Unless otherwise specified, all chemicals were purchased from Sigma-Aldrich (St. Louis, MO); tissue culture reagents and supplies from Life Technologies (Carlsbad, CA), Corning (Tewksbury, MA) or Thermo-Fisher Scientific (Waltham, MA). Draq5 was purchased from Biostatus (Shepshed, UK). Multiwell plates were purchased from Greiner BioOne (Monroe, CA), Corning or Matrical Bioscience (Spokane, WA). Bulk reagents (cells, medium, etc.) were dispensed to all plate wells with a Multidrop Combi (Thermo-Fisher) or a Matrix Wellmate fitted with plate stackers (Thermo-Fisher), and compounds transferred with a CyBi^®-Well 384-channel simultaneous pipettor (CyBio, Boston, MA). Serial dilutions of compounds were performed using a BioMek FX^® (Beckman Coulter, Brea, CA). Multi-well plates were washed with either a Skatron EMBLA-384 (Molecular Devices, Sunnyvale, CA) or Bio-TekEL405^TM (Winooski, VT) platewasher; washers were presterilized with bleach, then ethanol and flushed thoroughly with sterile water before use. The IN Cell Analyzer 3000, a confocal line-scanning instrument, along with cell analysis software modules, was purchased from GE Life Sciences (Piscataway, NJ).

EGFP-wild type LC3B plasmid was constructed as follows. The LC3B coding sequence was amplified from EST clone BC015810.1 (IMAGE) by PCR and then cloned into pEGFP-C1 (Clontech, Mountain View, CA). The resulting plasmid allowed the expression of EGFP-LC3B fusion protein under CMV promoter control. HeLa cells (ATCC, Manassas, VA) were stably transfected with EGFP-wild type LC3B plasmid using Fugene 6 (Roche Applied Science, Indianapolis, IN). Cells were then sorted based on EGFP intensity using a FACSDiva (Becton Dickinson, San Jose, CA) and single-cell derived clonal populations with high EGFP expression generated. HeLa EGFP-LC3 Clone C10 was selected based on fluorescence intensity (Fig. 2) using a FACSCalibur (Becton Dickinson) and on EGFP-LC3 grains observed by iCyte (Compucyte, Cambridge, MA). A lack of Pgp expression, desirable when screening compounds that may otherwise be pumped out of the cell, was confirmed using PE-conjugated anti-Pgp antibody (Becton Dickinson) with cells analyzed on the FACSCalibur. EGFP expression in these cells was monitored over time and found to be stable until at least passage 18. In preparation for screening, a large single batch of cells at passage 5 was frozen in vials in liquid nitrogen, and to maintain assay performance, cells were kept in culture only until passage 18 and then replaced with a fresh vial of cells.

Fig. (2) Flow cytometry analysis of the EGFP-LC3 HeLa cell population before clonal selection (left panel, blue line) and after (right panel, green line), where 8 % of the population were expressing EGFP-LC3, compared to WT HeLa cells (red line). Clone C10 (green line/green arrow) was produced by single-cell sorting of the mixed population, expanded, and selected because of its high fluorescent signal.

Three assays were set up for identifying active inhibitors of autophagy. For simplicity, the methods detailed in this section are the final protocols used in screening. Details of the assay development strategy are described in Results.

Data were analyzed using Microsoft Excel (Redmond, WA), Spotfire^® (TIBCO, Palo Alto, CA), and GeneData Screener^® (Lexington, MA), as well as automated image analysis using the software provided with the IN Cell as already described. Images and associated data were stored on the Cellomics^® Store database (Thermo-Fisher) and retrieved and further processed for analysis via Screener^®. Compound plate maps were retrieved from ActivityBase (IDBS, Bridgewater, NJ) and associated with image-based data via Screener^®. In the primary assay, compound plates were set up as shown in Fig. (3) and run at 10 µM in singlicate. Averaged intra-plate assay controls (maximal inhibition vs. no inhibition) were used to normalize results obtained with the compounds in each plate, and determine the plate Z’ [6Zhang JH, Chung TDY, Oldenburg KR. A simple statistical parameter for use in evaluation and validation of high throughput screening assays J Biomol Screening 1999; 4: 67-73.
[http://dx.doi.org/10.1177/108705719900400206] ].

Fig. (3) Plate map of controls and compounds for primary screening. Compounds were run at 10 μM in singlicate (320/plate), along with 16 of each control type on every plate. The results from the compounds were normalized as % plate controls, where appropriate for the feature measured; features such as Npass (nucleus number) were treated as raw data.

In the initial set-up for image analysis, HeLa-EGFP-LC3 Clone 10 cells were nutrient starved and then incubated for 2 hours with test compound in the presence of the lysosomotropic agent Hydroxychloroquine (HC). Inhibiting lysosome function with HC blocks the final degradation of EGFP-LC3 in the lysosome [4Klionsky DJ, Abeliovich H, Agostinis P, et al. Guidelines for the use and interpretation of assays for monitoring autophagy in higher eukaryotes Autophagy 2008; 4: 151-75., 5Mizushima N, Yoshimori T, Levine B. Methods in mammalian autophagy research Cell 2010; 140: 313-25.
[http://dx.doi.org/10.1016/j.cell.2010.01.028] ] and thus increases the numbers of fluorescently labeled autophagosomes available for imaging. Cells were imaged on the IN Cell Analyzer 3000, an automated, line-scanning confocal cell imager, with on-the-fly cell feature analysis using the Granularity Analysis Module GRN1. In this algorithm, cells are first identified by their nuclear stain and a measurement region for each cell is defined from the nucleus. Intense fluorescently areas (‘grains’) within this region are identified and analyzed (Fig. 4). In our assay, these grains represent the accumulated EGFP-LC3 labeled autophagosomes. The algorithm reports features for each cell selected for analysis in the well and well-level statistics are computed from these cell-level features. The features selected for analysis are shown in Table 1.

Fig. (4) Images of HeLa-EGFP-LC3 cells after 2 hr starvation; cells were stained with Draq5 to indicate nuclei (red, Channel 1), EGFP-LC3 is shown in green (Channel 2). The algorithm Granularity Analysis Module GRN1 was applied to first identify cell cytoplasm by expanding out a set number of pixels from the nucleus boundary designated in Channel 1 (large box) and then identify EGFP-autphagosomes located within the cell mask, which are counted as “grains” (small boxes). Other measurements are also made, e.g. brightness of objects in each Channel; nuclear size and cytoplasm area. Typical images of the assay controls are shown; the left panel shows cells after 2 hr starvation and the right shows cells after 2 hr starvation in the presence of the inhibitor wortmannin.

Table 1

Measurements Selected for Quantitation from the IN Cell GRN1 Module

As a first step, optimal concentrations of HC and wortmannin were determined. HeLa-EGFP-LC3 cells were set up in 384-well plates as described in Materials and Methods, transferred to EBSS and treated for 2 hr with a concentration range of both compounds as indicated. The images were analyzed to determine the % cells/well with, initially, >4 grains per cell. Incubation with either 5 or 10 µṂ HC produced similar IC₅₀ values for wortmannin (Fig. 5). We chose 33 nM wortmannin to provide the positive (max. inhibition) control. At this concentration of wortmannin, increasing HC was found, as expected, to improve the signal and the Z’ values [6Zhang JH, Chung TDY, Oldenburg KR. A simple statistical parameter for use in evaluation and validation of high throughput screening assays J Biomol Screening 1999; 4: 67-73.
[http://dx.doi.org/10.1177/108705719900400206] ] by inhibiting the dynamic lysosomal degradation of the autophagosomes: 10 µM HC was chosen for inclusion in the assay.

Fig. (5) Dose response to wortmannin and HC in Clone 10 cells. Each point represents Mean ± SD for 6 replicates. HC inhibits the lysosomal degradation of the autophagosomes so that more accumulate for imaging with increasing doses of HC. Wortmannin IC50 was similar at 5 and 10 μM HC (1.4 nM). A dose of 33 nM for wortmannin and 10 μM for HC was selected to produce a good signal: background and best Z’.

Next, the effects on the assay signal of increasing compound incubation times were compared, adding 10 µṂ HC immediately after compound addition. The response to wortmannin was already maximal by 2 hr starvation (Fig 6A). Therefore, we selected a starvation time of 2 hr post compound and HC addition.

Fig. (6) Assay development. (A): Time of starvation of Clone 10 cells in the presence of wortmannin. Points show Mean ± SD for 4 replicates. The signal was stable by 2 hr starvation and was not improved at 3 hr (Table). (B): Titration of number of Clone 10 cells/well, in the presence of wortmannin under starvation conditions for 2 hr. Bars show Mean ± SD for 4 replicates. The response to wortmannin was similar at all cell densities and the Z’ was stable at 0.80 ± 0.04 across the range. A cell concentration of 4,000/well was selected. (C): DMSO titration for Clone 10 cells ± 33 nM wortmannin under starvation conditions for 2 hr. Points are Mean ± SD for 4 replicates. Table: for each DMSO %, the Z’ was calculated from the 33 nM vs. 0 nM wortmannin value; Signal Window was the ratio of the same values. No effect of DMSO was seen ≤ 0.3% DMSO. (D): Effect of binning images of Clone 10 cells, 4000 cells/well in the presence of wortmannin under starvation conditions for 2 hr. The same cell images were used with and without applying 2x2 binning. Points are Mean ± SD for 4 replicates. No effect of binning on the Z’ (0.82 vs. 0.79) or wortmann in IC50 values was observed.

The IN Cell 3000 is a semi-confocal instrument and until this stage we had been using plastic imaging plates, but now switched to glass plates from Matrical, glass being a more suitable medium for high resolution imaging. The number of grains resolved per cell was observed to increase upon this plate change due to the sharper images obtained. We therefore increased the response threshold, from >4 to >6 grains/cell. Finally, a cell number titration and DMSO titration were completed. Similar statistics were achieved for all cell concentrations chosen (Fig. 6B), and DMSO was found to have no effect at ≤ 0.3% (Fig. 6C). Therefore, the screening assay protocol was set to 4,000 cells per well with a final DMSO concentration of 0.3%.

Image binning combines pixels together and sums their intensity. It increases scan and image analysis speeds and reduces the size of data files, and therefore is very beneficial for HTS. However, binning decreases image resolution. To determine if image binning was feasible for this assay, cells were treated as already described and imaged on the IN Cell 3000. The images were then analyzed using either no binning or 2x2 binning to determine % cells with > 6 grains per cell. Binning had no effect on Z’, or wortmannin IC50 values (Fig. 6D) and so 2x2 image binning was adopted for primary screening. Under these conditions each 384-well plate took 30 min to acquire and analyze.

Transitioning from assay development to primary screening involved an assay qualification step, whereby a “Validation Library” of about 8,000 pre-selected compounds was run on two separate days using reagents and equipment configured identically with the planned screening process. This step allowed us to check the reproducibility of compound activity in the assay, get a projected expected hit rate, and also assess the plate to plate variability in terms of Z’, before committing to an entire screen. A reasonable correlation was observed between % activities of compounds on the 2 days (Fig. 7A). From our assay development we were expecting a Z’ for controls well above our minimum acceptable level of >0.4 for a cell-based screen, and this was achieved for all plates, with an average Z’ of 0.66 (Fig. 7B). The assay was accepted into HTS and went on to test around 240,000 more compounds in the primary screen. The screening assay was run with the aid of a small robotics assembly consisting of a Thermo-Fisher CatX robot and plate “hotel” feeding a Combi and the plates were read off-line on the IN Cell 3000, usually overnight. Statistics for the primary screen followed those of the validation screen closely, with the excellent plate Z’ being maintained and very few repeat assays required. With a cut-off of >30% inhibition of grain formation, a hit rate of 2.8% was obtained, about 10,500 compounds (Fig. 8).

Fig. (7) Assay Qualification. Left: Correlation plot in Spotfire from % inhibition of autophagy on Day 1 vs. Day 2 of the Validation Screen. From 8,000 compounds (25 plates) screened and setting a threshold of >30% inhibition of the positive control, the hit rate on Day 1 was 3.0% vs. 3.2% on Day 2. Confirmation of positive compounds between Day 1 and 2 averaged 69%. Right: Z’ for control wells of individual plates (n=16 for each control, every plate) run for assay validation on Day 1 (pink) and 2 (blue). All plates were >0.5, with the Mean Z’ being 0.66. These Z’ were predictive of the HTS assay itself, indicating a robust assay. Where Z’ <0.5, the plate was repeated.

Fig. (8) Primary screen - Spotfire plots of results. Left: scatter plot (each point is a separate compound) with normalized results for primary screening (% inhibition of autophagy vs. compound tested). Horizontal lines represent Mean with ranges ± 2 x SD (Mean = 1.8 ± 15.9). Compounds were selected positive at ≥ 30% inhibition and are marked in red, rejected compounds in blue. Right: values binned as a histogram of activity distribution of all compounds marked as for the Scatter plot; inset is an expanded view of the distribution of selected positives.

As with many large compound libraries, a certain number of fluorescent compounds interfering with the GFP or Draq5 readout through actually binding to cells are present in the collection. Also, a number of compounds will be significantly cytotoxic even in a 2-hr assay. Example images of such compounds are shown in Fig. (9), A-F. Quantitative cell image analysis allowed us to set additional criteria to identify these categories. Therefore, following the completion of primary screening, fluorescent, quenching and acutely toxic compounds were picked out by applying a secondary data analysis. Highly fluorescent compounds staining cells are easily identified since they produce an increase in Icyt (average GFP intensity in cytoplasmic region), and this is often accompanied by an increase in Dmrk (average nuclear diameter) because fluorescent compounds can interfere with nuclear mask determination. GFP quenchers produce a decrease in Icyt values. Toxic compounds can be identified using the features Npass, Icyt and Dmrk. Acutely toxic compounds reduce the number of cells analyzed per well (Npass), and compounds that do not reduce cell number but produce undesirable morphological changes such as nuclear shrinkage and cell rounding, will decrease Icyt and Dmrk values significantly as compared to controls. Such false positive or undesirable compounds were filtered by eliminating those that reduce raw Npass values in 4 fields of view to below 200 cells (this represents about an 80% reduction in cell number) or increase, or decrease, normalized Dmrk and Icyt values by more than 3 standard deviations from mean control values (Fig . 9G). Example data from such compounds are shown in Table 2.

Fig. (9) Compound qualification: example images (20x) of cells after treatment with compounds excluded through the secondary data analysis; Green – EGFP channel fluorescence, Red – nuclear channel fluorescence. (A): Positive control (0.3% DMSO); (B): Negative Control (33 nM wortmannin); (C): Fluorescent compound; (D): EGFP quenching compound; (E): very toxic compound; (F): somewhat toxic compound. Data associated are shown in Table 2. (G): Re-analysis of primary screen image data to remove fluorescent compounds. Compounds were removed that increased the normalized Dmrk (average nuclear diameter) or Icyt (average intensity in cytoplasmic region) values by more than 3 SDs from mean values (Mean = 0.06±6.05 – horizontal lines).

Table 2

Example Data for Filtering Compound Interference Using 4 Features

After these filters had been applied, about 30% of the compounds were eliminated and a selection of 7,500 compounds were re-tested in triplicate at 10 µM in a confirmation of the primary assay, with the same ≥ 30% inhibition cut-off (Fig. 10). From this we confirmed about 2,000 (27%) of the compounds. Interestingly, this low confirmation rate was directly brought about by our original rather “relaxed” hit criterion of ≥30% inhibition: only 13% of compounds with between 30 and 50% inhibition in the primary assay were confirmed, whereas 46% of compounds with > 50% inhibition confirmed (data not shown).

Fig. (10) Confirmation of the primary screen compounds (Y axis) vs. viability counterscreen (X axis). Compounds were run at 10 μM in triplicate. Data are normalized as the % of controls on each test plate; each point is the Mean of 3 replicates. Unconfirmed screening hits are marked in blue; confirmed positives are marked in red. Compounds with ≥ 50 % inhibition of cell viability in the counterscreen are in the green box. Of the confirmed positives, 122 showed ≥ 50% inhibition of cell viability (red points in green box) and were eliminated from further consideration.

The primary screen selected compounds that inhibited autophagy when the cells were placed in starvation medium, which would ultimately lead to cell death although not in the 2 hr time-frame of the assay. Acutely toxic compounds that eliminated cells were removed in the secondary data analysis, but we also wished to eliminate compounds that were toxic to cells not undergoing autophagy and over a longer time period. Compounds successfully reaching the confirmation stage were therefore further tested for toxicity occurring in full growth medium, again at 10 µM in triplicate, using the WT HeLa cell line growing in full growth medium for 24 hr and a standard cell viability assay. Those compounds showing >50% toxicity (a further 122) were eliminated by this assay (Fig. 10).

At this stage, the remaining compounds were examined for structural attributes and their previously documented activities in other screens. Those with undesirable properties (frequent hitters, non-specific inhibitors, undesirable chemical structure, etc.) were identified and around 350 were removed from consideration.

Confirmed compounds (~1,400) were placed in a 10-point dose response (in triplicate) from 30 µM to 1.5 nM. Those compounds that were demonstrably dose responsive and also achieved a maximum activity of >30% inhibition were considered active, and about 1,200 compounds fit these criteria. A range of activities were observed (Fig. 11) from nM to µM. Nearly 300 compounds showed nM activity.

Fig. (11) Distribution of IC50 values of ~1,400 compounds tested in a dose response ranging from 2 nM to 30 μM. About 1200 compounds gave IC50 values of < 30 μṂ and of these, 280 compounds gave sub-μM values (green); 79 compounds were classed as inactive (IC50>30 μM) (orange).

Finally, we added a longer-term cell death assay into the screening tree in a second cell type, the tumor cell line H1299, to identify compounds that selectively brought about cell death under starvation conditions but not under normal growth conditions by 48 hr. From this about 400 compounds that were not toxic for either cell type in full nutrient conditions and only caused loss of viability under starvation conditions were identified (Fig. 12) and classified as “Active Specific”.

Fig. (12) Left: Dose response IC50 values (M) of ~1,200 compounds which were designated Active in the Primary and Counter screen assays in EGFP-LC3-HeLa cells plotted vs. their cytotoxicity activity in the Secondary starvation assay in H1299 cells. So that they appear on the plot, compounds are set to 100 μM activity if inactive (red, n = ~450), and those with IC50 values > 30 μM are set to 50 μM (n= ~50). For the two secondary assays in H1299 cells, compounds designated Active Specific (green) were compounds which gave IC50 values in the starvation medium which were 3x or more lower than in the parallel assay in full growth medium (~400 compounds). Active Non-specific compounds (yellow) have similar IC50 values in both formats. Right: Examples of compound dose responses obtained in the secondary assay (pink: starved; blue: full growth medium). (A): Active Non-specific, (B) and (C): Active Specific.