Results and Analysis
The Operetta High content imaging system is equipped with a spinning disk confocal microscope and is capable of up to 4 fluorescent colors and 3 different objective lenses. This allows for high resolution light or fluorescence microscopy in each individual well in a plate. The Operetta is also equipped with a robotic arm (Figure 1) which manipulates plates in/out of the imaging system and allows for full automation of the system. This factors greatly into the system’s utility, as a single run can take a huge amount of time to process. For example, the Harmony software that runs the machine estimated that our small run of 10x 384 well plates with 4 images per well would take ~16 hours to finish. You could imagine that larger screens would become uncomfortably time consuming with constantly switching out plates by hand.
The Operetta High content imaging system is equipped with a spinning disk confocal microscope and is capable of up to 4 fluorescent colors and 3 different objective lenses. This allows for high resolution light or fluorescence microscopy in each individual well in a plate. The Operetta is also equipped with a robotic arm (Figure 1) which manipulates plates in/out of the imaging system and allows for full automation of the system. This factors greatly into the system’s utility, as a single run can take a huge amount of time to process. For example, the Harmony software that runs the machine estimated that our small run of 10x 384 well plates with 4 images per well would take ~16 hours to finish. You could imagine that larger screens would become uncomfortably time consuming with constantly switching out plates by hand.
Figure 1. The Operetta imaging system. a) A robotic arm feeds plates in and out of the imaging system while scanning plate information, allowing continual autonomous screening once initiated. b) The trade off for convenience is the danger of the robotic arm becoming sentient and giving extreme hugs.
As
described in the previous post, our assay was designed to find molecules that
would inhibit TNFα induced NFκB nuclear translocation as
a means of discovering a novel anti-inflammatory therapeutic, particularly for
chronic inflammation. With this in mind, we had to set up the screen to examine
parameters that were relevant to this assay, as the Operetta platform is
capable of analyzing a large array of characteristics from apoptosis, cell
cycle, morphology change, internalization, protein expression to even counting
of viral factories. Using the Harmony
software, a few wells from a random plate were imaged and strategies for counting
and distinguishing individual cells were established using nuclear and
cytoplasmic (NFκB) stains (Figure 2). Using the intensity of green NFκB-
FITC fluorescence in the cytoplasm vs area overlapping with the nucleus
(stained by Hoescht), a threshold is then set for the program to distinguish
cells that have high nuclear NFκB vs those that do not,
while omitting partial cells or background fluorescence. Once these parameters
were set, the run was started with plates A-J taking 4 image fields per well.
The program then compiled all data collected in each field of view, analyzed
total cell numbers and the percentage of cells with high nuclear NFκB
is reported.
Figure 2. Fluorescent imaging of nuclear translocation in cultured cells. HeLa human cervical carcinoma cells treated with various small molecule inhibitors were treated with TNF-a then fixed and stained with anti NFkB (FITC) and Hoescht nuclear stain. Images were aquired using the Operetta high throughput imaging system. Orange arrow indicates a low nuclear NFkB cell, white arrow indicates a high nuclear NFkB cell.
Figure 3. Z scores are used to determine 'hits'. Histograms were generated for individual wells on technical replicate plates (A and B) using raw data (%high NFkB) or z-score. Orange and blue arrows represent cut off points for 2 standard deviations below the median for plates A and B respectively. Red arrow indicates the same cut off point in z-score.
Due to the ‘plate effect’, or variability between plates due to handling (by different lab groups in this case), relative humidity, or myriad other factors, a threshold of percent of translocated cells can’t be used to determine if a certain well or drug is a ‘hit’ and blocked NFκB translocation.The difference between distributions is illustrated in Figure 3, where plate A has a much tighter distribution than plate B. The figure also illustrates that choosing an arbitrary % NFkB translocated cells as a 'hit' would not be equal for each plate. Since each plate should in fall into a normal distribution, and in theory each well on individual plates are treated the same, standard deviation is instead used to determine hits. We can see that plates A and B have a similar number of 'hits' when z-scores are used to determine drug efficacy. The median number of percent translocated cells was used to determine a Z score for percent translocated cells for each well, which equates to the number of standard deviations away from the median. We then chose to look for Z scores below -2, which would equate to wells that had a number of high nuclear NFκB cells two standard deviations or more lower than the plate mean. This correlates to roughly 2.5% of a normal distribution, and will allow us to quickly find and sort through low % NFkB translocation outliers.
Due to the ‘plate effect’, or variability between plates due to handling (by different lab groups in this case), relative humidity, or myriad other factors, a threshold of percent of translocated cells can’t be used to determine if a certain well or drug is a ‘hit’ and blocked NFκB translocation.The difference between distributions is illustrated in Figure 3, where plate A has a much tighter distribution than plate B. The figure also illustrates that choosing an arbitrary % NFkB translocated cells as a 'hit' would not be equal for each plate. Since each plate should in fall into a normal distribution, and in theory each well on individual plates are treated the same, standard deviation is instead used to determine hits. We can see that plates A and B have a similar number of 'hits' when z-scores are used to determine drug efficacy. The median number of percent translocated cells was used to determine a Z score for percent translocated cells for each well, which equates to the number of standard deviations away from the median. We then chose to look for Z scores below -2, which would equate to wells that had a number of high nuclear NFκB cells two standard deviations or more lower than the plate mean. This correlates to roughly 2.5% of a normal distribution, and will allow us to quickly find and sort through low % NFkB translocation outliers.
Once
we found all of our ‘hits’, we disregarded any controls found in this cohort
and proceeded to cross-reference our wells to the LOPAC drug library to find
out which drugs were present in our target wells.
To look for the possibility of experimental error, a heat map was generated using z-scores obtained from our assay (Figure 4). Problems could arise from uneven evaporation, heating, other factors that could cause false positives. If there are no adverse effects on plates, hits should be distributed randomly unless the drugs are organized by category and a certain category of inhibitor can block TNF-a induced NFkB translocation.
To look for the possibility of experimental error, a heat map was generated using z-scores obtained from our assay (Figure 4). Problems could arise from uneven evaporation, heating, other factors that could cause false positives. If there are no adverse effects on plates, hits should be distributed randomly unless the drugs are organized by category and a certain category of inhibitor can block TNF-a induced NFkB translocation.
Figure 4. Z scores of Plate A wells.
Z-scores generated from %high nuclear NFkB were plotted relative to their spacial arrangement on the plate. Hits used were z = (-2) or lower. D and B denote wells containing 4% DMSO loading or Bortezomib control treatments respectively.
Another important factor of high throughput screening is consistency. Due to the large number of samples being run, combined with very small cell numbers (5200 cells /well). To this effect, each plate was run in duplicate and then each well was compared for z- scores (Figure 5). In theory, when these points are plotted against each other, if each plate gives identical values, a trend-line with R=1 will be achieved. Since there is always variability, especially in biology, we use this value to assess whether our variation is within an acceptable range where we can trust the assay and thus the generated data. An R2 =0.276 shows that our data is most likely fairly reliable. While there is deviation, the trend for each well tends to be fairly similar, and thus we can accept that our data is within an acceptable range of error. It can be noted, however, that Plate B wells tended to give less extreme Z-scores for the same wells in A, which could be due to the handling of individual plates or other factors previously discussed.
Another important factor of high throughput screening is consistency. Due to the large number of samples being run, combined with very small cell numbers (5200 cells /well). To this effect, each plate was run in duplicate and then each well was compared for z- scores (Figure 5). In theory, when these points are plotted against each other, if each plate gives identical values, a trend-line with R=1 will be achieved. Since there is always variability, especially in biology, we use this value to assess whether our variation is within an acceptable range where we can trust the assay and thus the generated data. An R2 =0.276 shows that our data is most likely fairly reliable. While there is deviation, the trend for each well tends to be fairly similar, and thus we can accept that our data is within an acceptable range of error. It can be noted, however, that Plate B wells tended to give less extreme Z-scores for the same wells in A, which could be due to the handling of individual plates or other factors previously discussed.
Figure 5. R2 analysis of technical replicate
results. Z scores for plates A and B were generated and graphed against each other. An R2 goodness of fit test was performed. Trendline indicated by hashed line.
After checking for plate effects, and for reliability of our data, all hits were collected from the dataset. Control wells were omitted, yielding the remaining experimental hits (Table 1). Wells were cross-referenced against the LOPAC drug library and cell images were assessed for morphology. Some cells experienced adverse morphologies such as syncitium formation, blebbing, reduced processes, or complete death (Figure 6) which was factored into selecting 3 potential drugs, since it can give clues to how the drug works. For example, some hits were discounted because they killed the vast majority of cells, preventing therefore preventing proper NFkB translocation. In addition, it is likely these compounds would likely be too toxic to an animal if they kill tissue culture cells so quickly. Using this information, and specific drug activity found in the LOPAC library (Table 1) and in the literature, 3 hits were selected as molecules of interest identified as myrecetin, stattic and Bay11-7082.
Figure 6. Morphologies of HeLa cells treated with various small molecules. Images from each hit were visually assessed for morphologies like a) healthy, b) shrunken, c) blebbing/ apoptotic, d) syncitia formation.
Table 1. Experimental hits garnished from the TNF-a inhibition small molecule screen. Hits listed after bortezomib control hits were removed from the pool. Hits from whole class data shown with z-scores for all technical replicates. Selected compounds of interest highlighted in green.
In addendum - more details of hit selection and molecule activity to come!
Many of the candadates that prevented NFkB translocation, while not adversely affecting cell health, were generally not selected due to a lack of specificity, or because of potential off target effects. For example, chelerythine chloride was not chosen because it is a protein kinase C inhibitor. Since these kinases are involved in a huge array of cellular processes, and broad spectrum blocking of these enzymes would likely cause huge complications in animal models. It is also unlikely that the structure could be easily modified to be specific for a certain NFkB related PKC (like PKC-theta). Another example of a drug with off target activity is BTO-1, which inhibits Polo-like kinases that are involved in NFkB translocation, but are also involved in mitotic spindle formation.
Myricetin, Stattic and Bay11-7082 were selected for their activity directly in the NFkB pathway. Myricetin blocks casein kinase II which phosphorylates IkB and allows it to be targeted for ubiquitin dependent degradation. Stattic inhibits STAT3 activation, preventing it from enhancing NFkB promoter binding. Finally, Bay 11-7082 inhibits IkB-alpha phosphorylation. IkB binds to NFkB binding sites in the nucleus, blocking transcription. Phosphorylation leads to degradation and frees these sites. Therefore we propose these candidate drugs for potential therapeutics.
In addendum - more details of hit selection and molecule activity to come!
Many of the candadates that prevented NFkB translocation, while not adversely affecting cell health, were generally not selected due to a lack of specificity, or because of potential off target effects. For example, chelerythine chloride was not chosen because it is a protein kinase C inhibitor. Since these kinases are involved in a huge array of cellular processes, and broad spectrum blocking of these enzymes would likely cause huge complications in animal models. It is also unlikely that the structure could be easily modified to be specific for a certain NFkB related PKC (like PKC-theta). Another example of a drug with off target activity is BTO-1, which inhibits Polo-like kinases that are involved in NFkB translocation, but are also involved in mitotic spindle formation.
Myricetin, Stattic and Bay11-7082 were selected for their activity directly in the NFkB pathway. Myricetin blocks casein kinase II which phosphorylates IkB and allows it to be targeted for ubiquitin dependent degradation. Stattic inhibits STAT3 activation, preventing it from enhancing NFkB promoter binding. Finally, Bay 11-7082 inhibits IkB-alpha phosphorylation. IkB binds to NFkB binding sites in the nucleus, blocking transcription. Phosphorylation leads to degradation and frees these sites. Therefore we propose these candidate drugs for potential therapeutics.
