Prostate cancers typically display an inverse correlation between the levels of p27 and SKP2 [29]. In contrast, the commonly used human prostate cancer cell line LNCaP, despite faithfully recapitulating many features of human prostate cancer [36], expresses relatively high levels of p27 but low levels of SKP2 when compared to HeLa cells (Figure 1a). In order to mimic the situation prevailing in primary prostate cancers, we created the LNCaP derivative cell line, LNCaP-S14, which stably overexpresses Myc-tagged SKP2 at six to eightfold excess over endogenous SKP2, a manoeuvre that led to maximal downregulation of p27 (Figure 1b). The same pattern was apparent by immunofluorescence staining; while nuclear levels of p27 are low in LNCaP-S14 cells, nuclear SKP2 levels are highly elevated (Figure 1c).

We also determined whether known inhibitors of p27 degradation could upregulate p27 in LNCaP-S14 cells. Upon normalization to the loading control actin, the proteasome inhibitors MG132, epoxomycin and bortezomib caused between 1.2 and 1.7-fold upregulation of p27, while the CDK inhibitor roscovitine led to 1.5-fold accumulation (Figure 1d). The induction of nuclear p27 by MG132 and roscovitine was also apparent by immunofluorescence (Figure 1e).

The immunofluorescence assay used in Figure 1c and 1e was adapted to 384 well plate format. All parameters, including the number of cells to be seeded, fixation, blocking conditions, antibody concentrations and incubation times with compounds were extensively optimized using positive (proteasome and CDK inhibitors) and negative (dimethyl sulfoxide; DMSO) controls, resulting in the reliable protocol described in the Methods section. Representative images of LNCaP-S14 cells treated with DMSO or roscovitine and stained with the above protocol in 384 well plates are shown in Figure 1f.

In order to evaluate the performance of our assay to reliably measure the percentage of p27 positive cells in a cell population, we determined the Z'-factor for the positive control reagent roscovitine. The Z'-factor measures the variation and separation bands of an assay thereby providing a statistical measure of its quality. It takes into account the dynamic range and variability of the positive and negative control measurements [37]. We determined a Z'-factor of 0.48 using the ImageXpress Micro and MetaXpress software from Molecular Devices (CA, USA) and 0.60 using the Cell Lab IC100 and Cytoshop software from Beckman Coulter (USA). Both values are deemed appropriate for a successful screen [37]. Figure 1g shows the signal-to-background ratios using both imaging platforms.

In order to assess the utility of the above protocol, we performed a pilot screen of 7368 compounds (derived from a variety of different compound libraries, see Methods) in duplicates for a total of 44 384-well plates. LNCaP-S14 cells were incubated with individual compounds (final concentration varied between ~5 and 35 uM depending on the molecular weights of the compounds) for 18 h, followed by fixation, staining, imaging (automatic microscopy) and analysis as described in Methods. We scored the percentage of cells positive for p27 in each well relative to the vehicle control (DMSO). Figure 2a shows the activity of positive control compounds (yellow triangles), the negative control compound DMSO (red squares), the non-specific staining (blue diamonds) and small molecules derived from compound libraries (blue circles) across all screening plates (44 plates) prior to normalization. Negative controls (DMSO) gave basal percentages of p27 positive cells ranging between 2% and 13% while the positive control roscovitine increased these populations up to 50%. As expected, the average of ~320 well measurements per plate - that is, the percentage of p27 positive cells in wells treated with individual compounds - was similar to the percentage of p27 positive cells in the negative control (DMSO), indicating that most compounds tested were inactive in causing p27 accumulation (Figure 2a).

Normalization of raw data for 7368 compounds was performed by calculating Z scores, which represent the number of standard deviations by which a signal for a given compound differed from the mean signal of the entire plate [38]. It excludes control measurements altogether under the assumption that most compounds are inactive and can serve as intrinsic controls. The summed Z scores for one set of duplicates (22 plates) showed clear separation between the positive control compound roscovitine (column 2) and the non-specific staining (cells treated with DMSO or stained with secondary antibody only, columns 23 and 24; Figure 2b). In addition, most summed Z scores had values similar to the negative controls (DMSO), indicating that we did not incur errors caused by plate position of compounds. This was further evaluated by visualizing every plate as a heat map of individual Z scores. As shown in the example of a single plate in Figure 2c, potential hits were randomly distributed over the plate (green and yellow positions). The analysis also identified several false-positive compounds with very high Z scores, namely known DNA binding compounds of red colour (doxorubicin and derivatives, crystal violet, homidium bromide, propidium iodine, pararosalinine pamoate).

In the initial primary screen, we identified 249 compounds that increased the percentage of p27 positive LNCaP-S14 cells. Upon manual examination of microscopic images, 21 duplicate compounds (occurring in more than one library), seven obvious false positives (DNA stains) and 45 compounds with low quality staining (few cells left in well, background, indiscriminate whole cell staining) were sorted out. This resulted in a list of 176 candidates (Additional File 1, Supplementary Table 1) with sixty compounds classified as strong (Z score > 5), 58 as medium (Z score 3-5) and the remainder as weak (Z score 2-3; Figure 3a). At a Z score of 3, a compound has a theoretical probability of 0.0013 of being a false positive hit.

A number of experimental and clinically used chemotherapeutics such as camptothecin, thapsigargin, 17-N-allylamino-17-demethoxygeldanamycin, mitomycin C and etoposide were identified (Additional File 1, Supplementary Table 1). The PI3 kinase inhibitor Wortmannin also scored positive, a finding that is consistent with the known role of the PI3K/AKT pathway in the downregulation of p27 through SKP2 [39]. We further validated some of the known bioactive compounds by immunofluorescence staining and immunoblotting (Figure 3b and 3c; data not shown).

The one hundred and nine most potent compounds (Z ≥ 3) were selected for confirmation by secondary screening using the primary screening assay. After statistical analysis, 60 compounds were still considered positive hits (Figure 3d). It is well established that compounds with high activity in the primary screen will appear less potent in the secondary screen because of a statistical effect known as 'regression towards the mean' [38]. Thus, weaker hits often fail to validate in the secondary screen due to random measurement error. Additional counter screening for autofluorescence identified another false-positive (pyrvinium pamoate).

Thirty-seven of the 60 validated compounds had known biological activity (for example, inhibitors of protein synthesis and topoisomerase activity, antibiotics and antihelmintics). Of the remaining 23 compounds, none of which had known bioactivity, 20 were prioritized for further characterization based on features of their chemical structures. These compounds were denoted small molecule inhibitors of p27 depletion, SMIPs.

Sixteen of the 20 SMIPs were obtainable as powders from commercial sources. The identity and integrity of all compounds was confirmed by high-performance liquid chromatography (HPLC) and mass spectrometry (data not shown). SMIPs were retested for activity in a dose-response experiment using the primary screening assay. Thirteen out of 16 purchased SMIPs induced a ≥ twofold increase in the percentage of p27 positive cells at the maximal dose with the remaining compounds having lower activity (data not shown). SMIP014 was identified as an additional false positive at this step (data not shown).

Several SMIPs also induced p27 accumulation as determined by immunoblotting (Figure 4a and 4b), although the effects were generally weaker (~twofold at 40 μM). While seemingly small, such an increase is biologically significant because p27 levels do not vary more than two- to threefold during a normal cell cycle ([40] and our own unpublished observation). This apparent discrepancy between the immunofluorescence and immunoblotting data, which we had encountered previously during assay development (with positive control compounds roscovitine and MG132; Figure 1c and data not shown), is explained by the different metrics of the assays: While the immunofluorescence assay determines the percentage of cells with nuclear p27 staining above a threshold determined by staining with secondary antibody alone, the immunoblotting assay measures the total amount of p27 that can be extracted from a cell population (no threshold).

Further prioritization of the remaining 15 SMIPs was achieved by counter screening against another unstable prostate tumour suppressor (NKX3.1) [41,42]. In order to address target specificity, LNCaP cells stably overexpressing NKX3.1 fused to yellow fluorescent protein (NKX3.1-YFP) were used. Cells were treated with a single concentration (40 μM) of SMIPs in 384-well plates and analysed by automated microscopy (see the Methods section). The Z' factors for this assay, as determined by positive and negative controls, were between 0.62 (epoxomycin) and 0.67 (MG132), while the signal-to-background ratio was five in both cases. Although some SMIPs (005, 006, 016) caused a ~twofold increase in the percentage of NKX3.1 positive cells, the effect was subtle in comparison to proteasome inhibitors (~fivefold; Figure 4c) and could not be confirmed by immunoblotting (data not shown). Likewise, when the expression of endogenous NKX3.1 was evaluated in DU145, another prostate cancer cell line with very low levels of NKX3.1, none of the SMIPs scored positive (data not shown). The minor effects of some SMIPs in the NKX3.1-YFP fluorescence assay were most likely an artifact of compound autofluorescence at the wavelength used to excite YFP.

We also determined the effect of SMIPs on a HeLa cell line stably expressing p27 fused to firefly luciferase (p27-luc) [40,43]. While proteasome inhibitors led to an increase in p27-luc expression, which was readily apparent by luciferase assay (Figure 4d) and immunoblotting (data not shown), none of the SMIPs were active. This finding suggests a certain degree of cell type specificity of SMIP action.

Since p27 can drive cell cycle arrest, senescence, and apoptosis [7,44,45], we assessed the effect of SMIPs on cell proliferation and viability. LNCaP-S14 cells and IMR90 normal human fibroblasts were exposed to increasing concentrations of SMIPs for 72 h and scored using the MTT assay followed by calculation of IC₅₀ values (concentration required for 50% inhibition) (Table 1). Whereas SMIP012 and 016 were moderately toxic in normal fibroblasts, SMIPs 001 and 004 showed substantial cancer cell specificity being at least five times more potent in LNCaP-S14 than in IMR90 cells. Although SMIP005 also showed good cell type selectivity, it was excluded at this point due to structural features that suggested potential unspecific reactivity with cellular macromolecules. Based on their apparent cancer cell specificity and their chemical structures, SMIPs 001 and 004 were selected for additional studies.

Figure 5a shows the chemical structures of the two compounds. We next confirmed the accumulation of p27 upon treatment with 40 μM SMIP001 and 004 in LNCaP-S14 cells (Figure 5b). The CKI p21, which was described as another ubiquitylation target of SKP2 [46], was also upregulated by SMIPs (Figure 5b). The effects on p27 and p21 were also observed with the CDK inhibitor roscovitine. Interestingly, SMIP004 caused a decrease in the levels of both the endogenous and the exogenous, stably expressed SKP2. In contrast, treatment of normal human fibroblasts with SMIPs did not affect p27 or p21 levels (see below).

In order to evaluate the dose-response of the effect on p27, LNCaP-S14 cells were treated with increasing concentrations of the respective SMIPs (5 - 40 μM), and the percentage of cells with nuclear p27 was measured by immunofluorescence and automated microscopy. Both SMIPs led to a dose-dependent increase in p27 positive cells (Figure 5c). Dose-dependent increases in p27 and p21 were also observed by immunoblotting (Figure 5d). Whereas SMIPs 004 led to ~threefold induction at 40 μM, SMIP001 was less active in this experiment causing only a modest accumulation of p27 (30%-40%).

In order to begin to determine the mode of action of SMIPs, we measured their effects on p27 protein stability in a cycloheximide chase experiment. Whereas the half-life of p27 was ~ 1.29 h in cells treated with DMSO, a value that is in good agreement with published reports (for example, [47]), treatment with either MG132 or SMIP004 increased p27 half-life to > 6 h (Figure 5f). No effect on p27 stability was seen for SMIP001. We also evaluated the stability of p21 and SKP2 but did not see any effects of SMIPs on these proteins. In summary, these data revealed a striking correlation between the ability of SMIP004 to downregulate SKP2 and to induce p27 stabilization. In contrast, SMIP001 appears to regulate p27 and p21 primarily at the level of mRNA expression (Figure 5f).

Since SMIP004 replicated the effect of MG132 on p27, we tested whether it acts as a proteasome inhibitor. Total cell lysate from LNCaP-S14 cells treated with 20 and 40 μM of the SMIPs were analysed by immunoblotting with anti-ubiquitin antibody to assess the abundance of polyubiquitylated proteins. None of the compounds caused the accumulation of polyubiquitylated proteins that was readily seen with the proteasome inhibitors MG132 and bortezomib (Figure 5g). Likewise, none of the SMIPs inhibited the proteolytic activity of purified 26S proteasomes toward fluorogenic peptide substrates even when assayed at 40 μM concentration (data not shown).

Since SMIPs robustly upregulated p27 and p21, we asked whether this resulted in G1 delay. Indeed, SMIPs 001 and 004 induced a G1 delay with a concomitant decrease in the S phase population (Table 2). We next asked whether the cell cycle delay coincided with inhibition of CDK2, one of the major cellular target kinases of p27 and p21. CDK2-cyclin complexes were immunoprecipitated from lysate of LNCaP-S14 cells treated with DMSO, the CDK inhibitor roscovitine or SMIPs and assayed for activity toward histone H1 in vitro. As with roscovitine, application of SMIPs led to strong inhibition of the CDK2 activity retrieved from cells (Figure 6a). In order to test the possibility that SMIPs act like roscovitine as direct inhibitors of CDK2 kinase activity, SMIPs were added to CDK2 complexes purified from untreated cells. Unlike roscovitine, none of the SMIPs inhibited CDK2 activity when added to the kinase reaction in vitro (Figure 6b), indicating that the two SMIPs are not kinase inhibitors.

We also evaluated the effect of SMIPs on the levels of p27, p21 and cyclins E and A complexed with CDK2 by coimmunoprecipitation. Both SMIP001 and 004 led to a strong increase in the recruitment of p27 to CDK2, while SMIP001 also slightly increased coprecipitation of p21 (Figure 6c). SMIP004 also reduced the amounts of cyclins E and A retrieved with CDK2 (Figure 6c). This was paralleled by a marked downregulation of cyclins E and A upon SMIP004 treatment (Figure 6c). A more variable downregulation of cyclin A and CDK4 was observed with SMIP001 (Figure 6e). Taken together, these findings suggested that SMIP-induced inhibition of CDK2 activity might be a combined consequence of p27/p21 upregulation and cyclin E/A downregulation.

In order to determine the specific contribution of p27 and p21 to SMIP-induced cell cycle delay, we performed siRNA-mediated knockdown studies. The depletion of p21 and p27 led to a decrease in the G1 population of untreated cells by 6 - 13 percentage points (Additional File 2), a finding that indicates biologically significant effects at the knockdown efficiencies achieved (~80%, Figure 6e). Surprisingly, however, neither individual nor combined knockdown of the two CKIs was able to abrogate SMIP-induced cell cycle delay (Figure 6d and Additional File 2). Likewise, CDK2 activity in SMIP-treated cells was not rescued by knockdown of p27 and p21 (data not shown). Finally, neither the downregulation of cyclins E and A nor that of CDK4 was consistently affected by knocking down the CKIs, suggesting that these effects of SMIPs probably account primarily for their G1 delay activity.

In addition to G1 delay, treatment of LNCaP-S14 cells with SMIPs for 24 h caused apoptosis as determined by the cleavage of poly-ADP ribose polymerase (PARP, Figure 6f and 6g). The apoptotic effect of SMIPs was independently evaluated by measuring the amount of cytoplasmic histone-associated DNA fragments. Cells treated with 40 μM SMIP001 and 004 from 24 to 72 h showed a three- to fivefold increase in the amount of mono and oligonucleosomes, thus confirming the apoptosis-inducing activity of SMIPs.

As shown in Table 2, SMIPs also induced robust G1 delay in parental LNCaP cells as well as in two other prostate cancer cell lines, DU145 and PC3. No such effect was seen with HeLa cells or IMR90 fibroblasts. In parental LNCaP cells, G1 delay correlated with an increase in the levels of p27 and p21 as well as a downregulation of cyclins A and E and CDK4, which was particularly prominent with SMIP004 (Table 3 and Figure 7a). Unlike in SKP2 overexpressing LNCaP-S14 cells, SKP2 was downregulated by both SMIP001 and 004 in parental LNCaP cells, although the suppression was substantially greater with SMIP004 than 001 (Figure 7a). Less pronounced effects on SKP2 were also observed in PC3 and DU145 cells and in IMR90 fibroblasts. In contrast, when averaged over four independent experiments, SMIPs had no consistent effect on p27 levels in PC3 and DU145 cells, although a minor increase in p21 levels was observed (Table 3) that coincided with a modest reduction in the levels of cyclin A and CDK4 (Figure 7a). None of the latter effects were apparent either in HeLa or IMR90 cells. Despite a lack of induction of apoptosis (PARP cleavage, Figure 7a) or overt cytotoxicity (Table 4), the SMIP-induced G1 delay of PC3 or DU145 cells resulted in the same robust inhibition of colony formation in soft agar as observed in LNCaP-S14 and parental LNCaP cells (Figure 7b).

NINDS, Prestwick Peakdale 1, Mixed Commercials, Antimitotic, and Enamine and Known Bioactive Compounds libraries were obtained from The Institute of Chemistry and Cell Biology (ICCB) at Harvard Medical School (MA, USA). Information about the libraries can be found at http://iccb.med.harvard.edu.

SMIP001 to SMIP018 were purchased from Ryan Scientific (SC, USA). SMIP0019 was purchased from ChemDiv, Inc (CA, USA). Roscovitine was purchased from Enxo Life Sciences (formerly Biomol International; NY, USA), bortezomib from LC Laboratories (MA, USA), cycloheximide (CHX) and camptothecin from Sigma (MO, USA), MG132 and epoxomicin from BostonBiochem (MA, USA). All drugs were dissolved in DMSO and kept at -80°C.

Anti-p27 antibody (610241, 1:5000 for immunoblotting and 1:1000 for immunofluorescence) was obtained from BD Biosciences (NJ, USA). Anti-SKP2 (32-3300, 1:2000 for immunoblotting.) and anti-ubiquitin (13-1600, 1:2000) was from Zymed Laboratories (CA, USA). Anti-SKP2 (sc-7164, 1:500 for immunofluorescence), anti-CDK2 (sc-163, 1:1000 for immunoblotting, 2 μg/500 mg protein for immunoprecipitation) and anti-CDK4 (sc-260, dil 1:500) from Santa Cruz Biotechnology, Inc (CA, USA). Antibodies for Cyclin E Ab-2 (clone HE12, dil 1:200) and Cyclin A Ab-6 (clone 6E6, dil 1:100) from Thermo Scientific (CA, USA). Anti-p21 (2947, 1:2000) and anti-PARP (9542, 1:1000) were from Cell Signaling Technology (MA, USA). Anti-Tubulin (T5168, 1:5000) was from Sigma Life Science. Anti-Actin (69100, 1:20000) was from MP Biomedicals, LLC (CA, USA). HRP Donkey anti-mouse IgG (715-035-150, 1:5000) and HRP Donkey anti-rabbit IgG (711-035-152, 1:5000) was obtained from Jackson ImmunoResearch Laboratories (PA, USA).

Parental LNCaP cells were stably transfected with a plasmid driving the expression of Myc epitope-tagged SKP2 or empty vector (pcDNA3). Briefly, 5 x10⁶ cells were transfected with 20 μg of each plasmid using Lipofectamine 2000 reagent (Invitrogen, CA, USA) following the manufacturer's protocol. Cells were selected in 600 μg/mL geneticin (G418) until individual colonies were visible. Colonies were picked, passed to 24 well plates and expanded. The levels of p27 and SKP2 were analysed by immunoblotting as described below. LNCaP-S14 (positive clone for SKP2 overexpression), LNCaP-P10 (vector control cell line), LNCaP (parental cell line), PC3 and DU145 cells were maintained in RPMI supplemented with 10% fetal bovine serum and 5% penicillin/streptomycin solution. IMR90 and HeLa cells were maintained in DMEM containing 10% fetal bovine serum, 5% penicillin/streptomycin and 4 mM glutamine. Cells were grown as a monolayer in a humidified incubator at 37°C and 5% CO₂.

The screen relies on the detection of differences in the levels of endogenous p27 by immunofluorescence. LNCaP-S14 cells were seeded at a density of 100,000 cells onto 15 mm glass cover slips and allowed to attach for 1 day. Cells were treated with vehicle (DMSO) or positive controls (roscovitine, MG132) for 18 h. After the incubation, cells were fixed by the addition of 10% paraformaldehyde in phosphate buffered saline (PBS; final concentration 4%) and incubated for 20 min at room temperature (RT). Cells were permeabilized by washing 3 × 5 min with PBS, 0.1% Triton X100, followed by blocking in 5% nonfat dry milk in PBS, 0.1% Triton X100 for 1 h at RT. Cells were incubated with 50 μl primary antibody (anti-p27, 1:1000 in blocking buffer) for 1 h, followed by one wash with blocking buffer. 50 μL of secondary antibody (Alexa Fluor 594 goat anti-mouse, Invitrogen A11005, 1:200 in blocking buffer) was added for 1 h. Cells were washed 3 × 5 min in PBS, 0.1% Triton X100 and stained with Hoechst dye (1 μg/mL in PBS) for 2 min. Cells were washed twice in PBS and mounted for imaging with a Nikon Type 120 inverted fluorescent microscope using 60× magnification. Compounds used as positive controls included MG132, epoxomycin and roscovitine, while DMSO was used as negative control. Cells stained with secondary antibody only were used to assess the non-specific (NS) staining background.

The above protocol was adapted to 384 well plates as follows: 4000 LNCaP-S14 cells per well were seeded into 384-well plates (black with clear bottom) in 30 μL RPMI, 10% fetal bovine serum (FBS). Nuclear p27 staining was done under the same conditions as above but with reducing the volume of solutions to 20 μL/well for 10% para-formaldehyde in PBS, 30 μL/well for blocking and wash solutions and 15 μL/well for primary and secondary antibody solutions and Hoechst dye. All liquid handling was done with an eight-channel multidrop liquid dispenser (Wellmate, OH, USA) and a wand aspirator (VP Scientific, CA, USA; VP-186L). After staining, plates were sealed and stored in the dark at 4°C until scanning.

Images from 384 well plates were acquired by using the ImageXpress Micro inverted epifluorescent automated microscope (Molecular Devices, CA, USA) at dual wave length to detect Hoechst and p27. The microscope was equipped with the Photometrics CoolSNAP ES digital CCD camera and an automated objective and filter cube changers. Two images per well at a 20× magnification were obtained at each wave length.

Images were analysed with the MetaXpress cellular imaging analysis software using the cell scoring module. Cells were scored positive for Hoechst, if the integrated pixel intensity was 210-fold above local background and positive for p27 when staining was 30-fold above background (signal obtained with secondary antibody). MetaXpress processing of the raw images provided quantitative measures of the total cell number and the number of p27 positive cells in a given field. The data from both images of each well were combined to get a single number of positive cells. The percentage of positive cells was calculated relative to the total number of cells. Background correction was done by subtracting the number of p27 positive cells in wells stained with secondary antibody only.

The staining protocol was also evaluated using another imaging and software package, the Cell Lab IC100 and Cytoshop software (Beckman Coulter, CA, USA). Minor changes were introduced to the protocol: The cell number was reduced to 3000/well and the secondary Alexa Fluor 568 goat anti mouse antibody was diluted 1:500 in blocking buffer. Four images per well were taken using 10× magnification with a numerical aperture of 0.25 and a camera binning of 2 × 2. A manual threshold was established by comparison of the total nuclear intensity in the positive control and the total intensity obtained in cells stained with the secondary antibody only (unspecific staining) and/or the vehicle control (DMSO). Processing of the raw images provided quantitative measures of the total cell number and the number of p27 positive cells in a given field.

Z'-factor calculation: This parameter was used to assess the quality of the assay in the HTP optimization (HTP Company, CA, USA) [37]. In three independent experiments, LNCaP-S14 cells were seeded in 384 well plates followed by treatment with 0.3% DMSO (192 wells, negative control) or 20 μM roscovitine (192 wells, positive control) for 18 h. The percentage of p27 positive cells was determined as described above and the Z' factor calculated from 576 replicates (each condition) as follows: Z'= 1- (3SD₊ + 3SD_-)/(Ave₊ - Ave_-), where SD₊ is the standard deviation for the positive control, SD_- the standard deviation for the negative control, Ave₊ the average for the positive control and Ave_- is the average for the negative control [37].

The cell-based screen was performed at the Institute of Chemistry and Cell Biology, Harvard Medical School, using libraries of uncharacterized compounds (Peakdale 1, Mixed Commercials, Antimitotic and Enamine, total of 4888 compounds) plated in 5 mg/mL stock solutions in DMSO, the NINDS library (1040 compounds, 10 mM in DMSO), the Prestwick Library (960 compounds, 2 mg/mL) and a library of 489 known bioactive compounds (plated at 5 mg/mL, 1.1 mg/mL, and 0.25 mg/mL in DMSO) for a total of 7638 compounds. LNCaP-S14 cells were seeded onto 384 well plates as described above. Compound stocks (100 nL) were pin-transferred using a robot-controlled stainless-steel pin array, resulting in a 300-fold-dilution (5 to 35 μM final screening concentration, depending on individual compound molecular weight) followed by incubation for 18 h. All compound plates were screened in duplicates. Each individual plate contained at least eight positive controls (roscovitine, 20 μM), eight negative controls (DMSO, 0.3%) and eight wells of cells treated with 0.3% DMSO and stained only with secondary antibody (non-specific staining). Staining of p27, imaging and analysis was performed as described above. Potential hits were classified according to their Z score. The Z score is a simple and widely known normalization method calculated as follows: Z = X_i - X/S_x, where X_i is the raw measurement on the ith compound, X and S_x are the mean and the standard deviation, respectively, of all measurements within each plate [38]. Potential hits were classified as weak if their Z score was between 2 and 3, medium if the Z score was between 3 and 5 and strong if the Z score was greater than 5.

Target specificity was analysed using LNCaP cells stably expressing the homeobox transcription factor NKX3.1 fused to YFP. In brief, 4000 LNCaP-NKX3.1-YFP (clone 4A) cells were seeded in a 384 well plate and treated with SMIPs for 18 h. Proteasome inhibitors were used as positive controls. Images from 384 well plates were taken using the Cell Lab IC100 automated microscope (Beckman Coulter), and the percentage of cells positive for NKX3.1-YFP was quantified using Cytoshop software (Beckman Coulter). The percentage of NKX3.1-YFP positive cells was determined relative to DMSO treated cells.

Cell specificity was evaluated using HeLa cells stably transfected with a p27-luciferase (p27Luc) construct [43]. Briefly, HeLa-p27Luc cells were seeded in 96 well plates and treated with 40 μM of the respective SMIPs or positive controls (proteasome inhibitors). After incubation for 18 h, cells were lysed using Cell Culture Lysis Reagent from Promega (WI, USA) and luciferase activity was determined by the Luciferase Assay System (Promega) using a Veritas microplate luminometer (Turner Biosystems, CA, USA). Protein concentration was measured in a parallel plate using Bio-Rad Protein Assay (Bio-Rad, CA, USA). Luciferase activity was normalized against protein concentration and compared to the activity recovered from DMSO-treated cells.

Total cell lysate from cells treated with vehicle (DMSO) or compounds were obtained by incubating cells in lysis buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.5% Triton X-100, protease inhibitors) for 15 min at 4°C. The supernatant was collected by centrifugation and the amount of protein was determined with a Bio-Rad protein assay (Bio-Rad) using BSA as standard. Equal amounts of protein were subjected to SDS-PAGE (4%-20% gels, Invitrogen) and transferred to PVDF membranes. The membranes were blocked with TBST (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.05% Tween-20) containing 5% non-fat dry milk or bovine serum albumin (BSA) for 1 h at room temperature, followed by incubation in primary antibodies 4°C overnight. Membranes were washed and incubated with secondary antibodies conjugated to HRP (Jackson ImmunoResearch Laboratories) for 1 h at room temperature. Signals were visualized using the ECL Western Blotting Substrate (Pierce Biotechnology, IL, USA) according to the manufacturer's instructions. Membranes were stripped using Restore™ Western Blot Stripping Buffer (Pierce Biotechnology). Films were scanned and quantification of the bands was performed using Image J software http://rsbweb.nih.gov/ij/.

Total lysates of cells treated with vehicle (DMSO), SMIPs or positive controls were prepared by incubation in immunoprecipitation (IP) buffer (25 mM Tris-HCl, pH 7.4; 150 mM NaCl, 0.5% Triton X-100, protease inhibitors) for 15 min at 4°C. Upon centrifugation, the supernatant was collected and the amount of protein was determined with a Bio-Rad protein assay (Bio-Rad Laboratories) using BSA as standard. Five hundred micrograms of protein was adjusted to 1 mL with IP buffer and incubated with 10 μL anti-CDK2 antibody (sc-163, Santa Cruz Biotech) at 4°C for 3 h with constant rotation, followed by addition of 100 μL protein G-sepharose beads (10% slurry) for 2 h. After four washes for 5 min each with 1 mL IP buffer, beads were resuspended in 2X-SDS Laemmli buffer, followed by SDS-PAGE and immunoblotting.

Histone H1 kinase assays were performed as described in reference [55]. Briefly, the total cell lysates from LNCaP-S14 cells treated with the respective SMIP (40 μM) for 24 h were prepared in IP lysis buffer supplemented with protease inhibitors followed by IP. Kinase reactions were performed by adding histone H1 (1 μg) and 7.5 μCi [γ-³²P]ATP (800 Ci/mmol Perkin Elmer Life Sciences, MA, USA) in kinase buffer (20 mM MgCl₂, 10 mM EGTA, 40 mM Hepes, pH 7). After incubation at 30°C for 20 min, the reaction was stopped by adding 20 μL 2× SDS gel loading buffer. Samples were separated by electrophoresis, gels were stained and dried, followed by exposure to X-ray film. DMSO was used as negative control and roscovitine (CDK2 inhibitor) as positive control.

LNCaP-S14 (10,000 cells), PC3, DU145 or IMR90 cells (5000 cells) were seeded in 96-well plates and treated with increasing concentrations of the respective SMIPs for 72 h. Cell viability was assessed using the MTT cell proliferation assay (ATCC^®, Middlesex, UK) according to the manufacturer's protocol. IC₅₀ was calculated using the BioDataFit 1.02 software http://www.changbioscience.com/stat/ec50.html.

LNCaP-S14 cells were treated with SMIPs (40 μM) for 18 h followed by the addition of 100 μg/mL cycloheximide (CHX). Cell extracts were obtained as described above at different times after CHX addition. Protein lysate was subjected to SDS-PAGE and immunoblotting for p27, p21 and SKP2. Tubulin was used as loading control. Protein half-lives were calculated using a web based calculator http://www.1728.com/halflife.htm.

Total RNA was isolated from LNCaP-S14 cells treated with 40 μM SMIPs or vehicle (18 h) using the RNeasy Mini Kit (Quiagen, CA, USA), followed by first-strand cDNA synthesis using Omniscript^® Reverse Transcription (Quiagen). Real-time PCR analysis was performed on the Stratagene Mx3000p detection system (Stratagene, Agilent, CA, USA) using the SYBR Green PCR Master Mix (ABI, CA, USA) at the Microarray and q-PCR Facility of the Sanford-Burnham Medical Research Institute. The primers used were:

5'-ACTGCAGAGACATGGAAGAG-3' and 5'-ATGCGTGTCCTCAGAGTTAG-3' for p27 (to amplify a region from nucleotides 608 to 852); 5'-CGATGGAACTTCGACTTTG-3' and 5'-GGGTACAAGACAGTGACAGG-3' for p21 (to amplify a region from nucleotides 142 to 355); and 5'-TGAGCTGCTAAAGGTCTCTG-3' and 5'-ATGTGCTGTACACGAAAAGG-3' for SKP2 (to amplify a region spanning from nucleotide 527 in exon No. 2 to 739 in exon No.4). Briefly, reactions were done in a 25 μL reaction volume of q-PCR mixture containing 2 μL of cDNA (40 ng) and 250 nM of each primer. Activation of the enzyme was done at 95°C for 10 min followed by 40 cycles of amplification at 95°C for 30 s, 56°C for 1 min and 72°C for 30 s. All reactions were done in duplicates and normalized using GAPDH as control. Primers were design using Primer 3 http://frodo.wi.mit.edu/primer3/ and synthesized by Valuegene, Inc, CA, USA.

Cells were exposed to SMIPs (40 μM) for 24 h and 48 h. Cells were trypsinized, washed with PBS and suspended in 600 μL PBS. Cells were fixed by adding 1.4 mL cold absolute ethanol and kept at -20°C for at least 12 h. After washing once with PBS, the cells were resuspended in 250 μL PBS containing 2.5 μg RNAse A and incubated for 45 min at room temperature. Staining was done by adding 40 μg/mL of propidium iodine followed by incubation for 15 min at room temperature. DNA-bound fluorescence was read at 564 to 606 nm using FACSort and FACSCanto flow cytometers (BD Biosciences) at the Flow Cytometry Facility of the Sanford-Burnham Medical Research Institute. Distribution of cells in the different cell cycle phase was determined with ModFit LT 3.2.1 or FlowJo 8.6 software. Aggregates were excluded from the analysis manually using pulse shape or identified by automated modelling in ModFit LT.

Apoptosis was measured using the Cell Death Detection ELISA (Roche, Basel, Switzerland) following the manufacturer's instructions. Briefly, LNCaP-S14 cells (0.1 × 10⁶) were seeded in 12-well plates and treated with SMIPs (40 μM) for 24 h. Cells were collected in medium, spun at 1000 rpm, and resuspended in 1 mL PBS at the specified time points (day 1 to day 3). The cell suspension was divided into two parts. The first half was used for determination of cytoplasmic histone-associated DNA fragments (ELISA), the second for protein determination to normalize the ELISA data to the amount of input protein.

LNCaP-S14 cells were seeded in 6 cm dishes or six-well-plates coated with poly-lysine the day before transfection. 10-20 nM of siRNA was transfected using DharmaFECT 3 (Thermo Scientific) according to the manufacturer's instructions. Briefly, siRNAs were dissolved in siRNA suspension buffer (Quiagen) at 20 μM, heated to 90°C for 1 min and incubated at 37°C for 1 h. siRNAs were added to the DharmaFECT 3 reagent diluted in Optimem media (1:100) and incubated for 20 min at room temperature. The mix was added to the cells for 24 h. Treatment with SMIPs (40 μM) or vehicle was carried on for another 24 h. Cells were obtained for FACS or lysed for immunoblotting analysis as described above. Untransfected cells as well as cells transfected with non-specific siRNA were used as controls. Silencer^® Negative Control siRNA No.1 from Ambion (Applied Biosciences, CA, USA) was used as non-specific control siRNA. siRNA target sequences for p27 and p21 synthesized by IDT (Integrated DNA Technologies Inc, Iowa, USA) were as follows; p27 siRNA: AAGGUUGCAUACUGAGCCAAG, and p21 siRNA: AACAUACUGGCCUGGACUGUU [56].

Agar Noble (Difco, MD, USA) was suspended at 6% in water and autoclaved. A dilution 1:10 was made with RPMI culture medium and added to six-well plates (2 mL). We suspended 20,000 cells were in 0.5 mL RPMI culture medium (containing drugs), added to 0.5 mL of agar (0.6%) and poured immediately into a six-well plate containing hardened bottom agar. Cells were fed with fresh medium containing DMSO, SMIP001, SMIP004 or bortezomib every third days. Images were taken after 14 days using a Nikon Eclipse E600 Microscope.