Single Well Outlier in qPCR Triplicates: Exclude It or Report With Error Bars?

You have three technical replicates. Two of them give you Ct values of 22.3 and 22.5. The third says 24.1. You know the answer you want — drop it and move on. But should you?

In most cases, a replicate that deviates by more than 0.5 Ct from the other two in a triplicate set is a legitimate candidate for exclusion, provided you can document a plausible reason (pipetting error, bubbles, poor well sealing, position effects on the block edge). The key is having a rule you set before looking at your data and applying it consistently — not just to the wells that inconvenience your hypothesis. If you don't have a pre-set rule, keep the outlier, report the mean of all three with the larger error bars, and let the data speak honestly.

What Counts as an Outlier in Technical Replicates

Technical replicates in qPCR exist to catch pipetting and instrument variability, not biological variability. For a well-optimized assay with decent pipetting technique, you should expect your triplicates to land within 0.3 Ct of each other. A CV of less than 0.5 Ct across triplicates is a reasonable benchmark. Once a single well deviates by more than 0.5 Ct — especially when the other two agree tightly — something likely went wrong mechanically.

Here's the intuition behind the math: a difference of 0.5 Ct corresponds to roughly a 1.4-fold difference in apparent starting quantity (2^0.5 ≈ 1.41). That's bigger than what normal pipetting jitter should produce when you're moving 1-2 µL of template into a 20 µL reaction. A full 1.0 Ct deviation means one well is reporting double (or half) the quantity of its siblings. At 1.8 Ct off, like in the opening example, the outlier well is reporting ~3.5-fold less template. That's not pipetting noise — something went wrong in that well.

Common culprits for single-well deviations:

• Partial aspiration or short pipetting — you got 0.5 µL of template instead of 1.0 µL
• Air bubble trapped under the seal — affects fluorescence reads or thermal transfer
• Seal not fully adhered over that well — evaporation shifts the Ct later
• Edge effects — wells on the perimeter of a 96-well block run slightly differently on some instruments, particularly older CFX96 or QuantStudio 3 units with uneven heating
• Cross-contamination from an adjacent high-concentration well — shifts the Ct earlier, not later

If the outlier is the later Ct (higher number), partial aspiration or seal failure is likely. If it's earlier, contamination or a splash from a nearby well is more probable. The direction matters for diagnosis, even if the exclusion decision is the same.

The Case for Excluding

Dropping an outlier is defensible when you meet three conditions simultaneously:

1. You defined your exclusion threshold before analyzing the data. This is the most important criterion. Write it into your methods: "Technical replicate wells deviating by >0.5 Ct from the median of the triplicate were excluded." Some labs use the Grubbs test (for triplicates with n=3, this has limited power, but it's a defensible statistical framework). Others use a simple fixed threshold. Either works, as long as it's predetermined.

2. The other two replicates agree well. If you have three values of 22.3, 23.1, and 24.1, you don't have "two good wells and one outlier." You have a noisy triplicate, and dropping the highest doesn't fix the underlying problem — it just masks it. Only exclude when the remaining wells have a spread of ≤0.3 Ct.

3. You report the exclusion. In your methods section, state the rule. In supplementary data, flag which wells were excluded. Reviewers may disagree with your threshold, but they can't disagree with transparency.

When you exclude a well from a triplicate, you're left with a duplicate. That's fine for calculating a mean Ct, but you've lost your ability to estimate within-sample variance for that particular sample-gene combination. If multiple samples in your dataset are dropping to duplicates, that's a red flag about your assay, your plate setup, or your pipetting — and you should troubleshoot rather than accept it.

Some labs run quadruplicates specifically to give themselves one expendable well. If you find yourself routinely excluding one replicate per triplicate, consider whether quadruplicates or a multichannel pipette would serve you better than post-hoc data trimming.

The Case for Keeping It

There are situations where you should leave the outlier in and report the wider error bars:

When you have no exclusion rule in place. If you didn't decide on a threshold before the experiment, dropping an outlier after the fact is cherry-picking, even if your intentions are good. Post-hoc exclusion is the single easiest way to introduce unconscious bias into qPCR data. You won't drop the outlier in the control group if it happens to make the fold change look bigger; you'll drop it in the treatment group where it dilutes your effect. This isn't malice — it's human pattern recognition doing what it does. A preset rule prevents it.

When you're comparing across conditions and the outlier doesn't change the conclusion. If your GOI shows a 4-fold upregulation with or without the outlier, just leave it in. The larger error bars actually strengthen your paper: they show the effect is robust to noise.

When the "outlier" might be real. This matters more for biological replicates than technical ones, but even in technical replicates, consider whether the well might be telling you something. If you see one divergent well specifically in your NTC (say, Ct of 36 when the others are undetermined), that's not an outlier to exclude — that's low-level contamination to investigate.

When the deviation is between 0.3 and 0.5 Ct. This is the gray zone. The well isn't great, but it's within what you'd expect from a slightly imprecise manual pipetting step. Including it and reporting a slightly larger SD is more honest than trimming it to make your replicates look tighter than your technique actually is.

How Outlier Handling Affects ΔΔCt Results

Let's walk through a concrete example to see how much a single outlier well actually moves your final fold-change number.

Reference gene (ACTB): Ct values of 18.2, 18.3, 18.4 → mean 18.3
GOI in control: Ct values of 25.1, 25.3, 25.2 → mean 25.2
GOI in treatment: Ct values of 23.0, 23.1, 24.8 → that third well looks off

With all three treatment wells included: mean Ct = 23.63. ΔCt(treatment) = 23.63 − 18.3 = 5.33. ΔΔCt = 5.33 − 6.9 = −1.57. Fold change = 2^1.57 ≈ 2.97-fold upregulation.

With the outlier excluded: mean Ct = 23.05. ΔCt(treatment) = 23.05 − 18.3 = 4.75. ΔΔCt = 4.75 − 6.9 = −2.15. Fold change = 2^2.15 ≈ 4.44-fold upregulation.

That's the difference between reporting a ~3-fold and a ~4.4-fold change from a single well decision. In a figure, those could look very different. In a manuscript, one might cross a significance threshold and the other might not, depending on your biological replicates (which is where your statistical test should actually be run — on ΔCt values across biological replicates, not technical ones).

This is exactly why the exclusion decision matters. It's not academic. It moves numbers.

A Practical Decision Framework

Here's what I'd recommend for any lab that runs qPCR regularly:

1. Set your threshold before the experiment. Write it into your lab's qPCR SOP. A >0.5 Ct deviation from the triplicate median is a reasonable default. The Grubbs test at α=0.05 is an alternative if you want a statistical criterion, but with n=3 it's not very powerful.

2. Flag, don't silently delete. In your analysis spreadsheet, mark excluded wells in a different color or with a flag column. When you write the paper, state how many wells were excluded out of how many total.

3. Look at the excluded wells in aggregate. If 15% of your wells are being excluded, your assay needs work. Check your pipetting, your plate sealing, and your instrument's block calibration. On a QuantStudio 5, run the built-in background calibration. On a CFX96, check whether edge wells are systematically worse.

4. Never exclude from NTC wells. If an NTC shows amplification, that's data, not noise. Investigate it.

5. Run your statistics on biological replicates. Technical replicate variance should be small enough that the outlier question rarely changes your biological conclusions. If it does, you need more biological replicates, not cleaner technical ones.

If you're processing a lot of plates and want consistent, automated handling of outlier wells, VoilaPCR flags single-well outliers for you based on a threshold you define, marks them in the output, and recalculates with and without — so you can see both numbers and make the call transparently.

The bottom line: exclude if you had a rule, the other two wells agree, and you document it. Otherwise, keep it and let your error bars tell the truth.