UHPLC Resolution System Suitability Specification Revision?

[carls]

I have come across a few examples over the past few years where normally acceptable variability in retention times (<5%) gave resolution values that fail system suitability limits (Rs>2). It seems the small peak widths obtained with UHPLC (and, therefore, the smaller difference in retention times) places higher demands on batch to batch reproducibility and UHPLC instrumentation. Below is an example using gradient elution where a 2.5% (or 14s) change in relative retention for a compound dropped the resolution value from 2.6 to 1.6:



The smaller peak widths in case 1 (2% and 4%, respectively) do not compensate for the change in retention/selectivity in case 2 (~2.5%).

The fact that a relatively small change in relative retention (~2.5% or 14s) can reduce the resolution value by a large amount (~50%) makes me think system suitability criteria should be reviewed in light of UHPLC capabilities.

Any comments or opinions?

[Rob Burgess]

A very interesting query you raise here. Are these results produced on the same or different instruments? What instrument is it may I ask? If it was on different instruments I'd be looking to answer why the relative drastic differences. As I understand it you are seeing some selectivity differences between the two runs that could point to differences in pump performance? I also presume these two sets of data are run on exactly the same column?

[carls]

The data in both cases were obtained on the same Agilent 1100 instrument (optimized for high performance) collected a few days apart. Cases1 and 2 are results from two different batches of sorbent both in 100x4.6mm format. The results for each column showed good reproducibility.

While there is a selectivity difference between the 2 batches, one objective of this post is to start a discussion regarding the "relatively small differences" in retention/selectivity that can result in drastic changes in resolution. The other objective is to determine if system suitability should be revised for UHPLC separations.

Perhaps Rs values obtained during method development should be increased to allow for these minor batch to batch and instrument to instrument differences.

[tom jupille]

"Relatively small" variations in selectivity can *always* have a large impact on resolution. Resolution is proportional to (alpha - 1)/alpha. UHPLC is no different than HPLC in that regard. Column-to-column variation is an unfortunate fact of life, which is why the compendia have moved toward allowing adjustment of method parameters (within default limits) in order to compensate.

A couple of additional comments:
- variation in retention time by itself is not that much of a problem if both peaks move in parallel.
- in a gradient system, the performance of the gradient proportioning system can have a huge effect on retention times (and selectivity). Given that there were several days difference between the two runs, have you verified that the system was performing the same on both days? By the way, I suspect that it was and that the difference is one of column variability.

[Rob Burgess]

With respect to setting system suitability citeria for this (or any method) then you would want to judge the merits of the criteria from a number of different runs to gauge an average figure. Have you tried more than two columns? IMHO the selectivity differences between these two columns would be unacceptable and I would look around for alternatives that didnt give such a difference.

[carls]

Tom, Rob,

Thanks for the replies.

What I find interesting is the high efficiency of UHPLC columns raises the bar not only on column and instrumentation manufacturers but also on method development scientists.

Assuming UHPLC sorbents give 2x higher plates counts than 3um columns then the allowable changes in selectivity/instrument accuracy&mixing giving a comparable change on a 3um sorbent are reduced by ~30%. This means method robustness, particularly control of secondary/unwanted interactions, must be investigated and controlled even more rigorously.

Is it realistic to expect that level of improvement or is it more realistic to relax requirements?

I agree the use of QbD in method development will help improve method robustness as well as allow modifications when necessary.

[carls]

I should also mention these batches were the extremes in a large number of batches. The vast majority of batches gave results within 0.2 Rs units of one another. Additionally, these are not optmized conditions and the analytes are not ideal (e.g. simple aromatic hydrocarbons).

[tom jupille]

Is it realistic to expect that level of improvement or is it more realistic to relax requirements?

Ideally, the system suitability resolution requirement is set by the need to reproducibly measure the peak areas (or to keep the FDA happy, but that was their original intent ) -- which means that unless somebody included a whopping big safety factor, you really can't relax the resolution requirement without compromising quantitation.

Based on what I've seen, UHPLC problems are much more likely to be linked to plumbing issues affecting efficiency (plate number) than they are to column chemistry issues affecting selectivity (alpha values); I'm not aware of any reason why small-particle columns should be more variable than larger-particle columns. In your specific case, I suspect that it's just a non-robust method with respect to column chemistry!

Method transfer between "HPLC" and "UHPLC" systems has its own set of problems, including the availability of identical column chemistry in both particle sizes and pressure-related selectivity changes. In any case, most of the UHPLC methods that I've seen focus on generating the same resolution (as HPLC) in a shorter time. In those cases, the relative contributions of selectivity and efficiency don't change. That said, they *do* place greater demands on user skills (kind of like driving a Ferrari on a racetrack at 300 km/hr versus a Corolla on the highway at 100 km/hr) .

[carls]

I should have been more specific in that statement.

What I "meant" was perhaps larger resolution values (>=3) should be targeted during method developement for UHPLC methods (than HPLC methods) due to the increased consistency required of the sorbent, instrument and conditions. I believe any loss in the speed offered by UHPLC by adopting larger Rs targets is more than compensated for by improved robustness.

Perhaps I'm missing something, but from the resolution equation, as N increases, smaller alpha values are required to maintain resolution. Working at smaller alpha values means smaller variations in selectivity will give larger variations in resolution, i.e. tighter control required in UHPLC. I showed data above demonstrating the consistency required of UHPLC sorbents is significantly higher than that of even 3um sorbents. Again, maybe I missed something.

Thanks for the input!

[tom jupille]

No you didn't miss anything, in fact as I re-read the whole dialog I see that *I* missed what you were driving at at first.

Still, I can see two limiting cases:
- UHPLC is used to improve speed at a given resolution (and pressure). In that case, the plate counts will about the same as in an HPLC method (somewhere around 10k for a good column on a good day), but they will be generated in less time by a shorter column, and the selectivity constraints will remain about the same. This, by the way, has characterized the evolution of HPLC over the past four decades: from 30-cm long columns with 10-micron packings generating about 10k plates in 20-40 minutes to 10-cm long columns with 3 micron packings generating about 10k plates in 3-5 minutes (in both cases for simple isocratic separations).
- UHPLC is used to determine resolution via increasing the plate count. In that case the run times will stay about the same, but somewhat longer columns will be used. Getting *huge* plate counts is still impractical on a day-to-day basis, with a doubling (to about 20k plates) about what you can expect (yes, you can get into the 40k range, but that's at the bleeding edge). A back-of- the envelope calculation suggests that for Rs = 2.0, a 10k plate column will need an alpha value of 1.08, and a 20k column will need an alpha value of 1.06.

Here's the table of required alpha values for a given resolution and plate count:


At this point, I suspect that 20k plates is about as high as most people will go, so the vulnerability to selectivity variations is there, but isn't catastrophic. You're right that it's a tighter constraint, but that's just one more of the "gotchas" with the small-particle columns (to go with fittings and filtration!).

[carls]

I have briefly discussed this with a few other people and their eyes started "glazing over" so I appreciate your comments.

I made a few different assumptions based on what I have seen over the last year and the results are shown below. I like to look at the difference in alpha (selectivity) between pass (Rs 2.0) and fail (Rs 1.5). This delta is 0.010 alpha units for UHPLC while it is 0.015 units for HPLC with a 3um sorbent. Therefore, for UHPLC the allowable delta is 30% lower than that for high efficiency HPLC.



I feel the time where this will become an issue is not too far off.