Saturday, December 01, 2007

Different Columns -- Detailed

In Different Columns yesterday, we passed on the breaking news that LNDD did not use the same GC column type on the Mass Spec that they used on the IRMS on every test done in the Landis case. They used an Agilent 19091s-433 on the MS, and a DB-17 on the IRMS. We'll call the one used for the MS the '433 from here on out.

Comments since have wondered whether this

  1. matters at all to results
  2. matters at all to the legal case
  3. late notice is a failure of the Landis defense team;
  4. late observance is indication even Landis thinks it doesn't matter.
As clarification, our conduit, Mr. Idiot, says that this information comes from Arnie Baker, and that Arnie pointed it out to him and it wasn't an independent observation. And Baker flags it as a "case dispositive" item. We think that answers points 3 and 4; it was a late catch by Team Landis, and they think it is important.

(We'd assume it is raised in the filed appeal brief, which we'd like to get released. USADA has it by now, don'tcha think?)


As to point 2, legal implications, let's look at the page that identifies the DB-17 as the SOP column for the MS, LNDD 664 from Exhibit 84.

LNDD 644: DB-17 in the SOP for the GCMS in MAN-52
(click for bigger)

The original LDP (the USADA pages) do not ever contain the method description page for MAN-52, nor was it produced during discovery, though it was requested. It only appeared in the Exhibits for the B sample tests. Was this an intentional omission in the LDP, and an accidental inclusion in the exhibits? We'll probably never know.

The legal implication seems direct. Their own SOP produced as documentation of runs says they should use the DB-17, but the run results show them using a different column. This would appear to be an ISO and ISL violation.

Baker says in mail to Mr. Idiot that USADA 104 also documents MAN-52 as the SOP for the MS to be used in the CIR test:

USADA 104: MAN-52 is the SOP for the MS in the C12/C13 testosterone test.

Baker alludes to COFRAC accreditation documents for the EC31 test identified above as specifying the DB-17, but we do not have a copy of that document to show.

Now, on to question 1, does this make any difference to the results?

The claim is that columns of different types can change the order of elution of compounds. Baker offers in the mail to Mr. Idiot the following support for that claim.

The Agilent website documents the columns and gives reference chromatographs for each. We're shown acenaphthylene eluting before acenapthene with the '433, and in the other order for the DB-17.

'433 Example from Agilent

DB-17 example from Agilent; reverse order of the a'ene's.

Baker notes for nitpickers that acenaphthalene and acenaphthylene are synonyms, and summarizes for us idiots:

HP-5ms '433 DB-17ms
(%-Phenyl)-methylpolysiloxane 5% 50% virtual
Polarity Non-polar Mid-polar
Acenaphthalene vs. acenapthene Elutes before Elutes after

OK, that's one example, but does that prove anything for the testosterone test? We don't know we have any acenaphthalene or acenapthene, so maybe it doesn't matter.

Baker doesn't have indisputable proof of switches with these columns of compounds known to be present in the samples. He does offer numerous examples of switches using similar columns, in particular a set from a drug screening test. He says the DB-5 is about the same as the '433, and pulls this from Agilent

The ordering on the DB-17 vs the DB-5 is:


Is this ordering relevant?
There are some huge differences in the ordering above: DB-17 peak 4 is DB-5 peak 16, and DB-5 peak 3 is DB-17 peak 13.

Baker finds a case from Skogsberg et. al with steroid analysis:

Skogsberg, U. et al. Investigation of the retention behaviour of steroids with calixarene-based stationary phases by modern NMR spectroscopy. Journal of Separation Science, vol. 26, p. 1119-1124. (2003).

In this example, steroid analytes 2 and 3 get switched. Ayalyte 1 is norethisterone, 2 is norethisterone acetate, 3 is chlormadinone acetate and 4 is testosterone acetate.

Let's pretend, for argument, that the column switch did not affect the elution order of the 4 primary analytes in the Landis F3, the5aAC, 5bA, 5aA, and 5bP. We don't know that, but let's pretend.

Given the switches seen in the examples above, how can we be sure that any of the other compounds haven't been moved, perhaps a lot, between the columns? Remember the big changes in the DB-5 vs DB-17 example above.

How can we be sure that any of the peaks in the IRMS are properly identified, and that they do not contain co-elutes that have arrived solely from the column shift? It is not clear that even full-scan mass-spec of the MS portion of this test will tell us anything about the peak contents in the IRMS. Given this column change, it would appear necessary to get full-scan MS from the column before combustion in the IRMS, and this wasn't done.

If the change in column is an ISL violation that causes a burden flip, it looks difficult to prove that the reported measurements were taken correctly, unaffected by the ISL violation.

That is why Baker considers this a "case dispositive" error.

Now, he thought the retention time argument was good too, and we saw how that was dodged. We'll have to see if there are holes in this argument as well.

Start shooting.


DBrower said...

Also note that USADA 104 identifies the MAN-52 as being a SCAN, where other GC/MS tests are identified as SIM.

That implies to me that the test is designed to collect a full-scan, where others are to use selective ion monitoring.

So why was the full-scan data not presented in the results, and wiped off the disk?


RBP said...


The above is a good summary of issues raised by this new information about the different columns. Just to clarify, from information on the Agilent website 19091S-433 is the part or stock number for 1 type of HP-5ms model (there are various HP-5ms models with differing column heights, inner diameters, etc.). DB-17ms is the model number for a different column made by Agilent. Part of the confusion may be that LNDD used the part number to identify the HP-5ms on the paperwork and the actual model number for the DB-17. As you have stated elsewhere, the main difference between the two seems to be the polarity (the HP-5 being non-polar, while the DB-17 is mid-polar). Interestingly, in a chart in the FAQ section of the site (regarding choosing the correct column), Agilent lists steroid identification for the DB-17, but not for the HP-5. From what I have read about the effect of column polarity, Neb's comments in another thread seem correct. Non-polar columns allow substances to elute according to boiling point, while polar columns retard the elution times of polar molecules. So it is possible for substances to change positions, as the Agilent examples show. Whether this has any bearing on the Landis results will probably be a matter for the experts to argue over (in addition to the legal issue of LNDD's use of the wrong column, if indeed its procedures specify the DB-17). Since this column issue now introduces a 2nd difference (polarity) between the two machines (the temperature difference being the 1st), I would be interested in hearing dailbob's opinion, since he has used GC-MS and was originally very concerned about the temperature issue.

GMR said...


This link for the Common Drug Screen DB-5 vs DB-17 provides a sharper detail than the link given above.

Unknown said...

First, let me make it clear that I don't operate our GC-MSs, nor did I ever mean to imply it. I work in product development for a large consumer products company. What I know about GC comes from submitting lots of samples, and reviewing the results with our Analytical Chemists, over a 27 year (so far) career. We have the same HP/Agilent GC-MS as LNDD. That said, my original concern with the different temperature ramps was that it created the possibility that either the peaks would switch or you could get co-elution of something that moved close/within the peak of interest. Fundamentally, there are two components of identification : elution (or retention) time and the mass spectra. Because the mass spectra from the IRMS only gives you C12 and C13 identification, the peaks need to be tied (or anchored)to a retention time of a known standard. LNDD did not include all of the metabolites in the cal-mix (which would have enabled them to identify them by retention time). So, in my view, they needed to be tied to the relative retention time to the peaks from the GC-MS. The different temperature ramp between the two machines, minimally, stretched the the IRMS output relative to the GC-MS output. Because it also created the possibility of a switch (even if the possibility is small), I've stated that I wouldn't be comfortable convicting anyone with this data. In the examples that TBV has posted above, it's stated that the switch in the example compounds is probably due to the different column types. However, these examples also use markedly different temperature ramps. So, it could also be due to that (or both).

In regard to the use of different columns, I don't know what to think. On the one hand, this further complicates trying to tie or anchor the IRMS peaks of interest to a known standard, because it makes the assumption of the peaks eluting in the same order indescribably more tenuous (maybe impossible). However, the fact that LNDD (evidently) did this repeatedly means it probably wasn't an accident. I'm wondering if there's something we're missing?

We use these same columns in our lab, and they aren't even close to each other in terms of their properties and what compounds we use them for, so it's not logical to me (from a scientific perspective) that you switch between these columns on the two machines.

For what it's worth...


RBP said...


Thanks very much for your reply, which was very understandable. Even if you don't actually operate the GC-MS's, you have valuable first-hand knowledge of their use. Your response confirms what I understand -- that the use of different columns, on top of the temperature ramp differences, adds to the possibility of peak switch and/or co-elution. And this contributes just another element of doubt about this whole case. Maybe LNDD had a good reason for the use of different columns, though this might be hard to defend if the written procedure states otherwise. Or maybe the report paperwork is wrong and the lab used the same columns on both machines (again not reflecting well on LNDD). Or maybe someone made a big mistake. Or there could be something else here. At any rate, it appears this issue will come up at the hearing. Thanks again for the reply.

DBrower said...

GMR, I used your better link in an update. It still works.



m said...


Take a look at the graphs below and tell me whether you think the 5A or P could have been misidentified. I think I've made the case that no peaks shifted, regardless of the GC columns. The proof is in the graphs and data, not some speculative possibility.

In the graphs below A=5B, B=5A, I=Internal Standard, P=pregnane. The a and d are unidentified small peaks.

Was the 5A identified in the IRMS F3 sample. Yes.



1) GCMS Mix Cal

2) GCMS F3


4) GC-IRMS F3 Blank Urine

5) GC-IRMS Mix Cal

The B in 1) and 2) match based on RTs, as well as the A, I and P.

The metabolites in the IRMS graphs in 3), 4), 5) eluted later than in 1) because the IRMS involved a combustion phase and because changed ramp conditions caused the mebolites to elute further apart probably to achieve greater peak separation. And possibly because of different column specs.

The RTs in the graphs in the IRMS all matched. So there is no question that the B in the IRMS F3, must be the same B as in the GCMS F3. Remember it comes right after the A, which we have fixed in the 5) graph.

For the GCMS F3, we know that the IS, A, B, P in the F3 GC are identified based on their MS and comparing their RRTs with the Mix Cal.

We know that the a A B d appear to follow in the same order in both IRMS F3 and the GCMS F3 and GCMS Mix based on RRTs and height of peaks. We expect a longer RT in the IRMS because of the combustion phase, and expect slightly different RRTs because of the lower ramp tempature and difficulty in making GC conditions identical, so we are not matching RTs.

We know from the GCMS F3 that all of the substances identified in the F3 have to be there and that the A and B have to be in the same order unless they switched position with some other substance.

We know that the I and A in the F3 IRMS are identified because their RTs match the Mix Cal and the blank urine.

For the F3 IRMS, can the B be anything else than the B? Answer no, unless it switched places with the d or a. There is no evidence that it switched based on the heights of the peaks or any other evidence.

DBrower said...

M, you wrote,

I think I've made the case that no peaks shifted, regardless of the GC columns.

Do you really mean no peaks shifter, or only that no peaks of interest shifted?

I've said that, for sake of argument, I'll concede no change in the relative positions of the 4 major analytes of interest.

On the other hand, I don't understand on what basis you can make the claim that none of the other peaks shifted or changed order. The literature would seem to support the proposition that many substances do change notably between the two columns.

Now, separating into two lines of thought, one legal and one Truth.

On the legal front, does the use of a different column than that specified in the Lab's SOP form an ISL violation? That's worth discussing from the lawyer's perspective.

Then there either is or isn't a burden flip about what presumption should be applied.

Second, on the matter of the Truth, how do we have any idea what is in the IRMS peaks if other substances might elute there, and we have no possible indication? The only angle we had for specificity was the GCMS, had we looked at the full-scan MS data. But that is no longer applicable when we changed column types.

I understand you are still mostly concerned about the identification issue. I remain of the belief that for Truth, specificity is also important.


RBP said...


You write, "The proof is in the graphical data, not some speculative possibility." I don't dismiss possibility quite so easily. Yes, the graphs and data show this and that and would tend to support a conclusion one way. But possibility plays a role in casting doubt on the accuracy of that data or at least the conclusions they seem to support. At first, the possibilities might seem unlikely, so it is easier to accept the presented data as true. But as the possibilities multiply (especially when they result from apparent errors or sloppiness), further doubt and questions creep in about how good that "proof" might be. TBV and dailbob have addressed the possibilities well. The legal issue might actually be more definitive and could have a larger impact on the final decision. At any rate, my point is that possibilities can fuel doubt, which in turn can be sufficiently high to tip a vote.

GMR said...

My clearer link has been removed. To find the PDF where you can really look at the data and see not only is order dramatically changed, but also peak height (quantification). Number 18 is half the size between the two different columns. Agilent Chromatograms Search for Common Drug. Registration is free.

Ali said...

Life would be so much more simple if the arbitration panel hadn't decided to ignore WADA rules. What ever happened to the lab having to prove that their mistakes didn't affect the results?

Maybe they could comment on the column issue? ... I won't hold my breath.


m said...


You are correct, I was only referring to the 4 peaks of interest.

I shall have more to say about minor peaks shifting later.

Unknown said...

Hey M,
I'm not ignoring you; I just haven't had time to respond properly. I will shortly.


Unknown said...

Sorry it took so long for me to get back. My wife and I have nutty jobs and my kid’s in all kinds of activities, so it’s been hard to find time. This is also why I haven’t been here much the past couple of months.

Before I start, a reminder that we don’t have an IRMS in our lab, so I’m “projecting” some of what I’ll call “GC 101” identification principles onto the IRMS. I think I’m correct in doing this, because this is essentially what the Landis team did via their retention time arguments. I also vetted my thinking with our Analytical Chem Manager, and he’s in the same place I am.

I first want to provide some history and an example to give you some background on what’s driving my perspective. I’m old enough that I remember getting our first GC-MS in the late ‘80s. The addition of the MS was a game changer, because it added another “layer” of identification, the value of which can’t be understated. Prior to that, you could only identify by elution time, and we were sometimes wrong. Now, we’re rarely wrong. This changed the analytical chemistry world to where I think most Chemists would tell you that they require both for identification, i.e., the MS upped the bar. I’m telling you this, because not having the full mass spectra to aid in identifying something on the IRMS places all the emphasis on elution time.

A few weeks ago, we got a sample in where one of our plants thought they accidentally put glycerin in a batch instead of what was supposed to go in. To determine if this happened, we ran a glycerin standard which eluted at 2.854 minutes. We then injected the sample, where a peak eluted at 2.862 minutes (0.008 minutes or about half a second difference). The mass spectra matched the glycerin standard, so these two pieces of information, taken together, identified the material as glycerin. I’ve provided this example to demonstrate how these machines identify compounds. This is the only way they identify compounds, and the identification is pretty “black or white” – either it’s identified, or it’s not (an exception is isomers, but that’s NA here).

I think the main issue as you move a sample from the GC-MS to the IRMS is maintaining the identity of the compound for which you want to determine the CIR. I say this for a couple of reasons. First, even if the GC-MS and IRMS use the same column, the fact that the columns are two separate physical units can cause differences in elution time, or even peak swapping, if one of the columns is contaminated. Second, as I mentioned, because the IRMS cannot identify the compounds with full mass spectra, the only way of identifying it is by elution time. Consequently, I think it is absolutely paramount that it be tied directly to a known standard, because without it, you have zero “layers” of identity.

I have expressed concern about the different temperature ramps between the two machines, which you have discounted (note: Larry has uncovered information that makes it look like the difference in temperature ramps is intentional, but this doesn’t negate my point). If you look at the Landis F3 from the GC-MS, there is a peak that elutes at about 13.2 minutes. This peak seems to have disappeared in the F3 IRMS. This peak either a) contained no carbon, b) moved due to the change in temperature ramps, or c) indicates a contaminated column. I can make this argument for any peak that is in the GC-MS output that doesn’t appear in the IRMS. If I can't account for a peak, and the peak I'm interested in is not directly identified by being tied to a standard, its identity becomes suspect.

You asked me whether I think the 5aA or P could be misidentified? If you can play a mental game for a moment and adopt the perspective that the only way things are identified are by elution time (minimally) and mass spectra (ideally), I think you can see that, technically speaking (i.e., by standard separation science), they aren't identified at all. What you have done in your example is arrived at the identification by logical deduction. While your logic is excellent, logic is not science. Because of the gravity of this case on a person’s life, I have to have the identification be done by standard separation science, particularly because it's capable of doing so in an incontrovertible way, if it's done carefully and correctly.

You asked me once whether I thought the science told us whether Floyd doped, and I replied that I really couldn't tell with any surety. This is why. While I can't prove he didn't dope, this science isn't good enough, in my view, to prove that he did. When I combine this with a) generally crummy chromatography, b) poor lab practices (e.g. white-out), c) the fact that the blank urine was “positive” (for me, indicating some possible machine bias) and d) the fact that I was influenced by Amory’s testimony, I still have enough doubt that the excellence of your logic doesn't push me over the hump.

M, I don't have time to engage in a huge running debate (it takes me a long time to write these things!). I feel that I've never effectively been able to communicate to you why I feel so strongly (and why I think it’s important technically) that all of the metabolites needed to be in the cal-mix (probably one at a time). Hopefully, I got a little closer this time.


Larry said...

D-Bob, great post! Thank you.

TBV, can you copy this so that it also shows under the 5bA anchor discussion?