and d18O of waters by TC/EA
i] Changes made to setup
II] Memory effects
III] Injection problems
IV] Data processing
I] Changes made to setup
NOTE: This page needs to be updated when we get
a little more time. For now, here is a brief summary of some of the
changes we have made that are not listed further down in this page
(posted to isogeochem list on Dec 1, 2006):
We now use an SGE 0.5uL syringe with a wide diameter (23 gauge I
think) and conical tip and find that this syringe works much better than
the Hamilton one. In addition to more reliable injections and fewer
problems with seizing plungers, it is less expensive and has a
syringe needle/plunger replacement kit that can be purchased for
about $28. Only catch is that you need to replace septa a little
more frequently due to the larger diameter needle. Injection volumes
All of the information listed below concerning memory effects were observed
with the Hamilton syringe and may no longer be applicable with the SGE
We now use a postinjection delay of 15 seconds instead of the 8
seconds mentioned in the page.
Our glassy carbon tube does not have a narrow diameter hole on top so
we cut a piece of graphite tubing that did have a narrow diameter hole
(about 4mm diameter hole)
and just set it on top of the glassy carbon tube.
We wrap the bottom fitting with about two to three layers of teflon tape to
get the seal (just need to feel that it is snug in the glassy carbon
We now have a 7mm I.D. glassy carbon tube and the peak shapes are a
little more narrow. I need to throw a screen shot of a data file up
We operate the TC/EA at 1400°C (no longer 1450°C) and the GC is at
90 to 110°C.
Below is the page that still needs to be
Waters can by analyzed using the TC/EA
(Thermo-Chemical Elemental Analyzer) at 1450°C. With our current method,
samples are placed into small vials. For small sample sizes it is
recommended to use vials with 150mL inserts.
These are good for sample sizes down to 50mL
or less. For these very small sample sizes it is highly recommended that
the user perform a series of tests in advance to make sure that their
sample transfer and storage techniques do not adversely effect the
isotope ratio of their samples. Please consult with us for this if you
have any questions.
There are several issues that arrise with
the use of a single syringe to inject a variety of samples via a
septum into the TC/EA. Most notable is that of sample memory: ie.
some remnant of the previously injected sample remains somewhere in
the system and biases the measurement for the next few injections.
The best way to see such effects and their duration is to perform a
series of injections of standards that have extremely different
isotope ratios. If the memory effects from standards with extreme
values can be nullified, then it stands to reason that such problems
will also be negated with real samples.
Here is an illustration of sample memory:
We tried varying the method used to prime
and clean the syringe but to no avail. This memory effect even
remained after flushing the syringe with relatively large volumes of
the next sample. We finally concluded that the memory effect that we
are seeing is coming from the injection port septum. Our solution
was to perform three 180
nL injections of
the sample in rapid succession into the TC/EA, after waiting
sufficient time for the resulting analysis gases to elute from the
system we perform four analyses of the sample or standard. Ideally
this gives up 4 measurements which we can average. Until recently, we
were experiencing many "failed" injections (perhaps 10-20%) which could
not be used. The precision of the
measurement for samples is not given as the precision of these injections,
instead we report the standard deviation of the mean value measured
for some laboratory standards which were run in replicates
throughout the sequence.
Here is an example of the file obtained
with the three rapid injections to clear the sample memory:
This is the result of our solution on
sample memory. There is one single measurement that came out notably
poor. It is the first run of the third sample (ie. the 9th point in
the plot). Can you see it?
Here is an example of a very nice injection of
300nL of water:
Conditions for the above data are:
TC/EA furnace at 1450°C
GC at 80°C
Carrier gas flow = 100mL/min (measured
at the exit from the GC)
GC column = 5A packed column,
1/4" O.D. by about 50cm.
no dilution on the conflo
injection volume = 300nL (using a 10uL
syringe holder on the PAL equipped with a 1.2uL syringe. Set PAL to
inject 2.5uL, this results in 2.5/10 * 1.2uL = 0.3uL)
Air volume = 0
No solvent precleaning
Sample precleaning = 2
fill speed = 5.0uL/s
fill strokes = 6
pullup delay = 500ms
preinjection delay = 0ms
postinjection delay = 8 sec (note, we
found that a long postinjection delay is necessary to allow time for
water to come off of the syringe needle. A short delay more
frequently resulted in significant peak tailing.)
injection speed = 5uL/s
(note: slowing down the injection rate from the default of 50uL/s to
this value has dramatically improved the quality of the injection)
postclean solvent = 2 (note that our
solvent bottle is empty, this is just an attempt to dry the syringe
a little bit)
Although each run is only 5 minutes, a sequence list will be very
long to allow for the multiple sample injections as well as the "memory
clearing" of the first rapid injections. Here is a
sequence list for our current setup. It is very long. The
user should replace the items labelled spl1, spl2, etc. on the left side
of the window with their real sample names. This will update the
sequence template in the columns to the right. A full tray of 98 samples
currently takes up 760 lines in the sequence. When we find more time to
improve our injection technique we hope to reduce the number of
replicate injections to 3 per sample and also get rid of the memory
Here are some examples of poor injections observed on our system:
The above injection was fairly common when using a short
This is probably the most common "poor" injection we have seen. It is easy
to find these in the exported results due to the unusually high
amplitude of the m/z 2 peak. With these injections, even though the CO
peak doesn't look so bad, the measured d18O
is usually far from what is measured with good injections. After
changing the preinjection delay from 500ms to 0ms and the injection
speed from 50uL/s to 5uL/s these poor injections have disappeared.
This injection was less common and appears to also have been a result
of a short postinjection delay
Precision and Accuracy
Since we finally figured out how to get the nice, reproducible
injections (Aug 9, 2006) we are getting very good precision and
accuracy. The section below on data processing will show how we achieve
these results. In brief, our first data set showed the following
results. Note that errors reported are two sigma.
||-63.9 +/- 0.34 ‰
||-7.93 +/- 0.03 ‰
||-107.6 +/- 0.18 ‰
||-14.57 +/- 0.07 ‰
All results above are given on the VSMOW scale. Our
water working standards
were first calibrated against SMOW/SLAP last year. The outstanding
quality of these numbers came as a surprise to us. However, even though
our standards look great it is necessary to ask about data quality for
samples. These standards are bottled waters (and antarctic water). They
will experience minimal matrix effects and so it is still necessary to
perform replicates of individual samples in order to get an appreciation
for the precision of the numbers for samples. Ideally, we'd also have
check standards that have matrices similar to those of the standards.
That is not the case yet, but we are hoping to make some in the future.
Here is the data
that led to the results posted in the table above. This data is part of
a set. The entire data set will be posted on the week of Aug 14th after
it has all finished running. At that point we will be able to check the
variation of replicates of samples as well as of more check standards.
A comment on statistics: The numbers given above are means and
standard deviations of three measurements taken at approximately 8 hour
intervals. However, it should be recalled that each measurement in our
data is itself the mean of four injections at one given time. So,
technically, we are actually showing the standard deviation of the means
of three measurements, ie. the standard error.
Here is another way to look at the numbers above:
From this representation of the data it is clear that the standard
deviation of individual measurements is about 0.3 to 0.4 permil for
d2H and about 0.06 to 0.10
permil for d18O. However, we
currently report a measured value as the average of 4 sequential
analyses, so the summary data shown further up this page shows the means
and standard deviations of each group of four injections of these
standards. Hence the standard deviations will resemble the standard
error of 12 measurements.
IV] Setting it up
The hardware setup - What you need to do this
You need a TC/EA or similar device that is capable of converting
water to H2 and/or CO. We use a PAL autosampler by CTC
Analytics with a 10uL syringe holder and a 1.2uL syringe (sold by
Hamilton specifically for the PAL). We use 2mL vials either filled with
minimal headspace or equipped with 150uL inserts (for small sample
Thermo Electron sells a "liquid injection kit" for circa $2500. This
includes a liquid injection port adapter for the top of the TC/EA, a
graphite tube with a narrow top to place on top of the existing glassy
carbon tube, a couple of septa (not very good ones), the 1.2uL syringe
and maybe one or two other small items. We initially used this setup and
obtained very good precision for hydrogen (< 1.5 ‰) but had inconclusive
results for oxygen as our small CO tank ran out and it took three months
to get a replacement one what was around natural abundance. As a side
note, the "research grade" CO sold by Air Liquide has a d18O and
d13C of about -5 to -10 ‰ vs VSMOW
and VPDB respectively.
An alternative setup using a sort of "wrap around" helium flow
discussed in the literature (Gehre et al. Rapid Commun Mass Spectrom.,
2004, 18, 2650-2660) reported very good precision for both
d2H (< 0.5 ‰) and
d18O (< 0.1 ‰). With funding from
John Sabo's group, here at ASU, we purchased the required bottom feed
adapter and glassy carbon tube from
IVA Analytical, in Germany. One
thing that this kit didn't come with was a narrow-topped tube, as
described in the literature. We had to take our narrow-topped graphite
tube that came with the initial Thermo liquid injection kit and cut it
to fit on top of the glassy carbon tube. If one does not do this then
peak shapes end up very poor due to the large dead volume at the top of
More to come... if you have any questions about this please contact
the lab manager.
The software setup
We program the PAL autosampler with one method for the "memory
clearing" part (internal no 8 in the sequence screenshot below) and one
for a typical sample run (internal no 7 in the sequence screenshot
below). In the memory clearing part, the only non-zero method parameters
sample volume = 1.5uL (with our 1.2uL syringe this translates to a
180 nL volume)
pullup delay = 500 ms
injection speed = 5 uL/s
postinjection delay = 6 s
The rest is set to zero
The sample run PAL method parameters are given further up in this
Here are some screen shots showing the time events for the isodat
methods for the sample runs or the memory clearing. Also there is one
screen shot of an active sequence. Note that we run 24 hours' worth of a
sequence and then perform a new H3 factor measurement. In the future we
may not need to do this as the sequence setup (sandwiching samples
between bracketing standards) will cancel out any slowly drifting
Here is an active sequence list. Line 400 of the sequence is about 24
hours after the start of this section of the list (we hope to improve
throughput later on). Note that the active sequence starts and ends with
Here is the timing events for the quick succession of sample
injections meant to clear any sample memory. Note that the 30 delay in
between each injection gives the PAL enough time to go back to its home
position. Also, the 290 second end time was selected so that we were
sure the CO had completely eluted from the column before proceeding to
the next line in the sequence.
Here is what the time events look like for a normal sample analysis.
The autosampler takes a while to do the precleaning so we trigger it
right at the beginning of the run.
V] Data Processing
Probably the most important thing to know about a reported number is
how it was obtained. There are many steps in the processing of stable
isotope data into final results. It is best to make minimal
manipulations of the data and to be explicit. If one is dropping values
for some reason, then that reason should be stated. An example (with
real data) of how we currently process results from the TC/EA is given
Download this file in order to better see what is being done in the
1) Reprocess the data using the "H_and_O" export template.
2) With all data processing, the first thing to do is to copy the raw
data into a second worksheet. Call the new worksheet "H_1".
3) In the new worksheet, highlight the "identifier 2" column then go
to Data > sort and say ok to expand to all data. Select sort by
"identifier 2" and delete all of the data with the identifier "clear
4) highlight the "Rt" column (retention time) and choose "A to Z"
sort option. This will separate the hydrogen data from the CO data. We
will process results for each isotope separately.
5) Copy the current worksheet to a new one called "O_1". In this new
worksheet, delete all of the lines with results for hydrogen isotopes
(line 2 to somewhere in the middle). In this same worksheet, delete the
two empty columns for hydrogen results.
6) Now go back to the H_1 worksheet and delete the rows of data
corresponding to oxygen isotope data. Delete the columns related to
oxygen isotopes. Your final product should look like this:
7) Highlight the "Rt" column again and select the "Z to A" sort
button. This will put the reference gas peaks at the end of the page.
Find the line where the results for the reference gas peaks occur and
insert 5 lines to separate them from the sample data. (Hint: the ref gas
peaks all have a d2H of 0).
8) Add a header in the leftmost cell above the ref gas peaks that
says "ref gas peaks". Now copy this worksheet into a new one and name it
This shows the new worksheet:
9) In the "H_2" worksheet, delete all of the reference gas peak data.
Also, sort the data by line number and delete the extraneous columns so
that the final product looks like this:
10) One very important factor in ensuring data quality with this
method is the injection. To test the reproducibility of the injection we
plot peak amplitude, peak area, peak width, and retention time as a
function of the line number in the sequence. In particular we are
looking for any misidentified peak tails, or spikes or dips. Here is
what you should have on the screen when done:
In the plots above one can see that there is a single data point with
a retention time around 100 seconds, whereas the rest are around 85
seconds. Similarly, there is a single peak with an ampl 2 that is well
above the norm and another with a very low amplitude. The peakwidth plot
also shows three points which stand out. Place the cursor over the odd
point and you will see the sequence line and value for that point:
When checking each plot in this data set, we see the high retention
time for line 236, low peak width for lines 190, 236, and 236 (yes,
there are two points for line 236: that suggests a peak tail was
misidentified), high/low peak amplitude and area for line 236 and 236.
Again two points for the same sequence line. This pretty much confirms a
split peak and the very high amplitude suggests a messed up injection.
Here is what the chromatogram looks like for sequence line 236:
One can see from the chromatogram that the tail of the hydrogen peak
has been integrated as well, hence the extra hydrogen peak in the plots.
Although there is nothing clearly wrong with the peak shapes here, for
consistency we will not include results from this data file as the
injection was clearly of a much larger volume than for the rest of the
analysis. Here is what the injection at line 235 looked like (note that
it is the same water sample):
Above is the injection just prior to the "odd" one at line 236. The
peaks are a little more narrow, and in particular, the hydrogen isotope
ratio differs by about 10 ‰ from the value at line 236.
Now that we have dealt with the oddity at line 236, we can go and
take a look at the data file for line 190, which gave rise to an
unusually short peak width. Here is what the chromatogram looks like:
Comparing this data file to the one just above it makes clear that
the integration has premuturely identified the peak end a little early.
In general, this variation has only a small effect on the final value,
however, we can modify the method file to get a more consistent peak
width if we want: Click on the "edit method" button (seen in upper left
corner of the above image) and go to the "peak detect at H2" tab. Change
the peak end slope from the default value of 0.4 mV/sec to 0.05mV/sec as
Click on "ok". Then click on the "reevaluate data" button that is
located right next to the "edit method" button. Here is the result:
The peak width is now 39 seconds, which matches well with all other
results in this sequence. The d2H measured here has changed from -10.39
in the earlier image to -10.56 here. That amount is not particularly
significant for hydrogen isotopes, however, for consistency we should
incorporate this value in the spreadsheet.
11) Copy the current results into a new worksheet and call it "H_3".
Delete the plots in this new worksheet and go flag any results which
should not be used. In the case of this data file delete the line
corresponding to the peak tail for sequence line 236 and flag the d2H
for the unusually large peak in line 236 in yellow.
Also, find the hydrogen isotope ratio for sequence line 190 and
change the d2H to -10.564 ‰. Flag this value with a different color and
put a comment in the cell next to it explaining why it is flagged:
12) Delete the extraneous columns so that the final product looks
As can be seen in the data above, there is still some memory
effect between samples. It is most notable between the PNZ and DSW-ANT
working standards, which span a wide range of isotope ratios. In
processing the data we will only use results acquired after the first
injection. For the working standards that means that we'll use the mean
of the second and third injection. Since we are only performing two
injections per sample in this data set, we'll only be using the value of
the second injection. As we have replicate analyses incorporated into
the sequence, we will still have a good estimate of the analytical
uncertainty of the measurements.
13) Label column E as "Mean d2H" and calculate the mean value for the
set of injections after the first one for the standards and samples. Be
sure not to include the flagged value for line 236.
14) Copy this worksheet into "H_4". Highlight all of the data
(easiest way is to click on the empty box above the row numbers and to
the left of the column letters). Now click on "paste special" >
"values". Sort the data by the mean d2H:
15) Delete the data that does not have a "mean d2H" value. Delete the
column headed as d2H/1H and sort the data by line number.
16) Add a column header saying "known d2H" and another saying "d2H vs
VSMOW". In the "known d2H" column, put the known values for the working
standards (found here):
17) Plot the "known d2H" values as a function of the "mean d2H"
values for the bracketing standards (the PNZ and DSW-ANT in this case).
Be sure to sandwich the data: see which data are highlighted in the
screenshot below. For the plot, choose "xy scatter" and be sure that
under the "data range" option (step 2 of 4) the "columns" box is
checked. Use the title and axes labels shown in the image below. Also,
be sure to remove the gridlines and legend.
18) In the final plot, right click on a data point and choose "add
trendline". In the trendline options choose "show equation". I would
also recommend right clicking in the area between the plot and the
border and choosing "edit chart area", then go to the font option and
turn of the autoscale. Shrink the plot and move it to the side.
19) Use the equation from the normalization plot to calculate the
hydrogen isotope ratios for the samples and check standards:
20) Copy the normalization plot and paste it between the next set of
bracketing standards. Change the title to normalization 2. Right click
on the plot area and select "source data". Highlight the new set of data
to use for the normalization:
(note that the screen capture option missed the outline that excel
makes on the highlighted source data so I added the boxes to highlight
them in the image above.)
21) Use the new equation to calculate the hydrogen isotope ratios for
samples in this section of the sequence.
22) Repeat this process for all of the data. After that, copy this
worksheet into another one called "H_results". In the new worksheet,
highlight everything, select "copy" then select "paste special" >
"values". Delete all of the plots as well as the column labeled "mean
23) Sort the data by identifier 1. We want to place all of the
results for standards on top of the sheet, and results for samples
further down. To do this, cut the rows containing sample data and choose
"insert cut cells" to paste it at the top of the sheet:
24) Delete the rows corresponding to the bracketing standards (they
should have no value for "d2H vs VSMOW"). Organize the data as shown
below. Also, calculate the mean and standard deviation for the
replicates of the standards.
25) Copy the values for the means and standard deviations for the
working standards into the top table. Note: don't just copy or drag the
formula over to the cell because of the next step: Delete the rows with
the detailed results from the working standards so that your sheet looks
26) For the sample resutls, delete the section called "known d2H".
Then sort the results for the samples by identifier 1. To do this,
highlight just the data for the samples and go to "Data" > "Sort".
Choose "identifier 1" and then in the second category choose "line":
27) Sample that were run in triplicate in the sequence will have the
word "replicate" in the "identifier 2" column. Calculate the mean and
standard deviation for these items. (Note triplicate runs will all have
the same "identifier 1"). Add a section between the standards and
samples in which you list the replicates and their associated standard
Analytical uncertainty for the replicate samples can be reported as
the standard deviation of those replicate measurements. For the rest of
the samples one should use the mean of the standard deviations of the
replicates as the estimated uncertainty.
The DAS, MSW, and DMSW check standards serve to demonstrate the
overall quality of the laboratory equipment and measurement. Ideally,
these standards would be closely matched to the sample matrix. In
general, the measured values should be within 2 standard deviations of
the known value, however, unusually good precision does sometimes occur
so if they are within 2 ‰ of the known value (assumes a stdev of 1 ‰)
then we will be happy with that result.
28) The same sort of data processing can be done with the oxygen
Page last updated: May 21, 2007