|NMRanalyst for 3D NMR Spectral Analysis|
The 3D shape of bio-molecules determines their function in organisms. Molecular structure determination by NMR requires the analysis of a series of multidimensional spectra. Varian, Inc. provides ProteinPack used below to facilitate the data acquisition. We have extended NMRanalyst for the automated analysis of 3D data. Given the number of spectra to be analyzed and the huge number of resonances per spectrum, this 3D analysis module provides an important advance for the routine use of 3D NMR.
Application I: 3D HNCO Spectrum of Human Ubiquitin
The analysis of all eight phase components of this 128 MByte spectrum takes 2 hours on a 3.2 GHz laptop. NMRanalyst exhaustively searches all spectral positions and produces a complete numerical description (signal integral, resonance frequencies, linewidths, phases, and corresponding error values) of identified spectral resonances. This information can be summarized in a simulated spectrum as shown below. The following residual spectrum illustrates which experimental signals are not completely explained by the reported spin systems.
Experimental HNCO Spectrum
Simulated HNCO Spectrum
Residual HNCO Spectrum
Application II: 3D HNCACB Spectrum of Human Ubiquitin
Experimental HNCACB Spectrum
Simulated HNCACB Spectrum
Residual HNCACB Spectrum
Graphical illustrations are convenient for evaluating the obtained analysis results. But most applications should
be based on the obtained numerical spectral description. The following table shows how the software summarizes the
backbone shift information determined from the HNCO and HNCACB spectral analyses. All shift values are given in ppm.
Neither the amino acid types nor the residue positions in the protein sequence are known. So determined residues
are sorted in increasing 15N shift and are sequentially numbered as shown in column
Traditionally, strip plots are used to graphically determine the amino acid sequence in a protein. Based on the
determined shifts given in the table, strong and weak C and
C resonances from residues can be matched numerically. The
This table does not contain 10 of the 76 ubiquitin residues. Prolin resonances (P19, P37, P38) are not detectable
in HNCO and HNCACB spectra due to the lack of an amide proton. The remaining unobserved ubiquitin resonances are
not sufficiently separated from other resonances for identification. The spectral areas in which such problems
occur can be identified from the residual spectra above. All reported backbone shift assignments are in agreement
with published values except for one challenge. Residues 1 and 8 in the table correspond to G47 and G75 and all
glycin residues have no C resonance. The reported 16.68 and 30.62
ppm shifts belongs to the sequential residue and should be listed in the
Correct sequential assignments (column
Application III: 3D HCCH-TOCSY Spectrum of Human Ubiquitin
The HCCH-TOCSY spectrum of the above described human ubiquitin sample was acquired in 9 hours and 42 minutes using ProteinPack's ghcch_tocsy pulse sequence (64 x 32 x 512 phase sensitive points, four transients per increment). The data was linear predicted to 128 x 64 x 512 points and no line broadening was used.
Experimental HCCH-TOCSY Spectrum
Simulated HCCH-TOCSY Spectrum
Residual HCCH-TOCSY Spectrum
The identification of amino acids based on the known C and C resonances and the observed HCCH-TOCSY peak patterns, and the derivation of their side chain carbon and proton assignments are not completed yet.
Application Results: NMRanalyst yields excellent spectral descriptions as shown by the HNCO and HNCACB residual spectra. Most residuals could be eliminated by using a spectrum acquired with higher resolution and by using more sample or longer acquisition times (further phase cycling). The generated table of determined backbone chemical shifts provides the HNCO and HNCACB analysis results in a format needed for the further protein structure determination.
Super Peak Picker: The acquisition time on a high-field NMR instrument is expensive. To maximize the information extracted from the acquired NMR spectra, the simplistic peak picking should be replaced by the more sensitive, selective, and reliable spin system modeling used by NMRanalyst. NMRanalyst's 3D module can analyze all types of phase sensitive 3D NMR spectra. Functioning as a "super peak picker", its results provide clean input for post-processing software or for the further human interpretation.
3D Phasing: Since NMRanalyst analyzes all spectral phase components, there is no need to optimize the data acquisition for minimal phase distortions or to attempt manual phase correction. NMRanalyst determines better phase functions than obtainable visually (U.S. Patents 5,218,299 & 5,572,125). The determination of phase functions is optional and if available they are used to increase the analysis speed.
Software Availability: The NMRanalyst 3D analysis is available for Red Hat Linux and Microsoft Windows. The Varian and Bruker (AMX and DMX) FID and spectrum formats are implemented. But the Bruker formats have not been tested. If you have access to 3D Varian or Bruker data and are interested in beta-testing this software, please contact dunkel@ScienceSoft.net.
Summary: The identification of spin systems (resonances) in real-world noisy 3D NMR data can be automated! A visual data analysis clearly excels compared to simplistic peak picking. But analyze the 3D data with NMRanalyst and take a look at the residual spectrum. All spectral signals not shown in the residual spectrum were properly characterized and reported. The structure determination of proteins by NMR involves further interpretation steps. But for the data acquisition and identification and characterization of spectral signals, we believe we have found an adequate solution.