I usually take peak heights at the same position in both spectra and divide both. Some of the mistakes I did earlier on leading to wrong values:

1) Did you divide presaturated/reference or the other way around (happens more often than you think)

2) Are both spectra scaled the same way (happens when you process both spectra seperately, in Bruker this is the nc_proc variable)

3) Did you use a long recovery delay (5-10s)?

Hope this helps!

You are right, this is not possible. Check your reference spectrum. Maybe "off-resonance" saturation pulse, which you use, is a troublemaker? Try switch off proton saturation in the reference spectrum.

Maybe it is a trivial S/N problem. If you have low S/N (as for large globular proteins) and values close to 1, you will typically get some ratios >1. Sometimes you can get more reliable data if you fit your cross-peak around the maximum and obtain your intensity from the fitted surface maximum. This should reduce the noise contribution by simple averaging.

Two reasons are most likely:

1) You did not take into account the scaling of the NMR data when processing with Bruker TOPSPIN software. Internally, scaling is taken into account, but when you export the data it might get lost, and CCPN does not know that the data have been scaled (the Bruker parameter "2s nc_proc" will show you the extend by which data have been scaled).

The second reason is likely integration errors, poor S/N, peak overlap etc. or a too-short relaxation delay (although this mainly effect the signals with negative H-NOE that have long T1s) as pointed out by the others.

My experience is that with large molecules you are not getting very precise H-NOE data.

I have never attempted to do NMR studies/analysis of protein structures, so I am unfamiliar with the NOE issue you raise.

I would like to insist on one point made here by Oliver Zerbe, you have to use very long relaxation delays in the reference experiment. Rieko Ishima demonstrated very nicely this effect in her paper "15N–{1H} NOE experiment at high magnetic field strengths" published in JBNMR (2007), 37 p147. In the presence of cross-correlated relaxation, the return to equilibrium can be extremely long for both 15N and 1H. Beware though, the solution proposed in this paper is not valid, your safer bet is to increase the relaxation delay to 12-15 s. Note that you do not need such long delays in the "saturated" experiment, for which a delay of about 4 times your longest 15N T1 will be enough.

If you have more questions on this, I invite you to read the recent literature on the topic to which I have contributed (see link and references therein) as well as:

-"Measurement of 15N relaxation rates in perdeuterated proteins by TROSY-based methods" by Lakomek, Ying and Bax , JBNMR (2012), 53, p209.

I hope this helps.

Article On the measurement of N-15-{H-1} nuclear Overhauser effects....

It should be mainly from S/N problem. In this case S/N problem can occur two ways: one being the case of large molecular size and the second if the recycle delay is not long enough. In such cases the chances of picking noise are prominent and hence leads to >1 values for the calculated NOE.

Perhaps the inherent NOE in those particular signals is getting more pronounced as happens in normal small molecule magnetization transfer due to the presence of nearby nuclei to get this enhancement, maybe near active site where acidic or basic sites are present. Hope this helps.

Wow, I'm overwhelmed by the amount of responses!

Scaling errors are unlikely: I paid extra attention to them while re-processing and everything looks the same. The original data set also is a pseudo-3D, whose planes have all the same "nc_proc" and which can be directly analysed in CCPN using the "follow intensities" function. I also double-checked I did not mix up the planes while determining the values (most of the values look about right).

I don't want to leave out the off-resonance saturation pulse of the reference spectrum. It is needed to account for temperature differences – the saturation pulse pumps quite some power into the sample.

I will consider obtaining intensities „from the fitted surface maximum“, but this could pose a problem if there is signal overlap.

There is the possibility that a too short relaxation / recovery delay may be also one cause of the problem. For our TROSY-based sequence and deuterated proteins I usually go for 10 seconds (value from the suggested Lakomek / Ying / Bax paper, which may be a bit excessive), but for nondeuterated proteins and our HSQC-based sequence I use the the default value of 1.3 seconds. Probably this is easily a bit too short. I will test going up with the delay length.

Has anyone tested for an ideal length of recovery delay? I'd imagine a series of 1D versions of my experiment, with different length of delays – and the first one with low intensity is the one to go for?

Probably it is also a problem of S/N. Unfortunately, we are limited in terms of protein concentration and overall amount. So I doubled the number of scans and the problem did not disappear, but was attenuated. The effect is largest on low-intensity peaks, which indicates my noise is really just too strong.

So I will try a longer recovery delay and repeat the measurement with a higher number of scans. I will report back here if this solved the issues!

Thanks again for the helpful suggestions!

While i was reading your last advise, I could advertise the following, "Has anyone tested for an ideal length of recovery delay? I'd imagine a series of 1D versions of my experiment, with different length of delays – and the first one with low intensity is the one to go for?" sincerely speaking i think that you should perform the 1D spectral set in order to determine the precise minimum relaxation delay/recycling time, this since going to the interior of the protein the relaxation times of certain nuclei can incredibly be varied. This indeed as well as the overpassed problem of number of repetitions could be of great help an improvement of spectral resolution and banning of artifacts. I guess that the entire problem resides on particular acquisition 1D conditions for your own system, develop T1 & T2 experiments, before you start in the multidimensional experiments as well. Remember that T1 resembles the "lifetime" of the first order rate process returning the magnetization

to the Boltzman equilibrium, this mainly along the +Z axis, which due to molecular or system anisotropy can vary, and i strongly suggest also to revise the solid state relaxation techniques since your problem seems to be quite similar as in solids. Also the T2 spin-spin relaxation is quite changed due to nuclei crowding as is present in the interstices of huge molecules such as proteins. Hope this helps.

Did anyone consider that (H)-N is an exchangeable proton ? In this case the NOE data become very unreliable. I don't work with proteins, but with the small molecules the exchangeable protons sometimes give very strange results in NOE.

@Vladimir Khlebnikov:

Yes, as far as I understand that this is can be a problem and one solution is to apply water flip-back pulses to keep the water „quiet“.

@all & especially Fabien Ferrage:

I got some time to test several different recycle delays and indeed, it makes quite a difference! I settled with a total waiting time of 10 s for the reference experiment as a compromise. I could wait even longer, but this is increasingly painful in terms of spectrometer time.

For the steady-state experiment you suggested a recycle delay of 4 times my longest 15N T1 (which if I recall correctly equals some 99% of recovery). My protein has T1 times of up to ~2.5 sec – that would mean I'd have to wait also 10 sec for the saturation experiment!

I have to admit, I never cared much about the recycle delay. I will also have to revise my T1 / T2 experiments I guess. Starting from, Gong & Ishima, Lakomek & Bax and your paper I still have a lot of reading to do though.

Thanks again for the help!

@Martin Ballaschak . I observe a similar effect in my NH NOE intensities that the NOE shows larger than 1 value. I have been through the comments. Apart from increasing the recycle delay did you have to try something else too to get rid of the positive values for the noe? I do not know the T1 values of my samples yet.

thank you and appreciate any feedback from you.

regards,