Generation of diced-siRNA pools
The protocol used to synthesize d-siRNA [8
] has been scaled-up to synthesize the 1,920 d-siRNAs library as previously reported [11
]. Specific primers for each selected gene (Additional data file 5) were automatically designed and were used to amplify from a cDNA library an approximately 600 bp PCR fragment of the 3' region of the coding sequence. A second amplification was performed with a set of nested primers bearing a T7 promoter sequence on their 5' extension. Nested PCR products with no products or products of unexpected size (116 genes among the 1,920) were eliminated from the analysis of the screen. Nested PCR products were transcribed in vitro
(T7 MEGA script kit, Ambion; Austin, TX, USA) and the resulting double-stranded RNAs were annealed and processed with 30 units/reaction of human recombinant Dicer (Invitrogen; Carlsbad, CA, USA) for 15 h at 37°C. The 21-mer d-siRNAs were separated from incompletely digested fragments using a succession of isopropanol precipitations and filtration on glass fiber plates (Nunc; Rochester, NY, USA). Synthetic siRNAs were from Dharmacon (Lafayette, CO, USA).
Transfection and transferrin internalization assays
HeLa cells were plated at a density of 1,800 cells per well in 96-well plates in DMEM media supplemented with 10% fetal bovine serum and antibiotics. The d-siRNAs (approximately 20 ng/well) in a final volume of 50 μl were transfected the following day using Gene Silencer (Genlantis; San Diego, CA, USA) according to the manufacturer's protocol. Two and a half days after transfection, cells were washed twice with warmed DMEM/HEPES and incubated with 5 μg/ml Alexa Fluor® 594-labelled holotransferrin (Invitrogen) for 10 minutes at 37°C with 5% CO2. Surface-bound transferrin was stripped off with a 1 minute incubation in NaCl (500 mM)/acetic acid (200 mM). Following pH neutralization with phosphate-buffered saline, cells were fixed with 4% formaldehyde and nuclei stained with Hoechst.
Time-courses of transferrin uptake (Figures and ) were determined as indicated above but transferrin accumulation was stopped at the indicated time. For recycling experiments, cells were first loaded with fluorescent transferrin (5 μg/ml) for 30 minutes at 37°C and then washed and incubated with non-fluorescent transferrin (50 μg/ml) for the indicated times. The rate constants, tau, were calculated from single exponential fits of the time-course data.
LDL and EGF uptake assays
HeLa cells were incubated with DiI-LDL (1 μg/ml; Invitrogen) or biotinylated EGF in complex with Alexa Fluor® 488 streptavidin (200 ng/ml of complex is equivalent to 15 ng/ml of EGF; Invitrogen) for 30 minutes before surface stripping and fixation.
Formaldehyde fixed HeLa cells were permeabilized with 0.1% Triton X100, blocked with 10% goat serum and 2% bovine serum albumin and stained with anti-TFRC antibody (5 μg/ml; M-A712, BD Pharmingen; San Diego, CA, USA), anti-actin antibody (1:200; MAB1501, Chemicon International; Temecula, CA, USA), anti-β-tubulin-Cy3 antibody (1:200; Sigma; St. Louis, MO, USA) or anti-phosphorylated (S235/236) ribosomal S6 protein (1:200; Cell Signaling Technology; Danvers, MA, USA) associated with appropriate secondary antibodies (Invitrogen).
Automated image acquisition and processing
Images were acquired on the ImageXpress 5000A automated epifluorescence microscope (Molecular Devices; Sunnyvale, CA, USA) using an ELWD 20×/0.45 Plan Fluor objective and a 1,280 × 1,024 pixels cooled CCD camera with 12-bit readout. Thirteen pairs of images (Hoechst and Alexa Fluor® 594 channels) were acquired per well. Image analysis was performed using the ImageXpress database and software. Images were corrected for non-uniform illumination and segmented using the nucleus-associated Hoechst fluorescence. Nuclear regions were then symmetrically expanded (+5 μm) in order to include the perinuclear endosomal recycling compartment. For each detected cell, the background-subtracted average intensities of the perinuclear regions were measured in the Alexa Fluor® 594 channel. This method is independent of the size of the cells but the size of nuclei might have influenced the scoring but only slightly since average intensities were considered. The scores of d-siRNAs strongly affecting cell viability did not affect the final scoring since 'toxic' d-siRNAs were generally outliers (for example, KIF11/Eg5) or did not score at all because the scores were calculated on few living cells, which were not probably transfected. COPB1 and ARCN1 affected cell viability but were considered as hits because transferrin internalization was greatly reduced even in the remaining cells.
For immunostaining, intensities integrated over the entire cell area were used to estimate the total amount of TFRC, as well as the total amount of actin or tubulin per cell. These values were extracted automatically in two steps: first, nuclear regions were expanded (+20 μm) in order to include the entire cells; and second, expanded regions were shrunk to fit the cell perimeters (using the antibody-associated intensities to threshold the images). The resulting cell areas were used to calculate single cell integrated intensities. To compare transferrin, LDL and EGF uptake, a texture detection algorithm implemented in the ImageXpress 1.0 software suite was used to identify fluorescent punctuate structures and to estimate the vesicular density per cell, as well as the average vesicle intensity. Single cell, average phospho-S6 immunofluorescence intensity was measured in a 3 μm ring around cell nuclei.
Analysis of the relative number of transferrin molecules per vesicle
For the results presented in Figure , a Gaussian filter adjusted for an average vesicle diameter of 2.2 pixels was used to subtract a local background intensity of each transferrin containing vesicle. The average of five pixel intensity values was used as a measure of the intensity of transferrin-loaded vesicles. The 30 brightest vesicles per cell were used for analysis. Rapamycin treatment reduced the average intensity of these 30 brightest vesicles by 25%. This was used to adjust the relative intensity threshold used for counting the number of vesicles in the control case as well as for the rapamycin treated cells.
Estimation of the number of hits in the primary screen: CAsH analysis
The experimental noise distribution was estimated using the distribution of the deviations of the duplicated F values. The data distribution fdata(x) was derived from the histogram of the experimental F values averaged with a low-pass filter (width = 2 × standard deviations of the noise). The noise distribution fnoise(x) was subtracted from the data distribution fdata(x) to create the hit distribution fhit(x). CAsH was defined as ±fhit(x)/fdata(x) to give an estimate of the expected fraction of hits for each F value; a sign was added to indicate that a hit is a suppressor (-) or an enhancer (+) (if 0 ≤ F < 1 ≥ CAsH ≤ 0; if 1 ≤ F ≥ CAsH > 0).
P values were based on two-tailed homoscedatic Student's t-distributions and reflect a comparison of control versus indicated siRNA, or control versus rapamycin treatment. The estimate for the off-target probability was based on the 10% hit rate of the initial screen, approximately 183/1,804. Hits occurred approximately 5% of the time in a given direction (increase versus decrease). Assuming that all the hits are off-target effects (worst case scenario), the upper boundary for off-target effects from the screen becomes 5% and the probability for two independent siRNAs to both be off-target and go in the same direction is therefore expected to be less than 0.25%.