A Lentiviral RNAi Library for Human
and Mouse Genes Applied to an
Arrayed Viral High-Content Screen
Jason Moffat,1,2,4,10Dorre A. Grueneberg,1,10Xiaoping Yang,1,10So Young Kim,1,3,7Angela M. Kloepfer,1
Gregory Hinkle,1,3Bruno Piqani,1Thomas M. Eisenhaure,5Biao Luo,1Jennifer K. Grenier,1Anne E. Carpenter,2,4
Shi Yin Foo,6Sheila A. Stewart,8Brent R. Stockwell,9Nir Hacohen,1,5,7,11William C. Hahn,1,3,7,11
Eric S. Lander,1,2,4,7,11David M. Sabatini,1,2,4,11and David E. Root1,11,*
1Broad Institute of MIT and Harvard, Cambridge, MA 02139, USA
2Whitehead Institute for Biomedical Research, Cambridge, MA 02142, USA
3Departmentof Medical Oncology andCenter forCancer SystemsBiology, Dana-Farber CancerInstitute, Boston,MA 02115,USA
4Department of Biology, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
5Center for Immunology and Inflammatory Diseases
6Division of Cardiology
Massachusetts General Hospital, Boston, MA 02114, USA
7Harvard Medical School, Boston, MA 02115, USA
8Department of Cell Biology and Physiology, Washington University School of Medicine, St. Louis, MO 63110, USA
9Department of Biological Sciences, Department of Chemistry, Columbia University, New York, NY 10027, USA
10These authors contributed equally to this work.
11These senior authors contributed equally to this work.
To enable arrayed or pooled loss-of-function
including primary and nondividing cells, we are
developing lentiviral short hairpin RNA (shRNA)
libraries targeting the human and murine ge-
nomes. The libraries currently contain 104,000
vectors, targeting each of 22,000 human and
mouse genes with multiple sequence-verified
constructs. To test the utility of the library for ar-
rayed screens, we developed a screen based on
high-content imaging to identify genes required
for mitotic progression in human cancer cells
and applied it to an arrayed set of 5,000 unique
shRNA-expressing lentiviruses that target 1,028
human genes. The screen identified several
known and ?100 candidate regulators of mitotic
progression and proliferation; the availability of
multiple shRNAs targeting the same gene facili-
tated functional validation of putative hits. This
work provides a widely applicable resource for
loss-of-function screens, as well as a roadmap
for its application to biological discovery.
to an increased need for tools that enable genome-scale
functional studies. Sequencing the Saccharomyces cere-
visiae genome fundamentally altered experimental ap-
proaches and led to the creation and widespread use of
itated studies of gene function (Winzeler et al., 1999). Sim-
ilarly, in model organisms such as Caenorhabditis elegans
terference (RNAi) can be exploited to suppress gene ex-
pression (Fire et al., 1998; Kennerdell and Carthew, 1998)
has led to the rapid identification of the genes underlying
many biological processes through powerful loss-of-func-
tion screens (Bettencourt-Dias et al., 2004; Boutros et al.,
2004; Fraser et al., 2000; Kamath et al., 2003; Kiger et al.,
availability of genome-scale libraries of RNAi reagents has
facilitated comprehensive and, at the same time, increas-
ingly complex loss-of-function screens.
RNAi also suppresses gene expression in mammalian
cells (Elbashir et al., 2001), and chemically synthesized
Indeed, screens in human cells using commercially avail-
able libraries of synthetic siRNAs targeting defined gene
families have identified modulators of TRAIL-induced ap-
optosis (Aza-Blanc et al., 2003) and cell survival (Mackei-
gan et al., 2005) as well as kinases required for clathrin-
and caveolae-mediated endocytosis (Pelkmans et al.,
2005). Unfortunately, many interesting mammalian cell
types are resistant to the transfection methods needed
to introduce synthetic siRNAs into cells.
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1283
An alternative approach is to transduce mammalian
cells with viruses carrying expression cassettes that en-
code short hairpin RNAs (shRNAs) to generate gene-
specific siRNAs within cells; this approach can achieve
stable and highly effective gene suppression in a variety
of mammalian cell types (Abbas-Terki et al., 2002; Brum-
melkamp et al., 2002; Paddison et al., 2002; Stewart et al.,
2003). Using large libraries of shRNA-expressing retroviral
vectors, one group screened pools of retroviruses and
identified components of the p53 pathway (Berns et al.,
2004). Another group screened by transfecting cells with
shRNA-expressing retroviral plasmids and identified
genes involved in proteasome function (Paddison et al.,
2004; Silva et al., 2005). Recently, the same two libraries
were used to identify two novel tumor-suppressor genes
(Kolfschoten et al., 2005; Westbrook et al., 2005). While
these reports establish the precedent that shRNA libraries
can be employed to perform loss-of-function screens in
mammalian cells, it is clear that further exploration of
the performance characteristics and limitations of such
approaches is necessary before such large-scale applica-
tions become routine.
The ideal resource for mammalian genetics would con-
sist ofawidely available shRNAlibrary thatcontains effec-
tive suppressors of all ?20,000 human and mouse genes
types, including nondividing cellsand primary cells in both
‘‘pooled’’ and ‘‘arrayed’’ formats. Arrayed screens, in
which each shRNA is tested in an individual well, allow
the study of biologically subtle and complex pheno-
types—for example, by high-content imaging of cells in
individual wells. This requires the development of proto-
cols for efficient production of a high-titer viral stock for
each shRNA. Such a resource would allow biomedical
researchers to perform comprehensive and reliable loss-
of-function screens to identify all genes that affect a
wide range of cellular processes.
We formed The RNAi Consortium (TRC) with the goals
of generating genome-scale shRNA libraries in viral vec-
tors and developing efficient protocols for arrayed viral
screens. The TRC library is designed to target most hu-
man and mouse genes, with multiple distinct constructs
targeting each gene. The lentiviral vectors in this library,
unlike their oncoretroviral counterparts, can infect nondi-
viding cells, a crucial asset for suppressing gene expres-
sion in tissues or cell lines refractory to transfection (Fed-
TRC lentiviral shRNA library and characterize its proper-
ties. The library (designated TRC1) currently contains con-
structs targeting 22,000 human and mouse genes, with
?5 distinct shRNA constructs per gene. We address sev-
eral significant challenges for efficient RNAi screening,
including the variable effectiveness of different shRNA
constructs, the potential for off-target effects, and the
technical requirements for producing the high-titer viruses
needed for arrayed screens. In addition, we have applied
a subset of the TRC1 library in an arrayed virus-mediated
shRNA screen to identify candidate regulators of mitotic
progression in human colon cancer cells, using high-
content imaging. We characterized the performance of
the library in the context of this screen and identified 100
genes for which at least two independent shRNAs pro-
duce substantial and consistent changes in mitotic index.
These genes represent a collection of candidate regula-
tors of mitosis that merit further cell biological study. The
TRC1 library offers a new resource for somatic-cell genet-
ics, and its application in this study provides insights into
the use of shRNA reagents in loss-of-function screens in
shRNA Library Production
To generate the TRC lentiviral library, we adapted the Len-
tihair vector (Stewart et al., 2003) to create pLKO.1, which
carries the puromycin-resistance gene and drives shRNA
expression from a human U6 promoter (Figure 1A).
Although significant levels of recombination are often
observed in retroviral vectors maintained in bacteria, we
found that pLKO.1 undergoes very low levels of recombi-
nation during the cloning and plasmid-purification manip-
ulations necessary for library construction. Specifically,
diagnostic restriction digests of plasmid DNA from 244 li-
brary clones showed no evidence of recombination even
after 10 rounds of sequential copying and regrowth (see
Figure S1 in the Supplemental Data available with this
article online), confirming that shRNA-containing pLKO.1
vectors remain structurally stable in bacteria.
We created a production pipeline (Figure 1B) to gener-
ate a library of sequence-verified shRNAs in pLKO.1. For
each shRNA, we designed stem sequences matching
a 21-base region of the target transcript with an interven-
ing 6-base ‘‘loop’’ consisting of an XhoI site (Figure 1A).
The 21-mer stem sequences were selected using previ-
ously described criteria for siRNA construction that at-
tempt to maximize knockdown (Khvorova et al., 2003;
Schwarz et al., 2003) and minimize off-target effects, as
well as to ensure that most genes in the library contain
shRNAs that target both the 30untranslated region (UTR)
and coding sequence (CDS) of their transcripts (see
Supplemental Data). Oligonucleotide pairs for 90 hairpin
sequences were annealed separately and ligated into
pLKO.1, and the ligations were transformed into compe-
tent bacteria in a 96-well microtiter plate. The 90 transfor-
mations in each plate were then pooled and plated onto
a large agar plate. A total of 672 colonies were selected
robotically for growth, plasmid purification, and sequenc-
ing. This process yields 94% of the designed clones (Fig-
ure S2); each gene has an average of 4.7 unique shRNA
The TRC1 library currently includes over 100,000 vec-
tors, targeting 12,000 human and 10,000 mouse genes.
We continue to generate ?4,500 additional constructs
per month. Detailed information on genes targeted in the
1284 Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc.
canbe foundat http://www.broad.mit.edu/
High-Throughput Lentivirus Production
To exploit this library, we developed a high-throughput
(HT) method to generate high-titer lentiviruses (Figure 1C).
Specifically, we optimized a semiautomated procedure in
96-well plates in which HEK293T cells were transfected
with library and packaging plasmids in a three-plasmid
lentivirus packaging system (Naldini et al., 1996; Zufferey
et al., 1997). We collected 300 ml of transfected cell super-
natants containing VSV-G pseudotyped lentiviruses over
36–60 hr and aliquoted and stored these lentivirus-
containing supernatants at ?80ºC. As described below, a
typical screen was performed in 384-well plates and used
only ?3 ml of lentiviral supernatant per well. Thus, the pro-
cedure above yields sufficient volumes of lentiviral super-
natants from a single 96-well plate for ?100 screens.
To monitor the infection efficiency of lentiviruses gener-
ated by this HT method, we measured the proportion of
cells that acquire resistance to puromycin treatment fol-
lowing infection. We infected A549 lung cancer cells with
an arrayed set of ?1,500 distinct shRNA-expressing lenti-
viruses. The experiment was performed in duplicate, with
puromycin added to one replicate and the other replicate
left untreated. We calculated the ratio of cell numbers in
paired wells (with and without puromycin treatment) after
4 days. Wells were designated as successfully infected if
this ratio exceeded 0.25. By this criterion, 87% of the
?1,500 lentiviruses yielded successful infections (Fig-
ure S3), consistent with the viral titers measured for a ran-
dom sampling of library lentiviruses of 2 ? 106–2 ? 107
cfu/ml (data not shown). These data indicate that this HT
process generates lentiviral stocks of sufficiently high titer
to infect target cells without the need to normalize titers
among wells or to concentrate the lentiviral stocks.
One attractive feature of this lentiviral library is its ability
to transduce a wide range of cell types, including primary
and nondividing cells. We successfully infected several
primary cell types, including mouse embryonic stem cells,
mouse embryonic fibroblasts, and rat neonatal cardio-
myocytes as well as extremely slow-growing or nondivid-
ing cells, including HCN-1A human cortical neurons, with
pLKO.1-based lentiviruses (Figure S4). We and others
have now successfully used pLKO.1 and its derivatives
to infect many cell types (Table S1), confirming that this li-
brary can be used to study a wide range of mammalian
We next asked whether the viral titers generated by our
HT process are sufficient to suppress gene expression.
Using quantitative RT-PCR (qRT-PCR), we determined
whether shRNAs specific for 12 tyrosine kinases affected
the mRNA levels of these targets in A549 lung cells in-
fected in duplicate experiments. For all 12 genes, at least
one of the shRNA viruses reproducibly decreased target
transcript levels (Figure 2A), and, for 10 out of 12 genes,
the shRNA virus that produced the best knockdown re-
duced mRNA levels greater than 4-fold. Overall, 31% of
the 54 lentiviruses reduced transcript levels by greater
than 4-fold (Figure 2B). We observed similar patterns
and levels of knockdown when infecting HT29 colon car-
cinoma cells with the same lentiviruses (data not shown).
Thus, lentiviruses produced by this HT process reproduc-
ibly and strongly suppress gene expression, suggesting
that virus-containing supernatants can be used directly
for primary arrayed screens.
For some screening applications, the use of viral pools
is advantageous; however, this format requires that infec-
tion of a cell with a single copy of an shRNA-expressing
Figure 1. Vector Features, Library Production, and Appli-
(B) Bacterial glycerol stock production method.
(C) Scheme for library production and use.
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1285
Figure 2. Knockdown Performance of HT-Generated Lentivirus in A549 Cells and Mitotic-Index Screen in HT29 Cells
(A) Knockdown performance of lentiviruses representing 54 shRNAs targeting 12 different tyrosine kinases. Transcript levels for duplicate experi-
ments were measured by qRT-PCR and are reported for each shRNA hairpin relative to average transcript levels for two control infections (i.e., an
shRNA targeting either lamin A/C or scrambled sequence). Knockdown for the first set of infections is shown by dark blue bars and the second
set of infections by light blue bars.
(B) Summary of knockdown levels for the duplicate infections of the 54 shRNA viruses from (A).
(C and D) Cell counts (C) and MI scores (D) following infection of HT29 cells with TRC1 as determined by automated image analysis with, versus with-
out, puromycin selection.
(normal), and green (high) diamonds. MI scores for 700 control shRNAs are displayed in gray in random order to indicate the background range of MI.
High and low MI thresholds for selection of MI hits are marked by the dashed lines.
1286 Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc.
tiviruses expressing shRNAs targeting FASTK or AKT3
(two essential genes) kill HT29 cells even at concentra-
High-Content Screen for Regulators of Mitosis
We next sought to characterize the utility of the shRNA li-
brary in an arrayed screen with high-content imaging. We
chose to focus on the regulation of mitosis in human HT29
colon cancer cells, a cell line that has been widely used for
the study of many normal and neoplastic processes. We
selected a subset of the TRC1 library consisting of 4,903
unique shRNA-expressing lentiviruses targeting 1,028
human genes (Table S2) with a single, distinct shRNA-
expressing lentivirus in each well. The targeted genes in-
cluded 476 protein kinases and 180 phosphatases that
represent 88% and 80%, respectively, of known NCBI
reference sequences assigned to these gene classes
(Manning et al., 2002). The remaining 372 genes included
nonprotein kinases, tumor suppressors, and DNA binding
and modification enzymes.
To detect cells in mitosis, we used automated fluores-
cence microscopy and image analysis to identify the cells
in each well that contain histone H3 phosphorylated on
Substantial evidence indicates that pH3 levels also corre-
late with proliferation rate and that the intracellular pattern
of pH3 staining differentiates between stages of mitosis
(Gasparri et al., 2004; Hendzel et al., 1997). In addition,
we visualized all cells with a DNA binding dye (Hoechst)
to identify nuclei and measure DNA content and an actin
stain (phalloidin) to detect cytoplasmic size and shape.
Wecalculatedthefraction ofcellsinmitosis(mitotic index,
or MI) by dividing the number of pH3-positive cells by total
cell number. As a second independent measure of the ef-
fect of gene suppression on mitosis, we extracted histo-
grams of DNA content from the Hoechst images.
A test of viral doses showed that the addition of 0.5–4.0
ml of lentiviral stocks per well of a 384-well plate yielded
high rates of infection in HT29 cells without reductions in
cell counts from toxicity (Figures S6A and S6B). To screen
for mitotic regulators, we used 3 ml of library lentiviruses to
infect HT29 cells in 384-well plates and cultured duplicate
sets in the presence or absence of puromycin. This dose
corresponded to an average moi of ?5. Four days after in-
fection, cells were fixed; stained for pH3, DNA, and actin;
and imaged using an automated fluorescence micro-
scope. The MI was determined by automated image anal-
ysis. We determined that MI did not depend on viral dose
for a number of control and MI-altering shRNAs (Fig-
ure S6C). The accuracy of the automated analysis was
verified by direct visual inspection of ?9% of the 13,551
composite images produced in the screen.
We successfully screened 4,903 distinct shRNAs.
Based on the same puromycin-selection test used for
the A549 infections, 80% of lentiviruses successfully in-
fected HT29 target cells; the correlation coefficient be-
tween cell numbers in puromycin-treated wells and un-
treated wells was r = 0.79 (Figure 2C). As expected
based on the high rate of infection, mitotic indices ob-
tained with and without puromycin selection were in
good agreement for each lentivirus (Figure 2D), and we
therefore averaged these measurements for subsequent
analyses. The average MI for all infected HT29 cells was
5.1. The data approximately fit a Poisson distribution in
its central regions, but with wider tails representing signif-
icant outliers in cell-cycle distribution (Figure 2E).
Based on visual inspection of 1,185 fluorescent images,
intensities and patterns of pH3 staining that are distinct
from typical wells(MI? 5).Moreover, the MIvalues and vi-
sually observed morphological changes were consistent
across repeat infections.
Analysis of Known Mitotic Regulators
We first examined whether shRNAs targeting genes
known to play important roles in regulating the cell cycle
induced changes in MI. For example, inhibition of CDC2/
CDK1, the canonical cyclin-dependent kinase that con-
trols progression through G2/M (Harborth et al., 2001),
was expected to cause a G2/M arrest with faint punctate
targeting CDC2 (shCDC2-820) induced a uniform faint
punctate pH3 staining pattern characteristic of G2/M
phase arrest (Figure 3A). Image analysis computed an
MI of 9.7, and visual examination of the images revealed
that, in fact, a majority of cells exhibited pH3 staining.
a dramatic G2/M shift (Figure 3A, right). Additional exper-
iments confirmed that shCDC2-820 suppressed the ex-
pression of the Cdc2 protein and, as expected, caused
decreases in cyclin B levels without affecting levels of
Cdk2 or a-tubulin (Figure 3B).
a kinase that directly phosphorylates serine 10 of histone
H3 during mitosis (Keen and Taylor, 2004). Three distinct
shRNAs targeting AURKB (shAURKB-1185, shAURKB-
468, and shAURKB-558) reproducibly induced low MIs
and characteristic multinucleate phenotypes in infected
cells (Figure 3C). Moreover, an obvious shift toward the
G2/M (shAURKB-1185) or polyploid state (shAURKB-
468 and shAURKB-558) was observed in DNA content
histograms extracted from the primary screening images
strongly reduced AurkB expression and pH3 levels with-
out affecting the expression of the closely related aurora
A gene (AURKA) (Figure 3D). We note that the lentiviruses
carrying shAURKB-468 and shAURKB-558 that induced
a more complete knockdown of AurkB also resulted in
more severe polyploidy.
A number of additional genes known to regulate the cell
cycle and mitotic progression showed high (>14) or low
(<0.3) MIs in the screen (Tables S3A and S3B). For exam-
ple, shRNAs targeting the cell-cycle effectors PLK1
(shPLK1-513) and CDK2 (shCDK2-923) caused large
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1287
Figure 3. Identification and Target Verification of Known Regulators of Mitosis
(A) Images of HT29 cells following shRNA-induced knockdown of CDC2 (shCDC2-820) that gave an elevated MI = 9.7 from the primary screen (all
channels for the same field are shown). DNA content histograms are shown to the right for shCDC2-820-induced knockdown of CDC2 (blue line)
and shRNA control (shCntrl) infections (gray line). The percentage of total events is shown on the vertical axis and the integrated nuclear intensity
on the horizontal axis. The control histogram is the average for ten images taken from control infections. The black solid triangle indicates the normal
G1 DNA content peak for HT29 cells.
(B) Immunoblot analysis of Cdc2, tyrosine 15-phosphorylated Cdc2, cyclin B, Cdk2, and a-tubulin protein levels following shRNA knockdown with
either shCntrl (targeting GFP) or shCDC2-820 (targeting CDC2) in HT29 cells. Cdk2 and a-tubulin were included as loading controls.
(C) Knockdown of aurora B in HT29 cells. Images are of aurora B (AURKB) knockdown cells from four distinct shRNAs targeting AURKB (shAURKB-
1288 Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc.
increases in MI (to 30 and 35, respectively), and images
from the primary screen show a concomitant drop in cell
numbers for both (Figure 3E). The shRNA shPLK1-513
caused a dramatic G2/M shift, and shCDK2-923 induced
an increase in S phase and G2/M phase cells (Figure 3E).
We confirmed that shPLK1-513 decreased Plk1 expres-
sion without significantly affecting Cdc2 or cyclin B (Fig-
ure 3F), whereas shCDK2-923 decreased Cdk2 expres-
sion and led to the expected drop in cyclin B and
PLK1 (shPLK1-836) that failed to affect MI did not de-
crease Plk1 protein levels (Figure 3F).
To determine whether these shRNAs induced a similar
phenotype in another cell type, we infected BJ-hTERT
fibroblasts (Hahn et al., 1999) with shPLK1-513- or
shPLK1-513 and shCDK2-923 efficiently knocked down
Plk1 and Cdk2, respectively, in BJ-hTERT cells (Fig-
ure 3G). Suppression of Cdk2 in BJ-hTERT cells led to in-
creased pH3 levels (Figure 3G), as was observed in HT29
cells (Figure 3E). In contrast, knockdown of Plk1 in these
cells did not affect pH3 levels (Figure 3G), corroborating
a report showing that small-molecule inhibition of Plk1 in
et al., 2005).
Finally, two distinct shRNAs targeting CDC25A, a phos-
phatase required for dephosphorylation of tyrosine 15 of
Cdc2 and progression from G2 to M, induced low MI in
the primary screen, and weverified thatthese shRNAs de-
crease CDC25A transcript levels (Figure 3H). The DNA
content histogram for cells expressing shCDC25A-1081
showed a significant G2 shift (Figure 3E, bottom). In addi-
tion, shRNAs targeting other genes with known roles in
cell-cycle regulation, including PLK2, PLK4, CHEK1,
SMAD4, and BUB1, also caused altered MI values (Tables
S3A and S3B). These findings demonstrate that our mi-
totic screening assay responds to suppression of known
cell-cycle regulators and thus is a sensitive tool to identify
additional mitotic regulators.
Novel Regulators of Mitosis
We then examined the results of the screen to identify po-
tential novel mitotic regulators. A gene was defined as
a ‘‘hit’’ if at least two independent shRNAs targeting the
gene showed notably high or low MI values. We required
that two independent shRNAs produce consistent pheno-
types to reduce the chance that genes identified in this
screen were due to off-target effects of shRNAs (Jackson
and Linsley, 2004). We required that both shRNAs exceed
the threshold noted above (MI > 9 or < 1) and that at least
one exceed a more stringent threshold (MI > 14 or < 0.3)
(Tables S3A and S3B). In addition, we measured changes
in the expression levels of three genes known to be in-
duced by interferon (INFB1, OAS1, and OAS2) after infec-
tion of cells with a selection of shRNAs that scored in our
screen. None of these shRNAs induced the interferon
pathway (Figure S7).
The screen yielded 87 genes associated with high MI
(Table 1 and Table S4) and 15 genes associated with
low MI (Table 2). To understand the cell-cycle effects
caused bysuppression of these genes,we reanalyzed pri-
mary screen images to assess changes in DNA content.
Figure 4 and Figure 5A show images for two distinct
responding DNA histograms superimposed in yellow.
These images reveal a rich panoply of morphological fea-
tures that accompany the changes in mitotic index and
cell-cycle distribution. Notably, some sets of genes
show strikingly similar phenotypes, suggesting that they
may function in the same pathway. For example, shRNAs
targeted against PAK7, FGR, and NTRK2 show high MI
and common changes in morphology, including enlarged
cell and nuclear sizes and intense actin staining on the cell
periphery (Figure 4). Nearly all of the shRNAs yielding high
MIalsoproduced substantivechangesinthe DNAcontent
distribution. The shRNAs targeting PDGFRB and U2AF2
resulted in a particularly marked G2/M arrest (Figure 4).
The low-MI hits showed an even greater diversity of cell
morphologies, most accompanied by altered DNA con-
tent profiles. The shRNAs targeted against GSK3b and
SGK3 resulted in cells with extended processes and
DNA content histograms with predominant G1 peaks,
while those targeted against BUB1B and PAK4 caused
greatly enlarged cell and nuclear sizes and a G2/M arrest
Wevisually inspected the patterns of pH3 staining in the
primary screening images to determine if the cell popula-
tions showed overrepresentation of specific phases of mi-
tosis relative to controls. A number of genes did show
red = actin) and lower panels show Hoechst staining. Bottom panels show the corresponding DNA content histograms for each shRNA. The solid
black triangle indicates the normal G1 DNA content peak for HT29 cells.
(D) Immunoblot analysis of AurkB, pH3, a-tubulin, and AurkA levels following shRNA knockdown with shCntrl virus (targeting GFP), shAURKB-1185,
shAURKB-468, shAURKB-227, or shAURKB-558 viruses in HT29 cells. Each infection was done at two viral doses (1 and 4 ml). a-tubulin served as
a loading control.
(shCDC25A-1081) MIs. Corresponding DNA content histograms are shown to the right. The solid black triangles indicate the normal G1 DNA content
peak in HT29 cells.
(F) Immunoblot analysis of Cdk2, Plk1, Cdc2, tyrosine-phosphorylated Cdc2, cyclin B, and a-tubulin levels following shRNA knockdown targeting
CDK2 (shCDK2-923) and PLK1 (shPLK1-513) in HT29 cells. shPLK1-836 served as a negative control for Plk1 knockdown.
(G)Immunoblot analysisof Cdk2,Plk1, pH3,anda-tubulin levelsfollowing shRNAknockdowntargeting CDK2 (shCDK2-923)andPLK1 (shPLK1-513)
in BJ-TERT fibroblasts.
HT29 cells. Error bars indicate the standard error for three qPCR measurements. Scale bars = 50 mm.
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1289
Table 1. Subset of Gene Targets for which Two or More shRNAs Induced an Increase in MI
Gene ID SymbolHairpin NameAverage MI Description
7145TNS shTNS-619743.0 tensin
2268FGR shFGR-38528.1 Gardner-Rasheed feline sarcoma viral (v-fgr)
5159 PDGFRB shPDGFRB-237130.6 platelet-derived growth factor receptor, beta polypeptide
4915 NTRK2shNTRK2-212334.1 neurotrophic tyrosine kinase, receptor, type 2
7525 YES1shYES1-90525.4 v-yes-1 Yamaguchi sarcoma viral oncogene homolog 1
7075 TIE1 shTIE1-3795 24.5tyrosine kinase with immunoglobulin-like and EGF-like
5571PRKAG1 shPRKAG1-15724.0 protein kinase, AMP-activated, gamma 1 noncatalytic
5997RGS2 shRGS2-51025.6 regulator of G protein signaling 2, 24 kDa
11183 MAP4K5 shMAP4K5-282623.6 mitogen-activated protein kinase kinase kinase kinase 5
11338U2AF2 shU2AF2-60631.2 U2 (RNU2) small nuclear RNA auxiliary factor 2
5563 PRKAA2 shPRKAA2-102831.0 protein kinase, AMP-activated, alpha 2 catalytic subunit
5651 PRSS7 shPRSS7-130621.2 protease, serine, 7 (enterokinase)
7010TEK shTEK-520 27.0 TEK tyrosine kinase, endothelial (venous malformations,
multiple cutaneous and mucosal)
55137 FIGN shFIGN-1661 21.3fidgetin
5922RASA2 shRASA2-57223.6 RAS p21 protein activator 2
2869GRK5 shGRK5-52622.5 G protein-coupled receptor kinase 5
9156EXO1 shEXO1-158625.2exonuclease 1
6197 RPS6KA3shRPS6KA3-982 23.9 ribosomal protein S6 kinase, 90 kDa, polypeptide 3
10733 PLK4shPLK4-43318.8polo-like kinase 4 (Drosophila)
1290 Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc.
such a pattern, suggesting that these genes are involved
in progression through the observed stage of mitosis
the investigation of potential mitotic regulators. Each pu-
tative hit requires further study to confirm that the ob-
served phenotype reflects knockdown of the targeted
gene (‘‘target confirmation’’) and to elucidate its biological
role. We suggest the following criteria for target confirma-
tion: (1) reproduction of the phenotype in multiple in-
dependent experiments, (2) verification that the shRNA
decreases the expression level of the target gene, and
(3) demonstration of a correlation between the observed
phenotype and the extent of gene suppression across
multiple shRNAs targeting the same gene.
We selected four genes (YES1, TIE1, ROCK1, and MET)
for which multiple shRNAs produced high MI and that had
not previously been implicated in the regulation of mitosis
forfollow-up experiments. For each shRNA,we confirmed
the initial phenotype and measured target-gene knock-
down. For the shRNAs targeting YES1, TIE1, and ROCK1,
we found a strong correlation between knockdown level
and increased MI as well as increased levels of pH3 (Fig-
ures 6A–6C). The shRNAs that induced greatest suppres-
sion of the target gene yielded the largest MI values, and
shRNAs that produced slight or no increase in MI induced
much weaker suppression of the target transcript. These
results strongly suggest that the observed phenotypic ef-
fects are due to suppression of these target genes.
In contrast, the shRNAs targeting MET did not show
a clear correlation between extent of gene knockdown
and MI phenotype (Figure 6D). While the shRNA that pro-
duced the most elevated MI (shMET-1651) did cause a
substantial knockdown of MET, another shRNA causing
a strong knockdown (shMET-502) failed to increase MI.
Additional work is needed to determine if changes in
We performed further biological characterization of
YES1, TIE1, and ROCK1. First, we found that infection of
immortalized BJ-TERT fibroblasts with shRNAs specific
for YES1, TIE1, and ROCK1 induced effective gene sup-
pression (Figure 6E). In the case of YES1 and TIE1, sup-
pression of these genes in BJ-TERT cells induced pH3
as was observed in HT29 cells. Suppression of ROCK1,
like PLK1, induced pH3 in HT29 cells but not in human fi-
broblasts. Thesefindingsshow thatsomegenesidentified
in this screen can regulate mitosis in both nonmalignant
and malignant cells while others may exhibit specificity
for cancer cells, suggesting possible cancer targets.
Second, we examined DNA content histograms for
HT29 cells expressing the shRNA targeting YES1, TIE1,
A substantial percentage of cells expressing shYES1-
1338, shYES1-905, and shTIE1-3795 were arrested in
G2/M (Figures 6F–6H). Because deregulation of the cell
cycle can lead to cell death (Golsteyn, 2005), we also
checked whether suppression of any of these genes
also induced apoptosis. We found that shRNAs that
strongly suppressed YES1 and TIE1 also increased levels
of the apoptotic marker cleaved PARP (Figures 6I and 6J),
while those that target ROCK1 did not induce apoptosis
(data not shown).
Finally, we examined the list of genes identified in this
screen to determine whether other genes obviously re-
lated to YES1, TIE1, and ROCK1 were present. The TIE1
receptor tyrosine kinase has rolesin angiogenesis and de-
velopment and is believed to function in a complex with
the TEK receptor tyrosine kinase (Marron et al., 2000;
Tsiamis et al., 2002). We found that three of the shRNAs
that target TEK also cause substantial increases in MI (Ta-
ble S3); we tested two of these shRNAs and verified that
they decrease transcript levels of TEK (Figure 6K) but
not of TIE1 (data not shown). Furthermore, cells express-
ing shTEK-1275 and shTEK-520 also showed altered DNA
content distribution, consistent with G2/M arrest (Fig-
ure 6L). These observations strongly suggest that the re-
ceptor complex that includes the products of TIE1 and
TEK plays a previously unknown role in the control of mi-
tosis in cancer cells.
The discovery of RNAi has revolutionized the study of
gene function in model organisms and promises to permit
large-scale loss-of-function studies in mammals. Mam-
malian siRNA and shRNA libraries have now been used
successfully (Berns et al., 2004; Kittler et al., 2004;
Kolfschoten et al., 2005; Paddison et al., 2004; Pelkmans
Table 1. Continued
Gene ID SymbolHairpin Name Average MIDescription
57144 PAK7shPAK7-191819.8 p21(CDKN1A)-activated kinase 7
6725SRMSshSRMS-1235 14.6src-related kinase lacking C-terminal regulatory tyrosine
and N-terminal myristylation sites
At least one shRNA induced an MI > 14, and at least one additional shRNA elicited an MI > 9. See Table S4 for a full list of genes
meeting these criteria.
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1291
et al., 2005; Silva et al., 2005; Westbrook et al., 2005), but
many practical and theoretical challenges remain before
such large-scale applications become routine. To create
a resource that will enable high-throughput screening in
mammalian cells, we formed the RNAi Consortium to
siRNAs to a broad variety of cells at high efficiency. We
focused initial efforts on enabling arrayed screening be-
cause this format offers some important advantages rela-
tive to pooled screens. Specifically, this format provides
increased sensitivity in the initial assay, reduces the num-
ber of false negatives, directly identifies active shRNAs for
follow-up without the need for postscreen deconvolution,
and enables use of complex and information-rich assays
Table 2. Gene Targets for which Two or More shRNAs Induced a Decrease in MI
Gene ID Symbol Hairpin Name Average MIDescription
7535ZAP70 shZAP70-23930.2 zeta-chain (TCR) associated protein kinase, 70 kDa
1608DGKG shDGKG-16850.0 diacylglycerol kinase, gamma 90 kDa
10298PAK4 shPAK4-285 0.0p21(CDKN1A)-activated kinase 4
2932 GSK3B shGSK3B-867 0.1glycogen synthase kinase 3 beta
7525YES1 shYES1-4270.0 v-yes-1 Yamaguchi sarcoma viral oncogene homolog 1
701 BUB1B shBUB1B-18220.0 BUB1 budding uninhibited by benzimidazoles 1 homolog
53904MYO3AshMYO3A-4214 0.1 myosin IIIA
23678SGK3 shSGK3-13860.3 serum/glucocorticoid regulated kinase-like
3656 IRAK2 shIRAK2-15630.0 interleukin-1 receptor-associated kinase 2
2585 GALK2 shGALK2-13300.2 galactokinase 2
51678 MPP6shMPP6-617 0.0 membrane protein, palmitoylated 6 (MAGUK p55 subfamily
5502 PPP1R1A shPPP1R1A-6120.0 protein phosphatase 1, regulatory (inhibitor) subunit 1A
1454 CSNK1E shCSNK1E-7660.2 casein kinase 1 epsilon
1859 DYRK1A shDYRK1A-39470.2dual-specificity tyrosine (Y) phosphorylation-regulated
8916HERC3 shHERC3-1556 0.3hect domain and RLD 3
At least one shRNA induced an MI < 0.3, and at least one additional shRNA elicited an MI < 1.1.
1292 Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc.
such as those involving cell-cell interactions and high-
ful tool to reveal genes that are critical for many biological
Here wedescribe a lentiviral shRNAlibrary and its appli-
cation to an arrayed screen in viral form. The features of
this library and the methods for its application developed
here enable effective arrayed screening in a wide range
of cell types. The TRC1 library currently contains over
100,000 sequence-validated arrayed shRNA constructs
targeting 12,000 human and 10,000 murine genes. We will
continue to generate additional constructs until nearly all
human and mouse genes are targeted. Methods for pro-
ducing DNA and lentiviruses from this library are routine
at a small scale, but many challenges exist in performing
these manipulations at the scale necessary to perform
HT studies. We report HT lentiviral production methods
that constitute a relatively small part of the total cost of
the screen when the reagents are distributed across
many screens. This library can thus serve as a cost-effec-
tive, renewable, and scaleable RNAi-screening resource
for the scientific community.
Quantitative assessment of library performance, mea-
sured on a sample of untitered library viruses, showed
that 83% of genes tested had at least one shRNA virus
that reduced transcript levels R4-fold. These results are
likely to underestimate the intrinsic shRNA knockdown ef-
ficacy due to variations in viral titer. We will continue to
measure library knockdown performance to rank con-
structs by level of knockdown efficacy. This information
will be useful for determining the effects of gene dosage
on phenotype for essential as well as nonessential genes.
shRNAs to contain at least three mismatches to all known
cDNAs in the human or mouse genome. However, this
does not eliminate the possibility of off-target effects with
shorter stretches of identity (Zamore and Haley, 2005).
To overcome this inherent property of shRNAs, we re-
shRNAs that induce a similar phenotype. Because distinct
shRNAs are expected to have nonoverlapping spectra of
off-target effects, this criterion should filter out most off-
target effects. We also investigated nonspecific effects of
viral infection using a small set of library shRNA vectors
and found no evidence for interferon induction.
The two issues above, differential effectiveness of
shRNAs and the possibility of off-target effects, under-
score the importance of using multiple shRNAs to mini-
mize false negatives and false positives in screens. For
this reason, the TRC1 library was designed to include
ing fewer shRNAs per gene by randomly removing one
construct from our data set. Using only four shRNAs, the
number of hits that would be detected in our screen would
fall from 102 to 75. Indeed, it would be desirable to use
even more thanfive shRNAs pergene—especially in order
to obtain ‘‘allelic’’ series with varying effects and to enable
testing of essential genes by inducing moderate levels of
We tested the utility of the TRC1 library in loss-of-func-
tion screening by infecting colon cancer cells with arrayed
A screen surveying kinases in Drosophila S2 cells identi-
fied 80 genes that cause cell-cycle dysfunction upon
downregulation (Bettencourt-Dias et al., 2004). Our
screen tested human homologs for 59 of these 80 Dro-
sophila genes, of which 21 were found to have altered
mitotic phenotypes in our screen, suggesting that the
function of many of these genes are evolutionarily con-
served. We found that three of the genes identified as mi-
totic regulators in our screen of HT29 cells also regulate
mitotic progression in human fibroblasts. Other genes
identified as mitotic regulators in HT29 cells did not have
similar effects in fibroblasts, suggesting that targeting
these genes may confer specificity for cancer cells. In-
deed, suppression of some of the genes identified in this
screen also leads to cell death, suggesting that they are
potential therapeutic targets. Further experiments are
necessary to determine the roles of each of these genes
in regulating mitotic progression. Although wehave begun
to investigate the role of some of these genes in other cell
lines, it is clear that a definitive investigation of genes that
regulate mitotic progression in normal and many types of
cancer cells will require performing this screen in dozens
of cell types. We believe that the library and methodolo-
gies described herein provide the means to undertake
such a study.
In summary, we have produced a genome-scale lentivi-
ral shRNA library to target human and mouse genes in
a wide range of cell types, developed a pipeline to effec-
forms in an automated fashion, and used a subset of the
mat for phenotypic screens. Future advances in RNAi bi-
ology are expected to improve our ability to design and
use RNAi libraries for genetic screening in mammals. In
addition, methodologies to use RNAi in animals to study
gene function are being developed by several groups
and promise to provide a critical tool for the follow-up of
genes identified in a cell-based RNAi screen (Dickins
et al., 2005; Sandy et al., 2005). The use of genome-
wide RNAi libraries for gene discovery should facilitate
rapid identification of the major regulators of many biolog-
ical processes, thereby annotating the genome and re-
vealing the first global views of mammalian genetic cir-
cuits. The lentiviral library described here will facilitate
comprehensive screening efforts and will be especially
useful in enabling arrayed screens that focus on primary
cells from mouse or human.
Details of the library-production methods are provided in the Supple-
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1293
Figure 4. Images of HT29 Cells Infected with shRNAs for 21 Genes that Induce High Mitotic Indices
Pairs of images are shown for knockdowns by two distinct shRNA viruses for each hit gene. Scale bars = 10 mm. Corresponding DNA content histo-
grams are superimposed as yellow traces in the left corner of each image. The percentage of total events is shown on the vertical axis and the in-
tegrated nuclear intensity on the horizontal axis. The control histogram is the average of ten images taken from control infections. The small yellow
triangles underneath each histogram indicate the G1 peak in HT29 cells.
1294 Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc.
HT DNA and Virus Production
Transfection-quality DNA was prepped using 96-well PureLink kits (In-
vitrogen) with average yields of 4 mg DNA/well,quantified using a Pico-
Green assay (Molecular Probes), and normalized robotically in each
plate. Lentiviruses were made in 96-well format by transfecting pack-
aging cells (293T) with a three-plasmid system (Naldini et al., 1996;
Figure 5. Images of HT29 Cells Infected with shRNAs for 13 Genes that Induce Low Mitotic Indices, and Association of High-
Mitotic-Index-Inducing Genes with Mitotic Phase
(A) Pairs of images are displayed for knockdown by two distinct shRNAs for each low-MI-inducing gene. DNA histograms are displayed for each im-
age as described for Figure 4.
(B) Genes identified by high MI for which a specific phase of mitosis is overrepresented were visually scored and are indicated below the correspond-
ing phases (blue = nuclei, green = pH3, red = actin). Asterisks indicate cases where condensed staining may also be representative of apoptosis.
Scale bars = 10 mm.
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1295
Figure 6. Correlation between Phenotype and Knockdown for Multiple Library shRNAs
(A–D) Immunoblot analyses of pH3, a-tubulin, and target protein levels following shRNA knockdown in HT29 cells targeting (A) YES1 (shYES1-3252,
shYES1-161, shYES1-1338, shYES1-287, shYES1-905), (B) TIE1 (shTIE1-3088, shTIE1-3795, shTIE1-3316, shTIE1-3206, shTIE1-1605), (C) ROCK1
(shROCK1-3241, shROCK1-1885, shROCK1-3377, shROCK1-1069), or (D) MET (shMET-4490, shMET-502, shMET-1651, shMET-1374, shMET-
345). Control infections using a hairpin sequence targeting GFP knockdown are shown on the left of each blot (shCntrl). MIs from the primary screen
data are indicated below each lane.
(E) Immunoblot analysis for indicated proteins and phosphorylation sites of BJ-TERT fibroblasts infected with shRNA viruses targeting YES1, TIE1,
(F–H) DNA content histograms from primary screening data in HT29 cells for knockdowns of (F) YES1 (shYES1-1338, shYES-905), (G) TIE1 (shTIE1-
3795, shTIE1-3206), and (H) ROCK1 (shROCK1-3241, shROCK1-1885). The black triangles indicate the G1 peak, and DNA histograms from control
infections are shown on the left of each panel for comparison.
(IandJ)Immunoblotanalyses ofpH3,PARP(full length,FL,or cleaved, CL,indicating apoptosis), anda-tubulinproteinlevelsfollowing shRNAknock-
down targeting YES1 (shYES1-1338, shYES1-905), TIE1 (shTIE1-3795, shTIE1-3088, shTIE1-3316), or a control shRNA targeting GFP (shCntrl) as
(K) Quantitative RT-PCR analysis of TEK transcript levels following lentiviral mediated RNAi with two different shRNAs that induced high MIs (shTEK-
1275, shTEK-520). Error bars indicate the standard error for three qPCR measurements.
(L) DNA content histograms following knockdown with a control shRNA (shCntrl), shTEK-1275, and shTEK-520, from primary screen data. The black
triangles indicate the G1 peak.
1296 Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc.
Zufferey et al., 1997; see also Supplemental Data and http://www.
HT Lentiviral Infections and Mitotic-Index Assay
Infection conditions were optimized in 384-well plates for growth con-
ditions, plate types, viral dose, and assay times prior to HT screening.
HT29 cells were seeded at adensity of 300–350 cells/well in a 384-well
concentrated shRNA lentiviral supernatant from the 96-well viral pro-
duction, and incubated for 4 days. All lentiviral infections were tested
in duplicate, one replicate using 2 mg/ml puromycin during the final 3
days of incubation and the other replicate with no selection. Cells
were ?50%–70% confluent at the time of fixation and fluorescent
staining for HT image acquisition. Images were analyzed using Cello-
mics software to extract MI. Data for each lentiviral sample were re-
the ratio of cell counts under +/? puromycin conditions exceeded
0.25, and the cell count was > 100 for the imaged area. MIs for +
and –puromycin conditions were averaged. DNA content histograms
were extracted from the same primary screening images using Cell-
Profiler Software (http://jura.wi.mit.edu/cellprofiler/). For follow-up
experiments, infections of HT29 and BJ-TERT cells were performed
using a similar protocol as for the primary screen, scaled up to 6 cm
dishes. Standard immunoblot analyses were performed for the hit pro-
teins and for pH3. Details of infection and assay conditions and data
analysis are provided in Supplemental Data.
mRNA was harvested in 96-well plates using GenePlate Hybridization
(RNAture). RT reactions were performed with a SuperScript II RT Kit
(Invitrogen). Quantitative PCR reactions were performed using As-
PCR Master Mix(Applied Biosystems). Quantification of GAPDH levels
in the same cDNA samples measured in separate qPCR reactions
served as an endogenous control. All qPCR reactions were run in trip-
licate, and the average Ct(cycles to threshold) was used for the com-
parative Ctmethod (ABI User Bulletin #2). Control infections using an
shRNA targeting lamin or an shRNA not targeting any human gene
were used to define 100% expression.
The RNAiConsortium(TRC)humanand mouselentiviralshRNAlibraries
are available from Sigma-Aldrich Company (http://www.sigmaaldrich.
com) and Open Biosystems (http://www.openbiosystems.com). Up-
dated contents of the library can be found at http://www.broad.mit.
Supplemental Data include Supplemental Experimental Procedures,
Supplemental References, four tables, and seven figures and can be
found with this article online at http://www.cell.com/cgi/content/full/
This work is a project of the RNAi Consortium (TRC). The TRC was ini-
tiated by N.H., W.C.H., E.S.L., D.E.R., D.M.S., S.A.S., and B.R.S. We
are grateful to the members of TRC—Academia Sinica, Bristol-Myers
Squibb, Eli Lilly, Novartis, and Sigma-Aldrich—for their financial sup-
port and scientific advice. We are indebted to Bristol-Myers Squibb,
Broad Institute, Dana-Farber Cancer Institute, and Whitehead Institute
for Biomedical Research for their support of early phases of this effort.
We thank S. Ali, N. Berkowitz, S. Bailey, J. Bridges, L. Brody, S.
Bulmer, A. Burds Connor, J. Davies, T.R. Jones, M. Lamprecht, M.
Lynes, H. Mizuno, J. Morawiak, C. Nguyen, S. Saif, D. Sarbassov,
and S. Yadav for technical assistance; and we are grateful to C. Nus-
baum for helpful discussions. We thank J. Evans, A. Davis, the White-
head-MIT Bioimaging Center, and N. Durso at Cellomics for providing
imaging resources and technical advice and Applied Biosystems for
providing reagents. This work was also supported by a Dana-Farber/
Harvard Cancer Center Core Grant Opportunity Award 2003-31C-
NOPA (W.C.H.), NIH P50 CA112962 (W.C.H.), the Tisch Family Fund
for Research in Solid Tumors (W.C.H.), an NSERC postdoctoral fellow-
ship (J.M.), NIH CA103866 (D.M.S.), Keck Foundation (D.M.S.), Edith
C. Blum Foundation, Stewart Trust (D.M.S.), NIH R01CA97061
(B.R.S), and a Career Award at the Scientific Interface (B.R.S).
Received: September 25, 2005
Revised: November 29, 2005
Accepted: January 4, 2006
Published: March 23, 2006
Abbas-Terki, T., Blanco-Bose, W., Deglon, N., Pralong, W., and
Aebischer, P. (2002). Lentiviral-mediated RNA interference. Hum.
Gene Ther. 13, 2197–2201.
Aza-Blanc, P., Cooper, C.L., Wagner, K., Batalov, S., Deveraux, Q.L.,
and Cooke, M.P. (2003). Identification of modulators of TRAIL-induced
apoptosis via RNAi-based phenotypic screening. Mol. Cell 12, 627–
Berns, K., Hijmans, E.M., Mullenders, J., Brummelkamp, T.R., Velds,
A., Heimerikx, M., Kerkhoven, R.M., Madiredjo, M., Nijkamp, W.,
Weigelt, B., et al. (2004). A large-scale RNAi screen in human cells
identifies new components of the p53 pathway. Nature 428, 431–437.
Bettencourt-Dias, M., Giet, R., Sinka, R., Mazumdar, A., Lock, W.G.,
Balloux, F., Zafiropoulos, P.J., Yamaguchi, S., Winter, S., Carthew,
R.W., et al. (2004). Genome-wide survey of protein kinases required
for cell cycle progression. Nature 432, 980–987.
Boutros, M., Kiger, A.A., Armknecht, S., Kerr, K., Hild, M., Koch, B.,
Haas, S.A., Consortium, H.F., Paro, R., and Perrimon, N. (2004).
Genome-wide RNAi analysis of growth andviability inDrosophila cells.
Science 303, 832–835.
Brummelkamp, T.R., Bernards, R., and Agami, R. (2002). A system for
stable expression of short interfering RNAs in mammalian cells. Sci-
ence 296, 550–553.
Dickins, R.A., Hemann, M.T., Zilfou, J.T., Simpson, D.R., Ibarra, I.,
Hannon, G.J., and Lowe, S.W. (2005). Probing tumor phenotypes us-
ing stable and regulated synthetic microRNA precursors. Nat. Genet.
Elbashir, S.M., Harborth, J., Lendeckel, W., Yalcin, A., Weber, K., and
Tuschl, T. (2001). Duplexes of 21-nucleotide RNAs mediate RNA inter-
ference in cultured mammalian cells. Nature 411, 494–498.
Federico, M. (2003). Lentivirus Gene Engineering Protocols, Volume
229 (Totowa, NJ: Humana Press).
Fire, A., Xu, S., Montgomery, M.K., Kostas, S.A., Driver, S.E., and
Mello, C.C. (1998). Potent and specific genetic interference by dou-
ble-stranded RNA in Caenorhabditis elegans. Nature 391, 806–811.
Fraser, A.G., Kamath,R.S., Zipperlen, P., Martinez-Campos, M.,Sohr-
mann, M., and Ahringer, J. (2000). Functional genomic analysis of
C. elegans chromosome I by systematic RNA interference. Nature
Gasparri, F., Mariani, M., Sola, F., and Galvani, A. (2004). Quantifica-
tion of the proliferation index of human dermal fibroblast cultures
with the ArrayScan high-content screening reader. J. Biomol. Screen.
Golsteyn, R.M. (2005). Cdk1 and Cdk2 complexes (cyclin dependent
kinases) in apoptosis: a role beyond the cell cycle. Cancer Lett. 217,
Gumireddy, K., Reddy, M.V., Cosenza, S.C., Boominathan, R., Baker,
S.J., Papathi, N., Jiang, J., Holland, J., and Reddy, E.P. (2005).
Cell 124, 1283–1298, March 24, 2006 ª2006 Elsevier Inc. 1297