Available via license: CC BY-NC 3.0
Content may be subject to copyright.
Resource
Extensive alternative polyadenylation during
zebrafish development
Igor Ulitsky,
1,2,3
Alena Shkumatava,
1,2,3
Calvin H. Jan,
1,2,3,4
Alexander O. Subtelny,
1,2,3
David Koppstein,
1,2,3
George W. Bell,
1
Hazel Sive,
1,3
and David P. Bartel
1,2,3,5
1
Whitehead Institute for Biomedical Research, Cambridge, Massachusetts 02142, USA;
2
Howard Hughes Medical Institute,
3
Department of Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA
The post-transcriptional fate of messenger RNAs (mRNAs) is largely dictated by their 39 untranslated regions (39 UTRs),
which are defined by cleavage and polyadenylation (CPA) of pre-mRNAs. We used poly(A)-position profiling by se-
quencing (3P-seq) to map poly(A) sites at eight developmental stages and tissues in the zebrafish. Analysis of over 60
million 3P-seq reads substantially increased and improved existing 39 UTR annotations, resulting in confidently identified
39 UTRs for >79% of the annotated protein-coding genes in zebrafish. mRNAs from most zebrafish genes undergo
alternative CPA, with those from more than a thousand genes using different dominant 39 UTRs at different stages. These
included one of the poly(A) polymerase genes, for which alternative CPA reinforces its repression in the ovary. 39 UTRs
tend to be shortest in the ovaries and longest in the brain. Isoforms with some of the shortest 39 UTRs are highly expressed
in the ovary, yet absent in the maternally contributed RNAs of the embryo, perhaps because their 39 UTRs are too short
to accommodate a uridine-rich motif required for stability of the maternal mRNA. At 2 h post-fertilization, thousands of
unique poly(A) sites appear at locations lacking a typical polyadenylation signal, which suggests a wave of widespread
cytoplasmic polyadenylation of mRNA degradation intermediates. Our insights into the identities, formation, and
evolution of zebrafish 39 UTRs provide a resource for studying gene regulation during vertebrate development.
[Supplemental material is available for this article.]
Alternative splicing and alternative cleavage and polyadenylation
(APA) act in concert to shape the eukaryotic transcriptome (Zhang
et al. 2005; Wang et al. 2008; Di Giammartino et al. 2011). APA
results from differences in recognition of polyadenylation signals
by the pre-mRNA cleavage and polyadenylation (CPA) machinery,
which leads to differences in 39 UTR identity, which can, in turn,
influence stability, translation and subcellular localization of mRNAs
(de Moor et al. 2005; Hughes 2006; Andreassi and Riccio 2009).
Widespread APA and the prevalence of either short or long 39 UTRs
in specific tissues or developmental stages has been reported in
human and mouse (Zhang et al. 2005; Evsikov et al. 2006; Liu et al.
2007; Flavell et al. 2008; Sandberg et al. 2008; Wang et al. 2008;
Ji et al. 2009; Mayr and Bartel 2009; Salisbury et al. 2009), but the
extent and the functional consequences of APA in vertebrate de-
velopment remain poorly understood. Furthermore, systematic
changes in usage of alternative 39 UTRs are yet to be studied in
detail in other vertebrates.
Zebrafish are model vertebrates used for studying RNA bi-
ology (Knaut et al. 2002; Giraldez et al. 2006; Choi et al. 2007;
Aanes et al. 2011), but their utility for studying various aspects of
post-transcriptional regulation, such as targeting by microRNAs
(miRNAs), has been hindered by incomplete 39 UTR annotation. In
version 66 of Ensembl zebrafish genes (Feb. 2012), only 64.4% of
the transcript models have any 39 UTR annotation, and only 28.9%
of the genes have more than one annotated 39 end. By comparison,
over96%ofthehumangeneshaveanannotated39 UTR in Ensembl
v66, 64.8% of which have multiple annotated 39 ends. Further-
more, 22.3% of ze brafish transcript models for protein-coding
genes end withou t a stop codon, indicating that both the C ter-
minus of their protein product and the 39 UTR are not annotated.
Maturation of germ cells and early embryonic development
in animals requires extensive post-transcriptional regulation of
mRNAs. Preparation of the maternal mRNA cargo that is deposited
into the zygote involves deadenylation of many maternal mes-
sages, as the cells suspend their metabolic and transcriptional ac-
tivity after entering a cell-cycle arrest (Tadros and Lipshitz 2009).
Shortly after fertilization, maternally deposited mRNAs play key
roles in orchestrating the early developmental sta ges, and the
translation, localization, and stability of these mRNAs are regu-
lated, at least in part, through their 39 UTRs. Eventually, the tran-
scriptome undergoes a maternal-to-zygotic transition (MZT), during
which maternal transcripts are degraded, probably in several waves
(Giraldez et al. 200 6; Schier 2007; Stitzel and Seydoux 2007),
alongside the commencement of transcription from the zygotic
genome. Zebrafish and Xenopus embryos are excellent models for
studying mRNA biology during early embryogenesis, as both pre-
MZT and post-MZT embryos can be easily separated and manipu-
lated. A recent study described significant changes in transcript
levels occurring between the one-cell and 16-cell stages of early
zebrafish development and provided evidence that hundreds of
genes undergo cytoplasmic polyadenylation during this period
(Aanes et al. 2011), a phenomenon resembling that described
previously in Xenopus, mouse, and fly embryos (Vassalli et al. 1989;
Simon et al. 1992; Salles et al. 1994; Richter 1999; Oh et al. 2000).
Differences in transcript isoforms expressed during those stages, and
in particular, differences in 39 UTR isoforms expressed in oocytes,
pre-MZT, and post-MZT embryos remain to be determined.
4
Present address: Department of Cellular and Molecular Pharma-
cology, Howard Hughes Medical Institute, University of California,
San Francisco and California Institute for Quantitative Biosciences,
San Francisco, California 94158, USA.
5
Corresponding author
E-mail dbartel@wi.mit.edu
Article published online before print. Article, supplemental material, and pub-
lication date are at http://www.genome.org/cgi/doi/10.1101/gr.139733.112.
Freely available online through the Genome Research Open Access option.
2054 Genome Research
www.genome.org
22:2054–2066 Ó 2012, Published by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/12; www.genome.org
To address these questions, we used poly(A)-position profiling
by sequencing (3P-seq) (Jan et al. 2011) to map poly(A) sites in five
different stages of zebrafish development and in three adult tis-
sues. These data allowed us to accurately map 39 UTRs for over 79%
of zebrafish protein-coding genes, define the C-terminal sequence
of 640 proteins, and identify widespread alternative polyadenylation
affecting transcripts from 55% of zebrafish genes. Comparison of
the patterns of 39 end usage in different stages revealed several sig-
nificant trends, such as the usage of shorter 39 UTRs in the ovaries,
preferential depletion of short 39 UTRs in the early embryo, and
widespread polyadenylation at noncanonical sites, which appears
to occur post-transcriptionally in the early embryo.
Results
39 UTR reannotation alters most zebrafish gene models
In order to map the poly(A) landscape in the zebrafish genome, we
obtained polyadenylated RNA from four embryonic stages (pre-
MZT embryo [1.5–2 h post-fertilization (hpf)], post-MZT embryo
[4.5–5.5 hpf], 24 hpf, and 72 hpf), mixed gender adults, and three
adult tissues (brain, testes, and ovaries) and subjected them to 3P-
seq (Jan et al. 2011). Each sample yielded 6,281,147–9,813,481
genome-mapped reads, which mapped to 231,580–954,595
unique genomic regions (Supplemental Table S1). We considered
as 3P tags only reads that mapped to no more than four loci in the
genome and possessed at least one 39-terminal adenylate that was
not templated in the genome. Positions supported by at least four
tags, out of which at least two were either distinct (i.e., had a dif-
ferent number of untemplated adenosines) or came from two dif-
ferent libraries, were carried forward as poly(A) sites.
After combining tags from all eight libraries and consolidat-
ing all tags ending within 10 nt of the most frequently implicated
site, we obtained 195,199 unique poly(A) sites (Supplemental Ta-
ble S2). Sequence composition around these sites was highly sim-
ilar to that reported in mammals, flies, and worms (Fig. 1A; Tian
et al. 2005; Retelska et al. 2006; Jan et al. 2011), and 62.3% had the
canonical cleavage and polyadenylation signal (PAS) motif
AAUAAA or one of ten related variants in a region 10 to 30 bases
upstream of the poly(A) site (Fig. 1B). These eleven PAS variants
Figure 1. Reannotation of zebrafish 39 UTRs. (A) Nucleotide sequence composition around all 197,350 3P-seq–identified poly(A) sites. (Black arrow)
Cleavage position. As previously noted (Jan et al. 2011), the sharp adenosine peak at position +1, the depletion of A at position –1, and blurring of
sequence composition at other positions was partly due to cases of cleavage after an A, for which the templated A was assigned to the poly(A) tail, resulting
in a –1-nt offset from the cleavage-site register. (Inset) Sequence composition around poly(A) sites in C. elegans (Jan et al. 2011), redrawn for comparison.
(B) Frequencies of sites containing the canonical PAS motif AA UAAA or one of its ten common variants in the region from –40 to –10 relative to the poly(A)
site. Known distal 39 ends are the distal-most poly(A) sites annotated in Ensembl, and known proximal 39 ends are all other annotated poly(A) sites. Novel
distal 39 ends are poly(A) sites more distal than the distal-most annotated 39 end. All other novel 39 ends were designated as proximal. (C ) Classification of
poly(A) sites as fractions of sites or as fractions of the 3P tags. The poly(A) site classification scheme is described in Supplemental Figure S1. (D) Genes with
alternative 39 UTR isoforms in Ensembl v66 and following 3P-seq-based annotation. (E) Maximal 39 UTR lengths in Ensembl v66 and following the 3P-seq-
based annotations. For the new models, the longest 39 UTR was supported by at least 10% of the 3P tags in at least one sample. (F)39 UTRs annotated for
the magi2 gene. 3P-seq and RNA-seq track s indicate all tags mapping to this locus. No 39 UTR was annotated for this gene in Ensembl v66.
Genome Research 2055
www.genome.org
Alternative polyadenylation in zebrafish
contained nine of the 14 most common variants found in Caeno-
rhabditis elegans (Jan et al. 2011) and all seven of the most common
variants reported in human and fly (Retelska et al. 2006). The re-
gion between the PAS and the cleavage site had two U-rich seg-
ments, and the region downstream from the cleavage site had a
GU-rich segment, followed by a U-rich segment (Fig. 1A; Supple-
mental Fig. S2). The nucleotide composition upstream of the
poly(A) sites resembled that of human and worm poly(A) sites.
Enrichment of GU-rich/U-rich sequences 5–30 bases downstream
from the sites resembled that of human sites and, to a lesser ex-
tent, C. elegans sites (Supplemental Fig. S2). Thus, the sequence
specificity of the CPA machinery appears to be conserved among
fish, mammals, and worms.
When related to the annotated gene models (Fig. 1C; Sup-
plemental Fig. S1; Supplemental Table S2), only 19.3% of t he
poly(A) sites were within 100 bases of an annotated 39 end, but
these accounted for 67.7% of all the 3P tags, indicating that, when
compared to the novel poly(A) sites, the previously annotated
poly(A) sites are for the more highly expressed transcripts or are
sites that are more efficiently processed. Overall, 88% of the tags
could be assigned to a known or novel 39 UTR of an annotated gene
model (Fig. 1C). Another 6.2% of the tags mapped to mitochon-
drial sequence indicating polyadenylated mitochondrial transcripts
and degradation intermediates (Shepard et al. 2011). As expected,
the mitochondrial sites rarely followed PAS motifs (Fig. 1B). Poly(A)
sites rarely occurred in coding sequence, near transcription start
sites, or in repetitive regions; each of these categories accounted for
<0.7% of the 3P tags (Fig. 1C). Less than 5% of the tags mapped
to previously unannotated regions, some of which were used to
identify long intervening noncoding RNAs in zebrafish (Ulitsky
et al. 2011).
The 3P tags from all the libraries except for the pre-MZT em-
bryo (which was atypical, as described below) were used for 39 UTR
reannotation (Supplemental Table S3). Although our sequential
annotation procedure did not allow assignment of sites to more
than on e category, sites were often assigned to mor e than one
transcript of the same gene because alternative start sites or alter-
native splicing generated different transcripts with the same 39-
terminal sequences. Therefore, to prevent double counting of sites
corresponding to multiple transcripts from the same gene, our
results are typically described with respect to unique gene models
(abbreviated as ‘‘genes’’) rather than to transcripts, even though we
recognize that CPA occurs to RNA transcripts, not genes. At least
one 39 UTR was assigned to 79% of the protein-coding genes, and
for 63% of the genes, at least one novel 39 UTR was identified. After
combining 39 ends appearing within 50 nt from each other, 69.7%
of the genes with assigned 39 UTR had at least two alternative
39 UTR ends, thereby increasing the incidence of APA by threefold
over Ensembl annotations. On average, these genes with alterna-
tive 39 UTRs had 2.8 distinct 39 ends (Fig. 1D).
3P-seq-based annotation substantially extended (>100 nt) the
length of the longest 39 UTR of 11,442 genes when using a cutoff in
which the longer UTR must account for at least 10% of the 3P tags
in at least one library (Fig. 1E,F; Supplemental Table S4). As a result,
the number of genes with 39 UTRs at least 2 kb long increased from
2104 to 6810. The reannotated 39 portions of the genes also allowed
us to extend the polypeptide sequences of 640 proteins by at least
three amino acids, adding an average of 55 amino acids to each
(Supplemental Table S5). In 797 cases, two poly(A) sites appeared
within 50 nt from each other on opposite strands, and in 336 of
these, the distance was 10 nt or less, suggesting that, similar to the
situation in C. elegans (Jan et al. 2011), palindromic arrangements
of bidirectional elements can lead to close cleavage positions of
convergent zebrafish tr anscripts, leaving little or no intergenic
space. Only 192 cases of convergent transcripts ending within 10 bp
of each other were annotated in Ensembl v66.
Our expanded and revised set of 39 UTR annotations provides
a rich resource for studying APA, miRNA function, and other types
of post-transcriptional regulation in zebrafish. miRNA target pre-
dictions based on our revised 39 UTR collection are presented
in TargetScanFish, beginning with TargetScan release 6.2 (http://
targetscan.org). Zebrafish 39 UTRs differed from human 39 UTRs in
several aspects that affect miRNA regulation. First, they were much
more likely to overlap repetitive elements (a s identified using
RepeatMasker), with 44% of the zebrafish 39 UTR sequence being
repetitive compared to just 16% for human 3 9 UTRs. The non-
repetitive fraction of zebrafish 39 UTRs is more A/U-rich than the
nonrepetitive human 39 UTRs (64% vs. 57%). This leads to a higher
density of miRNA target sites for miRNAs with A/U-rich seeds
(Supplemental Fig. S3), although t he overall mean frequency o f
heptamer miRNA target sites in the nonrepetitive fraction of the
39 UTR is similar for both species (0.167 sites per kb per miRNA
family in zebrafish vs. 0.142 in human).
Prediction of functional miRNA target sites in mammals, flies,
and worms has been greatly facilitated by conservation analysis
(Brennecke et al. 2005; Lewis et al. 2005; Lall et al. 2006; Ruby et al.
2007; Friedman et al. 2009; Jan et al. 2011). Unfortunately, con-
servation is currently of limited utility for predicting zebrafish
miRNA targets. Due to the large evolutionary distances between
zebrafish and other species with sequenced genomes, only 26.1%
of zebrafish 39 UTR bases are alignable to at least two species in
the eight-way whole-genome alignment available in the UCSC Ge-
nome Browser, which includes four other fish genomes, frog, hu-
man, and mouse. The aligned bases are preferentially located near
stop codons, suggesting that their alignability is driven primarily
by the alignment of coding sequences, which explains why longer
39 UTRs have significantly less alignable sequence (Pearson corre-
lation r = 0.27 between 39 UTR length and the fraction of the
39 UTR that is alignable). Thus, until more relevant genome se-
quences become available, TargetScanFish will predict zebrafish
miRNA targets based on seed-matched sites in 39 UTRs and will
rank them based on features of site context known to correlate
with target efficacy in mammals, but it will not consider site con-
servation because inclusion of conservation as a criterion for miRNA
target-site prediction would likely introduce bias against genes
with longer 39 UTRs.
Dynamics of poly(A)-site selection during zebrafish
development and organogenesis
The numbers of 3P tags assigned to each transcript was highly
correlated with transcript levels, as estimated using RNA-seq data
from the same or similar tissue or stage (Spearman correlation co-
efficients ranging between 0.77 and 0.52) (Supplemental Fig. S4A;
Supplemental Table S6), which indicated that the fraction of 3P
tags assigned to alternative poly(A) sites for the same gene reflected
the frequency of their utilization. Using this metric to estimate the
utilization of alternative poly(A) sites, the average 39 UTR lengths
varied as much as 1.8-fold between libraries, ranging from 912 in
the pre-MZTembryo to 1624 in the brain (Fig. 2A). Similar trends of
extreme 39 UTR lengths in the brain and gonads and of increasing
39 UTR length during embryonic development have been observed
in mammals using expressed sequence tags (Zhang et al. 2005; Ji
et al. 2009).
Ulitsky et al.
2056 Genome Research
www.genome.org
In human, transcripts with shorter 39 UTRs tend to accumu-
late to higher levels (Chiaromonte et al. 2003). This observation
could be explained in part by the presence of destabilizing ele-
ments, such as miRNA target sites, in the 39 UTRs (Sandberg et al.
2008; Mayr and Bartel 2009). We observed a similar trend in some
of the zebrafish samples. Strong negative correlations were ob-
served in 24 hpf and 72 hpf embryos (0.39 and 0.48, respec-
tively) (Fig. 2B) and weaker ones in ovaries and adults (0.26 and
0.18, respectively). However, there was little correspondence
between 39 UTR length and mRNA accu-
mulation in the brain and the early em-
bryo (0.08 and 0.037) (Supplemental
Fig. S4B; Fig. 2C). In the brain, the lack of
correlation could result from relatively
high expression of genes with long 39 UTRs
(Supplemental Fig. 4B), whereas in the
early embryo, lack of correlation could
stem from instability of isoforms with
short 39 UTRs (see below).
Prevalent use of proximal poly(A) sites
in the ovaries and pre-MZT embryo
Pairwise comparisons of libraries identi-
fied sets of genes with conspicuous changes
in preferred alternative 39 UTR isoforms
between samples. For 3111 genes, the
difference in the relative usage of at least
one poly(A) site differed by at least 30%
between two samples, and for 2619 of
those, the most commo n 3 9 end differed
between two samples (Supp lemental
Table S7). Consistent with the global
trend of shortest 39 UTRs expressed in
theovaryandlongestinthebrain,the
most change in preferred i soforms was
observed between these tissues: Com-
pared to brain, 1089 genes preferentially
used a poly(A) site that was more proxi-
mal to the transcription start site (for
simplicity, referred to as proximal sites)
in the ovary, whereas only 65 showed
the reverse trend and preferentially used
the distal sites in the ovary (Fig. 2D).
These APA events generated much shorter
39 UTRs for the affected genes (Fig. 2E).
Widespread differences also occurred be-
tween the ovary and the testis: Compared
to testis, 811 genes preferentially used the
proximal site in the ovary, whereas only
104 preferentially used the distal site, in-
dicating that shorter 39 UTRs were not
a shared characteristic of gonads. Finally,
comparing pre-MZT and post-MZT, 419
genes preferentially used the proximal
site pre-MZT, whereas only 80 had the
reverse trend. For some genes, such as
rilpl2 (Fig.2F),therelativechangeex-
ceeded fivefold, as also confirmed by qRT-
PCR (Fig. 2G; Supplemental Fig. S5). In
the tested cases, the predominance of
transcripts ending at the proximal site
appeared to persist until the beginning of the MZT (4 hpf), with
a rapid induction of longer isoform during MZT. These results
suggest that a less stringent CPA regime, which cleaves the pre-
mRNAs at the proximal sites, is prevalent in the ovaries and is
replaced during MZT with a more stringent one, which preferen-
tially skips the proximal sites and processes only the distal ones.
Proximal and distal sites were flanked by regions with similar
nucleotide preferences, with distal sites having a slightly higher
enrichment of A ’s in the 10 to 30 region and U’ s downstream
Figure 2. Changes in 39 UTR lengths in different developmental stages. (A) Distribution of 39 UTR
lengths in different stages and tissues. In each sample, for each gene with a single annotated or pre-
dicted stop codon and 3P-seq data, the mean 39 UTR length was computed by averaging the lengths of
all the 39 UTRs, weighted by the number of 3P tags supporting each of them. Box plots show the median
length, flanked by 25th and 75th percentiles. The whiskers are drawn to the fifth and 95th percentile. (B)
Negative correlation between 39 UTR length and transcript levels 24 hpf. For each gene, the mean 39
UTR length was computed as in A, and the RPKM was computed using available RNA-seq data from the
same developmental stage (SRA accession ERP000016), considering only protein-coding regions. (C )
Lack of correlation between 39 UTR length and transcript levels in the pre-MZT embryo. As in B, except
RNA-seq RPKM was computed using available RNA-seq data from the two-cell embryo (SRA accession
ERX008924). (D)39 UTR lengths of genes expressed in the ovary and in the brain. Lengths were com-
puted as in A.(E) Lengths of 39 UTRs resulting from proximal and distal poly(A) sites in analysis of genes
with substantial differences in isoform fractions (>0.3) when comparing ovary and brain samples. (F)
Poly(A) sites of rilpl2. The gene model shown is as annotated in Ensembl v66. 3P-seq tracks show tags
from clusters containing at least 10% of the tags in the indicated samples. (Red and black arrows)
Position of the qRT-PCR primers for the constant and the alternative regions of the transcript, re-
spectively. (G) qRT-PCR analysis of changes in 39 UTR usage during early embryogenesis. RT was per-
formed with random primers and expression levels were computed using probes located in the constant
and alternative regions of the transcript (cUTR and aUTR, respectively) (Supplemental Fig. S4) and
normalized to expression at 6 hpf. (n.d.) aUTR could not be detected at that time point.
Alternative polyadenylation in zebrafish
Genome Research 2057
www.genome.org
(Supplemental Fig. S6A,B). Comparison
of the prevalence of 15 CPA-associated
motifs in the regions surrounding the
proximal and distal sites using the polya_
svm algorithm (Cheng et al. 2006) showed
that 14 of the motifs were associated with
the two site classes with similar frequen-
cies, but that the motif that captured the
canonical PAS was associated with proxi-
mal sites much less frequently than it was
with the distal sites (P = 2.9 3 10
11
)
(Supplemental Fig. S6C). Direct compari-
son of defined hexamer motifs 10 to 30
bases upstre am of the proximal sites
showed a pronounced reduction (>33%)
in the use of AAUAAA but not of its com-
mon variants (P < 1 3 10
15
) (Supple-
mental Fig. S6D).
If the CPA machinery was more ef-
ficient in the ovaries and acted on sites
that were not used in other tissues, then
intronic poly(A) sites, which would lead
to formation of alternative last exons,
would also be preferentially utilized in
the ovary. Indeed, we found that such
sites were preferentially used in both the
ovary and the pre-MZT embryo (Supple-
mental Fig. S5E).
Differential expression and alternative
polyadenylation of cleavage
and polyadenylation factors
in ovaries and pre-MZT embryo
The increased use of weaker poly(A) sites
is typically concomitant with increased
expression of CPA factors (Liu et al. 2007;
Sandberg et al. 2008; Mayr and Bartel
2009). When comparing ovary and brain, most CPA-associated
factors (taken from Liu et al. [2007]) were more highly expressed in
the ovary (Fig. 3A). The difference was most pronounced for cstf1-3
genes, whose mRNAs were 15- to 60-fold higher in the ovary than
in the brain and appeared to be specifically up-regulated in the
ovary and the early embryo (Fig. 3B). Adding another dimension
to the regulation, some CPA factors underwent alternative poly-
adenylation. In two cases, sub1 and clp1, an ovary-specific short
39 UTR isoform accumulated, and in four others, papolb, pcf11,
cstf3, and pabpn1, the APA determined the identity of the last exon
of the transcript, thereby affecting the coding sequence (Fig. 3C;
Supplemental Fig. S7). One of these, papolb, encodes one of the
zebrafish poly(A) polymerases. The zebrafish genome has two
poly(A) polymerase genes, papolb and papolg, which encode
orthologs of mammalian PAP and PAP gamma, respectively. Al-
though the two genes are expressed at similar levels in most tissues
and developmental stages, papolg is up-regulated in the ovary and
the early embryo, whereas papolb is repressed (Fig. 3D). In mouse,
PAP is alternatively spliced and polyadenylated, which generates at
least five isoforms, the shorter of which (PAP III, IV, and V) contain
only about half of the exons of the full-length mRNA and lack
polymerase function (Zhao and Manley 1996). Likewise, in zebrafish,
alternative polyadenylat ion generates three abundant isoforms—a
long isoform with 23 exons, and shorter isoforms containing just 11
or 12 exons (Fig. 3C). These isoforms have different expression
patterns, with ovary and testis preferentially accumulating the
shortest isoform. Thus, in zebrafish ovaries and early embryo, APA-
mediated truncation of the papolb coding sequence acts concor-
dantly with altered mRNA levels to shift productive expression
nearly exclusively to the papolg homolog.
Preferential loss of ovary transcripts with very short 39 UTRs
Paucity of transcription in the early embryo allowed for a direct
comparison of the post-transcriptional fate o f different iso-
forms transcribed in the ovary, many of which are deposited
intotheoocytes.Wefocusedon1351genesthathadasingle
annotated or predicted stop codon (and thus a unique 39 UTR 59
end) a nd two different poly(A) sit es, separated by at least 100 nt,
expressed in the ov ary ($20 3P tags each). For most of these
genes, the 39 UTRs resulting from the use of the proximal CPA
sites were very s hort, often <200 nt (Fi g. 4A). The proximal sites
had a lower propensity for appearing downstream from the
AAUAAA hexamer (31.5%), although 43% of them appeared
downstream from one of its close variants, indicating that their
formation was likely relate d to the global increase in usage of
weaker CPA signals in the ovary. For 576 of the 1351 genes,
preference for t he longer isoform increased more than twofold
Figure 3. Changes in expression and polyadenylation of CPA factors during zebrafish development.
(A) Relative expression of CPA-related genes (listed in Liu et al. 2007) in ovary and brain. Mammalian
homologs of CstF factors are in parentheses. (B) Expression of mRNA for CstF factors in different stages
and tissues. (C ) Differential alternative polyadenylation of papolb. The transcript models shown are as
annotated in Ensembl v66. The height of each plot indicates the number of 3P tags ending at each
position, normalized to the maximum value, which is indicated at the top of each axis. (D) Expression of
mRNA for poly(A) polymerases in different stages and tissues.
Ulitsky et al.
2058 Genome Research
www.genome.org
in the pre-MZT relative to the ovary (e.g., rnase t2)(Fig.4B),
compared to just 51 genes for which preference for the shorter
isoform i ncreased twofold.
A possible explanation for the pref-
erential depletion of the shorter isoforms
in the early embryo is that they lack se-
quence elements required for either long-
term stability of the mRNA in the oocyte
or protection against degradation of ma-
ternal mRNA following fertilization. In-
deed, for genes with two 39 UTR isoforms
in the ovary, the drop in relative abun-
dance in the pre-MZT embryo was in-
versely correlated with the length of the
shorter 39 UTR isoform (Spearman r =
0.44, P < 10
15
)(Fig.4C).Moreover,all
158 distinct mRNAs in the ovar y with
asingle39 UTR that was shorter than
50 nt were down-regulated (11-fold on
average), compared to 81% of the 381
mRNAs with 39 UTRs between 50 and 100
nt, and just 41% of the 1187 mRNAs with
39 UTRs longer than 1000 nt (Fig. 4D).
Overall, for single-isoform genes, there
was a highly significant inverse correla-
tion between the length of the 39 UTR in
the ovary and the underrepresentation of
3P-tags at the transcript in the pre-MZT
embryo (Spearman r = 0.33, P < 10
15
).
We observed a similar correlation when
using RNA-seq data from the ovary and
a two-cell embryo and considering only
reads that mapped to the coding sequence
(r = 0.28). These results suggest that
a39 UTR of 50–100 bases is required for
the stability of a maternal transcript in
oocytes or during the first cell divisions
after fertilization. For example, 13 zona
pellucida genes, which encode compo-
nents of the fish egg membrane (Wang
and Gong 1999), all had 39 UTRs <50 nt
long, were expressed very highly in the
ovary (over 750,000 3P-seq tags mapped
to two clusters of these genes on chro-
mosomes 17 and 20), and were almost
completely absent in the pre-MZT em-
bryo [e.g., the poly(A) site of zp2.6 yielded
114,089 3P tags in the ovary but only 427
in the pre-MZT embryo].
In order to identify regulatory ele-
ments that might contribute to mRNA
stability, we compared 39 UTRs of single-
isoform genes that were decreased at least
twofold (‘unstable’) to those that either
did not change or increased (‘stable’), as
measured by comparing 3P tags in the
ovary and pre-MZT embryos (note that by
this measure either complete degradation
or deadenylation would render transcripts
unstable). The two sets had very similar
patterns of enrichment for the AAUAAA
PAS ;20 nt upstream of the poly(A) site
and similar sequence composition downstream from the poly(A)
site (Fig. 4E). The difference came in the region 30–90 nt upstream
of the poly(A) site, where a marked enrichment of U’s and U-rich
Figure 4. Differential accumulation of transcripts with short 39 UTRs in the ovary and pre-MZT
embryo. (A) Comparison of 39 UTR lengths of the short and the long isoforms in genes with exactly two
isoforms in the ovary. (B) Poly(A) sites of rnaset2 in the indicated samples. The shown 39 UTR structure is
as annotated in Ensembl v66. The height of each plot indicates the number of 3P tags ending at each
position, normalized to the maximum value, which is indicated at the top of each axis. (C ) Relationship
between length of the shorter isoform and relative abundance of the shorter isoform in the pre-MZT
embryo, as inferred from 3P tags for genes with two alternative poly(A) sites in the ovary. (D) Re-
lationship between the length of the 39 UTR in the ovary and the change in mRNA observed in the pre-
MZT embryo relative to that in the ovary, as inferred by 3P tags. Analysis was for genes with a single
39 UTR supported by at least 20 3P tags in the ovary. (E) Frequency of the indicated motifs or nucleotides
flanking poly(A) sites of isoforms that were not reduced in the pre-MZT embryo compared to the ovary
(stable) and those that were reduced at least twofold (unstable). For the hexamer motifs, the frequencies
shown at each position are averages of a window of 11 consecutive nucleotides centered at that po-
sition. (F) The motif identified by Amadeus (Linhart et al. 2008) as significantly enriched in regions
upstream of the stable sites. (G) Destabilization in the pre-MZT embryo of shorter isoforms lacking the
U-rich motif. Genes with two UTR isoforms in the ovary and a U-rich motif 20–90 nt upstream of only one
of the poly(A) sites were stratified based on the location of the motif—upstream of the distal site (red
line) or upstream of the proximal site (blue line). A U-rich motif was defined as present if a decamer with
up to two mismatches from the UK(U)
8
consensus appeared 20–90 nt upstream of the poly(A) site.
Alternative polyadenylation in zebrafish
Genome Research 2059
www.genome.org
hexamers was observed in the stable transcripts compared to the
unstable ones (Fig. 4E). This enrichment was observed even when
considering only 39 UTRs at least 200 bases long (to avoid bias due
to the proximity to coding sequence). Analysis of sequences from
these upstream regions using the de novo motif finder Amadeus
(Linhart et al. 2008), which accounts for differences in G/C con-
tent, identified a single U-ric h motif UKUUUUUUUU (P = 6.3 3
10
19
) (Fig. 4F), which, together with its minor variants, was
present in 28% of the stable and only 8.5% of the unstable mes-
sages. This motif resembled the recently identified motif associ-
ated with cytoplasmic polyadenylati on in zebraf ish (Aanes
et al. 2011), and stretches of 12–27 uridines positioned up-
stream of the PAS have been associated with cytoplasmic poly-
adenylation after fertilization (but not during ooge nesis) in
Xenopus (Si mon et al. 1992). This poly(U) motif is thought to be
distinct from the UUUUUAU CPE element that directs cytoplasmic
polyadenylation during meiotic maturation of the oocyte (Richter
1999). A more detailed analysis using a sliding window of 15 bases
showed that, in the stable transcripts, the maximal U-enriched
window contained nine or 10 uridines (Supplemental Fig. S8A)
and that the second base in the motif was preferentially G, whereas
the other bases were more likely to be A or G in cases in which they
were not U (Supplemental Fig. S8B). We did not observe a signifi-
cant location preference for this motif with respect to the PAS or
the poly(A) site, suggesting that its exact position in the region up
to 70 bases upstream of the PAS does not have as much influence as its
mere presence in the 39 UTR. Among the 1351 genes with two poly(A)
sites in the ovary, the relative change in shorter and longer isoforms
when comparing ovary and the pre-MZT embryo was also dependent
on the motif, and presence of the U-rich motif exclusively near the
distal poly(A) site coincided with less depletion of the longer iso-
form (P = 1.75 3 10
13
) (Fig. 4G). Taken together, our results indicate
that the U-rich motif helps stabilize (or prevents deadenylation of)
maternally loade d transcripts in the pre-MZT embr yo.
A wave of polyadenylation at many noncanonical sites
in the early embryo
When each library was separately compared to others, many
poly(A) sites appeared exclusive to one sample (the adult sample
was excluded from this analysis as it inherently overlapped with
other samples) (Fig. 5A). Testis had the largest number of sample-
specific poly (A) sites (11,894), many of which could not be explained
by the existing gene models. Some testis-specific transcripts are not
yet annotated (testis RNA-seq data are not yet available) and pre-
sumably account for many of these sites. The testis-specific sites
did not differ significantly from other sites in their surrounding
base composition (Fig. 5B), or prevalence of an upstream PAS
(40.4% followed AAUAAA, and 36.9% followed one of its 10
common variants).
Figure 5. Many noncanonical poly(A) sites in the pre-MZT embryo. (A) Abundance of sample-specific poly(A) sites. Site classification was as described in
Supplemental Figure S1. Downstream sites are those appearing up to 8 kb downstream from the annotated 39 ends but without the support for con-
nectivity with the stop codon required for assignment as a novel 39 UTR. (B) Sequence composition near poly(A) sites specific to the testis. (C ) Poly(A) sites
of edc3. The 39 UTRs shown are as annotated in Ensembl v66. (D) Density of poly(A) sites occurring within 39 UTRs from the indicated samples. Poly(A)-site
density was defined as the ratio between the number of poly(A) sites and the length of the longest 39 UTR. (E) Densities of poly(A) sites in the coding
sequence and 39 UTRs plotted for genes expressed in the pre-MZT embryo. (F) Sequence composition near poly(A) sites specific to the pre-MZT embryo.
Ulitsky et al.
2060 Genome Research
www.genome.org
Another 3913 library-specific events appeared in the pre-MZT
embryo, and in contrast to the testis sites, these were pre-
dominantly novel sites that could be assigned to existing gene
models (Fig. 5A). The poly(A) site density [defined as the number of
distinct poly(A) sites as a function of the transcript length] was the
highest in this sample (Fig. 5C,D). Although these pre-MZT-spe-
cific poly(A) sites were significantly more dense in 39 UTRs than in
the coding sequence (Fig. 5E), they differed dramatically from
other sites with respect to surrounding base composition (Fig. 5F)
and rarely followed a canonical PAS or its common variants (3.4%
and 13.8%, respectively). These observations indicated that many
of the pre-MZT-specific poly(A) sites were not generated by the
canonical CPA mechanism. Furthermore, these sites were specific
to a stage in which transcription is not thought to occur, which
suggests that they were formed post-transcriptionally on messages
transcribed previously in the oocytes. Thus, we propose that these
sites were formed by cytoplasmic polyadenylation of mRNAs un-
dergoing 39 exonucleolytic degradation. Consistent with the idea
that these transcripts underwent degradation in the early embryo,
the fraction of poly(A) sites that were pre-MZT-specific was highly
correlated with decreased abundance of the transcript in the post-
MZT compared to the pre-MZT embryo (Spearman r = 0.28, P <
10
15
, expression changes based on RNA-seq data taken from
Aanes et al. 2011).
If cytoplasmic polyadenylation generates these isoforms, it
would differ from that of previous reports in that previously de-
scribed cytoplasmic polyadenylation acts by elongating existing
short poly(A) tails (Richter 1999). To further characterize the CPA
landscape in the pre-MZT embryo, we performed additional se-
quencing and obtained a total of 39,051,782 3P tags from the pre-
MZT sample, which mapped to 2,921,536 unique genomic posi-
tions. Focusing on 6838 genes that had at least 50 tags in the pre-
MZT embryo and at least 20 tags in the ovary and reducing our
stringency to allow a single tag to define a site, we defined pre-
MZT-specific poly(A) sites as those that were at least 10 nt away
from a position present in any library except that of the ovary and
at least 10 nt away from positions covered by at least 10 tags in the
ovary. Overall, 641,377 tags mapped to 254,155 pre-MZT-specific
poly(A) sites, an average of 2.3 tags per site, compared to the global
average of 59.9 in the pre-MZT embryo and 46.6 for all the posi-
tions in the ovary. Of these poly(A) sites, 38.9% were defined by
a single tag. Thus, the pre-MZT-specific poly(A) sites are typically
each used rarely but collectively corresponded to a nontrivial
fraction of the polyadenylated messages at the pre-MZT embryo.
For 182 genes, most of the 3P tags came from pre-MZT-specific loci
(85 loci on average), and for 1942 genes (almost a third of the genes
passing our expression cutoffs for analysis), $10% of the 3P tags in
the pre-MZT embryo were from loci exclusive to that time point.
These exclusive positions were 15-fold more likely to appear in the
39 UTR compared to the coding sequence.
In bacteria, yeast, plants, mammals, and the mitochondria,
addition of a short oligo(A) sequence marks an RNA for degrada-
tion (Anderson 2005; Slomovic et al. 2006; West et al. 2006; Lange
et al. 2009; Shcherbik et al. 2010; Chang and Tong 2011). To test
whether the noncanonical poly(A) sites in the pre-MZT embryo
have short tails resembling those of marked RNAs or longer poly(A)
tails resembling those of typical mRNAs, we developed 3P-PEseq,
a variation on 3P-seq that uses paired-end Illumina sequencing to
estimate the lengths of poly(A) tails on a global scale (Fig. 6A). 3P-
PEseq readouts consist of two reads. The first (read #1) starts at the
39 end of the RNA (in antisense orientiation), which is typically
the 39 end of the poly(A) tail. The other (read #2) starts within the
mRNA and often contains the 39 portion of the 39 UTR, followed by
untemplated adenylates. 3P-PEseq readouts are informative when
read #1 begins with T’ s, which provides information on the number
of terminal adenylates in the amplicon [i.e., an estimate of the length
of the poly(A) tail], and read #2 contains a sequence that can be
mapped to the genome, followed by untemplated A’s, which iden-
tifies the mRNA and its poly(A) site. As implemented, 3P-PEseq
identified the precise lengths of tails up to 70 bp long and a lower
bound on the lengths of longer tails. This dynamic range was more
than adequate to distinguish between short tails thought to mark
RNAs for degradation and the longer tails typically found on mRNAs.
We applied 3P-PEseq to total RNA from pre-MZT (1.5–2 hpf)
and post-MZT (6 hpf) embryos and obtained 3,874,353 3P-PEseq
readouts from 452,817 sites and 4,803,851 readouts from 365,752
sites, respectively. The fraction of all poly(A) sites that mapped to
noncanonical positions within 39 UTRs pre-MZT was greater than
fivefold higher than that fraction in post-MZT embryos, which
further indicated the prevalence of noncanonical poly(A) sites pre-
MZT (Fig. 6B). Indeed, the 82,280 3P-PEseq pre-MZT sites that
mapped within 39 UTRs but not near poly(A) sites observed in
other stages generally appeared in noncanonical sequence con-
texts with no more than chance association with a PAS (2.9% and
14.6% downstream from the AAUAAA PAS or one of its derivatives,
respectively), whereas the 18,598 post-MZT sites were more fre-
quently in canonical contexts (31% downstream from the AAUAAA
PAS or one of its derivatives).
The distribution of poly(A) tail lengths at canonical and
noncanonical sites appeared to be practically indistinguishable
(Fig. 6C), as 61% and 64% of the canonical and noncanonical
poly(A) si tes, respectively, had poly(A ) tails of at least 6 0 adeny-
lates pre-MZT. This suggests that the mechanism that gen erates
the noncanonical sites ultimate ly endo ws them with long
poly(A) tails and is not simply th e mechanism that marks RNA for
degradation.
To further characterize the dynamics of noncanonical sites,
we acquired 8,736,478 3P tags from the one-cell embryo and
7,423,152 tags from the pre-MZT embryo at 4 hpf and compared
the relative numbers of 3P tags coming from noncanonical sites at
different stages (Fig. 6D). The fraction of these sites is highest at
1.5–2 hpf and 4 hpf, which suggests that the bulk of noncanonical
poly(A) sites are generated in the pre-MZT embryo rather than
inherited as part of the maternal RNA and that transcripts with
noncanonical 39 ends are degraded before or during the MZT.
A correlation between the evolutionary divergence in 39 UTR
length and changes in gene expression
A whole-genome duplication in the teleost lineage, which oc-
curred soon after divergence from tetrapods, generated numerous
paralogous gene copies in the zebrafish genome, many of which
have diverged in spatial and/or temporal expression during em-
bryogenesis (Talbot et al. 2005; Semon and Wolfe 2007; Kassahn
et al. 2009). Although the length of the coding sequence between
paralogous pairs was highly consistent (Spearman r = 0.91 across
754 pairs) (Fig. 7A), 39 UTR length was much less conserved
(Spearman r = 0.41) (Fig. 7B). Furthermore, the relative change in
39 UTR length was inversely correlated with similarity of gene ex-
pression profiles across 16 different developmental stages/tissues
(Spearman r = 0.12, P = 5 3 10
4
), suggesting that sequence di-
vergence in the 39 UTR, even when evaluated very crudely as
change in its length, contributes to divergence of gene expression
patterns following gene duplication. Consistent with the typically
Alternative polyadenylation in zebrafish
Genome Research 2061
www.genome.org
negative effect of sequence elements in the 39 UTR on transcript
levels, the paralog with the longer 39 UTR had globally lower ex-
pression levels (P = 2 3 10
7
, sign test).
Discussion
Comparison of poly(A) sites across diverse developmental stages
and tissues allowed us to increase the estimate of the number of
zebrafish genes that undergo alternative polyadenylation from
26.1% to over 80%, a major fraction of which appears to be regu-
lated. The most pronounced differences in 39 UTR length occurred
in the ovaries and in the brain, which resembled trends observed in
flies and mammals (Zhang et al. 2005; Ji et al. 2009; Hilgers et al.
2011). The efficiency of a poly(A) site is correlated with the
strength of the polyadenylation signal (Zhao et al. 1999). A mech-
anism that takes advantage of this correlation appears to account
for the most widespread alternative polyadenylation trends, as
differences in sequence composition that are analogous to the ones
we identified in zebrafish were previously reported to account for
other significant trends of differences in 39 UTR length (Tian et al.
2005; Zhang et al. 2005; Liu et al. 2007; Sandberg et al. 2008; Ji and
Tian 2009; Ji et al. 2009; Mayr and Bartel 2009; Hilgers et al. 2011).
In these diverse systems, alternate poly(A) sites that are more
proximal to the transcription start site appear to share most of the
sequence characteristics of the distal ones but are significantly less
likely to appear after the AAUAAA motif. Widespread use of alter-
native proximal sites in a specific stage or tissue, such as the ovary,
thus appears to rely in part on increased activity of the CPA ma-
chinery toward weaker poly(A) sites. Specific genes presumably use
additional strategies to either enhance or counter the prevailing
trend, but these strategies were not apparent in our genome-wide
analysis.
Although our analysis provided some hints as to the mech-
anism r esponsible for widespread shortening of 39 UTRs in t he
ovary, the purpose of this shortening is unclear. Perhaps transcripts
with short 39 UTRs are more suitable for long-term storage of
Figure 6. Genome-wide estimation of poly(A)-tail lengths. (A) Outline of 3P-PEseq. See text for description. (B) Poly(A) sites and poly(A)-tail lengths pre-
MZT and post-MZT at the acsl4 39 UTR. The 39 UTR shown is as annotated in Ensembl v66. The height of the 3P-Seq plots shows the number of 3P tags at
each position, normalized to the maximum value, which is indicated at the top of each axis. The height of the 3P-PEseq plots shows the average poly(A)-tail
length measured at each position. The length is zero at positions for which no 3P-PEseq tags were obtained. (Arrow) The poly(A) site corresponding to the
paired reads shown below. (Bold) Untemplated nucleotides. In this example, the cleavage position is within three genomically encoded A’s, and thus, at
least 77 of the 80 T’s of read #1 correspond to untemplated A’s of the the poly(A) tail. The adapter sequence in read #2 is in green italics. (C ) Distribution of
pre-MZT poly(A)-tail lengths at poly(A) sites mapping to the indicated loci and RNA classes. Canonical 39 UTR ends are tallied as 3P-PE tags that map within
20 nt of a 39 UTR end annotated using samples from other stages. Noncanonical 39 UTR ends are tallied as 3P-PE tags that map within a 39 UTR annotated in
other stages but not within 20 nt of a poly(A) site defined at any other stage. rRNA and mitochondrial ends are tallied as tags that map within rRNA repeats
and the mitochondrial chromosome, respectively. (D) Abundance of noncanonical isoforms. Plotted is the ratio of 3P or 3P-PE tags mapping to non-
canonical 39 UTR ends (defined as in C) relative to those mapping to canonical 39 UTR ends.
Figure 7. Analysis of paralogous gene pairs generated in the teleost
whole-genome duplication. (A) Relationship of coding-sequence lengths
for pairs of paralogous genes. Genes of each pair were arbitrarily assigned
to each axis. (B) Relationship of 39 UTR lengths for pairs of paralogous
genes. 39 UTR lengths are the weighted average across all the samples in
which the transcript had at least one 3P tag.
Ulitsky et al.
2062 Genome Research
www.genome.org
maternally deposited transcripts. Alternatively, the cell might use
alternative 39 UTR isoforms for potent regulation of gene expres-
sion during oocyte maturation and pre-MZT development. For
example, robust generation of transcripts with very short 39 UTRs,
such as the zona pellucida genes, which then become very un-
stable once massive degradation of maternal mRNA takes place,
could be a useful strategy for providing the first coordinated wave
of degradation of maternal gene products.
The thousands of poly(A) sites specific to the pre-MZTembryo
represent an unexpected set of transcripts whose 39 ends appear to
be generated post-transcriptionally, as they are rare in the ovary
sample, and have no enrichment of the sequence elements asso-
ciated with other poly(A) sites. These features would be expected if
these sites represent either degradation intermediates of a 39!59
exonuclease or products of endonucleolytic cleavage, followed by
extension of the 39 end by polyadenylation, most likely by
a cytoplasmic polyadenylation activity known to be active at
this developmental stage (Aanes et al. 2011). The prevalence of
noncanonical sites in 39 UTRs compared to coding sequence would
be explained if loss of the stop codon triggered rapid decay and
consequent underrepresentation of mRNA fragments ending in
the coding sequence (Frischmeyer et al. 2002), or if actively
translating ribosomes protected the coding sequence from degra-
dation. As these readenylated messages might be subject to addi-
tional rounds of shortening and readenylation, the levels of ma-
ternal transcripts may be governed by a previously unknown
competition between mRNA decay and polyadenylation. In this
scenario, transcripts with longer 39 UTRs would persist for longer
time periods, which would help explain why we did not observe
a negative correlation between 39 UTR length and expression levels
at the pre-MZT embryo. Also helping to explain this lack of cor-
relation is our finding that 39 UTRs shorter than 100 bases are al-
most exclusively at reduced levels in the pre-MZT embryo relative
to the ovary. Interestingly, a similar trend of preferential degrada-
tion of transcripts with short 39 UTRs has been reported in mouse
during GV-oocyte to two-cell stage embryos (Evsikov et al. 2006),
suggesting that the underlying mechanisms might extend to other
species, including mammals.
Our results also point to the benefits of using a dedicated
method for mapping poly(A) sites as part of standard genome an-
notation. Prior to our work, zebrafish had a rather extensively
characterized transcriptome, with more than 1,400,000 ESTs in
dbEST (Boguski et al. 1993) (more than for the rat, chicken, and
Xenopus genomes), and more than 300 million RNA-seq reads that
were already integrated into the Ensembl transcription annota-
tion. Still, eight 3P-seq samples, each with less than 10 million
genome-mapping reads were sufficient to identify novel 39 UTRs
for >60% of all annotated protein-coding genes, substantially in-
creasing the 39 UTR length for over 5000 of them, as well as
pointing to novel layers of regulation of 39 formation at canonical
and noncanonical sites. The 3P-seq results were also an important
starting point for identifying long noncoding RNAs in zebrafish
(Ulitsky et al. 2011). Thus, application of a dedicated method for
identifying 39 ends should become a routin e feature of f uture
transcriptome annotation projects.
Methods
3P-seq
Zebrafish were maintained and staged using standard procedures
(Kimmel et al. 1995). Ovaries and testes were obtained as described
(Gupta and Mullins 2010). 3P-seq was performed as described ( Jan
et al. 2011). Sequencing was performed using Illumina Genome
Analyzer II, except for one of the two additional pre-MZT emrbryo
libraries, which was sequenced using Illumina Hi-Seq. Read num-
bers and genome mapping statistics are presented (Supplemental
Table S1). 3P-seq data were processed as described ( Jan et al. 2011).
Briefly, reads were mapped to the genome using Bowtie (Langmead
et al. 2009). Those that mapped to no more than four loci in the
genome and possessed at least one 39-terminal adenylate that was
not templated in the genome, were considered 3P tags, unless the 39
end of the mapped region fell within three bases of an annotated
splice site. (Those mapping near splice sites were excluded from
further analysis because the reads that supported them could end
with adenylates templated in downstream exons). Positions sup-
ported by at least four tags, out of which at least two were either
distinct (i.e., had a different number of untemplated adenosines) or
came from two different libraries, were carried forward as poly(A)
sites. To deal with the microheterogeneity of sites, 3P tags were
iteratively joined into poly(A) sites that contained all the 3P tags
implicating CPA within 10 bases of the most frequently supported
site.
Genome annotations and RNA-seq data
Zebrafish genome assembly Zv9 (danRer7) was used throughout
this study. Gene models were obtained from Ensembl 66. Predicted
transcripts, GenBank cDNAs and ESTs, whole genome alignments,
and repetitive elements (excluding simple repeats) were obtained
from the UCSC Genome Browser ( July 2011). RNA-seq reads were
obtained from SRA (accessions ERP000016, ERP000400, and
SRP003165) and processed using TopHat (Trapnell et al. 2009) and
Cufflinks (Trapnell et al. 2010) with default parameters. Protein
homology relationships were taken from Ensembl v66. Genes were
considered as paralogs if they were (1) annotated as Clupeocephala
paralogs in Ensembl, (2) had exactly one ortholog in mouse, and
(3) came from distinct genomic loci.
Poly(A) site classification
Our pipeline for poly(A) site annotation is shown (Supplemental
Fig. S1). After clustering to consolidate tags within 10 nt of the
most frequently supported site, each poly(A) site was tested for fit
to possible categories in the following order:
1. Sites mapping to the mitochondria were annotated as mito-
chondrial poly(A) sites and were not processed further.
2. Sites within 100 bases of the 39 end of a known or predicted
gene model were annotated as known poly(A) sites and were
not processed further.
3. Sites downstream from the 39 end of a known or a predicted
gene, but upstream of the beginning of the coding sequence of
the next nonoverlapping gene model on the same strand were
considered as potential end points of extended 39 UTRs. The
extension was restricted to 8 kb from the annotated 39 end.
These poly(A) sites were further tested for connectivity with the
stop codon (if available) or with the 39 end of the gene model.
Connectivity was tested by screening sets of potential transcript
fragments obtained by combining RNA-seq-based reconstruc-
tions done using Cufflinks or Exonerate (Slater and Birney 2005;
Trapnell et al. 20 10) (the latter were obtained from Ens embl
v66), ESTs and cDNAs from GenBank. Because the RNA-seq-
based reconstr ucti ons were based on data that was not strand-
specific, the terminal exons of these transcript models could be
used to support connectivity on both strands. If connectivity
was supported by a fragment spanning the end of the annotated
Alternative polyadenylation in zebrafish
Genome Research 2063
www.genome.org
39 end and the poly(A) site, the site was annotated as a novel
poly(A) site.
4. Sites overlapping 39 UTR exons were annotated as novel poly(A)
sites, and sites overlapping coding sequence or 59 UTRs were
annotated as coding sequence and 59 UTR sites, respectively.
5. Sites overlapping introns and appearing downstream from the
39 end of the coding sequence were tested for connectivity with
the 39 end of the coding sequence. If connectivity was con-
firmed by a transcript model, the site was annotated as a novel
poly(A) site.
6. Sites overlapping introns and appearing upstream of the an-
notated stop codon were tested for connectivity with the end
of the preceding exon. If connectivity was supported and the
poly(A) site was not annotated as a nonintronic poly(A) site for
any other gene model, the site was annotated as the end point
of a novel terminal exon. Sites annotated in one of the steps 4–6
were not processed further.
7. Sites appearing within 500 bases of the promoter in reverse
orientation with respect to the gene model were annotated as
promoter-associated poly(A) sites and were not processed further.
8. Sites overlapping repetitive elements were annotated as repeat-
associated poly(A) sites.
9. Sites not fitting any of the categories were classified as unknown
poly(A) sites.
Alterative PAS motifs
In order to identify alterative PAS motifs, we focused on regions
between 10 and 30 bases upstream of poly(A) sites. We iteratively
identified the most common hexamer in this region across all the
poly(A) sites, tested its significance, removed from consideration
all the sequences containing this hexamer, and repeated the pro-
cess. The search was terminated when the most common hexamer
appeared in <1% of the remaining sequences. Significance was
tested by comparing the fraction of sequences containing the hex-
amer with that found in 100 randomly shuffled sequences with
the same dinucleotide frequencies. Eleven hexamers (AAUAAA,
AUUAAA, UAUAAA, AGUAAA, UUUAAA, CAUAAA, AAUACA,
AAUGAA, AAUAUA, GAUAAA, UGUAAA) each had P-values <0.05
and were enriched at least twofold compared to random sequences.
Gene models
For annotation of 39 UTRs, known protein-coding gene models
were obtained from Ensembl v66 and were supplemented with
predicted coding genes obtained from alignments of RefSeq gene
models from other organisms. We first obtained clusters of known
genes by dividing the genes into groups of genes that appeared on
the same strand and shared at least one exonic nucleotide. Pre-
dicted genes were added to clusters of known genes if they over-
lapped only one cluster. Predicted gene models that did not overlap
any known gene models were then clustered using the same pro-
cedure. This procedure resulted in 26,348 clusters.
Alternative polyadenylation between tissues
In order to identify genes with conspicuous differences in usage of
alternative poly(A) sites between samples, we first computed for
each poly(A) site the number of 3P tags that mapped within 10 bp
of it in each sample. These were used to compute f
ik
, the fraction of
3P tags in sample i mapping to poly(A) site k. Genes with poly(A)
sites for which |f
ik
f
jk
| > 0.3 in pairwise comparisons of samples
were defined having a difference in usage of poly(A) site k between
samples i and j.
qRT-PCR
Total RNA from embryos was isolated using TRI Reagent (Ambion).
For each sample, 100 ng of total RNA were used in reverse tran-
scription reactions using oligo-dT primers (IDT) and SuperScript III
Reverse T ranscriptase (Invitrogen). For each gene, two gene-specific
primer sets were designed, a ‘‘constitutive’’ set targeting exons shared
between the long and the short isoforms, and the ‘ ‘alternative’’ set
targeting only the 39 UTR fragment that was alternatively used based
on the 3P-seq data. DDCt values were calculated for each gene by
normalizing the alternative set against the constitutive one and
normalizing this ratio to that observed at 6 hpf.
3P-PEseq
To selectively capture polyadenylated ends for paired-end sequenc-
ing, 2.5 mg of total RNA from 2 to 2.2- or 6-hpf zebrafish embryos
were splint ligated to a 39 biotinylated adapter (p-AGATCGGAA
GAGCGTCGTGTAGGGAAAGAGTGTAGACACATAC-biotin, IDT)
in the presence of a bridge oligo (TTCCGATCTTTTTTTTT, IDT)
using T4 Rnl2 (NEB) in an overnight reaction at 18°C. Following
partial digestion with RNase T1 (Ambion), 115- to 750-nt RNAs
were isolated from a denaturing polyacrylamide gel, and ligation
products were captured on streptavidin M-280 Dynabeads (Invi-
trogen). RNAs were 59 phosphorylated on beads using Poly-
nucleotide Kinase (NEB) and subsequently ligated to an adapter
(C3.spacer-GTTCAGAGTTCTAcaguccgacgauc, uppercase, DNA;
lowercase, RNA; IDT) using T4 Rnl1 (NEB) in an overnight reaction
at room temperature. Complementary DNA was synthesized on
beads using SuperScript II (Invitrogen) primed with AATGATA
CGGCGACCACCGAGATCTACACTCTTTCCCTACACG, liberated
from the beads by base hydrolysis, size-selected (155–790 nt) on
a denaturing polyacrylamide gel, and amplified by PCR for 15 cy-
cles. One hundred eighty- to 750-nt products were isolated from
a formamide gel and amplified for four additional cycles using
primers that contain the Illumina paired-end sequencing primer-
binding sites. After a final size selection (220–800 nt) and purifi-
cation on a formamide gel, 80 3 80 paired-end sequencing was
performed on the Illumina Hi-Seq platform.
3P-PEseq data analysis
The 3P-PEseq yielded 219,846,644 and 207,519,359 paired-end
80-nt reads for the pre- and post-MZT samples, respectively. Because
of the longer reads compared to 3P-seq and diminishing sequencing
quality in longer reads, we used slightly more stringent criteria for
defining poly(A) sites. Read #2 began with an adapter of 26 bases.
Reads with more than 10 mismatches to the adapter were dis-
carded, and the first 26 bases were removed before further pro-
cessing. A read pair was considered informative only if read #1
began with T’s and read #2 contained 2–39 terminal A’s. After
leading T’s and terminal A’s were removed from reads #1 and #2,
respectively, they were mapped to the genome using Bowtie,
allowing for up to two mismatches and requiring a unique map-
ping position in the zebrafish genome. When mapping read #2,
7,901,731 and 8,009,618 of the pre-MZT and post-MZT reads, re-
spectively, could be uniquely mapped to the genome. The length
of the tail encoded in read #2 was defined as the maximum number
of trailing A’s, allowing for up to one mismatch. This number was
compared to the number of A’s in the genome at the corresponding
position, accounting for the possibility that an A-rich genomic
segment with a single sequencing error might be misclassified as
part of a poly(A) tail, and only cases with at least two untemplated
A’s were carried forward. The poly(A) site was defined as the last
non-A base in read #2. The length of the poly(A) tail for that
Ulitsky et al.
2064 Genome Research
www.genome.org
amplicon was estimated using the corresponding read #1 and de-
fined as the maximal i for which >90% of the bases in the first
i bases of read #1 were T’s. This criterion allowed for some se-
quencing errors expected when sequencing long homopolymers.
Data access
Sequencing data have been submitted to the NCBI Gene Expres-
sion Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/) under
accession number GSE37453.
Acknowledgments
We thank Olivia Rissland and Noah Spies for comments on the
manuscript, the Whitehead Institute Genome Technology Core
for sequencing, Wendy Johnston for technical support, and the
Wellcome Trust Sanger Institute for the availability of unpublished
RNA-seq data. This work was supported by a grant (GM067031)
from the NIH (D.P.B.), an EMBO long-term fellowship (I.U.), a
Human Frontiers Science Program long-term fellowship (A.S.), and
an NSF predoctoral fellowship (C.H.J.).
References
Aanes H, Winata CL, Lin CH, Chen JP, Srinivasan KG, Lee SG, Lim AY, Hajan
HS, Collas P, Bourque G, et al. 2011. Zebrafish mRNA sequencing
decipher novelties in transcriptome dynamics during maternal to
zygotic transition. Genome Res 21: 1328–1338.
Anderson JT. 2005. RNA turnover: Unexpected conseq uences of being
tailed. Curr Biol 15: R635–R638.
Andreassi C, Riccio A. 2009. To localize or not to localize: mRNA fate is in
39UTR ends. Trends Cell Biol 19: 465–474.
Boguski MS, Lowe TM, Tolstoshev CM. 1993. dbEST–database for ‘‘expressed
sequence tags.’’ Nat Genet 4: 332–333.
Brennecke J, Stark A, Russell RB, Cohen SM. 2005. Principles of microRNA-
target recognition. PLoS Biol 3: e85. doi: 10.1371/journal.pbio.0030085.
Chang JH, Tong L. 2011. Mitochondrial poly(A) polymerase and
polyadenylation. Biochim Biophys Acta 1819: 992–997.
Cheng Y, Miura RM, Tian B. 2006. Prediction of mRNA polyadenylation sites
by support vector machine. Bioinformatics 22: 2320–2325.
Chiaromonte F, Miller W, Bouhassira EE. 2003. Gene length and proximity
to neighbors affect genome-wide expression levels. Genome Res 13:
2602–2608.
Choi WY, Giraldez AJ, Schier AF. 2007. Target protectors reveal dampening
and balancing of Nodal agonist and antagonist by miR-430. Science 318:
271–274.
de Moor CH, Meijer H, Lissenden S. 2005. Mechanisms of translational
control by the 39 UTR in development and differentiation. Semin Cell
Dev Biol 16: 49–58.
Di Giammartino DC, Nishida K, Manley JL. 2011. Mechanisms and
consequences of alternative polyadenylation. Mol Cell 43: 853–866.
Evsikov AV, Graber JH, Brockman JM, Hampl A, Holbrook AE, Singh P, Eppig
JJ, Solter D, Knowles BB. 2006. Cracking the egg: Molecular dynamics
and evolutionary aspects of the transition from the fully grown oocyte
to embryo. Genes Dev 20: 2713–2727.
Flavell SW, Kim TK, Gray JM, Harmin DA, Hemberg M, Hong EJ,
Markenscoff-Papadimitriou E, Bear DM, Greenberg ME. 2008. Genome-
wide analysis of MEF2 transcriptional program reveals synaptic target
genes and neuronal activity-dependent polyadenylation site selection.
Neuron 60: 1022–1038.
Friedman RC, Farh KK, Burge CB, Bartel DP. 2009. Most mammalian mRNAs
are conserved targets of microRNAs. Genome Res 19: 92–105.
Frischmeyer PA, van Hoof A, O’Donnell K, Guerrerio AL, Parker R, Dietz HC.
2002. An mRNA surveillance mechanism that eliminates transcripts
lacking termination codons. Science 295: 2258–2261.
Giraldez AJ, Mishima Y, Rihel J, Grocock RJ, Van Dongen S, Inoue K, Enright
AJ, Schier AF. 2006. Zebrafish MiR-430 promotes deadenylation and
clearance of maternal mRNAs. Science 312: 75–79.
Gupta T, Mullins MC. 2010. Dissection of organs from the adult zebrafish.
J Vis Exp 37. doi: 10.3791/1717.
Hilgers V, Perry MW, Hendrix D, Stark A, Levine M, Haley B. 2011. Neural-
specific elongation of 39 UTRs during Drosophila
development. Proc Natl
Acad Sci 108: 15864–15869.
Hughes TA. 2006. Regulation of gene expression by alternative untranslated
regions. Trends Genet 22: 119–122.
Jan CH, Friedman RC, Ruby JG, Bartel DP. 2011. Formation, regulation and
evolution of Caenorhabditis elegans 39UTRs. Nature 469: 97–101.
Ji Z, Tian B. 2009. Reprogramming of 39 untranslated regions of mRNAs
by alternative polyadenylation in generation of pluripotent stem cells
from different cell types. PLoS ONE 4: e8419. doi: 10.1371/journal.
pone.0008419.
Ji Z, Lee JY, Pan Z, Jiang B, Tian B. 2009. Progressive lengthening of 39
untranslated regions of mRNAs by alternative polyadenylation
during mouse embryonic development. Proc Natl Acad Sci 106: 7028–
7033.
Kassahn KS, Dang VT, Wilkins SJ, Perkins AC, Ragan MA. 2009. Evolution
of gene function and regulatory control after whole-genome
duplication: Comparative analyses in vertebrates. Genome Res 19:
1404–1418.
Kimmel CB, Ballard WW, Kimmel SR, Ullmann B, Schilling TF. 1995.
Stages of embryonic developm ent of the zebrafish. Dev Dyn 203:
253–310.
Knaut H, Steinbeisser H, Schwarz H, Nusslein-Volhard C. 2002. An
evolutionary conserved region in the vasa 39UTR targets RNA
translation to the germ cells in the zebrafish. Curr Biol 12: 454–466.
Lall S, Grun D, Krek A, Chen K, Wang YL, Dewey CN, Sood P, Colombo T,
Bray N, Macmenamin P, et al. 2006. A genome-wi de map of conserved
microRNA targets in C. elegans. Curr Biol 16: 460–471.
Lange H, Sement FM, Canaday J, Gagliardi D. 2009. Polyadenylation-assisted
RNA degradation processes in plants. Trends Plant Sci 14: 497–504.
Langmead B, Trapnell C, Pop M, Salzberg SL. 2009. Ultrafast and memory-
efficient alignment of short DNA sequences to the human genome.
Genome Biol 10: R25. doi: 10.1186/gb-2009-10-3-r25.
Lewis BP, Burge CB, Bartel DP. 2005. Conserved seed pairing, often flanked by
adenosines, indicates that thousands of human genes are microRNA targets.
Cell 120: 15–20.
Linhart C, Halperin Y, Shamir R. 2008. Transcription factor and microRNA
motif discovery: The Amadeus platform and a compendium of
metazoan target sets. Genome Res 18: 1180–1189.
Liu D, Brockman JM, Dass B, Hutchins LN, Singh P, McCarrey JR,
MacDonald CC, Graber JH. 2007. Systematic variation in mRNA
39-processing signals during mouse spermatogenesis. Nucleic Acids Res
35: 234–246.
Mayr C, Bartel DP. 2009. Widespread shortening of 39UTRs by alternative
cleavage and polyadenylation activates oncogenes in cancer cells.
Cell 138: 673–684.
Oh B, Hwang S, McLaughlin J, Solter D, Knowles BB. 2000. Timely
translation during the mouse oocyte-to-embryo transition. Development
127: 3795–3803.
Retelska D, Iseli C, Bucher P, Jongeneel CV, Naef F. 2006. Similarities and
differences of polyadenylation signals in human and fly. BMC Genomics
7: 176. doi: 10.1186/1471-2164-7-176.
Richter JD. 1999. Cytoplasmic polyadenylation in development and
beyond. Microbiol Mol Biol Rev 63: 446–456.
Ruby JG, Stark A, Johnston WK, Kellis M, Bartel DP, Lai EC. 2007. Evolution,
biogenesis, expression, and target predictions of a substantially
expanded set of Drosophila microRNAs. Genome Res 17: 1850–1864.
Salisbury J, Hutchison KW, Wigglesworth K, Eppig JJ, Graber JH. 2009.
Probe-level analysis of expression microarrays characterizes isoform-
specific degradation during mouse oocyte maturation. PLoS ONE 4:
e7479. doi: 10.1371/journal.pone.0007479.
Salles FJ, Lieberfarb ME, Wreden C, Gergen JP, Strickland S. 1994. Coordinate
initiation of Drosophila development by regulated polyadenylation of
maternal messenger RNAs. Science 266: 1996–1999.
Sandberg R, Neilson JR, Sarma A, Sharp PA, Burge CB. 2008. Proliferating
cells express mRNAs with shortened 39 untranslated regions and fewer
microRNA target sites. Science 320: 1643–1647.
Schier AF. 2007. The maternal-zygotic transition: Death and birth of RNAs.
Science 316: 406–407.
Semon M, Wolfe KH. 2007. Rearrangement rate following the whole-
genome duplication in teleosts. Mol Biol Evol 24: 860–867.
Shcherbik N, Wang M, Lapik YR, Srivastava L, Pestov DG. 2010.
Polyadenylation and degradation of incomplete RNA polymerase I
transcripts in mammalian cells. EMBO Rep 11: 106–111.
Shepard PJ, Choi EA, Lu J, Flanagan LA, Hertel KJ, Shi Y. 2011. Complex and
dynamic landscape of RNA polyadenylation revealed by PAS-Seq. RNA
17: 761–772.
Simon R, Tassan JP, Richter JD. 1992. Translational control by poly(A)
elongation during Xenopus development: Differential repression and
enhancement by a novel cytoplasmic polyadenylation element. Genes
Dev 6: 2580–2591.
Slater GS, Birney E. 2005. Automated generation of heuristics for biolog ical
sequence comparison. BMC Bioinformatics 6: 31. doi: 10.1186/1471-
2105-6-31.
Alternative polyadenylation in zebrafish
Genome Research 2065
www.genome.org
Slomovic S, Laufer D, Geiger D, Schuster G. 2006. Polyadenylation of
ribosomal RNA in human cells. Nucleic Acids Res 34: 2966–2975.
Stitzel ML, Seydoux G. 2007. Regulation of the oocyte-to-zygote transition.
Science 316: 407–408.
Tadros W, Lipshitz HD. 2009. The maternal-to-zygotic transition: A play in
two acts. Developmen t 136: 3033–3042.
Talbot WS, Woods IG, Wilson C, Friedlander B, Chang P, Reyes DK, Nix R,
Kelly PD, Chu F, Postlethwait JH. 2005. The zebrafish gene map defines
ancestral vertebrate chromosomes. Genome Res 15: 1307–1314.
Tian B, Hu J, Zhang H, Lutz CS. 2005. A large-scale analysis of mRNA
polyadenylation of human and mouse genes. Nucleic Acids Res 33:
201–212.
Trapnell C, Pachter L, Salzberg SL. 2009. TopHat: Discovering splice
junctions with RNA-Seq. Bioinformatics 25: 1105–1111.
Trapnell C, Williams BA, Pertea G, Mortazavi A, Kwan G, van Baren MJ,
Salzberg SL, Wold BJ, Pachter L. 2010. Transcript assembly and
quantification by RNA-Seq reveals unannotated transcripts and isoform
switching during cell differentiation. Nat Biotechnol 28: 511–515.
Ulitsky I, Shkumatava A, Jan C H, Sive H, Bartel DP. 2011. Conserved
function of lincRNAs in vertebrate embryonic development despite
rapid sequence evolution. Cell 147: 1537–1550.
Vassalli JD, Huarte J, Belin D, Gubler P, Vassalli A, O’Connell ML, Parton LA,
Rickles RJ, Strickland S. 1989. R egulated polyadenylation controls
mRNA translation during meiotic maturation of mouse oocytes. Genes
Dev 3: 2163–2171.
Wang H, Gong Z. 1999. Characterization of two zebrafish cDNA clones
encoding egg envelope proteins ZP2 and ZP3. Biochim Biophys Acta
1446: 156–160.
Wang ET, Sandberg R, Luo S, Khrebtukova I, Zhang L, Mayr C, Kingsmore SF,
Schroth GP, Burge CB. 2008. Alternative isoform regulation in human
tissue transcriptomes. Nature 456: 470–476.
West S, Gromak N, Norbury CJ, Proudfoot NJ. 2006. Adenylation and
exosome-mediated degradation of cotranscriptionally cleaved pre-
messenger RNA in human cells. Mol Cell 21: 437–443.
Zhang H, Lee JY, Tian B. 2005. Biased alternative polyadenylation in human
tissues. Genome Biol 6: R100. doi: 10.1186/gb -2005-6-12-r100.
Zhao W, Manley JL. 1996. Complex alternative RNA processing generates
an unexpected diversity of poly(A) polymerase isoforms. Mol Cell Biol
16: 2378–2386.
Zhao J, Hyman L, Moore C. 1999. Formation of mRNA 39 ends in
eukaryotes: Mechanism, regulation, and interrelationships
with other steps in mRNA synthesis. Microbiol Mol Biol Rev 63: 405–445.
Received February 27, 2012; accepted in revised form June 14, 2012.
Ulitsky et al.
2066 Genome Research
www.genome.org
