NUS for HMBC and LR-HSQMBC experiments

Non Uniform Sampling (NUS) of multidimensional NMR data can greatly reduce the time taken to record a spectrum by recording only a subset of the normal data. A variety of algorithms are available to reconstruct the omitted data based on the data that was recorded. The most commonly used algorithm is Iterative Soft Thresholding (IST). Most implementations of the IST algorithm rely on the peaks in the detected dimension being phased correctly and positive. For most modern experiments this is not a problem, but in the HMBC experiment it is not possible to phase the peaks. For this reason I have not recommended using NUS with HMBCs. The LR-HSQMBC experiment, however, can be phased and I recommend using NUS with it. In this post, spectra recorded with different levels of NUS were recorded to determine how NUS affects HMBC and LR-HSQMBC experiments.

The sample used to record the spectra was a halogenated steroid dissolved in chloroform-d. All of its ¹H peaks appear between 0 and 5 ppm, and its ¹³C peaks between 0 and 70 ppm, so the spectral widths were reduced. The experiments were optimised to detect long range couplings of 8 Hz. ASAP versions of the pulse sequences were used with a relaxation delay of 30 ms, a DISPI-2 mixing period of 25 ms, and an acquisition time of 513 ms. ¹H decoupling was not used. All experiments were run with 16 scans and nominally 256 t₁ increments. The fully sampled (100%) spectra took 50 minutes, while the 12.5% sampled spectra took 6.5 minutes. Spectra were processed identically with Topspin 4.0.8 using the IST algorithm. The figure below shows the spectra. All spectra are plotted with the same contour threshold and level spacing. Click on the figure to see a larger version.

The spectra clearly show increased noise as the amount of sampling is reduced from 100% to 12.5%. In particular the tallest, sharpest peaks show increasing streaks of truncation artifacts as the sampling is reduced. The HMBC spectra appear to show more noise than the LR-HSQMBC spectra, but the HMBC signals are more intense and the noise is more obvious. Close inspection reveals the two experiments show little difference in the noise as sampling is decreased.

To make comparison easier, ¹³C traces running through the large signals near 1.1 ppm in the ¹H dimension were extracted and are shown below.

The extracted ¹³C traces show the appearance of artifacts as the sampling is reduced. At 12.5%, in particular, the artifacts become numerous enough to be problematic. Surprisingly, the HMBC experiment does not appear to have any more artifacts than the LR-HSQMBC. I was expecting the HMBC to be much worse.

These spectra indicate that HMBC and LR-HSQMBC experiments perform equally well with NUS. The spectra also indicate that reducing sampling to 12.5% produces lots of artifacts. The default sampling level on Skaggs Facility NMR experiments is 25%, except for the ASAP-HSQC where it is 50% because of the reduced acquisition time. I will update the HMBC parameters to include NUS with 25% sampling, but this value can easily be changed in the IconNMR interface when setting up experiments.

Acknowledgements

The sample for this work was kindly prepared and provided by Prof. Ted Molinski.