Friday, January 24, 2014

NMR Thermometry Part 2

Yesterday I wrote a part 1 of a blog post regarding a Nature paper from October 2013.  You might want to check out that post before reading this one.

These two posts review/critique the paper "Thermal maps of gases in heterogeneous reactions" by Jarenwattananon et al.

My post yesterday was an attempt to review the publication.  I tried to minimize my comments and let the authors speak for themselves.  Today I plan to air my grievances and concerns about this paper.  

Before I begin I want to note that this paper got tons of press.  For example ..,229E6,E1MX8G,7GOMF,1

With so much attention, I'm sure my puny critiques will not be too rough for Jarenwattananon and co-workers.  I will divide them into three categories: accuracy, precision and sources of error.

#1) Accuracy.

In the body of the paper the authors state "The disagreement between NMR-derived temperatures and fibre-optic sensor measurements was at most 4%."  In the abstract they say "measurement error [is] less than four per cent of the absolute temperature."  I feel that it a bit misleading use agreement with external temperature sensors as a metric of accuracy.

Drill down into the paper and really ask yourself "what is the limit of accuracy in the temperature measurement?"  It has to be how accurately you can measure line widths of the analyte.  For example, Fig 1c shows data and Lorentzian fit for propylene in temperature calibration experiment.

Look carefully and tell me, what is the smallest line width difference you think that you can assess on such a spectrum?  Could you distinguish two spectra with line widths that differ by 1 Hz, 5 Hz or 10 Hz?    By inverting the linear equation (I'll use calibration curve for PtNP) , it is possible to calculate how much the temperature changes if line width changes by 1, 5 or 10 Hz.

T =  1 Hz / 0.16 Hz K^-1 = 6.25 K
T =  5 Hz / 0.16 Hz K^-1 = 31.25 K
T = 10 Hz / 0.16 Hz K^-1 = 62.5 K

4% of 400 K is 16 K, so the authors seem to acknowledge an accuracy of line width measurement of about 3 Hz.  I question if they can really measure the line width so accurately.  If the error is closer to +/- 25 K, then their false color thermal maps, particularly on the microreactor in Fig. 4b, could be quite misleading.

#2) Precision

In the caption for Fig. 1 and 3, the authors state that "temperature calibrations were performed for over 30 different systems." and "thermal maps were generated for more than 15 systems."  How much do the results vary from system-to-system?  Are these variations an indication of the variation in each reactor or the precision of the measurement.  If you repeat the calibration or mapping experiment several times how much variation would you see?

#3) Sources of error

I think of this experiment as almost an anti-DOSY.  In this experiment, the NMR signal in the presence of a gradient is sharpened not attenuated by molecular motion.  I surmise that this experiment works because gas molecules at elevated temperatures diffuse fast enough average gradients.  The problem as I see it is that this experiment only works if everyone agrees that the only factor that impacts the line widths is molecular motion.  You could fool yourself and think you see a cool spot if regions of the sample have any reason other than the gradient to have broad lines.  Some examples that spring to mind with heterogeneous catalysts is local concentration of paramagnetic substances. 

A related issue that troubles me is that the slope of the calibration curve (deltaF vs. T) is different for the two catalysts (PtNP and Pd-MOF).  I don't understand why.  It seems like if gradient-induced line width is narrowed by motional averaging (which is related to T) the specific catalysts should not matter.  Why is deltaF equal to -87 Hz at 400 K and G equals 0.05 G cm^-1 for PtNP and -79 Hz for Pd-MOF.  Either the material impacts the line width (which is clearly what the authors think) or these values represent errors in line width measurement.


In the end, I really love this paper.  It really made me think about stuff I don't get to think about too much line mean free path and the Maxwell distribution.  I get persnickety about a few issues, but you shouldn't let me distract you from the unprecedented view into reactor energetics this new NMR thermometry technique offers.    

Thursday, January 23, 2014

NMR thermometry

The paper for this week was published back in late October in Nature.  I'm just getting to it now though,  because a) I'm always a bit behind and b) the title "Thermal maps of gases in heterogeneous reactions" does not alert the reader that the content of this paper is highly relevant to anyone who wishes to be fluent in contemporary NMR research.  The full citation is

"Thermal maps of gases in heterogeneous reactions"


Jarenwattananon, Gloggler, Otto, Melkonian, Morris, Burt, Yaghi and Bouchard

Nature 2013 502, 537

This paper describes a novel NMR thermometry technique to measure gas temperatures at gas-solid interfaces in heterogeneous catalysts.  For engineers who optimize reactor design, temperature gradients on catalyst beds contain vital information on reaction energetics (so I am told).  The key insight of the authors is that in the presence of a weak magnetic field gradient (<< 1 G cm^-1) motional averaging will result in an inverse relationship between line width of a given peak and temperature of the system.  Figure 1a provides an excellent pictorial description of this technique.

To demonstrate the feasibility of NMR thermometry, the authors design a simple calibration experiment using a 10 mm NMR tube with catalyst-loaded glass wool (more on their catalysts later) and an analyte of propylene gas flowing at 15 cc min^-1, 40 PSI.  A 1D 1H spin echo spectrum is recorded with gradient applied during acquisition on a 400 MHz NMR (with microimaging capabilities) and the probe temperature set from 303 to 413 K in 10 K steps.  A sum of Lorentzians is fit to the olefin region of the propylene spectrum to get line widths.  Figure 1b shows linear fit of a plot of the change in linewidth relative to 303 K (deltaF) versus temperature (T) for three different gradient strength.

In the absence of a gradient, the slope of deltaF versus T is negative but small.  For a gradient strength of 0.1 G cm^-1, the slope is -0.13 Hz K^-1.  A stronger gradient results in a larger slope.

The authors settle on 0.005 G cm^-1 for gradient strength for their NMR thermometry setup.  They measure deltaF vs. T for two catalysts: Pt nanoparticles (PtNP) and Pd metal-organic frameworks (Pd-MOF).  Least squares fit results in a slope and intercept of -0.16 Hz K^-1 and 151 Hz and -0.10 Hz K^-1 and 119 Hz for PtNP and Pd-MOF, respectively.  Now the authors have the tools in place to measure T by NMR.  Simply measure the line width (of propylene olefin peaks) in 1D 1H spectrum with a gradient of 0.05 G cm^-1 during acquisition and use the appropriate linear equation to calculate temperature.

Having convincing demonstrated feasibility, the authors move to an application: thermal maps of lab-scale demo reactors with each catalysts.  In this experiment, propylene and hydrogen gas (actually para-hydrogen) are flowed through the reactor at 15 cc min^-1, 40 PSI and catalysis (hydrogenation) takes place.  The microimaging experiment (which takes ~30 min) divides the reactor into voxels and in each voxel the spin echo/gradient experiment is performed.  Line widths and, subsequently, temperature is determined for each 0.73 x 0.73 mm pixel in the reactor.  Figure 3 shows false color image of axial, coronal and sagittal views of the reactors.

Significant variations in the temperature of the catalyst bed is observed.  There are spots with T > 600 K and other spots with T < 400 K!  To validate these results, the authors carefully place a fiber optic temperature sensor at 3 spots in the demo reactor (you can see where in the coronal view of PtNP and sagittal view of Pd-MOF reactors).  The sensor agrees well with NMR thermometry-derived T.  The error is at most 4%.

Finally, the authors demonstrate NMR thermometry on a 1 mm microreactor with catalyst supported on silica gel.  Para-hydrogen preheat to 418 K is flowed into the reactor.  I assume that there is also propylene.  I have no clue what the flow and pressure is because the authors do not go out of their way to bore us with any details.  The temperature calibration experiment results in a slope of -0.2 Hz K^-1.  No mention of the intercept.  Figure 4 show a false color image thermal map from the microimaging experiment.

Once again, NMR thermometry reveals significant temperature variations in the reactor from 30 K higher than gas T at the inlet to 30 K lower at the outlet.

This is an amazing paper!  It is easy to understand why it was published in Nature.  The thermal maps in Fig. 2 and 4 are incredible.  I am no expert, but  don't see any other way to measure these types of temperature gradients on this scale other than NMR thermometry.  In Fig. 4 for instance, there is a gradient of ~60 degrees over a space of ~2.5 mm (which is less than 1/10 of 1 inch).


I'm going to make this a two-part blog and return tomorrow and sharpen my critique up a little bit.  See you then.


Thursday, January 9, 2014

Wrapping up 2013

There were a lot of papers that I read last year and really liked.  I would like to write a full critique, but I just don't have time.  So here is my twitter critique (<150 words - I know that twitter uses characters, but give me a break.)

The quiet renaissance of protein nuclear magnetic resonance


Paul Barrett and a cast of thousands

Super well-written review.  A call to arms to NMR spectroscopists to embrace NMR as a broad problem-solving tool rather than machine that generates PDB files.

Mycalol: A natural lipid with promising cytotoxic properties against human anaplastic thyroid carcinoma cells


Adele Cutignano and co-workers;jsessionid=E7C1B2F402EF5A4BA96E06EFFEF70D1F.f04t03

A great natural products structure elucidation paper.  This paper is what really inspired me to start this blog.  The authors isolated a lipid from a sponge they found in Antarctica.  The molecule has five stereocenters and it would have been a pain in the ass to assess stereochemistry using NOE and J coupling.  The authors use CD and Mosher's method and a lot of clever logic.  A+

Characterization of Cyclopentyllithium and Cyclopentyllithium Tetrahydrofuran Complex


Su, Hopson and Williard

The first of three JACs papers published by these authors this fall.  Good grief.  I won't get three JACS papers in a lifetime.  These guys got three during the football season!  This one is probably my favorite for the cool use of DOSY to explore the equilibrium between a hexameric and tetrametric molecule.   

The influence of electronic modifications on rotational barriers of bis-NHC-complexes as observed by dynamic NMR spectroscopy


Kolmer, Kaltschnee, Schmidts, Peeck, Plenio and Thiele (so many Germans and no umlauts!)

A great physical organic chemistry paper regarding Grubbs catalysts.  Quantitative dynamics to measure rotational barriers and a clear relationship between redox properties and rotational barrier.  Nice work.

Direct Measurements of the Mn(II) hydration state in metal complexes and metalloproteins through 17O NMR linewidths


Gale, Zhu and Caravan

NMR tricks to assess the hydration state of metal complexes.  Relaxivity measurements.  Sign me up. 

Xenon-based Molecular Sensors

This weeks paper is a hot-off-the-press JACs article by Garimella et al.  The full title is

Hyperpolarized Xenon-Based Molecular Sensors for Label-Free Detection of analytes.


Garimella PD, Meldrum T, Witus LS, Smith M, Bajaj VS, Wemmer DE, Francis MB, Pines A.
The full citation is

A few months ago I reviewed a very nice paper by Perrone et al. regarding NMR-based sensors (see  This weeks paper describes an alternative approach to designing an NMR-based sensor.  Garimella et al. design a sensor with two binding sites.  One site will bind a specific analyte, in this case the dye Rhodamine 6G.  The other site will bind xenon gas molecules dissolved in solution.  It turns out that xenon NMR is so sensitive that subtle changes on the other end of the sensor, such as the binding of a ligand, alter the chemical shift.  Below it the TOC graphic to explain the concept visually.


What do the author's actually do?

They begin by creating a peptide receptor library with 1.6e5 (=20^4) molecules based on the sequence KXXPGXXGWKKG.  Their hope is that some of these molecules (~2% or so - which is 3,200) will bind with a dye.   These peptides are bound to polystyrene beads at the C-terminus and pyrene at the N-terminus.  Since the peptides are attached to beads, the authors incubate the bead library with 10 uM dye for 1 hour, wash away the dye and manually sort the colored beads.  (I am glad I don't have that job.)  They screen the bead library multiple times and sequence the hits by mass spectrometry.  I am sure that this part is pretty clever, but the details are a bit lost on me.  In the end, their consensus sequence to bind Rhodamine 6G is H2N-KDDPGDEDWKKG-CO2H, which they call the "D-peptide."  The make another peptide as a control with the sequence H2N-KNNPGNQGWKKG-CO2H.  They call this peptide the "N-peptide."  The figure below shows visual confirmation that the D-peptides binds with Rhodamine 6G using the bead incubation assay, whereas the N-peptides does not show any color change.  So are we the audience to assume that the N-peptide does not bind the dye?  At this stage, I think the answer is yes. 


Garimella et al. can now move away from the beads and focus on the peptides.  They do some smart controls like replacing the pyrene with the Xenon binding cryptophane cage at the N-terminus to confirm that the dye binds to the peptide, not the pyrene.  They also confirm specificity by trying many other dyes, none of which bind with the D-peptide (no word about the N-peptide, though).  

Now we can get into some spectroscopy.  The authors use NOESY to confirm the interaction between the dye and D-peptide.  Their sample is 250 uM dye and 200 uM D-peptide-cage.  They also record a NOESY of 250 uM dye and 200 uM N-peptide-cage.  Here is what the authors report:

"The spectra of both the D- and N-peptides in the presence of dye revealed cross peaks in the region corresponding the dye-peptide interactions.  However, there were many more cross peaks in the D-peptide-cage sample, and they were more intense than those in the N-peptide-cage sample."

and then a bit later

"The NOE data .. verifie(s) that there was a difference in interaction of the dye with the D- and N-peptides, confirming our visual evidence that the library produced a receptor for the Rhodamine 6G analyte."  
So much for a negative control!  Still you have to give the authors a lot of credit for turning a bug into a feature.  (I am stealing that phrase from my friend Todd, who used it to describe a research projects around here.  It seems to me that the ability to turn a bug into a feature is a key trait in good science.)  So the N-peptide binds Rhodamine 6G, albeit with less affinity than the D-peptide.  As a final note on this topic, it seems to me that there is something funny going on with the interactions between the dye and peptide.  I spent some time staring at the NOESY spectra in the SI and I can't make heads or tails of this interaction. 

On to the Xenon-based detection - The Pines lab has developed a really cool system to deliver optically hyperpolarized xenon gas into their NMR samples.  The net result is a HUGE signal.  In the introduction to this paper the authors say that the NMR signal has the "strength compatible to those of water in conventional experiments."  I am assuming that they only need one scan!

At any rate, the authors make ten NMR samples, five for the D-peptide-cage, five for the N-peptide-cage each with increasing dye concentration from 0 to 1000 uM in steps of 250 uM.  The peptide concentration equals 200 uM.  Let's first consider the Xe NMR spectrum for the peptide-cage samples with no dye.  There are three peaks.  One for hyperpolarized Xe bound to the cage at ~60 ppm.  There is a sample for aqueous Xe (some the gas pumped into the solution dissolves in the solvent like CO2 in soda pop) at ~200 ppm.  There is also so undissolved Xe gas signal, which is set to 0 ppm.  (Since these signals are in slow exchange, I guess the dissolution process is slow on the NMR time scale.)  The gaseous signal is used as a reference and set to zero ppm in all spectra.  What I find strange is that the exact chemical shift of the aqueous Xe is at ~190.75 for the D-peptide and 191.25 for the N-peptide.  This is not a good quality for a sensor to possess.  I think that you would want any extra unbound sensor to be impervious to other conditions in the solution.  Is Xe sensing the pH of the solution (presumably the pKa of the D- and N- peptide is different)?  The authors did warn us that Xe NMR was sensitive.  The exact chemical shift of the Xe bound to cage equals ~60.5 ppm for the D-peptide and 62.25 ppm for the N-peptide.  I'll say it again - The authors did warn us that Xe NMR was sensitive.

Let's first consider the D-peptide.  When dye is added to the solution both peaks shift.  The aqueous Xe moves upfield and the Xe-cage moves downfield.  I think it is strange that the aqueous Xe shifts, but it is what it is.  The Xe-cage moves quite a bit more than the aqueous, Xe by the way.  Now let's consider the N-peptide.  As the dye concentration increases, the Xe-cage peak move upon binding with the N-peptide, albeit less than with the D-peptide.  The authors advocate using the difference between the peaks as a readout of the assay.  Using this metric, the Xe peaks move 0.78 and 0.33 ppm for the D-peptide-cage and N-peptide-cage upon binding 1 mM dye.  See figure below ...  

I am going to criticize this paper a bit, so I'll issue my usual caveat: I am a nobody doing LN2 fills and teaching 1st year graduate students how to acquire 1D 1H spectra, whereas the authors of this paper are real scientists doing the hard, creative work of producing scientific knowledge.  My purpose in starting this blog is not to trash people's work so that I feel better about my life.  I wanted to force myself (on a very public forum) to critique great science in hope of improving my science.  So here is my critique in one sentence: the negative control is not so negative.  Traditionally, in a paper like this one, you would have one positive and one negative control, proving that you get a signal when the analyte is present and no signal when the wrong analyte present.  In this paper the authors see a delta delta of 0.78 ppm for a carefully chosen strong binder and a delta delta of 0.33 ppm for a weaker binder.  If they could make a non-binder would the delta delta equal 0 ppm?  What would they see if they tried another dye molecule? 

Overall, I give the authors a lot credit for using a SELEX type procedure to hunt out a binder.  I kind of glossed over this point earlier to get to the spectroscopy, but it was a substantial effort to find a needle in a stack with 20^4 pieces of hay.  Clearly, this paper is a starting point.  The authors set out to show that their sensor-cage-Xe trick would is feasible.  I think they succeeded.  I wonder if their problem isn't going to be that Xe NMR is just too sensitive, though.