Non-linearity of UPLC method

[aceto_81]

Hi,

While developing a UPLC method for an API, I came accross something strange.
My sample concentration is 0.09mg API/ml, and I would like to determine my related compounds on the same solution as for my assay.
This means that my method should be linear between 80% and 120% of that concentration for the assay, but also from 0.1 - 5% as this is the range for my related compounds.

The lower range is giving me some serious troubles, here are my areas and concentration:

% Response
0.1 2459.051963
0.25 6745.959772
0.5 15412.28079
1 35414.32085
3 124463.9612
5 216200.0982

If you plot this data (I don't know how to post a graph ), you will see that the lower end of the data isn't linear.

Any ideas about how this come, and more import, how to solve this problem?

Best regards

Ace

[unmgvar]

what is the injection volume?
are you diluting the concentrations?
can you tell us your steps for the different std preparation of the conc.?

[fandaga]

I plotted it and got a linear regression of

y = 43945x - 5360.4

R^2 = 0.99914

The correlation coefficient isn't too bad. There is some slight bias for the y-intercept. How is the low level recovery? Perhaps, there is a matrix effect at low levels? Also, did you do triplicate injections at each level?

[aceto_81]

@ unmgvar:
injection volume is 10µl (full loop injection)
For my concentrations:
I made up a stock solution of 1mg/ml, which I further dilute x ml / 100ml
One of these solutions is then further diluted to give the lower range.
I have made up such solutions in the past without any problems (for other projects), and I repeated all this dilutions already 3 times to verify if something went wrong with my dilutions.

@fandaga:
if you take a close look at the 3 lower points, you will see that it isn't linear.
correlation coefficient isn't bad, but has nothing to do with linearity (see previous discussions on this board). There is a bias for the y-intercept, which even increases when you ignore the lowest points.
If you see %residuals for the lowest points, this is unacceptable.
a matrix effect is possible, but when I inject blank solution, I don't see any peaks.
Assume there is a matrix effect then my y-intercept should be positive?
I didn't triplicate injections for each level, but I already made up my solutions 3 times, all with the same result, so the problem isn't injection accuracy. Pipette handling could be the cause, but I already did such things many times without any problem, so I don't think that's the problem.
But to be sure, a colleague has made up the solutions, and also got the same non-linear regression line.

Thanks for your time

Ace

[aceto_81]

Some more information (just acquired):
When I switch my method from UPLC to HPLC, same buffer, lookalike column (X-Terra MS C18 instead of Acquity BEH C18), all of a sudden my linearity is fine...

Ace

[DSP007]

I plotted it and got a linear regression of

y = 43945x - 5360.4

R^2 = 0.99914

The correlation coefficient isn't too bad. There is some slight bias for the y-intercept. How is the low level recovery? Perhaps, there is a matrix effect at low levels? Also, did you do triplicate injections at each level?

Yes of course it's a good koeffetsient.
But on the other hand
3 = 124
5 = 216 5% mistake
but
0.1=24.5
0.5=144 25% mistake
isn't good.
Two primitive "hands" problems need to exclude
1) The procedure for preparing calibration solutions (volumetric flack & dilution)
2) It turns out there beyond the linearity of the D detector (T> 90%)

Another error, the peak of the lowest concentration a jump (noise jump)
or in solvent may be substace a similar retention time

[aceto_81]

....
Two primitive "hands" problems need to exclude
1) The procedure for preparing calibration solutions (volumetric flack & dilution)
2) It turns out there beyond the linearity of the D detector (T> 90%)

1) Stock solution = 1 mg/ml, 100% solution = 9ml stock solution diluted to 100ml.
5ml of 100% solution to 100ml = 5% solution
3ml of 100% solution to 100ml = 3% solution
1ml of 100% solution to 100ml = 1% solution
10ml of 1% solution to 20ml = 0.5% solution
5ml of 1% solution to 20ml = 0.25% solution
2ml of 1% solution to 20ml = 0.1% solution
even the 0.1 - 0.25 - 0.5% range isn't linear, so it isn't an serial dilution problem in case you might think.

2) you mean above the linear range of the detector? I doubt, because the range 80-120% of the concentration is linear (11ml stock solution to 100, 10 ml stock solution to 100, ... 7 ml stock solution to 100)

Thanks for your response

Ace

[unmgvar]

we are talking of a UPLC acquity right? not the new system right?
this is the one with the pulled loop system, so it is importan that you check that "flush", "wasted" volumes are well set.
i would check the repeatibility of the system.
inject each of the 4 lower volumes 5-6 times and check RSDs
if I am guessing right and it is the pulled loop design, the lower ones will look terrible and the higher conc should be good.

[aceto_81]

we are talking of a UPLC acquity right? not the new system right?
this is the one with the pulled loop system, so it is importan that you check that "flush", "wasted" volumes are well set.
i would check the repeatibility of the system.
inject each of the 4 lower volumes 5-6 times and check RSDs
if I am guessing right and it is the pulled loop design, the lower ones will look terrible and the higher conc should be good.

We are talking about the good old Acquity UPLC, not the new H-class, injecting full loop.
Repeatibility is ok, RSD's on the lower points is 8-10, but this is normal for near LOD/LOQ concentrations.
It's not a reproducibility/repeatibility problem, it has something to do with the injector mechanism and interactions with washes, or sticking somewhere, I don't know...

Ace