Look at attached chromatogram- peaks or noise?

[EmpowersBane]

Im running a related substances method using a gradient method that starts 13% Acetonitrile in A Line and 87%Buffered Mobile phase in B Line over a 42 min run time. I need to use a low wavelength of 225nm for extraction. At 225nm the baseline can be a bit noisy so its important I don't mix peaks up with noisy baseline. Problem is, even when I set my peak width and threshold to capture all relevant peaks, there is still some integration near the end of the gradient (after 38 or so mins when it starts to "return to initials" of 13:87).
My problem is when I overlay my sample with my blank for this region, although both plots are noisy, they don't match up enough for me to conclusively write off the peaks as "Blank" or "Over integrated baseline" and even though the UV profile of said peaks don't match my active peak or any of its related compound. My data reviewer insists "Impurities could be anywhere in the chromatogram even near the end, you cant prove its blank or noisy baseline".

Can anyone suggest a way to make a compelling argument to disregard these peaks? Thanks in advance.

[Peter Apps]

If they are baseline noise they will not be repeatable from run to run in replicate injections from the same vial. If they appear in replicates they are in the sample - therefore impurities.

Peter

[HPLCaddict]

Is there a reporting threshold or disregard limit or whatever defined in the method? If yes, set the minimum area accordingly - probably this would exclude these noise/peaks.
There is no reporting threshold/disregard limit? Bad luck, there should be one

[EmpowersBane]

Thanks very much for the replies.

I don't know if they are repeatable as we only perform one injection of the sample so I don't get to see further injections. The baseline in the blank is noisy in that region as well, it just doesn't match exactly when overlaid, making it hard for me to write off any peaks as blank.

Yes we do have a disregard limit but those peaks are higher than that so they have to be acknowledged. The method is specific so we have established that there are no peaks that routinely show up in the 38-40 min region, plus the peaks can be an odd shape as if its taking the natural rise in baseline of a gradient as the starting point of a peak, resulting in an elongated "peak" with no distinct UV profile. Surely at that point in the gradient I can eliminate any peaks because this condition at 38 min (30:70) was ALREADY experienced at 25 mins so any peaks that may elute under this condition have already eluted?

[lmh]

You're not going to want to hear this (or your boss isn't, anyway) but if those peaks are bigger than your disregard limit, then your method isn't adequate for purpose.
I'm completely unconvinced that peaks that size, on a base-line that wobbly, will be reliably detectable by the integrator; they're the sort of size where tiny variations in integrator settings will affect whether they're found or not, and how big they are, and the same chromatogram integrated by the algorithm of another manufacturer would give a different answer.
The disregard limit implies that any peak bigger than the limit is important enough that it must be detected and reported. If the method/equipment can't reliably see peaks at the disregard limit, then the limit is pointless (what I really mean is that the disregard limit must be at least as good as the LOD, and those peaks, I think, are unlikely to be as large as your LOD).
Given that the peaks are after the end of the gradient it's unlikely they're relevant, but unfortunately you can't ignore data you've collected. A less fastidious person might have turned their detector off at the end of the rising part of the gradient!
As a matter of interest, I have very occasionally encountered genuine peaks that don't come off during the gradient but come out later. It can happen where an analyte is very bi-functional, and can stick to the column by two different mechanisms (e.g. some sort of hilic or silanol interaction at high organic and reverse-phase hydrophobic interactions at low organic). This can give it a U-shaped curve of binding-versus-%B, meaning that it moves only during the middle stage of a gradient. During the gradient it moves onto the column - as the gradient goes back again, it moves a bit more (and maybe comes off), and during the next run it carries on again and probably comes off!

[tom jupille]

A few comments here:
1. It may look like something is happening "after" the gradient, but remember that what the detector sees is offset from what the pump generates by the column dead time and the system dwell time (gradient delay time). Depending on your system, those offsets may be several minutes.
2. As has been pointed out, you really need to run duplicates. If the "peaks" show up at the same times, then the chances are that they are real. That said, the fact that they don't show up in the blank suggests that they *are* real.
3. However, a reasonable working estimate of LOD ("limit of detection") is a peak 3X the baseline noise. Just eyeballing the noise in your blank, it looks to me like those peaks would be below LOD.

[James_Ball]

I would overlay 3 injections of the blank and 3 injections of the sample and see if the replicates for each reproduce in that region. You may find that you have the same "peaks" in one of the blanks and they are missing in at least one of the sample replicates. If so then they are noise. If the replicates of the blank and sample reproduce results then you could possibly call them peaks in the sample.