Problem with Reproducibility - #2

[tpmcginnis]

Hello. I'm experiencing a reproducibility problem somewhat similar to Damien's in the previous thread, but with a twist. Here's the rundown;

First off, I know I have an injection volume issue, as I'm using a GC PAL autosampler, and the syringes we've been getting are extremely substandard. It goes primarily to the manufacturer of the syringes, and the area reproducibility for a given peak/standard has a %RSD of 10% on ten replicate injections of the same standard. Obviously not good at all. But with that said, I am using on-column injection with very simple non-polar compounds on a simple non-polar column, and I would expect area ratios to not be a problem. They are.

Injection solution is ~ 1000 ug/L each in C11 and C22 alkanes, dissolved in MeCl2. Injection volume is 1 uL.

Instrument is HP5890 Series II, with EPC, and cool on-column injection.
Column is 20 m x 0.53 mm DB-5, with 1 um film thickness.
Carrier is He @ 10 cc/min

Inlet is programmed as
60 C initial temp., held for 0.5 min.
Ramped to 320 C @ 100 C per min., with zero hold time

Oven is programmed as
60 C initital temp., held for 1.5 min.
Ramped to 320 C @ 10 C per min., no hold time

Detector is FID @ 320 C
Detector gases are He makeup @ 20 cc/min., H2 @ 30 cc/min., Air @ 400 cc/min.

(I should note that I've been using this instrument for years, with similar ramps and rates, flows, etc. without problems. I like to use the on-column inlet with the thin film non-polar megabore columns because I often analyze TMS derivatives of wide molecular weight range samples vs. internal standards.)

My peak shapes are excellent. Visually everything looks very good. However, the area ratios of the C22 peak to the C11 peak are as follows;

0.890546373
0.849581975
0.857527365
0.843012247
0.824673196
0.833617403
0.917500427
0.923461612
0.862258644
0.91404731

(please forgive the inordinate amount of sig figures)

The %RSD of these is > 4%. As I mentioned earlier, I know I have non-reproducible injection volumes. However, the error I see there should not come into play for area ratios. Also, flash vaporization should not be an issue with this system in the cool on-column mode. That's the beauty of the on-column design, as long as the initial inlet T is appropriate.
The construction of the inlet for a 5890 on-column system really doesn't lend itself for compounds to be "lost".

There is no undue noise in the baseline. Retention times are reproducible.

Does this seem like a FID problem? Or perhaps an electrometer or main board problem? The instrument is quite old (~ 15 years) but the beauty of some of these 5890's is that they seem to run forever.

I don't think I've overlooked anything obvious, but if I have I'm more than willing to accept that I'm wrong!

Any thoughts or replies are greatly appreciated.

Best Regards to all.

[Peter Apps]

Tricky one. In principal an on-column injection is the most robust of all. Just to be sure, because there have been threads where other injection techniques were described as on-column - the end of the needle goes into the top of the column ?

You are actually doing a vapourizing injection, since the BP of dichloromethane is 40C and you inlet is at 60C. Getting everything cooled to 40 will take a lot of time, so how about using a higher boiling solvent ?

What are the mechanics of the 5890's on-column inlet, does it hold pressure while the injection is being made ?, if not how long does the pressure take to get back up ? As an outside chance I see a possibioity that if the inlet pressure is low the flow through the column reverses during injection, and then you get chromatographic migration of the solutes back out of the column as the inlet heats (the major puzzle with your problem is how to explain an eratic discimination between two compounds). Does it help if you delay inlet and oven heating for say an extra 2 min to give the pressure a chance to stabilise ?

Peter

[tpmcginnis]

Thanks for the reply, Peter.

Yes, I agree that there is a vaporization effect at 60C, and 40C is a problem with cool down time. I've got time, though, so it probably is a good excercise to re-perform the analyses with an initial column and oven temp of 35C. The instrument can run overnight.

The design of the 5890's "cool" on-column inlet is really simple. It's just pretty much a heating block with a hole drilled in it. The megabore column extends up into the block and the column end is placed up flush against a small metal sleeve that is concave. There is a small spring on the other side (right below a small septum). You put enought pressure on the column during insertion from the underside to feel the pressure from the spring, and then tighten down the nut. The needle goes in through the septum and through the sleeve, which acts as a needle guide, and then goes into the column with a depth of several mm.

As far as pressure changes in the inlet.....I'm just not sure. These older 5890's that run stand alone (I'm not running Chemstation or doing any computer control on this instrument) aren't big on ability to monitor things, although I could probably at least check to see if there is a way to check the head pressure at the static initial temp of the inlet in the flow displays on the front panel.

Your points about potential backflash of solvent are well noted, and in a related point, also bring to mind a potential issue of needle discrimination, based on the boiling point of the solvent combined with the initial inlet temp, as well as a syringe problem that is known to deliver imprecise volumes that may include air bubbles (the variables here are adding up, aren't they ) If the inlet is at 60C, and solvent is bleeding out during injection from the needle, when it should be deposited as a discrete plug, are there analytes (in this case C11 and C22 alkanes) that are being irreproducibly left in the needle?

Regardless, it looks like you've given me some good ideas and thoughts for a couple of more experiments. I'll do some things and get back.

Thanks again, and best regards.

- Tim

[Peter Apps]

Hi Tim

Needle discrimation ! of course and I should have thought of it because I've seen the same thing during PTV injections, even below the BP of the solvent unless I turned the gas flow down as slow as it would go. A faster injection rate might help, at the possible cost of some band spreading, but with your rapid temperature programme that should not be too much of a problem.

Peter

[tpmcginnis]

Here's an update, after repeating the experiment, but with some minor modifications;

All operating conditions are identical to that in previous posts, except that Heptane is used as the solvent (bp = 98C), and a new syringe is installed. The concentration of analytes is 1000 ug/mL for both C11 and C22 alkanes. Fifteen injections were performed overnight using the autosampler.

(1) The injection volume reproducibility problem appears better with the new syringe. Except for one point that was about 10% high in area (injection #14, why this one was so off I have no idea) the %RSD for C11 alkane areas is 0.3%. The area reproducibility for C22 alkane, excluding the high area injection, is less precise, with a %RSD of ~ 2%.

(2) The area ratios for all 15 injections are

Area
Inj. # Ratio
1 0.933
2 0.923
3 0.888
4 0.886
5 0.886
6 0.897
7 0.894
8 0.898
9 0.918
10 0.893
11 0.890
12 0.924
13 0.924
14 0.892
15 0.893

Ave. = 0.903
sigma = 0.017
%RSD = 1.84

This is not rock solid, but is greatly improved over the %RSD of > 4% that was achieved when MeCl2 was used as the solvent.

(3) Since the C11 areas are very consistent, the variation in the area ratios pretty much tracks the variability in the C22 areas. This is somewhat telling in that it seems to point to a variation in the transfer of the higher molecular weight / less volatile compound from the syringe to the column.

(4) If I step back and take a look, I'm really talking about the comparison of two compounds which are very similar, yet very different, both of which are being injected, vaporized, separated, and eluted from a column. Yes, they're both linear alkanes, but one is twice the size and much less volatile than the other one. Fundamentals of matching internal standards and analytes might be worth reviewing here.

Maybe the next step is to add one more twist; Add two more alkanes to the mix (perhaps C12 and C21?) and see how these track against each other. Don't want to beat the dead horse here, but I want to better understand what kind of discrimination I might have been overlooking is taking place, and also discount the odd theory that there is some sort of transient detector variation.

Back to the lab...........

[gpronger]

For heavier compounds, I've had success in improving their mass transfer by increasing the head pressure or (if you have available) a injection pulse. Try tweaking this part of your configuration to see if can help.

[Peter Apps]

That's a drastic improvement. It's ages since I used a 5890 and as far as I recall it does not have a constant flow control, so as the column temperature increases the volume flow goes down. This just might affect the FID - it is getting less total gas flow when C22 elutes than when C11 elutes. Why this would be variable I can't say, but you are looking at pretty small effects.

Peter

[tpmcginnis]

Okay, this time there is a mixture of 5 alkanes (C11, C12, C17, C21 and C22). Each is dissolved in heptane at ~ 500 ug/mL each.

Identical HP5890 conditions as the previous experiments.

This time I let the instrument just go for a lot of injections, with the final number being n = 20.


Below are the responses of the different alkanes to C11, using the areas and concentrations to calculate the numbers. As such, the relative standard deviations noted are identical to those for the area ratios of the components in question.


C12 vs. C11 Response (C11 = 1.000)
Ave. = 0.997
%RSD = 0.1

C17 vs. C11 Response
Ave. = 0.993
%RSD = 0.4

C21 vs. C11 Response
Ave. = 0.964
%RSD = 1.7

C22 vs. C11 Response
Ave. = 0.917
%RSD = 1.8

C21 vs. C22 Response
Ave. = 1.052
%RSD = 0.2

So..........this tells me that

(1) The C11 and C12 responses are almost identical, and the variation in values is essentiall nil. These two very similar compounds are not being discriminated from one another.

(2) The C17 compound has a similar response to the lower molecular weight C11, but the variation in responses is larger. It's still good, but there is more scatter to the data. Perhaps some more variable discrimination going on here.

(3) The C21 and C22 responses drop off in relation to C11. There are %RSD's for each of these at a level of almost 2%. Not horrific, but the trend is that as the molecular weight increases, the apparent magnitude and variability of the discrimination increases.

(4) The C21 and C22 responses track each other closely. Over 20 injections, the %RSD of the response between the two is < 0.2%.


So the two lower molecular weights behave identically to one another, and the two higher molecular weights behave identically to one another. The one in the middle shows signs of increasing variability in response.

This supports the notion that, for this particular instrument at least, it is important to match internal standards to analytes in terms of volatility and elution. There is a gradient relationship that could be due to injector issues and/or inlet issues.

The HP5890 instrument I have does have constant flow, so in theory the flow should be reproducibly constant across the temperature profile. Whether or not the initial inlet pressure in the inlet is stable at the time of injection, that I don't know. I suspect that it might not be, and that this may be part of the problem with transfer of heavier compounds.

So I think that Peter's suggestion to change the solvent definitely helped improve the precision. However, there appear to be other factors at play that are a function of MW/volatility of the analytes under study. I think that my initial notion of the detector being the problem is not being supported by the data. It seems more like a good example of how internal standards and analytes should be matched well in order to have the best precision and confidence in the data.

Thanks to all who have chimed in. I think this is the kind of thing you have to do once in awhile to convince yourself of what is really going on with your system.

Best Regards,

Tim

[Pragmatist3]

Okay, this time there is a mixture of 5 alkanes (C11, C12, C17, C21 and C22). Each is dissolved in heptane at ~ 500 ug/mL each.

Identical HP5890 conditions as the previous experiments.

This time I let the instrument just go for a lot of injections, with the final number being n = 20.


Below are the responses of the different alkanes to C11, using the areas and concentrations to calculate the numbers. As such, the relative standard deviations noted are identical to those for the area ratios of the components in question.


C12 vs. C11 Response (C11 = 1.000)
Ave. = 0.997
%RSD = 0.1

C17 vs. C11 Response
Ave. = 0.993
%RSD = 0.4

C21 vs. C11 Response
Ave. = 0.964
%RSD = 1.7

C22 vs. C11 Response
Ave. = 0.917
%RSD = 1.8

C21 vs. C22 Response
Ave. = 1.052
%RSD = 0.2

So..........this tells me that

(1) The C11 and C12 responses are almost identical, and the variation in values is essentiall nil. These two very similar compounds are not being discriminated from one another.

(2) The C17 compound has a similar response to the lower molecular weight C11, but the variation in responses is larger. It's still good, but there is more scatter to the data. Perhaps some more variable discrimination going on here.

(3) The C21 and C22 responses drop off in relation to C11. There are %RSD's for each of these at a level of almost 2%. Not horrific, but the trend is that as the molecular weight increases, the apparent magnitude and variability of the discrimination increases.

(4) The C21 and C22 responses track each other closely. Over 20 injections, the %RSD of the response between the two is < 0.2%.


So the two lower molecular weights behave identically to one another, and the two higher molecular weights behave identically to one another. The one in the middle shows signs of increasing variability in response.

This supports the notion that, for this particular instrument at least, it is important to match internal standards to analytes in terms of volatility and elution. There is a gradient relationship that could be due to injector issues and/or inlet issues.

The HP5890 instrument I have does have constant flow, so in theory the flow should be reproducibly constant across the temperature profile. Whether or not the initial inlet pressure in the inlet is stable at the time of injection, that I don't know. I suspect that it might not be, and that this may be part of the problem with transfer of heavier compounds.

So I think that Peter's suggestion to change the solvent definitely helped improve the precision. However, there appear to be other factors at play that are a function of MW/volatility of the analytes under study. I think that my initial notion of the detector being the problem is not being supported by the data. It seems more like a good example of how internal standards and analytes should be matched well in order to have the best precision and confidence in the data.

Thanks to all who have chimed in. I think this is the kind of thing you have to do once in awhile to convince yourself of what is really going on with your system.

Best Regards,

Tim

I guess my question would be was the improvement in results due to changing the solvent? Or that needle? My guess would be the solvent.

[tpmcginnis]

I think you are right. Changing the solvent probably plays the biggest role.

Changing the syringe did improve overall area reproducibility.
However, barring some volume effects that manifested themselves as a function of inlet temperature, one would expect the ratios of peaks to be relatively constant regardless of injection volume variations.

However, changing the solvent is likely the biggest factor in getting more reproducible ratios.
MeCl2 boils at 40C.
Heptane boils at 98C.
My inlet has been initially at 60C....too high for MeCl2, as Peter suggested.