Negative intercept for Chloride analysis by IC

[philcous]

Hi.
My lab is validating a chloride analysis using IC for the linearity we are running 5 points 10, 20, 25, 30, 40 ppm. We get a respectable regresion value but a negative intercept approximately 10% of the 25 ppm level.
The usual parameters are
Column : AS4A-SC with guard (250 x 4mm) @ 30C
Detection: DS6 heated conductivity @ 35C
Eluent: 1.8mM Na2CO3/ 1.7mM NaHCO3 @ 1.8ml/min
ASRS 300 Suppressor @ 24mA (recycle mode)

The peak size is good for the lowest and highest levels not sinking into the baseline or overloading.

I understand how a positive intercept can occur but I'm pretty stumped by a negative one.

Has anyone seen this before or can anyone offer any advice?

Cheers Phil

[tom jupille]

I would guess that you have some chloride contaminating your mobile phase.

[philcous]

Thanks Tom.
I've had a look at the possibility of Chloride in the eluent. I made up fresh eluent using new reagents and HPLC-MS grade bottled water. I ran the linearity again getting a similar negative intercept. I also injected a portion of the original eluent which showed no response for chloride.

I have now also extended the linearity using 10 points from 0.5 ppm to 60ppm just incase there was a skew that I hadn't seen before or I was approaching the LOQ. The response is still nice and linear, the LOQ looks like it is a long way below the bottom level of 0.5ppm, 0.5ppm is a good sized peak and well above 10 x the baseline noise. There doesn't appear to be a curve to the linearity.

Phil

[tom jupille]

Hmmm

If I read your original post correctly, you are getting a negative intercept equivalent to about 2.5 ppm of chloride. That implies that when you inject standards containing less than that, you should be seeing a negative peak (which would be indicative of contamination with chloride -- or something else that elutes at the same time). Since you are *not* seeing negative peaks, then the problem lies in your calibration curve.

If this were my problem, the next thing I would do is to look at the standard error associated with the y-intercept. If it's large enough to include zero, then you simply have a lot of imprecision in your data.

If the standard error of the y-intercept is small (i.e., if the non-zero intercept is "real"), then the next thing to do is to run replicates at each level, re-do the least squares fit and look at the distribution of the residuals (which should be more or less even across the entire range).

You may have to use weighted least squares or a log-log fit.

[adam]

I think of two possible explanations.

One would be that you are integrating in such a way as to get a reduced area count for the peaks. For example if the peaks are tailing and you end the peak early (not accounting for all of the tail). This might cause it.

Second possibility is that there is some negative response coming from your diluent. For example, if you inject in water and use a conductivity detector, the conductivity will drop somewhat when the water goes through. If the water coelutes with your chloride this might cause what you are seeing.