the same time as its reduced counterpart.
The design of the on-line setup is still under development
and may not incorporate the additional LC column after the
reactor chamber.
[5] D. Fenyö, J. Qin, B. T. Chait. Protein identification using
mass spectrometric information. Electrophoresis 1998,
19, 998.
[6] A. Shevchenko, O. N. Jensen, A. V. Podtelejnikov,
F. Sagliocco, M. Wilm, O. Vorm, P. Mortensen, A. Shevchenko,
H. Boucherie, M. Mann. Linking genome and proteome by
mass spectrometry: large-scale identification of yeast proteins
from two dimensional gels. Proc. Natl. Acad. Sci. USA 1996,
93, 14440.
[7] O. N. Jensen, A. Podtelejnikov, M. Mann. Delayed extraction
improves specificity in database searches by matrix-assisted
laser desorption/ionization peptide maps. Rapid Commun.
Mass Spectrom. 1996, 10, 1371.
[8] J. Eng, A. L. McCormack, J. R. Yates. An approach to corre-
late tandem mass spectral data of peptides with amino acid
sequences in a protein database. J. Am. Soc. Mass Spectrom.
1994, 5, 976.
[9] D. N. Perkins, D. J. Pappin, D. M. Creasy, J. S. Cottrell.
Probability-based protein identification by searching sequence
databases using mass spectrometry data. Electrophoresis
1999, 20, 3551.
[10] L. Y. Geer, S. P. Markey, J. A. Kowalak, L. Wagner, M. Xu,
D. M. Maynard, X. Yang, W. Shi, S. H. Bryant. Open mass
spectrometry search algorithm. J. Proteome Res. 2004, 3, 958.
[11] R. Craig, R. C. Beavis. TANDEM: matching proteins with
tandem mass spectra. Bioinformatics 2004, 20, 1466.
[12] J. Allmer. Algorithms for the de novo sequencing of pep-
tides from tandem mass spectra. Expert Rev. Proteomics
2011, 8, 645.
[13] A. Michalski, J. Cox, M. Mann. More than 100 000 detectable
peptide species elute in single shotgun proteomics runs but
the majority is inaccessible to data-dependent LC-MS/MS.
J Proteome Res. 2011, 10, 1785.
CONCLUSIONS
[14] S. Houel, R. Abernathy, K. Renganathan, K. Meyer-Arendt,
N. G. Ahn, W. M. Old. Quantifying the impact of chimera
MS/MS spectra on peptide identification in large scale pro-
teomics studies. J. Proteome Res. 2011, 9, 4152.
[15] T. P. Conrads, G. A. Anderson, T. D. Veenstra, L. Pasa-Tolić,
R. D. Smith. Utility of accurate mass tags for proteome-wide
protein identification. Anal. Chem. 2000, 72, 3349.
[16] C. C. Wu, M. J. MacCoss. Shotgun proteomics: tools for the
analysis of complex biological systems. Curr. Opin. Mol.
Ther. 2002, 4, 242.
[17] R. D. Smith, G. A. Anderson, M. S. Lipton, L. Pasa-Tolić,
Y. Shen, T. P. Conrads, T. D. Veenstra, H. R. Udseth.
An accurate mass tag strategy for quantitative and
high-throughput proteome measurements. Proteomics
2002, 2, 513.
In the present work a new three-dimensional approach combin-
ing the AMT-tag strategy with partial chemical degradation
for bottom-up proteomics was proposed. It can be used for
unambiguous MS/MS-free peptide identification by employ-
ing novel high-performance and throughput MS technologies
such as LC-HRT. The results of in silico studies on yeast
and human tryptic peptides, as well as yeast peptides with
modifications, demonstrated both possible benefits and
constraints of this extended AMT approach. The feasibility of
rapid Edman degradation with the reaction rate compatible
with on-line implementation of the proposed method was also
shown experimentally.
[18] T. Baczek, R. Kaliszan. Predictions of peptides’ retention
times in reversed-phase liquid chromatography as a new
supportive tool to improve protein identification in proteo-
mics. Proteomics 2009, 9, 835.
[19] O. V. Krokhin, S. Ying, J. P. Cortens, D. Ghosh, V. Spicer,
W. Ens, K. G. Standing, R. C. Beavis, J. A. Wilkins. Use of
peptide retention time prediction for protein identification
by off-line reversed-phase HPLC-MALDI MS/MS. Anal.
Chem. 2006, 78, 6265.
Acknowledgements
This work was supported by the Russian Foundation for
Basic Research (Grant No. 11-04-00515) and Prot-HiSPRA
Project No.282506 funded by the 7th Framework Program of
the European Commission. The authors also thank Prof.
Albert T. Lebedev from Moscow State University for useful
comments and discussions on the results.
Rapid Commun. Mass Spectrom. 2013, 27, 391–400
Copyright © 2012 John Wiley & Sons, Ltd.
wileyonlinelibrary.com/journal/rcm

A. A. Lobas et al.
[20] C. T. Mant, T. W. Lorne Burke, J. A. Black, R. S. Hodges.
Effect of peptide chain length on peptide retention
behaviour in reversed-phase chromatography. J. Chromatogr.
1988, 458, 193.
[21] M. Gilar, H. Xie, A. Jaworski. Utility of retention prediction
model for investigation of peptide separation selectivity
in reversed-phase liquid chromatography: impact of
concentration of trifluoroacetic acid, column temperature,
gradient slope and type of stationary phase. Anal. Chem.
2010, 82, 265.
[22] A. V. Gorshkov, I. A. Tarasova, V. V. Evreinov, M. M. Savitski,
M. L. Nielsen, R. A. Zubarev, M. V. Gorshkov. Liquid chroma-
tography at critical conditions: comprehensive approach to
sequence-dependent retention time prediction. Anal. Chem.
2006, 78, 7770.
[23] A. V. Gorshkov, V. V. Evreinov, I. A. Tarasova, M. V. Gorshkov.
Applicability of the critical chromatography concept to
proteomics problems: Dependence of retention time on the
sequence of amino acids. Polym. Sci. B 2007, 49, 93.
[24] I. A. Tarasova, A. V. Gorshkov, V. V. Evreinov, C. Adams,
R. A. Zubarev, M. V. Gorshkov. Applicability of the critical
chromatography concept to proteomics problems: Experi-
mental study of the dependence of peptide retention time
on the sequence of amino acids in the chain. Polym. Sci. A
2008, 50, 309.
[25] R. Kaliszan, T. Baczek, A. Cimochowska, P. Juszczyk,
K. Wiśniewska, Z. Grzonka. Prediction of high-performance
liquid chromatography retention of peptides with the use of
quantitative structure-retention relationships. Proteomics
2005, 5, 409.
[26] N. Pfeifer, A. Leinenbach, C. G. Huber, O. Kohlbacher.
Statistical learning of peptide retention behavior in chroma-
tographic separations: a new kernel-based approach for
computational proteomics. BMC Bioinformatics 2007, 8, 468.
[27] K. Petritis, L. J. Kangas, P. L. Ferguson, G. A. Anderson,
L. Pasa-Tolić, M. S. Lipton, K. J. Auberry, E. F. Strittmatter,
Y. Shen, R. Zhao, R. D. Smith. Use of artificial neural net-
works for the accurate prediction of peptide liquid chroma-
tography elution times in proteome analyses. Anal. Chem.
2003, 75, 1039.
[31] M. Bartlet-Jones, W. A. Jeffery, H. F. Hansen, D. J. Pappin.
Peptide ladder sequencing by mass spectrometry using a
novel, volatile degradation reagent. Rapid Commun. Mass
Spectrom. 1994, 8, 737.
[32] G. Van der Rest, F. He, M. R. Emmett, A. G. Marshall,
S. J. Gaskell. Gas-phase cleavage of PTC-derivatized electro-
sprayed tryptic peptides in an FT-ICR trapped-ion cell: mass-
based protein identification without liquid chromatographic
separation. J. Am. Soc. Mass Spectrom. 2001, 12, 288.
[33] F. L. Brancia, A. Butt, R. J. Beynon, S. J. Hubbard, S. J. Gaskell,
S. G. Oliver. A combination of chemical derivatisation and
improved bioinformatic tools optimises protein identification
for proteomics. Electrophoresis 2001, 22, 552.
[34] K. S. Sidhu, P. Sangvanich, F. L. Brancia, A. G. Sullivan,
S. J. Gaskell, O. Wolkenhaue, S. G. Oliver, S. J. Hubbard.
Bioinformatic assessment of mass spectrometric chemical
derivatisation techniques for proteome database searching.
Proteomics 2001, 1, 1368.
[35] Available: http://www.uniprot.org/.
[36] A. A. Goloborodko, L. I. Levitsky, M. V. Ivanov, M. V. Gorshkov.
Pyteomics – a Python framework for exploratory data
analysis and rapid software prototyping in proteomics.
J. Am. Soc. Mass Spectrom. 2012. DOI: 10.1007/s13361-012-
0516-6.
[37] Available: http://pypi.python.org/pypi/pyteomics.
[38] Available: http://pypi.python.org/pypi/pyteomics.biolccc.
[39] A. N. Vilkov, C. M. Gamage, A. S. Misharin, V. M. Doroshenko,
D. A. Tolmachev, I. A. Tarasova, O. N. Kharybin, K. P. Novoselov,
M. V. Gorshkov. Atmospheric pressure ionization permanent
magnet Fourier transform ion cyclotron resonance mass spec-
trometry. J. Am. Soc. Mass Spectrom. 2007, 18, 1552.
[40] S. Gay, P.-A. Binz, D. F. Hochstrasser, R. D. Appel. Modeling
peptide mass fingerprinting data using the atomic composi-
tion of peptides. Electrophoresis 1999, 20, 3527.
[41] C. F. Klitzke, Y. E. Corilo, K. Siek, J. Binkley, J. Patrick,
M. N. Eberlin. Petroleomics by ultrahigh-resolution time-
of-flight mass spectrometry. Energy Fuels 2012, 26, 5787.
[42] V. I. Babushok, I. G. Zenkevich. Retention characteristics of
peptides in RP-LC: Peptide retention prediction. Chromato-
graphia 2010, 72, 781.
[28] P. Edman. Method for determination of the amino acid
sequence in peptides. Acta Chem. Scand. 1950, 4, 283.
[29] P. Klemm, in Methods in Molecular Biology: Proteins, (Ed:
J. M. Walker). The Humana Press Inc., Clifton, 1984, pp. 243–254.
[30] B. T. Chait, R. Wang, R. C. Beavis, S. B. Kent. Protein ladder
sequencing. Science 1993, 262, 89.
[43] T. Y. Perlova, A. A. Goloborodko, Y. Margolin, M. L. Pridatchenko,
I. A. Tarasova, A. V. Gorshkov, E. Moskovets, A. R. Ivanov,
M. V. Gorshkov. Retention time prediction using the model
of liquid chromatography of biomacromolecules at critical
conditions in LC-MS phosphopeptide analysis. Proteomics
2010, 10, 3458.
wileyonlinelibrary.com/journal/rcm
Copyright © 2012 John Wiley & Sons, Ltd.
Rapid Commun. Mass Spectrom. 2013, 27, 391–400