Large variation in some analyte responses

[psychemist]

First post. I searched the forum for something similar to my particularly befuddling problem concerning reproducibility of some analytes, but found nothing. The web at large is useless for this sort of thing.

I'm working with a 5890 with FID, using a 7694 headspace autosampler. The method I setup is for the high sensitivity (>1 ppm of ~20 mg solid or liquid sample) quantitation of various hydrocarbons with a BP range from -40 C up to over 250 C. The methods is splitless, which I only considered because HS is much gentler to columns, and the samples I run don't generally out-gas nasty gook at the 140 C sampler oven temp. I get very good resolution on most of my analytes, except the very heaviest (I need to tweak the temp program some), and judging by peak shape my data are mostly problem-free.

Except that I can't reproduce some analytes, specifically the more volatile ones, between days. Quadruplicate injections of gas-volume-sampled analyte gas are repeatable to the error of the GC syringe I'm using (a 10 uL, so about 1.5%), so I know that my vial seals are good and my sampling is consistent. But a week after calibrating over a wide range with Rsq > 0.999, another quadruplicate set of the same samples comes in ~78% larger in measured peak area, again with the same fairly narrow spread within the set. The heavier analytes don't have the same level of day-to-day variance, but there's still more there than I'd like.

At this point, I can't tell if it's the detector or the sampling. I've not changed any parameters (of course).

Detector: I've measured flow rate of all three gases (He carrier) on several occasions on days where results make no sense, and they agree to within about 1-2% of the original values after calibration. I've read that perturbations in flow of the FID gases can cause a change in response, but I've deliberately altered the H2 flow from about 16 mL/min down to 6 mL/min and the response change on my problem analytes was multiplicatively much smaller than I'm dealing with here. I know my FID gas ratio isn't perfect, but I've tons of tracking data (FID idle response 6.6 to 6.9 counts every day for many weeks, noise on the order of 0.05 counts with my FID chimney in place, and lab atmosphere sampling data that shows how well we keep our solvent bottles closed) for this method and I only want to change what will fix the issue. I will redesign the method in a few months when we switch to H2 carrier.

Sampler: If it was a consistency of sampling issue, I would expect the heavier, less volatile analytes to be less reproducible, but the opposite is the case. All my concerning analytes are very much gases at room temperature, and definitely from the sampler oven all the way into the column head. I've had some trouble with the thermocouple on the transfer line having a circuit gap or cruddy connection, which leads the instrument to think the transfer line is warmer (more thermocouple resistance) than it actually is. Almost always the temperature fluctuates wildly and prevents the instrument from thinking it is stable and ready to inject, and again this would manifest as a loss that's selective for the less volatile compounds.

At this point I'm lost, and I have no idea what the problem actually is. I could start using daily replicate external standards with my analytes of concern, but I still think my instrument shouldn't provide such variable results. 5% or even 10% accuracy error I can understand, but 78%??? Something is wrong. Please help!

[rb6banjo]

Are you injecting the calibration gas into a vial, through the septum, to get your sample into the 7694?

You have given a lot of detail about what you've done but some details on your instrument and sampling conditions (to get the sample into the GC) would also be helpful to us. Also, what are the identities of your analytes?

I doubt it's your detector but generally, the support gases for an FID should be more like 300 cc/min. air and 30 cc/min. hydrogen. The newer Agilent systems run more like 400:40. I have a 5890A and a Series II, both equipped with flame detectors that work well at 300:30 so that's what I use on those. You want to make sure that small changes in support gas flows do not cause perturbations in the mass-flow-sensitive flame.

Is it possible that you have active sites in your transfer line between the sampler and the GC inlet? Have you tried changing that out to make sure that the flow is continuous? No leaks.

What if you disconnect the transfer line from the GC and do manual injections. Do you get reliable injections with the gas-tight syringe? If you could do that, perhaps you could narrow your problem to somewhere in the headspace sampler.

I'll keep thinking about this....

[Peter Apps]

Welcome to the forum.

I'll echo the call for the details of the method, we can't troubleshoot unless you tell us what you are doing.

A couple of anomalies in your procedures: you are using a valve and loop autosampler, but you refer to 10 ul syringe injections ?? A thermocouple indicates temperature by generating voltage, not resistance.

I presume that you are aiming for complete evaporation of the sample, given the high equilibration temperature ?

If the samples are solid or liquid at room temperature then they must contain a much higher fraction of high boilers than of compounds that are gasses at room temperature. What are the peak areas for the volatiles compared to the peak areas for the heavy components ? The light components are in the C1 to C4 range - it is tricky to get these to focus at the head of a column that can also elute the heavier compounds in a reasonable time. Samll. poorly focussed peaks are vulnerable to unreliable integration.

Peter

[psychemist]

Thank you guys for a warm welcome and a rapid response!

I figured my FID flow was too low. I've been on the same hydrogen bottle for 5 months, and the cylinder pressure has only fallen about 300 psi from the starting 2700 or so. I also assume that stoichiometric mixing produces best results, so I'll look into that when I go back to work tomorrow.

My apologies for using "injection" in two contexts, one of the injection of a gas sample into a headspace vial, and another for the obvious injection of the heated sample in the headspace system.

The gas mixes I've used (injected via 10 uL GC syringe into a sealed headspace vial through the septum) are small alkanes, propane, isobutane, and butane, with trace impurities (judging by peak area). Due to the inertness of my problematic analytes, active binding sites are probably not the issue. I've used several different sources, the one I've been using the most is about 99.5% n-butane by peak area, the remainder being mostly propane and isobutane. I have a small beaker covered in tape with a pinhole to discharge an upside down gas cylinder into, and for each gas sample I discharge at least 2 mL of liquid into the beaker, allow it to boil completely, warm the beaker bottom with my hand, and check temperature with an IR thermometer. Once the glass temp is above 20 C, I withdraw 10 uL of gas with the syringe needle about 1 cm from the bottom of the beaker to mitigate the diffusion mixing near the pinhole. This provides same-day repeatability less than 2% in all cases, with better than 1% often the result. Considering how error-prone this bootstrap sampling is, I'm fairly pleased with the results.

I admit that I would much prefer a cryo-equipped GC to aide analyte focusing, but I'm able to almost completely resolve each isomer with more than 3 carbons. Peak widths are about 4 seconds or so, and very consistent. I'm using a column and temperature program suitable for residual solvent analysis, somewhat consistent with the method reported for the USP 467. Initial temperature is 35 C, about as low as I'm willing to go without forced cooling. Carrier flow is about 66 cm/s, based on a dead time of about 0.76 min (0.755 to 0.757 is the ET range for the tiny blip of methane that shows up in every sample) in the 30 m column. Isobutane and n-butane elute at 0.91 and 0.98 minutes, respectively with no resolution problems until a very large amount is used. At the lower amounts I'm using for R&R tests, response returns to baseline within three seconds, making decently well-defined peaks. Shouldering or baseline lifts on the peaks of interest are almost never a problem as far as I can tell.

The actual autosampler loop volume is 10 mL. I can't recall all of the time parameters for sampling, but I believe I have it set for 18 second pressurize time, 6 second vent time, 3 second equilibration (pause) time, and 6 second injection. I will add the actual method details when I return to work tomorrow. The very brief injection time, coupled with splitless injection, massively improved the resolution of the lightest analytes when I was developing the method, so I stuck with these changes.

Most of the routine samples of concern are fairly low-melting solids (<80 C) with very low volatility, with some amount of light residual solvents (alkanes, alcohols, ketones, esters) and other medium-weight hydrocarbon content (C10). Full evaporation of the lightest important analytes is the goal, and can be guaranteed when they are introduced to the vial as a gas! I added calibration for some of the heavier analytes when I noticed I resolved them very well on accident. I modified the temperature program to resolve them even better, and I've calibrated most of the heavier hydrocarbons to similar linear tightness (always forcing a zero intercept). My quantitation for these varies more than I'd like, but they are still quite reproducible as opposed to the gaseous analytes.

Again, I am able to quantify calibration samples same-day with pretty good repeatability (less than 2% of the 10 uL volume placed inside the headspace vial), and I can calibrate over a wide range - including gas dilution (removing gas sample from a headspace vial) with a pretty good Rsq (0.99960 with a forced origin). If the flaw is in the sampling - that is, anywhere before the sample is introduced into the column - I have a hard time understanding where, unless it's a random, intermittent issue which stays consistent for the entire duration of a run (up to 14 hours). With the amount of probing I've been doing, I hope that I would have caught an intermittent problem that crops up while the instrument is running. The problem seems to be when it goes to standby and then back to running again.

Also, if there was a gas leak before the column head, wouldn't all analytes be affected similarly? Wouldn't I also notice changes in carrier total flow from the three outlets? I wasn't around to witness the plumbing of the system, but the carrier pressure regulator on the HSS and the column head pressure regulator on the GC seem to share the same pressurized space on both sides of the regulator, i.e. adjustment of one can be compensated by adjustment of the other, according to the digital carrier pressure gauge on the HSS and the analog column head pressure gauge on the GC.

I understand how slight inconsistencies in peak width might manifest as integration problems, but not when peak heights agree with peak areas, and when the variance of concern is FAR larger than the variance of peak shape.

To sum up where my head is right now:
- Probably not sample preparation
- Probably not stationary phase bias or active sites
- Could be HSS inconsistent sampling, if it exaggerates minor inconsistencies in gas flow between days
- Could be FID acquisition (gas flow perturbations), but that doesn't explain the bias for some analytes
- Could be integration, but peak shape is good and consistent
- If gas flow inconsistency is the source, the instrument is exaggerating the inconsistencies quite profoundly (2% variation in flow results in 78% variation of peak area)

Again, thanks guys! I'm already learning and improving the quality of my self-criticism!

Edit: I'll be sure to post my method change history as it applies. I was very good about noting every change as I developed the method. On a side note, my Restek capillary column (with cyanopropyl, phenyl, and C18 functionality) snapped about 0.2 m from the detector inexplicably. It looked like it had been resting on the oven surface based on microscope investigation, but I still can't tell if that was it. The column was only 5 months old, max temp is 240 C, 40 C lower than the limit on the column. Column bleed is only a few counts, shifting baseline from about 6.6 to about 11 or so. I can provide hard data here tomorrow.

[Peter Apps]

"The actual autosampler loop volume is 10 mL. I can't recall all of the time parameters for sampling, but I believe I have it set for 18 second pressurize time, 6 second vent time, 3 second equilibration (pause) time, and 6 second injection. I will add the actual method details when I return to work tomorrow. The very brief injection time, coupled with splitless injection, massively improved the resolution of the lightest analytes when I was developing the method, so I stuck with these changes."

I am almost certain that the loop volume is 1 ml, not 10 ml. This is default on the instrument, and with splitless transfer there is no way you would get any resolution trying to inject 10 ml onto a capillary column - unless it is a 530 um megabore perhaps. Even with a 1 ml loop the 6 s injection will not transfer the whole loop volume to the column unless you have a flow rate of more than 10 ml/min, which again would need a megabore column. The problem with partial loop transfer with gas is that it is vulnerbale to pressure difference sin the loop, which can be a consequence of your short loop fill time.

The full method details will help.

Peter

[psychemist]

Here's the whole tamale.

Gas sources:
He - 30 psi
H2 - 22 psi
Air - 20 psi

Headspace:
Zone Temps: Oven - 140 C, Loop - 160 C, Transfer Line - 170 C
Event Times: Cycle Time - 22.0 min, Vial equilibrate - 20.0 min, Vial pressurize - 0.10 min, Loop fill - 0.30 min, loop equilibrate - 0.05 min, Inject - 0.10 min
Gases Carrier gas pressure - 10.0 or 10.1 psi, usually right at the halfway point, Vial pressure - 10.0 psi

GC:
Column: 0.53 mm ID fused silica, 3 um functional polysiloxane film (cyanopropyl, phenyl, and C18), 30 m
Gases: He carrier, H2 and ultrapure air for FID; Carrier flow 15.5 to 15.7 mL/min from FID (with all other gas off); forgot to write down H2 and air, but the sum of all of them is 202 mL/min, and H2 is around 25 mL/min
Inlet purge starts off, opens after 0.2 min (after entire loop has injected)
Zone temps: inlet - 200 C, detector - 250 C
Temp program: 35 C for 1.5 min, ramp of 30 C/min until 240 C, then hold for 5 min
Empirical carrier linear velocity: 66 cm/s
FID gas valves - both all the way open, no makeup gas
FID lit with a Bic lighter, the glow igniter is a piece of ####

Sampling: small portion (20 +/- 2 mg) of solid sample placed in tared clean headspace vial, mass recorded on (impressively repeatable) 4-place balance, crimped with PDMS/PTFE septum in aluminum crimp cap. Leaks here have been tamed, I had some issues with other analysts not crimping properly and it REALLY showed in the results.

I didn't have time today to record the peak picking and integration parameters for ChemStation, but I really don't think the issue lies here. As I mentioned before, the problem is almost certainly before data acquisition. The misestimated peaks are consistent with good quality peaks, and the peak areas and heights are covariant.

If I missed anything, I'm sure you guys will ask for it. At this point, you've probably established that this is the first GC I've ever spent lots of quality time with, and this is only my second method, which is a marked improvement over my first method which was designed for the same analysis.

Cheers!

[Peter Apps]

With 15 ml/min going down a megabore column nearly all of that looks OK.

There may be a problem with the vial pressure. At 140 C you might have a pressure in the vial that is very close to the pressurization setting in the method. You then get erratic additions of carrier gas and losses of headspace. Try increasing the pressurization setting a bit.

Peter

[psychemist]

I varied the vial pressure with the regulator built into the headspace from 5 psi (half the normal value) to 15 psi. I anticipated that at higher vial pressures, the extra pressure (and thus mass of carrier gas) would dilute the analytes enough that a decrease in measured signal would be observed. Only at 15 psi did I notice a significant change, in the direction anticipated, but the change was quite minor, about 4% decrease from the mean of the first four injections.

Basic ideal gas calculations show that the pressure of just air in the vial at 140 C is about 22.2 psi, a scant 2.3 psi lower than the applied vial pressure (10.0 psi above ambient), so I definitely know why this is a potential for serious errors in sampling. This explains why I've noticed lower quantitation at higher sample masses - the pressure in the vial can exceed the applied vial pressure, thus letting some sample gas backflow into the sample needle, and escape quickly when vented during loop filling. A much higher vial pressure is the only way to beat this problem.

All of this said, my challenge remains. The masses of standard gas and air in each headspace vial do not change very much, and so any impacts vial pressurization may have day-to-day are likely much smaller than the observed margin of irreproducibility.

At this point, my supervisor and I are leaning strongly towards leaving all gases at the running flow level, instead of going to standby mode (shut off FID gas at the valves at the top left of the GC, reduce carrier supply pressure to about 1 psi). I've already more or less verified that changes to the gas flows within normal operating limits do not significantly change response to the analytes of concern, so it seems almost superstitious to stop gas-saving in favor of keeping day-to-day changes to the absolute minimum, but we'll see. We will also be switching carriers to hydrogen soon, which should help us in a number of ways; I'm certainly not concerned about the cost of hydrogen gas!

I'm pretty much completely stumped at this point. Over the course of the last two weeks, response for the analytes of concern have increased each day at random intervals. I will continue tracking response changes by injecting replicate standards with every run; perhaps this will provide enough data to notice a causative trend.

If you guys have any more input, I would love to read it.

[Peter Apps]

Have you tested repeatability within days and between days with the vial pressure set to 15 psi ?

Another thing to try is to set the pressurization time to zero and use just the excess pressure in the vial to fill the loop.

With both settings you should make the loop fill time long enough for the pressure to bleed down to atmospheric - until no more gas comes out of the vent at the back of the sampler - as shown by connecting a piece of pipe, dipping the end into liquid and watching for the bubbles to stop.

Peter

[psychemist]

Vial pressure was indeed related to the problem. I was poking around the instrument this morning and noticed that a small portion of sample residue had been dripping out of the vent line out the back of the HS.

Problem solved: Clogged vent line!

I assumed my usual samples weren't so goopy, and now I'm going through an extensive cleaning. Agilent's steam clean procedure but with multiple passes of alcohol first.

I'm posting mostly for an online record of problem -> solution, I'm sure most of the community here is face-palming at my lack of adherence to standard maintenance protocols!

[rb6banjo]

Nope. Happens to us all. It's almost always something simple. I hope this fixes your problem.

[Peter Apps]

Vial pressure was indeed related to the problem. I was poking around the instrument this morning and noticed that a small portion of sample residue had been dripping out of the vent line out the back of the HS.

Problem solved: Clogged vent line!

I assumed my usual samples weren't so goopy, and now I'm going through an extensive cleaning. Agilent's steam clean procedure but with multiple passes of alcohol first.

I'm posting mostly for an online record of problem -> solution, I'm sure most of the community here is face-palming at my lack of adherence to standard maintenance protocols!

Good feedback - an exactly similar problem afflicts the split vent line with conventional liquid injections. The underlying problem is that the lines are not heated, so sample condenses in them. With the usual headspace conditions of aqueous samples and lower vial temperatures this rarely causes a significant problem.

Peter