Newsletter Features: September 2003

USING LINEAR PREDICTION TO PROCESS 2D HMQC/HMBC DATA

For nmr structure elucidation of compounds with 25 or more carbons with closely spaced and overlapping proton resonances, 2D HMQC (or HSQC) and HMBC experiments have now become routine. These heteronuclear correlation nmr experiments give one bond HC connectivity (in HMQC/HSQC) and two and three bond HC connectivity (in HMBC), helping make structural assignments.

The majority of these 2D experiments are done using an inverse probe. The acquired fids, collected along the “f2 axis”, are proton frequencies and the second dimension (f1 axis) are a time incremented set of fids that contain the carbon information incoded in phase and amplitude modulation of the proton frequencies. This time incremented (f1) axis appears to involve a compromise between resolution and experiment time since doubling the number of time incremented spectra (default is 256) more than doubles the total experiment time. This need for compromise can be avoided by using the technique of ‘linear prediction’(LP). This is a processing technique and can be applied to any raw data files 1D or 2D.

The forward linear prediction (LPfr) is usually only applied to the f1 dimension of a 2D data set for obvious reasons, since this is where the least amount of digitization is, (i.e. 1024 points (TD2) in f2 and 256 points (TD1) in f1 dimension). The LPfr algorithm looks at the fid and continues to predict the digital points in time from the end of the fid. The parameters are set in >edp in XWINNMR…e.g. ME_mod=LPfr; NCOEF=100; LPBIN=1K (for 4 fold prediction if TD1=256). Remember, SI (in f1) is 1K which means that without LP the f1 dimension is being ‘zero filled’ from the 256th point up to 1024th point.

Page P164 in XWINNMR’s processing manual explains this in more detail, as does Ref 5 below.

Ref 5: Reynolds,W.F. Mag. Res. Chem. 35, 505-519 (1997)
The spectra below show variations on processing, using LP for a 2D HMQC for strychnine (30mg in CDCl3) but note TD1=only 122 increments collected in f1; ns=4; d1=1s (total acquisition time ~10 min!).

Fig 7a. LPfr=off (both H1 and C13 profiles are 1D spectra imported from standard 32K fids)
Fig 7b. LPfr=on ( NCOEF=50; LPBIN=1K; SI=1K for TD1 (i.e. f1 axis only)) (C13 profile is now the 2D positive projection across f1 axis (i.e. the sum of the contours in C13 dimension) with obvious resolution improvement along f1 axis by LP).
Fig 7c. LPfr=on (in TD2 SI=2K;LPBIN=1500;NCOEF=100) and (in TD1 SI=2K;LPBIN=1K;NCOEF=50) and WDW=SINE/TRAF (now LP is applied to both f1 and f2 dimensions and both profiles are the 2D positive projections along f1(C13) and f2(H1)). Note the excellent H1 resolution in the projection by using the ‘window functions’ WDW=SINE (in f2) and TRAF (in f1).

Back to Newsletter Table of Contents