CAT | Edge effects
In a previous post on this topic, I discussed ways of preventing edge effects in plate-based potency assays. But how do you know you have them in the first place? Sometimes it can take a long time and many runs to figure out that these problems exist. At that point you may have generated results for several samples and by tampering with the assay you lose connection to historical data or at the very least cause the previous results to be looked at with suspicion.
Instead of waiting for bias to appear in your data, I advocate doing edge effect studies early in assay development. The results you find can be invaluable in determining your plate layout, assay testing strategy and assay analysis.
There are several different ways you can characterize your assay for edge effects during development. A simple and useful experiment is to run a “heatmap” on your plate. This type of experiment is as simple as picking a single dose response, usually from the reference standard, and applying it to every well on the plate. The data is then graphed using a color code to visually pick up patterns in the results and/or statistical tests on the row and column data can be performed.
An important part of this experiment is to pick a suitable dose response. Ideally this response should come from the part of the dose response curve with the steepest slope (see the red circle in the image below). This is to ensure that you will have the optimum sensitivity to small changes in signal.
The resulting data from the experiment can be visualized using a graph such as this one:
As you can clearly see in this "heatmap" the center wells of the plate are lower in signal than the outside of the plate.
Most plate based assays will have some sort of edge effect due to the asymmetrical nature of 96 well plates. It isn’t always clear if the effect will actually have a significant impact on the results of the assay. If the differences in signal between rows and columns are much smaller than the variability of the signal itself, then the effect may not be noticeable in the final results. One way to determine if this is the case is to do a two-way ANOVA for statistical significance. In this case, the ANOVA clearly shows that the effect is significant across both rows and columns (p value is less than 0.05):
Two-way ANOVA: Response versus Row, Column
Source DF SS MS F P
Row 7 9088182 1298312 7.87 0.000
Column 10 4776165 477616 2.90 0.004
Error 70 11540595 164866
Total 87 25404942
S = 406.0 R-Sq = 54.57% R-Sq(adj) = 43.54%
Heat map experiments generate a lot of information about your assay, but it only deals with detecting an edge effect in a single dose. Since the ultimate readout of these assays, the potency, relies on the entire dose response range, you may want to do some further characterization of the edge effect by running the reference standard in multiple sample positions on several plates. The graph below shows the results of such a study where the reference standard was placed in every row of five plates (positions 2-8) and a potency was generated relative to the reference in position 1. The expected potency is 1.0 since every position contains the reference standard. It is clear that in this assay, the potency is severely biased by the position effect. It is much higher in the middle of the plate. This type of analysis can also indicate if there is a bias in the variance of the response. In this case, positions 4 and 5 clearly have much less variability than the others.
As you can see, there is much information about your assay that can be gained from just a few experiments early in the development process. Once you have this information, the assay protocol can be designed to prevent a bias in your results. But that’s a topic for another post.
Thanks for reading,
Dan
If you work in the potency assay field long enough you will eventually run into problems with edge effects. Sometimes they are blatantly obvious, and in some assays you might not notice them at first, but over time they show up as a bias in your potency results.
What causes edge effects? There is not always a straight forward answer to this question and even when you know the answer it’s not always possible to prevent edge effects.
In cell based assays, edge effects are often caused by evaporation in the outer wells of the plate. This leads to changes in salt concentrations in the cell medium, affecting cell metabolism in those wells. They can also be caused by temperature gradients when the plate is placed in a heated incubator; the outer wells tend to heat up faster than the inner wells. Often, this will affect cell attachment and adhesion.
Lundholt et. al. write about an elegant solution here: http://jbx.sagepub.com/cgi/content/abstract/8/5/566. They investigated several different ways to prevent uneven cell-seeding of their plates and they came up with a simple solution. They simply pre-incubated the plates at room temperature until the cells adhered, and then transferred them to the incubator.
Anther common way of preventing edge effects in 96 well plates is to use various humidity and temperature chambers. One such example can be found here: http://www.btc-bti.com/product_literature/stabilitychamber.htm. There are also special gas permeable membranes that can be attached to the cell plate to help prevent evaporation.
Immunoassays can also suffer from edge effects. Because of the shorter incubation times of these assays, evaporation is probably not the culprit, but instead they are often due light exposure and temperature gradients. These causes can be exacerbated in assays where plates are stacked on top of each other during incubations.
Like in cell-based assays, common ways to prevent edge effects in immunoassays include temperature controlled chambers and plate seals. In this case, cheaper plate seals can be used since gas exchange isn’t necessary in assays without cells. Goetz et. al. discusses some of these solutions here: http://gateway.nlm.nih.gov/MeetingAbstracts/ma?f=102268104.html
These are some of the physical causes and preventions of edge effects in potency assays. However, in come cases, none of these help remove the bias from your assay, and your only recourse is to deal with it using statistical measures. This will be the topic of future post. I also want to try to convince you that waiting for these effects to show up during testing isn’t necessary, instead I will write about ways to discover the bias of edge effects during assay development.
Thanks for reading,
Dan
