Content uploaded by Sara Jawdy
Author content
All content in this area was uploaded by Sara Jawdy on Mar 26, 2015
Content may be subject to copyright.
Method
Discovery and annotation of small proteins using
genomics, proteomics, and computational approaches
Xiaohan Yang,
1,2,6
Timothy J. Tschaplinski,
1,2
Gregory B. Hurst,
3
Sara Jawdy,
1,2
Paul E. Abraham,
2,4
Patricia K. Lankford,
1
Rachel M. Adams,
2,4
Manesh B. Shah,
1
Robert L. Hettich,
2,3
Erika Lindquist,
5
Udaya C. Kalluri,
1,2
Lee E. Gunter,
1,2
Christa Pennacchio,
5
and Gerald A. Tuskan
1,2,5,6
1
Biosciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee 37831, USA;
2
BioEnergy Science Center, Oak Ridge
National Laboratory, Oak Ridge, Tennessee 37831, USA;
3
Chemical Sciences Division, Oak Ridge National Laboratory, Oak Ridge,
Tennessee 37831, USA;
4
Graduate School of Genome Science and Technology, University of Tennessee–Oak Ridge National
Laboratory, Oak Ridge, Tennessee 37830, USA;
5
DOE Joint Genome Institute, Walnut Creek, California 94598, USA
Small proteins (10–200 amino acids [aa] in length) encoded by short open reading frames (sORF) play important regu-
latory roles in various biological processes, including tumor progression, stress response, flowering, and hormone sig-
naling. However, ab initio discovery of small proteins has been relatively overlooked. Recent advances in deep transcriptome
sequencing make it possible to efficiently identify sORFs at the genome level. In this study, we obtained ~2.6 million
expressed sequence tag (EST) reads from Populus deltoides leaf transcriptome and reconstructed full-length transcripts from
the EST sequences. We identified an initial set of 12,852 sORFs encoding proteins of 10–200 aa in length. Three compu-
tational approaches were then used to enrich for bona fide protein-coding sORFs from the initial sORF set: (1) coding-
potential prediction, (2) evolutionary conservation between P. deltoides and other plant species, and (3) gene family
clustering within P. deltoides. As a result, a high-confidence sORF candidate set containing 1469 genes was obtained. Analysis of
the protein domains, non-protein-coding RNA motifs, sequence length distribution, and protein mass spectrometry data
supported this high-confidence sORF set. In the high-confidence sORF candidate set, known protein domains were
identified in 1282 genes (higher-confidence sORF candidate set), out of which 611 genes, designated as highest-confidence
candidate sORF set, were supported by proteomics data. Of the 611 highest-confidence candidate sORF genes, 56 were new
to the current Populus genome annotation. This study not only demonstrates that there are potential sORF candidates to be
annotated in sequenced genomes, but also presents an efficient strategy for discovery of sORFs in species with no genome
annotation yet available.
[Supplemental material is available for this article. The sequence data from this study have been submitted to GenBank
(http://www.ncbi.nlm.nih.gov/Genbank/) under accession nos. HP451655–HP451687, HP451690–HP451709, and HP451711–
HP451725. Mass spectrometry data have been uploaded to the Proteome Commons Tranche repository (https://
proteomecommons.org/tranche/).]
In recent years, individual experiments have demonstrated that
small proteins (<200 amino acids [aa] in length), encoded by short
open reading frames (sORF), play a major role in plant and animal
development, e.g., the TAL protein (11 aa) influencing fruit fly
development (Galindo et al. 2007), the Cg-1 protein (<33 aa)
controlling the tomato–nematode interaction (Gleason et al.
2008), the CLE family proteins (75–140 aa) (Fletcher et al. 1999;
Trotochaud et al. 2000; Muller et al. 2008; Oelkers et al. 2008)
involved in Arabidopsis meristem development, the galectin-1
protein (;130 aa) associated with the malignant human tumor
progression (Camby et al. 2006), the lipid-binding protein AZI1
(161 aa) involved in priming plant defenses (Jung et al. 2009),
and the FLOWERING LOCUS T (FT) protein (175 aa) acting as
a long-range signal regulating flowering (Notaguchi et al. 2008).
Although small proteins play important roles in regulation of
biological processes, genome-wide identification and character-
ization of sORFs has been limited. Typically, an arbitrary minimum
open reading frame (ORF) cutoff (e.g., 100 aa) is applied in gene
annotation algorithms to reduce the likelihood of falsely catego-
rizing non-protein-coding RNAs (ncRNAs) as mRNAs (Dinger et al.
2008). As a result, sORF genes are under-represented in many
current genome annotations. By searching for ORFs, Lease and
Walker (2006) identified 33,809 unannotated Arabidopsis ORFs
encoding small proteins between 25 and 250 aa in length, out of
which 10,247 (30%) had expression evidence from genome-wide
tiling hybridization experiments. Hanada et al. (2007) performed
a large-scale search for sORFs encoding proteins of 30–100 aa in the
intergenic regions of the Arabidopsis genome using a simple gene-
finding method. They identified 7159 sORF candidates, of which
3241 had either transcriptional evidence or indication of purifying
selection. Based on this research, Hanada et al. (2010) developed
a program package, sORF Finder, for identifying sORFs according
to the nucleotide composition bias among coding sequences and
the potential functional constraint at the amino acid level through
evaluation of synonymous and nonsynonymous substitution rates.
6
Corresponding authors.
E-mail yangx@ornl.gov
E-mail tuskanga@ornl.gov; fax: (865) 576-9939.
Article published online before print. Article, supplemental material, and pub-
lication date are at http://www.genome.org/cgi/doi/10.1101/gr.109280.110.
Freely available online through the Genome Research Open Access option.
634 Genome Research
www.genome.org
21:634–641 Ó2011 by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/11; www.genome.org
However, only 2% of unannotated sORFs predicted by Hanada
et al. (2007) were confirmed by the Arabidopsis proteomic data
(Castellana et al. 2008). The sORF-finding approaches used by
Lease and Walker (2006) and Hanada et al. (2007) are solely in
silico gene predictions based on genomic DNA sequence, with
the assumption that small proteins are encoded by intronless
(i.e., single-exon) genes. In silico prediction of full-length tran-
scripts based on genomic sequences is challenging and has low
accuracy (sensitivity ranging from 41% to 68% and specificity
from 20% to 53%) (http://augustus.gobics.de/accuracy). Thus,
an alternative strategy is needed for identifying protein-coding
sORFs.
Here we report the outcome of an integrative procedure based
on transcriptomics, proteomics, and computational biology for
the discovery of sORFs that encode small proteins <200 aa in
length in Populus deltoides. Our strategy for large-scale discovery of
small proteins is outlined in Figure 1. Briefly, a three-step approach
was used to reconstruct transcription units (TU) using expressed
sequence tags (EST) obtained from deep sequencing of the P. del-
toides leaf transcriptome. Since a true protein-coding transcript is
more likely to have a long and high-quality ORF compared with
a non-coding transcript (Kong et al. 2007), we established an initial
sORF candidate set by selecting the longest ORF-encoding protein
sequence of <200 aa in length for each TU. Then we applied three
computational approaches sequentially to enrich for protein-
coding sORFs: (1) coding-potential prediction based on known
protein sequences, (2) evolutionary conservation between P. del-
toides and other plant species, and (3) protein sequence clustering
within P. deltoides.
The efficiency of our sORF discovery strategy was validated by
both bioinformatics (e.g., protein domain-scanning) and experi-
mental approaches (e.g., protein mass spectrometry). This study
not only demonstrates that there are many potential sORF candi-
dates yet to be annotated in sequenced genomes, but also presents
an efficient strategy for sORF discovery in species with as yet un-
annotated genomes.
Results
Establishment of an initial sORF candidate set
We sequenced the transcriptome of six Populus deltoides leaf sam-
ples and generated ;2.6 million ESTs with a medianlength of ;240
nucleotides. We comparatively examined the representation of
100 annotated P. trichocarpa gene models encoding proteins <200
aa in length (http://genome.jgi-psf.org/Poptr1_1/Poptr1_1.home.
html) using the ;634,000 ESTs from one of the six leaf samples.
Twenty-five of these selected gene models were found in the EST
data, with 80% (=20/25) of them having full-length or alternative
splicing coverage (Supplemental Fig. 1; Supplemental Table 1),
indicating that our EST data provided an appreciable full-length
coverage of transcripts encoding proteins <200 aa in length. To
minimize the number of truncated TU reconstructed from all ESTs,
a three-step approach was utilized to create full-length TUs: (1)
high-stringency de novo assembly, followed by (2) genome-loca-
tion-based assembly, and then (3) medium-stringency assembly.
Since there was only ;1% divergence between P. trichocarpa and P.
deltoides at the genomic sequence level (data not shown), the ge-
nomic resources (genome sequence and annotated mRNA se-
quences) in P. trichocarpa were used as references for P. deltoides EST
assembly. As such, the P. deltoides ESTs, pooled with the annotated
P. trichocarpa mRNA sequences (Tuskan et al. 2006), were assembled
into TUs that each contained at least three ESTs. From these TUs,
an initial sORF candidate set (Fig. 1) encoding 12,852 proteins of
10–200 aa in length was created by including the longest possible
complete ORF that contained start and stop codons in six-frame
translations from each TU.
Enrichment for protein-coding sORFs from the initial sORF
candidate set
Three computational approaches were used to enrich for protein-
coding sORFs to address the challenge of identifying protein-cod-
ing genes from a large number of short TUs assembled directly
from ESTs. First, we interrogated the initial sORF candidate set
using the Coding Potential Calculator (Kong et al. 2007) trained
with protein sequences obtained from the UniProt database (The
UniProt Consortium 2009). This approach identified 4918 sORF
candidates with high protein-coding potential, designated as
Subset A (Fig. 2A). Second, we compared the initial sORF candidate
set derived from P. deltoides with 14 additional plant genome se-
quences ranging from algae to angiosperm species (Supplemental
Fig. 2) and identified 2649 sORFs that are conserved between
P. deltoides and at least one other plant species, designated as Subset
B (Fig. 2A). The number of conserved sORFs between P. deltoides
and the 14 tested species ranged between 300 and 2076 and as
expected, the number of conserved sequences was inversely pro-
portional to evolutionary distance (Supplemental Fig. 2). Finally,
we performed a clustering analysis of the initial sORF candidate set
and detected 3372 sORFs that clustered into families with 3–51
members, designated as Subset C (Fig. 2A). The 1469 sORF candi-
dates shared by Subsets A, B, and C were designated as the high-
confidence sORF candidate set (Figs. 1, 2A).
Figure 1. The strategy for large-scale discovery of small proteins in
Populus deltoides.
Genome Research 635
www.genome.org
Small protein discovery using an integrated method
Length distribution of protein sequences in the
high-confidence sORF set
We examined the protein length distribution to assess whether the
putative small protein sequences occurred more often than ex-
pected by chance alone. The frequency of protein sequences <100
aa in length in the high-confidence sORF candidate set was, as
expected, lower than that in the random sequence set (Fig. 3),
suggesting that sORFs in the high-confidence sORF candidate set
are likely not randomly generated as a result of assembly errors.
Moreover, the length distribution of the high-confidence sORF
candidate set was similar to that of the small protein set (<200 aa)
in the current Arabidopsis genome annotation (v9) (Fig. 3).
The most recent annotations of the Arabidopsis genome
(v8–9) include more small proteins relative to the earlier versions
(v5–7) (Supplemental Fig. 3A). Out of the 1694 new protein se-
quences added to Arabidopsis genome annotation v8, 1079 (64%)
gene models encode proteins of <200 aa in length (Supplemental
Fig. 3B), indicating that incorporation of a large number of new
small protein sequences is a key feature of the improved annota-
tion. Similarly, the frequency of sequences <120 aa in length in the
high-confidence sORF candidate set was greater than that found in
the current annotation of Populus (Fig. 3), suggesting that sORFs
are under-represented in the current annotation data sets in Pop-
ulus, particularly for those proteins between 30 and 120 aa in
length. Similar results were found within the rice genome (Fig. 3).
To evaluate the possibility that the sORF sequences were not
full-length (i.e., truncated), we cloned full-length TUs for 15
sORFs (Supplemental Table 2) in the high-confidence sORF can-
didate set using the Rapid Amplification of cDNA Ends (RACE)
technology. All of the 15 tested sORFs were confirmed as full-
length sequences. Thus, it is likely that the sORFs in the high-
confidence sORF candidate set are predominantly full-length
transcript sequences.
Validation of the high-confidence sORF candidate set
by analysis of protein domain and mass spectrometry data
We subsequently surveyed the high-confidence sORF candidate set
for known protein domain(s) using InterProScan, which integrates
results obtained from searching 14 databases (Mulder and Apweiler
2007). Approximately 23% of the initial sORF candidate set had
known protein domains. The protein domain discovery rate was
increased by sequential filtering using coding potential prediction,
interspecies conservation, and protein sequence clustering (Fig.
2B). We detected known protein domains in 1282 (87%) sORFs
in the high-confidence sORF candidate set (Fig. 4A). These 1282
sORF candidates were designated as a higher-confidence sORF can-
didate set (Fig. 1).
To further evaluate the validity of the sORF candidates, we
surveyed existing P. deltoides xylem (Kalluri et al. 2009) as well as
new xylem, phloem, and leaf protein mass spectrometry (MS) data
(Supplemental Tables 3, 4). TheMS analysis led to identification of
4943 different tryptic peptides, which were assembled into 1158
sORF-encoded proteins (Supplemental Table 3). Unique peptides
were detected in one or more experiments for 307 sORF-encoded
proteins (see ‘‘distinct’’ [DS] or ‘‘differentiable’’ [DF] sets in Sup-
plemental Table 3). Only 9% (=1158/12,852) of the initial sORF
Figure 2. P. deltoides small protein-coding candidate genes enriched
from transcription units. (A) Number of genes in different sORF candidate
subsets. (B) Proportion of the sORF subsets having known protein domains
detected by InterProScan. Subset A contains the sORF candidates with
high protein-coding potential predicted using known proteins as training
sequences. Subset B contains sORF candidates conserved between
P. deltoides and at least one other plant species. Subset C contains sORF
candidates clustered into families. (Initial) The initial sORF candidate set
(Fig. 1). (AB) The intersection of Subsets A and B. (ABC) (i.e., the high-
confidence sORF candidate set) The intersection of Subsets A, B, and
C. The value in parentheses represents the number of sORFs in each
individual subset.
Figure 3. Length distribution of predicted protein sequences. (Random)
The random sORFs; (At-annotation) the small proteins in Arabidopsis ge-
nome annotation (v9); (Pt-annotation) the small proteins in Populus ge-
nome annotation (v2.0); (Os-annotation) the small proteins in rice
genome annotation (v6.1); (Pd-sORF) the high-confidence sORF candi-
date set (Fig. 1) shared by Subsets A, B, and C in Figure 2.
Yang et al.
636 Genome Research
www.genome.org
candidate set had proteomics matches, with the length of matched
proteins ranging from 20 to 200 aa (Supplemental Fig. 4). The size
of the sORF subset containing known protein domains was three
times that of the sORF subset with proteomics support (Subset
D vs. Subset P in Supplemental Fig. 5), indicating the possibility
that deeper proteome coverage would provide additional evidence
supporting more sORFs.
The proteomics-matching rate was increased by sequential
filtering using coding potential prediction followed by analysis of
interspecies conservation and protein sequence clustering (Fig.
4B). Our proteomics analyses revealed that ;43% of the high-
confidence sORF candidate set matched the xylem, phloem, or leaf
protein MS data (Fig. 4B). Filtering of the high-confidence sORF
candidate set by InterProScan search for known protein domains
increased the proteomics-matching rate to 48%, with 611 sORFs
in the higher-confidence sORF candidate set having proteomics
support (Fig. 4B). These 611 proteomics-supported ORF candidates
were designated as the highest-confidence sORF candidate set (Fig.
1), with protein length ranging from 40 to 200 aa (Fig. 5). Fur-
thermore, 373 small proteins encoded by sORFs in the highest-
confidence sORF candidate set were detected in proteomics mea-
surements of conductive tissues (i.e., in phloem or xylem), but not
in leaf (Fig. 6). Approximately 9% (56 protein sequences) of the
highest-confidence sORF candidate set were misannotated in or
missing from the P. trichocarpa genome annotation (v2.0; http://
www.phytozome.net/) (Supplemental Table 5).
Possibility of sORF candidates as non-protein-coding RNA
Many short RNA sequences have been identified as ncRNAs, as
documented in the Rfam database (Gardner et al. 2009). To de-
termine whether the high-confidence sORF candidate set contains
potential ncRNAs, we conducted an Rfam-based search with all of
its 1469 TU sequences using the Infernal program (Nawrocki et al.
2009) and found that only 0.3%–2.1% (4–31 TUs) of the high-
confidence sORF candidate set were potential ncRNAs (Fig. 7),
suggesting a low probability of ncRNAs (Rfam database) in high-
confidence sORF candidate set.
Discussion
Although small proteins have been shown to play important roles
in various biological processes, they have largely escaped detection
because it is difficult to predict sORFs (Kastenmayer et al. 2006;
Dinger et al. 2008). Previous large-scale ab initio discovery of sORFs
(Lease and Walker 2006; Hanada et al. 2007, 2010) identified
thousands of single-exon genes directly from genomic sequences.
Interestingly, nearly 50% of the annotated Arabidopsis genes en-
coding small proteins <100 aa in length contain introns (un-
published observation). For example, the Arabidopsis RCI2A gene
encoding a small protein of 54 aa in length contains two introns
(http://www.arabidopsis.org). The major limitation in these pre-
vious in silico sORF prediction efforts is that their sORF predictions
were not designed to detect multiple-exon sORF genes. As a result,
only 155 (;2%) of 7442 sORFs predicted by Hanada et al. (2007)
were verified by proteomics data (Castellana et al. 2008). Our ap-
proach integrates experimental data (transcriptome), coding po-
tential prediction, evolutionary conservation, and gene family
clustering. We reconstructed full-length transcription units (i.e.,
mRNAs) directly from the large volume of EST sequences obtained
from deep sequencing of the transcriptome. In other words, ex-
perimental evidence provided the initial candidate set for our
predictions. The sORF candidates predicted in this study had
a relatively high rate of proteomics support. In our high-confi-
dence sORF candidate set, ;43% have protein MS data support.
This rate is similar to the Arabidopsis whole-genome annotation,
Figure 4. Protein domain annotation of sORF candidates. (A) Venn di-
agram showing the number of sORF candidates in four different subsets
and their intersections. (B) Proportion of the sORF subsets having protein
mass spectrometry data support. The ‘‘Initial’’ set, Subsets A, B, C, AB, and
ABC are as described in Figure 2. Subset D contains sORF candidates with
known protein domains detected by InterProScan. (ABCD) (i.e., the
higher-confidence sORF candidate set) The intersection of Subsets A, B, C,
and D. The value in parentheses represents the number of sORFs in each
individual subset.
Figure 5. Size distribution of the 611 sORF-encoded proteins in the
highest confidence set. (All) All of the 611 proteins. (Novel) The 56 sORF-
encoded small proteins new to the Populus genome annotation (v2.0;
http://www.phytozome.net/).
Small protein discovery using an integrated method
Genome Research 637
www.genome.org
in which 40% of the gene models have protein MS support
(Castellana et al. 2008).
In this study, we reconstructed TU directly from EST se-
quences, avoiding the uncertainty caused by ab initio prediction
from genomic sequences. The EST sequences obtained from deep
transcriptome sequencing provided numerous full-length tran-
scripts (Supplemental Fig. 1; Supplemental Table 1) and all 15
RACE-tested TUs from the high-confidence sORF candidate set
were shown to be full-length messages (Supplemental Table 2).
This high-quality reconstructionof full-length TUs suggests that the
majority of the predicted sORF-encoded proteins are not false pos-
itive predictions of truncated portions of long protein sequences.
Our first computational filter, prediction of protein-coding
potential of transcript sequences based on known protein se-
quences, markedly increased the proteomics-matching rate for the
sORFs. Still, the remaining two filtering approaches, based on
interspecific conservation and protein family clustering, respec-
tively, identified additional sORF candidates with protein support.
These data suggest that small protein sequences are under-repre-
sented in the current protein databases. Thus, we anticipate that
coding potential prediction can be improved as additional in-
formation is deposited in public protein databases.
It is well known that many genes are conserved among species
across different evolutionary distances (Kriventseva et al. 2008;
Ostlund et al. 2010). We identified sORFs that encode protein se-
quences conserved between P. deltoides and 14 other plant species
(Supplemental Fig. 2), suggesting that interspecific conservation
is a valid approach to enrich for protein-coding genes. As addi-
tional plant genome sequences become available, the interspecific
conservation approach should become more useful in small pro-
tein discovery.
Matching sORF candidate sequences with proteomics data
can provide direct evidence for small protein discovery. In this
study, we demonstrated that ;43% of the high-confidence sORF
candidate set had supportive MS data. However, the number of
sORFs having experimental proteomics support was lower than the
number of sORFs with predicted protein domains. sORFs with
protein MS data support are limited by protein sampling depth.
Our analysis showed that the vast majority of protein MS data were
represented by protein domain data (Supplemental Fig. 5), sug-
gesting that computational validation of sORFs using a protein
domain could be complementary to the more expensive experi-
mental validation approach based on protein MS analysis.
By using EST data obtained from deep transcriptome se-
quencing, this study revealed more sORFs than predicted in the
current Populus annotation (Fig. 3). One possible reason for this
bias against small proteins in the current annotation of Populus is
that the EST sequences were obtained by traditional Sanger se-
quencing of cloned cDNA libraries, in which cDNAs smaller than
400–500 bp were typically eliminated by size selection (Lease and
Walker 2006). A key feature of recent improvements in the Arabi-
dopsis genome annotation is the incorporation of a large number of
short protein sequences (Supplemental Fig. 3). Small proteins are
proportionately under-represented in the current Populus genome
annotation compared with the most recent Arabidopsis annotation
(Fig. 3), which reflects the more mature nature of the Arabidopsis
annotation (v9.0) relative to Populus (v2.0). The length distribu-
tion of our predicted sORFs is similar to that of Arabidopsis (Fig. 3),
indicating that our prediction offers a potential improvement in
small protein annotation in Populus.
Some small proteins have been reported to be involved in cell-
to-cell communications in plants. For example, a small protein of
94 aa called CAPRICE (CPC) is a transcription factor involved in
intercellular signal transduction associated with root hair devel-
opment in Arabidopsis (Kurata et al. 2005). It was also recently
demonstrated that a membrane-associated thioredoxin (140 aa)
moves from cell to cell, suggestive of a role in intercellular com-
munication (Meng et al. 2010). Our proteomics measurements
identified hundreds of sORF-encoded proteins in P. deltoides in
phloem or xylem, but not in leaf. These sORFs represent a candi-
date pool of putative proteins that may be mobile molecules
mediating intercellular signal transduction.
We have been able to demonstrate that deep RNA sequencing
can be used in combination with computational approaches to
predict high-likelihood protein-encoding sORFs that have not
typically been annotated in most plant genomes. These results,
supported by protein domain and proteomics evidence, suggest
that the integrative approach used in this study to create the high-
confidence sORF candidate set is effective in identifying protein-
coding sORFs.
Methods
Plant material and RNA extraction and sequencing
Total RNA was isolated from leaf tissue of 6-mo-old Populus deltoides
plants grown under normal and drought conditions using a Sigma
Figure 6. Venn diagram showing the number of sORFs from the 611-
sORF set with the highest confidence that were detected in P. deltoides
leaf, phloem, and xylem tissue based on analysis of the trypsin-digested
whole proteome using two-dimensional HPLC interfaced with tandem
mass spectrometry.
Figure 7. Number of sORFs in the high-confidence sORF candidate set
classified as potential ncRNAs by an Rfam database search. The e-value
cutoff was used in the Rfam search.
Yang et al.
638 Genome Research
www.genome.org
Spectrum Plant Total RNA kit. RNA was extracted from 10 bi-
ological replicates. Equal amounts of total RNA from each of the
biological replicates were pooled. The samples were run on an
Experion (BioRad) to verify RNA quality and a Nanodrop (Thermo
Fisher Scientific) to determine sample concentration. A total of 500
mg of total RNA from each sample was sent to the Joint Genome
Institute (JGI), where transcriptome sequencing was performed
using the Roche 454 Genome Sequencer FLX System (GS FLX). The
raw expressed sequence tag (EST) sequences generated by tran-
scriptome sequencing were trimmed for vector, adaptor/linker,
poly(A) or T tails. The trimmed ESTs were edited for length and
ESTs <100 nt were removed. ESTs with low complexity sequence
greater than the threshold (default =50%) were also removed. The
EST sequence s were then blasted against the GenBank nucleoti de
database in order to identify and eliminate contaminants. ESTs
found to match nontarget sequences (e.g., non-nuclear) were
removed.
Transcription unit assembly
Transcription units (TU) were created through three rounds of as-
sembly using the P. trichocarpa genome and annotated mRNAs
as references. Whole-genome resequencing of P. deltoides revealed
that there was only ;1% divergence between P. trichocarpa and
P. deltoides at the genomic sequence level (data not shown). Thus,
the reference P. trichocarpa transcript sequences (GeneCatalog_
frozen20080522; ftp://ftp.jgi-psf.org/pub/JGI_data/Populus_
trichocarpa/v1.1/) were pooled with the filtered P. deltoides EST
sequences obtained from the 454 sequencing data and clustered
using sclust implemented within the tgicl software (Pertea et al.
2003) using 97% identity and 80% sequence coverage criteria.
Sequence clusters were then assembled using the CAP3 software
(Huang and Madan 1999) to form consensus sequences with an
overlap length cutoff of 40 and an overlap identity of 97%. The
second-round assembly was alignment based. The consensus se-
quences were aligned onto the P. trichocarpa genome sequence
version 1.1 (http://genome.jgi-psf.org/Poptr1_1/Poptr1_1.home.
html) using BLAT (Kent 2002) with a minimum coverage (i.e.,
minimum fraction of query that must be aligned) of 80% and
a minimum identity of 92%. Only the ‘‘best match’’ position was
selected as the genomic location for each query consensus se-
quence. The genomic locations (i.e., GFF) of the annotated gene
models were obtained from http://genome.jgi-psf.org. The con-
sensus sequences and/or the JGI gene models that have over-
lapping genomic locations were reassembled using the CAP3
software with an overlap length cutoff of 30 and an overlap
identity of 75%. In the final round of assembly the contigs
obtained from the second-round assembly were mixed with the P.
trichocarpa genome annotation v2.0 (http://www.phytozome.net/)
mRNA sequences and assembled using the CAP3 software with an
overlap length cutoff of 30 and an overlap identity of 92%. We
empirically examined the influence of overlap identity (75%, 80%,
85%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, and 98%) on
CAP3 assembly and found that with 92% or 93% identity we were
able to distinguish gene duplications while being capable of tol-
erating sequencing error or single-nucleotide polymorphisms.
Similarly, Masoudi-Nejad et al. (2007) found that 92% was an ap-
propriate overlap identity for CAP3 assembly of plant EST se-
quences.
sORF/small protein analysis
The TUs assembled from EST and JGI gene models were translated
in 6-frame using the Emboss package (Rice et al. 2000). An initial
sORF candidate set encoding proteins of 10–200 aa in length was
created by including the longest possible complete ORF that
contained start and stop codons in six-frame translations from
each TU.
Clustering of protein sequences
All-vs-all BLASTp (Altschul et al. 1997) of the protein sequences
were performed with an e-value cutoff of 10. The BLASTp data was
then used to cluster the protein sequences into groups using the
FORCE program (Wittkop et al. 2007) with a cutoff of 3.4.
Assessment of protein coding or non-coding potential
The assessment of coding potential of the sORF candidate set was
conducted using the Coding Potential Calculator (Kong et al.
2007) based on a training set of proteins obtained from the UniProt
database (The UniProt Consortium 2009). To determine whether
the sORFs correspond to the non-coding RNAs, the ORF sequences
were used as queries to search against the Rfam database (Griffiths-
Jones et al. 2005; Gardner et al. 2009) using Infernal (Nawrocki
et al. 2009).
Protein motif analysis
Protein sequences were scanned for domains using blastprodom,
coils, gene3d, hmmpanther, hmmpir, hmmpfam, hmmsmart,
hmmtigr, fprintscan, patternscan, profilescan, superfamily, seg,
signalp, and tmhmm implemented in InterPro (Zdobnov and
Apweiler 2001; Mulder and Apweiler 2007).
Conservation analysis of protein sequences between species
The putative P. deltoides small protein sequences were used as queries
to search against the genome sequences of Arabidopsis lyrata (http://
www.phytozome.net/), A.thaliana (http://www.arabidopsis.org/),
Brachypodium distachyon (http://www.brachypodium.org/), Carica
papaya (http://asgpb.mhpcc.hawaii.edu/papaya/), Chlamydomonas
reinhardtii (http://www.phytozome.net/), Cucumis sativis (http://
www.phytozome.net/), Glycine max (http://www.phytozome.net/),
Medicago truncatula (http://www.medicago.org/), Oryza sativa
(http://rice.plantbiology.msu.edu/), Physcomitrella patens (http://
www.phytozome.net/), Selaginella moellendorffii (http://www.
phytozome.net/), Sorghum bicolor (http://genome.jgi-psf.org/
Sorbi1), Vitis vinifera (http://www.genoscope.cns.fr), and Zea
mays (http://maizesequence.org/) using BLAT (Kent 2002) with
a minimum coverage (i.e., minimum fraction of query that must
be aligned) of 80% and a minimum identity of 60%.
Generation of random sORF sequences
Random coding sequences were generated using GenRGenS
(Ponty et al. 2006). Specifically, a Markov model was first
constructed based on the P. t r i c h o c a r p a coding sequences
(GeneCatalog_frozen20080522; ftp://ftp.jgi-psf.org/pub/JGI_data/
Populus_trichocarpa/v1.1/) with an order of 2 and a phase of 3.
Then, 20,000 random sequences of 600 bp starting with ATG (the
start codon in protein-coding sequences) were generated. Finally,
the complete coding sequences containing the first start codon
(ATG) and a stop codon (TAA, TAG, or TGA) were selected as the
random sORFs.
Mass spectrometry analysis of proteins
Protein was extracted from fully expanded P. deltoides leaves fol-
lowing the method of Lee et al. (2009). Tissues from each plant
Small protein discovery using an integrated method
Genome Research 639
www.genome.org
were ground separately under liquid N
2
and stored at 80°C. Ap-
proximately 600 mg of leaf powder from each plant was suspended
in 2.5 mL of lysis buffer (100 mM Tris HCl at pH 8.5; 5 mM DTT; 1
mM EDTA; 1 mM PMSF; 0.1 mg/mL leupeptin) and homogenized
using a glass dounce tube. Each homogenate was centrifuged at
1000gfor 10 min; each supernatant was further centrifuged at
30,000gfor 60 min. Protein concentration in each final superna-
tant was measured by the Lowry method (Lowry et al. 1951) and
equal protein amounts from the separate supernatants were com-
bined to yield three pooled extracts, each containing a total of 3 mg
of protein. Proteins were precipitated using 25% trichloroacetic
acid; the resulting pellets were washed with acetone and resolu-
bilized in 6 M guanidine/100 mM Tris HCl (pH 8.5) with sonica-
tion. Aliquots corresponding to ;1 mg of protein were reduced
by incubation with 10 mM DTT for 20 min and carboxy-
amidomethylated by incubation with 100 mM iodoacetamide for
15 min in the dark, both at ambient temperature. Samples were
diluted to decrease guanidine concentration to 1 M with 50 mM
Tris HCl/10mM CaCl
2
. Proteins were digested by incubating
overnight at 37°C with trypsin (10 mg/mg protein; Promega se-
quencing grade), followed by the addition of a second identical
amount of tr ypsin and an additio nal 4-h incubation. D igests were
desalted (SepPak Lite C18, Waters) and analyzed in triplicate
using two-dimensional HPLC interfaced with tandem mass
spectrometry as described previously (Kalluri et al. 2009).
Protein extraction and quantification from xylem and phloem
tissue was performed using a method essentially identical to that
recently applied for proteomic analysis of xylem tissue (Kalluri et al.
2009). An additional centrifugation step (3000gfor 10 min) was
performed following trypsin digestion to remove cellular debris
from solution. Digests (100-mg aliquots, based on protein mea-
surement) were analyzed using two-dimensional HPLC interfaced
with tandem mass spectrometry.
Tryptic peptide identifications were extracted from the tan-
dem mass spectra from leaf, xylem, and phloem tissues, as well
as from previously published data on P. deltoides xylem proteins
(Kalluri et al. 2009) using Sequest. Peptide identifications were
filtered and compiled using DTASelect to provide protein identi-
fications; a protein required evidence from two or more tryptic
peptides per protein, or identification of a single tryptic peptide
in two or more charge states. The protein database for the Sequest
searches contained protein sequences in P.trichocarpa annotation
v2.0 (www.phytozome.net), the initial sORF candidate set (12,852
sequences) (Fig. 1), a sequence-reversed analog of each protein for
estimation of false discovery rates, and commonly observed con-
taminant proteins. Sequest was executed with no enzyme speci-
ficity, and non-tryptic peptide identifications were subsequently
removed from the data set using DTASelect. False discovery rates
among the remaining tryptic peptides were typically 1% or less.
Parsimony analysis was performed to identify tryptic peptides
shared among several proteins (Yang et al. 2004). Further details
are provided in the caption for Supplemental Table 4.
The proteomics data have been deposited in Proteome Com-
mons Tranche repository, https://proteomecommons.org/tranche/
(hashes Fv9zgC97mv0bld5KMOM7ww9mP24qhchGLvS7Cx4ddY
CJgm28KiUsD5xp0UQCglivuBSz59Tdes8+auDPWYDOleix/vUAAA
AAAAAMhA== and /WZkinVg1kkkYxvkqAQKAXSW5ujyhPCjp9W
FBdzXoLAh6qnH50N+Tl1sekqV9XVWLCssVOGS63e9OyhkAvCLG
49wAeQAAAAAAAAOlA==).
Full-length cDNA cloning using Rapid Amplification
of cDNA Ends
Full-length cDNA was synthesized from total RNA using a GeneRacer
Kit (Invitrogen). The resulting cDNA template was used in PCR
reactions to amplify both the 59and 39end of each gene of interest.
59ends were amplified using a GeneRacer 59primer and a reverse
gene-specific primer (GSP). 39ends were amplified using a 39
GeneRacer primer and a forward GSP. GSPs (Supplemental Table
2) were designed according to specifications provided in the
GeneRacer protocol. After PCR amplification the 59and 39frag-
ments were run on an agarose gel, excised out, purified using a
Minelute Gel Purification kit (Qiagen), and sequenced. Sequences
were aligned using Sequencher version 4.5.
Comparison of sORF-encoded proteins with the Populus
genome annotation
The sORF coding sequences were mapped to the P. trichocarpa ge-
nome v2.0 (http://www.phytozome.net/) by BLAT (Kent 2002)
with a minimum coverage (i.e., minimum fraction of query that
must be aligned) of 80% and a minimum identity of 92%. Only the
‘‘best match’’ position was selected as the genomic location for
each query sequence. The genomic locations (i.e., GFF) of the
gene models in annotation v2.0 were obtained from http://www.
phytozome.net/. In cases where there were overlapping genomic
locations between sORF CDS and the annotated Populus gene
models, the sORF CDS sequences were compared with annotated
CDS using the MAFFT alignment program (Katoh et al. 2002, 2005).
Acknowledgments
We thank S.D. Wullschleger and D.J. Weston for thoughtful and
insightful comments on the manuscript. Transcriptome sequenc-
ing was supported by the U.S. Department of Energy Joint Genome
Institute Laboratory Science Program project with X.Y. and T.J.T.
The work conducted by the U.S. Department of Energy Joint Ge-
nome Institute is supported by the Office of Science of the U.S.
Department of Energy under Contract No. DE-AC02-05CH11231.
Proteomics and bioinformatics analysis was supported by the U.S.
DOE Office of Biological and Environmental Research, Genomic
Science Program and the U.S. DOE BioEnergy Science Center. The
BioEnergy Science Center is a U.S. Department of Energy Bio-
energy Research Center supported by the Office of Biological and
Environmental Research in the DOE Office of Science. Oak Ridge
National Laboratory is managed by UT-Battelle, LLC for the U.S.
Department of Energy under Contract Number DE–AC05–
00OR22725.
References
Altschul SF,Madden TL, Schaffer AA, Zha ng J,Zha ng Z, Miller W, Lipman DJ.
1997. Gapped BLAST and PSI-BLAST: A new generation of protein
database search programs. Nucleic Acids Res 25: 3389–3402.
Camby I, Le Mercier M, Lefranc F, Kiss R. 2006. Galectin-1: A small protein
with major functions. Glycobiology 16: 137R–157R.
Castellana NE, Payne SH, Shen Z, Stanke M, Bafna V, Briggs SP. 2008.
Discovery and revision of Arabidopsis genes by proteogenomics. Proc
Natl Acad Sci 105: 21034–21038.
Dinger ME, Pang KC, Mercer TR, Mattick JS. 2008. Differentiating protein-
coding and noncoding RNA: challenges and ambiguities. PLoS Comput
Biol 4: e1000176. doi: 10.1371/journal.pcbi.1000176.
Fletcher JC, Brand U, Running MP, Simon R, Meyerowitz EM. 1999.
Signaling of cell fate decisions by CLAVATA3 in Arabidopsis shoot
meristems. Science 283: 1911–1914.
Galindo MI, Pueyo JI, Fouix S, Bishop SA, Couso JP. 2007. Peptides
encoded by short ORFs control development and define a new
eukaryotic gene family. PLoS Biol 5: e106. doi: 10.1371/
journal.pbio.0050106.
Gardner PP, Daub J, Tate JG, Nawrocki EP, Kolbe DL, Lindgreen S, Wilkinson
AC, Finn RD, Griffiths-Jones S, Eddy SR, et al. 2009. Rfam: Updates to the
RNA families database. Nucleic Acids Res 37: D136–D140.
Gleason CA, Liu QL, Williamson VM. 2008. Silencing a candidate nematode
effector gene corresponding to the tomato resistance gene Mi-1 leads to
acquisition of virulence. Mol Plant Microbe Interact 21: 576–585.
Yang et al.
640 Genome Research
www.genome.org
Griffiths-Jones S, Moxon S, Marshall M, Khanna A, Eddy SR, Bateman A.
2005. Rfam: Annotating non-coding RNAs in complete genomes.
Nucleic Acids Res 33: D121–D124.
Hanada K, Zhang X, Borevitz JO, Li WH, Shiu SH. 2007. A large number of
novel coding small open reading frames in the intergenic regions of the
Arabidopsis thaliana genome are transcribed and/or under purifying
selection. Genome Res 17: 632–640.
Hanada K, Akiyama K, Sakurai T, ToyodaT, Shinozaki K, Shiu SH. 2010. sORF
finder: A program package to identify small open reading frames with
high coding potential. Bioinformatics 26: 399–400.
Huang X, Madan A. 1999. CAP3: A DNA sequence assembly program.
Genome Res 9: 868–877.
Jung HW, Tschaplinski TJ, Wang L, Glazebrook J, Greenberg JT. 2009.
Priming in systemic plant immunity. Science 324: 89–91.
Kalluri UC, Hurst GB, Lankford PK, Ranjan P, Pelletier DA. 2009. Shotgun
proteome profile of Populus developing xylem. Proteomics 9: 4871–4880.
Kastenmayer JP, Ni L, Chu A, Kitchen LE, Au WC, Yang H, Carter CD,
Wheeler D, Davis RW, Boeke JD, et al. 2006. Functional genomics of
genes with small open reading frames (sORFs) in S. cerevisiae.Genome Res
16: 365–373.
Katoh K, Misawa K, Kuma K, Miyata T. 2002. MAFFT: a novel method for
rapid multiple sequence alignment based on fast Fourier transform.
Nucleic Acids Res 30: 3059–3066.
Katoh K, Kuma K, Miyata T, Toh H. 2005. Improvement in the accuracy
of multiple sequence alignment program MAFFT. Genome Inform 16:
22–33.
Kent WJ. 2002. BLAT–the BLAST-like alignment tool. Genome Res 12:
656–664.
Kong L, Zhang Y, Ye ZQ , Liu XQ, Zhao SQ, Wei L, Gao G. 2007. CPC: Assess
the protein-coding potential of transcripts using sequence features and
support vector machine. Nucleic Acids Res 35(Web Server issue): W345–
W349.
Kriventseva EV, Rahman N, Espinosa O, Zdobnov EM. 2008. OrthoDB:
The hierarchical catalog of eukaryotic orthologs. Nucleic Acids Res 36:
D271–D275.
Kurata T, Ishida T, Kawabata-Awai C, Noguchi M, Hattori S, Sano R,
Nagasaka R, Tominaga R, Koshino-Kimura Y, Kato T, et al. 2005. Cell-to-
cell movement of the CAPRICE protein in Arabidopsis root epidermal cell
differentiation. Development 132: 5387–5398.
Lease KA, Walker JC. 2006. The Arabidopsis unannotated secreted peptide
database, a resource for plant peptidomics. Plant Physiol 142: 831–838.
Lee J, Feng J, Campbell KB, Scheffler BE, Garrett WM, Thibivilliers S, Stacey
G, Naiman DQ , Tucker ML, Pastor-Corrales MA, et al. 2009.
Quantitative proteomic analysis of bean plants infected by a virulent
and avirulent obligate rust fungus. Mol Cell Proteomics 8: 19–31.
Lowry OH, Rosebrough NJ, Farr AL, Randall RJ. 1951. Protein measurement
with the folin phenol reagent. J Biol Chem 193: 265–275.
Masoudi-Nejad A, Goto S, Jauregui R, Ito M, Kawashima S, Moriya Y, Endo
TR, Kanehisa M. 2007. EGENES: Transcriptome-based plant database of
genes with metabolic pathway information and expressed sequence tag
indices in KEGG. Plant Physiol 144: 857–866.
Meng L, Wong JH, Feldman LJ, Lemaux PG, Buchanan BB. 2010. A
membrane-associated thioredoxin required for plant growth moves
from cell to cell, suggestive of a role in intercellular communication. Proc
Natl Acad Sci 107: 3900–3905.
Mulder N, Apweiler R. 2007. InterPro and InterProScan: Tools for protein
sequence classification and comparison. Methods Mol Biol 396: 59–70.
Muller R, Bleckmann A, Simon R. 2008. The receptor kinase CORYNE of
Arabidopsis transmits the stem cell-limiting signal CLAVATA3
independently of CLAVATA1. Plant Cell 20: 934–946.
Nawrocki EP, Kolbe DL, Eddy SR. 2009. Infernal 1.0: Inference of RNA
alignments. Bioinformatics 25: 1335–1337.
Notaguchi M, Abe M, Kimura T, Daimon Y, Kobayashi T, Yamaguchi A,
Tomita Y, Dohi K, Mori M, Araki T. 2008. Long-distance, graft-
transmissible action of Arabidopsis FLOWERING LOCUS T protein to
promote flowering. Plant Cell Physiol 49: 1645–1658.
Oelkers K, Goffard N, Weiller GF, Gresshoff PM, Mathesius U, Frickey T.
2008. Bioinformatic analysis of the CLE signaling peptide family. BMC
Plant Biol 8: 1. doi: 10.1186/1471-2229-8-1.
Ostlund G, Schmitt T, Forslund K, Kostler T, Messina DN, Roopra S, Frings O,
Sonnhammer EL. 2010. InParanoid 7: New algorithms and tools for
eukaryotic orthology analysis. Nucleic Acids Res 38: D196–D203.
Pertea G, Huang X, Liang F, Antonescu V, Sultana R, Karamycheva S, Lee Y,
White J, Cheung F, Parvizi B, et al. 2003. TIGR Gene Indices clustering
tools (TGICL): A software system for fast clustering of large EST datasets.
Bioinformatics 19: 651–652.
Ponty Y, Termier M, Denise A. 2006. GenRGenS: Software for generating
random genomic sequences and structures. Bioinformatics 22: 1534–
1535.
Rice P, Longden I, Bleasby A. 2000. EMBOSS: The European Molecular
Biology Open Software Suite. Trends Genet 16: 276–277.
Trotochaud AE, Jeong S, Clark SE. 2000. CLAVATA3, a multimeric ligand for
the CLAVATA1 receptor-kinase. Science 289: 613–617.
Tuskan GA, DiFazio S, Jansson S, Bohlmann J, Grigoriev I, Hellsten U,
Putnam N, Ralph S, Rombauts S, Salamov A, et al. 2006. The genome
of black cottonwood, Populus trichocarpa (Torr. & Gray). Science 313:
1596–1604.
The UniProt Consortium. 2009. The Universal Protein Resource (UniProt)
2009. Nucleic Acids Res 37: D169–D174.
Wittkop T, Baumbach J, Lobo FP, Rahmann S. 2007. Large-scale clustering
of protein sequences with FORCE - A layout based heuristic for
weighted cluster editing. BMC Bioinformatics 8: 396. doi: 10.1186/1471-
2105-8-396.
Yang X, Dondeti V, Dezube R, Maynard DM, Geer LY, Epstein J, Chen X,
Markey SP, Kowalak JA. 2004. DBParser: Web-based software for shotgun
proteomic data analyses. J Proteome Res 3: 1002–1008.
Zdobnov EM, Apweiler R. 2001. InterProScan – An integration platform for
the signature-recognition methods in InterPro. Bioinformatics 17: 847–
848.
Received April 18, 2010; accepted in revised form December 29, 2010.
Small protein discovery using an integrated method
Genome Research 641
www.genome.org
