Application of a New Expert System for the Structure Elucidation of Natural
Products from Their 1D and 2D NMR Data
Mikhail E. Elyashberg,† Kirill A. Blinov,† Antony J. Williams,*,† Eduard R. Martirosian,† and
Sergey G. Molodtsov‡
Advanced Chemistry Development Inc., 90 Adelaide Street West, Suite 702, Toronto, Ontario, M5H 3V9 Canada, and
Novosibirsk Institute of Organic Chemistry, Siberian Branch of Russian Academy of Science, Lavrentiev Avenue 9,
Novosibirsk 630090, Russia
Received June 29, 2001
Described herein are applications of the latest version of the StrucEluc expert software system, enhanced
to use 2D NMR data, to the structure elucidation of 60 recently isolated natural products. In this study,
selected molecules containing between 15 and 65 skeletal atoms and having molecular masses ranging
from 200 to 900 amu have been investigated. The correct structure was determined unambiguously for
58 of these molecules. The structures for 75% of the data sets were determined in less than one minute,
while 90% of the analyses required no more than 30 minutes. The strategy of structure elucidation by
this expert system is described, and several examples are discussed. These illustrate that StrucEluc is a
powerful and versatile analytical tool for the structure elucidation of natural products.
A large number of reports have been devoted to the
structure determination of natural products using 2D NMR
spectroscopy. Not surprisingly, a number of computer-
assisted methods, particularly expert systems, have been
devised to build on the structure-solving techniques devel-
oped in these investigations. By increasing the speed and
reliability of structure elucidation, these expert system
methods are poised to become significant contributors in
the development of new pharmaceuticals. The first-genera-
tion expert systems (such as EXPERT,1 CHEMICS,2 SES-
AMI,3 SpecSolv,4 and the first version of StrucEluc5) used
only 1D NMR data in conjunction with other types of
analytical information and did not provide a general
solution to the structure elucidation issue. Reported data
suggest that these applications will usually fail to deter-
mine the structure of molecules containing more than about
25 skeletal atoms. This limitation is largely due to the
severe underdetermination of the problem using only 1D
NMR data in conjunction with other analytical methods.
In the early 1990s, the first publications describing second-
generation expert systems that used 2D NMR data for
structure determination of organic molecules began to
appear. New versions of SESAMI,6,9 CHEMICS,10 and
StrucEluc11,12 are now available that can use 2D NMR data.
In addition, several new programs that can use 2D NMR
information have been released. These include CISOC-
SES,13-16 LSD,17 LUCY,18 and COCON.19 A review of a
number of the 2D NMR expert systems has been recently
published by Jaspars.20
Studies of the structure determination of specific natural
products using some of these systems have been pub-
lished.6-9,14-18,21-22 Unfortunately, only some of these
reports present detailed information on the structure
elucidation process.7,14-16 In others, this information is
limited or missing, making meaningful comparisons be-
tween these programs almost impossible. The authors
believe a thorough and systematic analysis of the features,
molecular size capacity, computational metrics, and limita-
tions of these expert systems using a large number of
natural product test cases has not been published. Few
summaries of actual structure elucidation case studies exist
that discuss the difficulties encountered in the elucidation
process and their possible resolutions. In particular, the
issue of contradictions generated from 2D NMR data has
received relatively little attention and calls for detailed
study.
The aim of the present investigation was twofold: first,
to address the above gaps in the literature of expert system
structure elucidators; and, second, to report on tests of the
efficiency of a 2D NMR-based expert system as a routine
tool for structure determination of newly isolated natural
products. To meet these goals, the StrucEluc system11,12
has been applied to the structure elucidation of natural
compounds using 1D and 2D NMR, IR, and mass spectra.
To make these results more easily reproducible, molecules
from recently published natural product structure deter-
minations were chosen that contained listings of 2D NMR
data.
The 1D version of StrucEluc, comprising two interacting
processes, was described previously.5 This system relied
on a knowledge base consisting of 140 000 assigned 13C
NMR spectra, a fragments library containing about 500 000
fragments with assigned 13C NMR subspectra, and spec-
trum-structure correlations for NMR and IR spectra. The
program could use all of the 1D NMR, IR, and mass spectra
available. Two structure generators utilized different prin-
ciples, one library-based, the other molecular formula-
based, to provide the system with operational flexibility.
However, the 1D StrucEluc shared the limitation common
to all similar systems constrained to using only 1D NMR
data: it failed to analyze molecules containing more than
20-25 skeletal atoms due to the severely underdetermined
nature of the structural search.
Recently, a new process11,12 has been developed that can
be used either independently or in conjunction with the
two original processes. It is intended primarily for structure
elucidation of large organic molecules and is based on the
use of 2D NMR data. The use of these data, in combination
with a knowledge base, spectrum-structure correlations,
candidate structure spectral prediction, and numerous
other software aids, has significantly extended the system’s
* To whom correspondence should be addressed. Tel: 919-570-0217.
Fax: 425-790-3749. E-mail: Tony@acdlabs.com.
† Advanced Chemistry Development Inc.
‡ Novosibirsk Institute of Organic Chemistry.
693J. Nat. Prod. 2002, 65, 693-703
10.1021/np0103315 CCC: $22.00 © 2002 American Chemical Society and American Society of Pharmacognosy
Published on Web 04/25/2002

ability to handle molecules containing more than 25
skeletal atoms.
This work describes the results of application of the
StrucEluc for structure determination of 60 new com-
pounds from natural sources. 1D and 2D NMR spectra of
most of these products were published in the Journal of
Natural Products during the year 2000. These studies
demonstrate that StrucEluc can be successfully applied to
the structure elucidation of new and novel natural product
molecules containing more than 60 skeletal atoms.
Results and Discussion
The 2D NMR module is based on a number of programs
developed for deducing the molecular structure from a
combination of 2D NMR spectra. The most typical combi-
nation providing an appropriate basis for the structure
determination includes 1H-13C HMQC and HSQC (see ref
23; more recently the ADSQC experiment has been applied
due to superior line shape and resolution relative to
HMQC24), 1H-1H COSY, and HMBC. The StrucEluc sys-
tem presently operates with the following 2D NMR meth-
ods: HMQC, HETCOR, HSQC, HMBC, COLOC, INAD-
EQUATE, COSY, TOCSY, ROESY, and NOESY. The
program can utilize data from both 1H-13C and 1H-15N
heteronuclear correlation experiments. In addition to the
spectral data, the program also needs at least one possible
molecular formula.
The following section illustrates the use of the 2D module
for the structure elucidation of polycarpol (C30H48O2;
compound 1, Figure 1). The initial data were taken from a
literature description of the structure elucidation of poly-
carpol using the LUCY program.18
To begin, the molecular formula, the 1D 13C NMR
spectrum, and the coordinates of all cross-peaks (COSY,
HMQC, and HMBC) were entered into the program. The
2D data18 shown in Table 1 were entered into StrucEluc
manually. However, it should be noted that, when avail-
able, raw data can be processed, analyzed, and easily
transferred to StrucEluc using a companion software
program.25 This program allows the processing of both 1D
and 2D NMR data as well as reading in preprocessed
spectral data files from the primary spectrometer vendors.
The program allows automated peak-picking through
specific algorithms, but ultimately it is the responsibility
of the user to ensure that the appropriate input data set
is fed into the StrucEluc program. This can include
identification and removal of appropriate artifacts or
spectral responses including one-bond responses from
HMBC experiments. An advantage of using these compan-
ion programs is that the cross-peak intensities are auto-
matically transferred to StrucEluc. When available, Struc-
Eluc uses cross-peak intensity data to estimate the number
of skeletal bonds separating correlated nuclei. This is
generally more appropriate for COSY experiments rather
than long-range heteronuclear experiments. In those situ-
ations where short-range couplings may be weak, for
example, two-bond HMBC peaks in olefinic or aromatic
fragments, the default values will account for the range of
possible values. In those cases where the actual detail of
the couplings has been proven to be 2-,3- or even higher
bond range, the actual bond order of the coupling can be
explicitly defined in the input table even though these data
are rarely published. Because the cross-peak intensity data
were unavailable for the test cases reported here, the
program default values were relied on. These assign the
ranges of 2-3 and 1-3 intervening bonds between cor-
related nuclei in COSY and HMBC experiments, respec-
tively. In those situations where a two-bond peak is so weak
that it cannot be observed, this may lead to an incomplete
representation of the skeletal framework since certain
connectivities will not be available. In these cases the
structure generation mode of the problem will attempt to
fill in the appropriate gaps left by these absences.
Once the data were entered, the NMR cross-peak data
tables were transferred into tables of carbon atom connec-
tivities. The program is supplied with convenient aids for
detecting and removing user mistakes that can arise in the
connectivity data during the data entry process. Since the
information in these tables provides the foundation for the
structure generation process, they must be examined for
consistency (i.e., the absence of contradictions). The main
cause of contradictions arising from 2D data is during the
interpretation of cross-peak information: if the maximum
number of intervening bonds assigned to a cross-peak is
less than the true number of bonds, a contradiction will
arise. Because contradictions can produce wrong connec-
tivities, particularly for heteronuclear long-range correla-
tions across five or more bonds (see primary review),26
methods have been devised to resolve them.
Contradictions arising from 2D NMR data sets have been
discussed in the literature, and an approach that allowed
one to perform structure generation in the presence of
definite contradictions and ambiguous connectivities was
offered.13-16 Unfortunately, the robustness of this approach
was not proven by a significant number of examples, so a
method was developed for automatically detecting and
resolving contradictions prior to the actual structure
generation process. In so doing, a series of specific criteria
for contradiction detection was created.11 In the polycarpol
(1) example, the contradiction detection routine utilized
found that the data taken from the article were fully self-
consistent.
Table 1. NMR Data Used for Elucidating the Structure of
Polycarpol (1)
DEPT äC
C,H-COSY
cross-signal
with äH
HMBC
cross-signal
with äH
H,H-COSY
cross-signal
with äH
C 146.0 6.00, 2.00/2.26, 1.01
C 142.0 5.30, 1.04
C 131.0 1.66, 160
CH 125.5 5.20 1.66, 1.60
CH 122.0 6.00 2.05/2.14 2.05/2.14
CH 116.3 5.30 2.00/2.26 2.00/2.26
CH 79.0 3.15 1.68
CH 74.8 4.30 1.04 1.83/1.93
C 52.5 1.04, 0.64
CH 49.7 1.14 1.00, 1.01, 0.91
CH 49.3 1.65 0.87, 0.64 1.83/1.93
C 44.4 5.30, 1.65, 1.04, 0.64
CH2 39.8 1.83/1.93 4.30, 1.65
C 39.0 1.14, 1.00, 1.68, 0.91
CH2 38.8 2.00/2.26 0.64 5.30
C 37.7 1.14, 1.68, 1.01
CH2 36.8 1.07/1.45 0.87
CH 36.4 1.32 0.87 0.87
CH2 36.3 1.38/1.92 1.01 1.68
CH3 28.3 1.00 0.94
CH2 27.9 1.68 3.15, 1.38/1.92
CH3 25.9 1.66 1.60
CH2 25.3 2.00
CH2 23.3 2.05/2.14 6.00
CH3 23.0 1.01 1.14
CH3 18.7 0.87 1.32
CH3 17.8 1.60 1.66
CH3 16.1 0.64 1.65, 2.00/2.26
CH3 16.0 0.91 1.14, 1.00
694 Journal of Natural Products, 2002, Vol. 65, No. 5 Elyashberg et al.

Once the data were determined to be free of contradic-
tions, structure generation was conducted without altering
any other program default constraints. The generator
produced six structures in an elapsed time of 6 s (Celeron
operating at 500 MHz, Window98, RAM 128 Mb). Then
StrucEluc was used to predict the “accurate” 13C NMR
spectra for the six generated structures, ranking the
structures in order of increasing average deviation between
these accurate estimates and the experimental data, dA.
The correct structure of polycarpol (1) had the lowest dA;
the ranked structures are shown in Figure 1. The deviation
of the polycarpol molecule was 1.71 ppm; the dA value of
the next-best candidate structure was 2.49 ppm. It should
be noted that the LUCY program produced the same six
structures, but took 2 h (Pentium PC operating at 100
MHz) and left the selection of the correct structure to the
chemist (the presence of spectral prediction aids was not
mentioned in ref 18). In StrucEluc, stereochemistry is
utilized in the prediction of both 1H and 13C chemical shifts.
Certainly E/Z isomers can be clearly distinguished and
extracted automatically. At present, 2D frameworks are
provided for the molecules assembled through the elucida-
tion process without specific spatial stereochemistry de-
fined other than for E/Z isomers. In those cases where the
input data provide a direct hit in the assigned structure
database the stereochemistry may be distinguished di-
rectly.
Features of StrucEluc. StrucEluc provides a graphical
user interface (GUI) that allows easy and intuitive visual
analysis and editing of connectivity data. Carbon resonance
multiplicity, heteroatoms, primary structural blocks (PSBs),
and “unattached” hydrogen atoms (usually belonging to a
heteroatom) are displayed. The default display properties
color-code the different atoms and groups, as well as the
different connectivities; the user can easily change the color
assignments. This scheme permits “at-a-glance” knowledge
of whether two atoms are linked by exactly one bond,
exactly two bonds, or a specific range of bonds. Information
from different experiments can be viewed or suppressed
by clicking appropriate buttons on the toolbar. Values of
the associated 1H and/or 13C NMR chemical shifts can also
be displayed, if desired, as shown in Figure 2. This figure
represents the StrucEluc interface. The upper left corner
represents the molecular formula of the molecule repre-
sented as distinct nuclear centers. The numbers of qua-
ternary, methane, methylene, and methyl fragments have
already been deduced using a combination of the spectral
data available. The display options used include the display
of the 1H NMR chemical shift below each nuclear center,
and the lines drawn between fragments represent the
connectivities deduced using the COSY data. The chemist
is also able to draw bonds of any multiplicity between
appropriate atoms to set user fragments (CdO, O-CdO,
O-H, etc.) This provides a quick and intuitive mechanism
for entering structural information evident from 1H NMR
and/or IR spectra without the need for typing exhaustive
tables of numerical data. The rest of the figure represents
the chemical shift projections of both the 1H and 13C data
extracted from the spectral data inputs. The upper right-
hand corner displays a reconstruction of the 2D HMBC
data set.
The program analyzes 13C and 1H NMR chemical shifts
of CHn groups (n ) 0-3) and automatically sets atom
properties for each carbon atom. These properties are the
Figure 1. Polycarpol (1) structure elucidation. Output structural file ranked in order of increasing dA values.
Structure Elucidation of Natural Products Journal of Natural Products, 2002, Vol. 65, No. 5 695

allowed hybridizations (sp3, sp2, sp, not defined) and the
possibilities of heteroatom connectivity (obligatory, forbid-
den, not defined). To set these atom properties, StrucEluc
generates special correlation tables from appropriate frag-
ments and structures stored in the system database. It is
very important that both 13C and 1H NMR chemical shifts
are taken into account when setting the atom parameters,
so this requirement has been built into the program. The
chemist can edit these parameters to introduce a priori
background information via the connectivity GUI. The data
generated at this stage are used as input for the specialized
2D structure generator, which assembles atoms and frag-
ments into larger fragments and structures using the
constraints imposed by the topological distances between
correlated atoms. While NOESY and ROESY data can be
included in the input data, they provide only supporting
connectivity information to the COSY and TOCSY homo-
nuclear correlation experiments. At present, these experi-
ments cannot be used for determining stereochemistry
directly, but this capability is being examined.
Structures are generated from the structural blocks,
PSBs, and user-defined fragments (if entered) under
constraints imposed by the tables of carbon atom connec-
tivities. This includes the map of connectivities and any
additional constraints imposed by the chemist (maximum
bond multiplicity, allowed ring sizes, etc.).
Generated structures can be verified by means of the aids
common for all processes of the StrucEluc system (filtering
with libraries of spectrum-structure correlations, GOOD-
LIST, BADLIST, etc.). The GOODLIST and BADLIST may
be formed automatically, or manually as a result of a
fragment search using the approach described in the
literature.5 In addition, application of spectral filtering is
very effective and allows the program to reject an enormous
number of invalid structures.
As the generation is carried out from structural blocks
containing assigned carbon and hydrogen atoms, all the
structures generated intrinsically possess assigned spectra.
Even within the constraints imposed by long-range con-
nectivity limits, the structure generator may produce
identical structures with slightly different carbon atom
assignments. The reason for this is due to the fact that a
series of structure-spectral pairs can be created using the
spectral data available to the program. These may result
from the combination of the data in subtly different ways,
often with the interchange of only two resonances in close
chemical shift proximity, to give rise to the same structure
but with different nuclear assignments. For this reason,
the structural file was checked for structure identity, and
all but one structure in each set of duplicated structures
was eliminated. The structure retained in each group of
duplicates was the one whose intrinsic 13C NMR spectrum
best matched the 13C NMR spectrum calculated “from
scratch”. This process may be performed automatically
upon user request through a menu selection.
One of the most important and distinctive attributes of
the StrucEluc system is its array of powerful predictive
tools. These include several types of spectral estimation
and routines for the prediction of many physicochemical
properties. StrucEluc is capable of estimating the spectra
of candidate structures using the following methods: (1) a
fast 13C NMR chemical shift prediction based on increment
rules; (2) an accurate 13C NMR chemical shift prediction;
(3) 1H NMR chemical shift and, optionally, full spectral
prediction including coupling patterns correctly modeling
dihedral angle dependencies and second-order effects; (4)
calculation of candidate structure correspondence to the
experimental mass spectrum with fragmentation (if avail-
able).
In the authors’ experience, the fast 13C NMR chemical
shift prediction provides the best compromise between
speed and accuracy for the first level of candidate structure
filtering. The fast 13C NMR spectrum prediction is per-
formed for all structures included into the output file, and
Figure 2. Polycarpol (1) 1H-1H COSY NMR connectivities.
696 Journal of Natural Products, 2002, Vol. 65, No. 5 Elyashberg et al.

the average deviation, dF, between each of these estimated
spectra and the experimental 1D spectrum (or the chemical
shifts derived from suitable projections of the 2D data) are
calculated. The structures are then ranked by increasing
dF. The speed of the fast spectral prediction (1-2 s per
structure, even for large molecules) makes it practical for
use as an initial screen of all structure candidates, even
for lists containing thousands of structures. This ability is
very important, as it frees the user from imposing other-
wise unjustifiable constraints (which may exclude the
correct structure) in order to make the program finish in
a reasonable amount of time.
The correct structure usually falls near the top of the
dF-ranked list. To better discriminate between the most
likely candidates, a more accurate 13C NMR spectral
calculation is performed for the first 10-20 structures of
the dF-ranked list. A second statistic, dA, is calculated for
each structure from the average deviation between the
more accurately predicted and experimental spectra. As
discussed below, the correct structure is almost always the
one possessing the smallest dA statistic. Once computed,
the user can have StrucEluc sort the structures in ascend-
ing dA order to bring the most likely candidate to the top
of the list.
Three more criteria can be used to evaluate structures
possessing similar dA values or to provide independent
confirmation of the most likely structure indicated by this
parameter. First, the 1H NMR chemical shifts can be
predicted and ranked by deviation from the corresponding
shifts obtained from suitable projections of the 2D data.
Second, if an experimental 1D 1H NMR spectrum is
available, full 1H NMR spectra, including not only chemical
shift information but expected coupling patterns (derived
from database values and adjusted for dihedral-angle
dependencies and second-order perturbations) can be pre-
dicted for each candidate structure of interest. Visual
comparison between these predicted proton spectra and the
experimental one provides another criterion for accepting
or rejecting candidate structures. During this work it was
observed that, when present, the near first-order multiplets
associated with methyl groups in a calculated 1H NMR
spectrum are diagnostically valuable and frequently pro-
vide convincing fingerprint-like confirmation of the most
likely structure. Third, if an experimental mass spectrum
is available, the program can analyze each structure for
assignable fragments. The resulting percentage of assigned
experimental MS peaks provides another criterion for
ranking proposed structures. Finally, StrucEluc also in-
corporates routines for the prediction of several physico-
chemical parameters (log P, pKa, boiling point, solubility,
surface tension, etc.) When experimental physicochemical
data are available, prediction of these properties provides
several more independent criteria for potential structure
evaluation and aids in further characterizing the compound
once its structure is determined.
All but the fast 13C NMR spectral calculations can be
improved by prior 3D optimization of the structure under-
lying the predictions. The 3D optimization algorithm allows
the planar (2D) structure to be rapidly translated into a
realistic three-dimensional structure. It is based on modi-
fied molecular mechanics which take into account bond
stretching, angle bending, internal rotation, and van der
Waals nonbonded interactions. Modifications include minor
simplification of potential functions and enforcement of the
minimization scheme by additional heuristic algorithms for
dealing with poor starting conformations. The proprietary
3D optimization algorithm uses a force field initially based
on classical CHARMM parametrization. The modifications
involve some simplification and were intended to increase
the stability and speed of computation. It should be
emphasized that the 3D optimization, as well as all of the
spectral predictions, is performed automatically. They may
be applied to the entire list of possible structures generated
by the program (or to any user-defined subset of this list)
by using a menu to choose the spectra and options required,
then clicking a button.
StrucEluc can also be used in a “reverse” mode to check
the validity of 2D NMR assignments of a given structure
from multiple experiments. If any contradictions are found,
the program displays a conflict notification message ex-
plaining the cause(s) of the conflict(s).
One of the unique advantages of the StrucEluc is its
ability to allow the user to utilize the system knowledge
base effectively in solving underdetermined structures with
2D NMR data. In some cases, the total number of 2D NMR
correlations is insufficient for structure elucidation. This
is usually revealed by a seemingly endless structure
generation stage. If such a situation arises, the user has
an opportunity to perform fragment searches in the knowl-
edge base, which can be combined with user-generated
auxiliary databases (described in the next paragraph), if
these were not provided to the program at the start of the
data analysis. In such cases, one may use a menu command
to initiate a search for fragments whose spectra match
parts of the experimental spectra. Any fragments found can
be integrated into the 2D connectivities map using another
menu command. This can significantly accelerate the
structure generation process.
StrucEluc also provides for the automatic creation of a
user database containing fragments obtained from a set
of assigned structures similar to the molecule under
investigation (e.g., derived from the same or related
biological sources). These fragments can be employed in
two ways: they can be directly drawn into the 2D connec-
tivities map, and they may also be used in automatic
fragment set generation. These fragment sets can be used
for structure generation as described previously.5 This
approach of creating and using a user fragments database
is analogous to the “common sense” method frequently
utilized by chemists: to elucidate the structure of a new
compound, a chemist often makes use of comparisons and
contrasts between data from a new compound and that
from similar compounds (or compounds from similar
sources). The evident advantage of the system application
is that all procedures are performed either automatically
or in an interactive mode.
Selection of Test Data. The performance of StrucEluc
was initially tested by elucidating the structures of 10
natural products, using 2D NMR data taken from refs 27
and 28 and a series of articles published by authors of
previous expert systems.6-9,14-18 Articles were carefully
chosen where the 2D NMR spectral data were presented
clearly and without evident mistakes. The goal of these
initial studies was not only to test StrucEluc but also to
compare the results described here with those obtained by
other expert systems. Utilizing StrucEluc, all 10 problems
were successfully solved in reasonably short time periods;
the results are summarized in Figure 3. The correct
structure was identified unambiguously by dA-ranking in
all but one case: structure 3.8 was ranked second.
Comparison of the speed of StrucEluc with other systems
was difficult because documentation of processing times
was available only for CISOC-SES13 and LUCY18 systems.
The results described in this work showed that the
Structure Elucidation of Natural Products Journal of Natural Products, 2002, Vol. 65, No. 5 697

structure generation speed of CISOC-SES was comparable
to that of StrucEluc, while the LUCY system was several
orders of magnitude slower than StrucEluc.
To further characterize the performance of StrucEluc in
the structure determination of natural products, 2D NMR
data from 50 natural products studies29-62 published in the
Journal of Natural Products mainly during 2000 were used.
Structural formulas of these compounds, together with the
StrucEluc results, are shown in Figures 4-6. The 60 mole-
cules totally discussed in this article covered a broad range
of masses and skeletal sizes, as illustrated by the histo-
grams presented in Figures 1 and 2 (Supporting Informa-
tion). In summary, the molecules were reasonably large
and fairly complex, ranging in mass from 200 to 900 amu
and having between 15 and 65 skeletal atoms.
In these studies, StrucEluc was configured to attempt
automatic contradiction resolution whenever the contradic-
tion check found an inconsistency. To resolve these prob-
lems, the automated routine increments (by one bond) the
upper limit on the number of linking bonds. Ideally, one
could preclude contradictions by unambiguously associat-
ing each cross-peak with the number of intervening bonds.
When available, data from INADEQUATE-type experi-
ments are invaluable in this regard. Unfortunately, the
concentration and instrument time requirements of this
experiment render such information inaccessible in the
majority of cases. When additional experiments are not
feasible, attempts to resolve contradictions by educated
trial-and-error can be quite successful. Methods for recog-
nizing likely sources of contradiction are discussed below,
and strategies for alleviating the most common of these
problems are discussed in the next section.
The 2D NMR data in the source articles were presented
either in tables of chemical shifts or as structures with
graphical correlation schema. Since cross-peak intensities
are usually not published, it was not possible to use them
as criteria to determine the most probable bond separation
between correlated atoms. Therefore, each solution was
started using the StrucEluc default values mapping cross-
peaks to bond separation; these default values depended
Figure 3. Results of the preliminary system testing. Designations: n ) number of skeletal atoms, k ) number of structures in an output file,
r ) position of the right structure in an output file, t ) elapsed time of structure generation. Figures in square brackets show the references from
which the 2D data were taken.
698 Journal of Natural Products, 2002, Vol. 65, No. 5 Elyashberg et al.

only on the type of 2D data (COSY, HMBC, etc.). These
defaults were appropriate for about 75% of the cases tested.
This ratio indicated that the defaults set by the program
represent a reasonable compromise between thoroughness
and required computational time. Investigations of the
source of the problems in the remaining cases indicated
that they all had exhaustive lists of long-range connectivi-
ties, one or more of which (per structure) fell outside the
default limits used by StrucEluc.
An in-depth investigation of those cases that generated
contradictions was performed. Most often, the contradictory
data arose from COSY correlations (up to five per structure
in some cases) or HMBC correlations (again, up to five per
structure) across more than three bonds.
Some of these exceptions were very easy to recognize.
For instance, it is readily apparent that a problem exists
for a CH2 group assigned three one-bond connectivities to
non-hydrogen atoms, or for a CH3 group assigned more
than one one-bond connection to non-hydrogen atoms,
based simply on the tetravalent nature of carbon. These
contradictions can be resolved manually by replacing all
connectivities to a “defective” atom (e.g., one where the
tetravalency rule was in violation) with fuzzy (or fuzzier)
connectivities. [A fuzzy connectivity is one where a range
Figure 4. Structures recognized when 2D NMR data either contained no contradictions or the contradictions were removed by the program.
Designations: k ) number of structures in the output file, r ) position of the right structure in an output file, t ) elapsed time of structure
generation, HMBC ) only HMBC long-range correlations were available. Figures in square brackets show the references from which the 2D data
were taken.
Structure Elucidation of Natural Products Journal of Natural Products, 2002, Vol. 65, No. 5 699

of values is specified, as opposed to a definite connectivity
such as one expects in HMQC data, where only one-bond
(direct) H-C connectivities are expected.]
By increasing the maximum allowed number of inter-
vening bonds by one for all the correlations to the offending
atom, contradictions in the data set may be resolved. This
action effectively tells the program that one or more of the
correlations may represent a 4JHH or a 4JCH rather than a
3JHH or a 3JCH cross-peak. However, the loosened restraints
come at the expense of increased computational time and
Figure 5. Structures recognized when contradictions were manually removed from the 2D NMR data. Designations: k ) number of structures
in the output file, t ) elapsed time of structure generation, HMBC ) only HMBC long-range correlations were available. Figures in square brackets
show the references from which the 2D data were taken.
Figure 6. Elucidation of the structure of Cryptolepis alkaloids. Designations: k ) number of structures in the output file, t ) elapsed time of
structure generation, B ) only HMBC long-range correlations were available. Figures in square brackets show the references from which the 2D
data were taken.
700 Journal of Natural Products, 2002, Vol. 65, No. 5 Elyashberg et al.

a larger output list of possible structures. The number of
possibilities to be checked increases nonlinearly as con-
straints become fuzzier, and only one (or a few, allowing
for duplication) of the structures found to fit the data can
actually be correct. This is the basis of the earlier statement
about the appropriateness of the StrucEluc default values
for these parameters.
Other sources of contradictory data could be found by
analyzing the connectivities map for several simple, well-
known situations where coupling can be observed across
more bonds than usual. One or more of the well-known
mechanisms that propagate J-couplings beyond the “nor-
mal” number of bonds could usually explain these excep-
tional cases: allylic, homoallylic, or conjugated systems,
and zigzag (W-geometry) arrangement of the intervening
bonds.
In general, a preliminary contradiction check was run
on each test case before launching the fully automated
elucidation program, and any simple contradictions were
resolved manually. All contradictions must be resolved
before the program will start to generate structures.
Because manual correction of obvious contradictions such
as those presented above is faster than relying on the more
rigorous, fully automated resolution protocol, this manual
pretreatment of the connectivity parameters helped to
decrease the time required to reach a solution.
Several levels of rigor are available for the contradiction
tests; normally the “not checked” mode was employed. This
meant that there were no constraints (beyond valence
considerations) in the bonding to and between carbon
atoms. The defaults for the other types of allowed bonding
were kept (no heteroatom-heteroatom bonds at all, no
increase in connectivity link when merging connectivities).
In addition, the “Use NMR Shifts Correlation Table” option
was used for all spectra except those with very unusual
chemical shifts.
In most cases, if the program did not detect contradic-
tions, the solution set included the correct structure.
However, in some cases, the program failed to detect
contradictions present in the 2D data, and in those cases
the program either refused to generate any structures or
produced a solution set that did not contain the correct
structure. This occurred because the search for contradic-
tions is based on heuristic principles that do not exhaus-
tively check for all possible contradictions. For instance,
connectivities across more than five bonds that are in
conflict with other connectivities may not be detected. This
represents a design tradeoff: exhaustive contradiction
analysis yields diminishing returns and increases processor
time.
Contradiction Resolution Strategies. The number of
correlations obtained from 2D NMR experiments can
number in the hundreds. Therefore, attempts to resolve
contradictions by arbitrarily increasing the limit on the
number of intervening bonds for all correlations will be
grossly inefficient. Since there can be only one correct
structure, nearly all structures generated by this action
will be incorrect. Furthermore, the calculation time re-
quirements would become impractical. Fortunately, experi-
ence has shown that the majority of contradictions can be
resolved by judicious choice of the structure blocks for
which the upper bound on the number of intervening bonds
is increased.
This work demonstrated that, in most cases, long-range
correlations propagating through more than the “usual”
number of intervening bonds involve methyl groups. The
intensity of a long-range cross-peak to the three equivalent
protons comprising a methyl group is usually much more
intense than the corresponding cross-peaks to methylene
or methine protons. This is not only because of the
increased number of protons in the methyl group but also
because this intensity is concentrated in a narrower
spectral region, as the number of nuclei coupled to a methyl
group with large (>3 Hz) couplings is usually much smaller
than for the other types of proton environments. As a
result, 4JHH and 4JCH involving the protons of a methyl
group are frequently visible in 2D NMR data sets with a
cross-peak intensity comparable to those of 3JHH and 3JCH
cross-peaks involving only methine and/or methylene
groups. This knowledge was applied to those test cases
where StrucEluc was unable to automatically resolve
contradictions. By incrementing the upper limit on number
of linking bonds to CH3 groups only, nearly 50% of the
contradictions were resolved. The resulting increases in the
number of possible structures incurred a small time penalty
for structure generation and resulted in a substantially
larger output file.
In addition to the intensity phenomenon associated with
methyl groups, the presence of allylic or conjugated systems
and the presence of a W-geometry can all extend the
transmission range of scalar coupling interactions. Treat-
ment of the structural blocks dCH2 and dCH (usually
identifiable on the 2D NMR connectivities map) as de-
scribed above for methyl groups was also helpful. However,
these strategic approaches are most effective when the
number of structural blocks requiring adjustment is small,
although they may be the only option available if additional
NMR experiments are infeasible.
Identification of the Best Candidate Structure. In
the experience of the authors, an unusually large dA value
(>5 ppm) for the first structure in a dA-ranked file often
indicates an invalid solution. This conclusion is supported
by the histogram presented in Figure 3 (Supporting
Information), which shows the distribution of correct
answer structures as a function of dA. It is clear from this
figure that, for 85% of the test cases, the deviation did not
exceed 4 ppm. Only in a small number of cases did a dA
statistic greater than 5 ppm correspond to a correct
structure; one example is cryptomisrine (structure 6.8),
which had a deviation of 5.5 ppm. These results indicate
that a dA > 5 ppm can be used as a criterion indicating
that the solution obtained must be thoroughly checked.
Such checks may include a priori information, known to
the chemist but not provided to the program, that explains
differences between the observed and predicted 13C NMR
spectra (e.g., unusually strong solvent effects). They may
also include one or more additional NMR experiments, or
acquisition of a mass spectrum if one was not already
provided to StrucEluc. The use of the percentage assign-
ability of an experimental mass spectrum by StrucEluc as
a structure discriminator was discussed above.
Utilization of a User Fragments Database. This
investigation showed the high efficiency of the combined
use of all three (library, MF, and 2D NMR) processes in
the structure generation system described in an earlier
section of this article. In particular, compound 6.960-62
proved especially challenging; it was not identifiable using
2D H-H, C-H, and N-H correlations due to a large
number of missing and/or overlapped cross-peaks. Attempts
to generate structures using only these available data
proved to be unsuccessful: no structure was generated
during several tens of hours. To augment the basic Struc-
Eluc program, it was decided to create a user fragments
library from eight biosynthetic precursors of compound 6.9.
Structure Elucidation of Natural Products Journal of Natural Products, 2002, Vol. 65, No. 5 701

The 1D and 2D NMR data for these compounds53-59 were
entered in StrucEluc, and the elucidation process was
performed on each one in turn. StrucEluc determined the
structures of compounds 6.1-6.5 and 6.7 by 2D NMR
correlations; molecule 6.6 was recognized from its 1D 13C
NMR spectrum, which was already present in the Struc-
Eluc internal database. Molecules 6.1-6.7, with their 13C
NMR assignments, were put into a separate reference file
from which StrucEluc automatically created a user frag-
ment library. The eighth compound was more challenging
than the first seven. StrucEluc required the user library
just described to reach a solution. During the structure
generation process for compound 6.8, 20 fragments were
selected from the combined internal and user fragments
library, and these fragments were used to generate 5996
structures. The correct structure was identified unambigu-
ously by performing an accurate prediction of the 13C NMR
spectra identified as the top candidates by dF rank. The
structure with the lowest dA statistic, 0.27 ppm, was the
correct structure.
The assigned cryptolepinone molecule (compound 6.8)
was added to the reference file, allowing StrucEluc to
enhance the user’s fragments database. The search through
this enhanced user database using the 1D 13C NMR
spectrum of cryptospirolepine (structure 6.9) detected
several fragments whose addition to the 2D connectivities
map provided much of the information missing from the
experimental 2D data. As a result, the structure generation
process took only three minutes and produced only one
structuresthe correct one, cryptospirolepine. This approach
is of significant importance, and its details will be published
separately.64
The example of cryptolepinone illustrates that StrucEluc
can successfully detect the correct structure in an output
file of nearly 6000 structures, underscoring the high
efficiency of the StrucEluc strategy for correct structure
identification. Interestingly, it has been shown that large
output files were more often the exception than the rule.
The high selectivity of the system can be realized by
inspection of the histogram in Figure 4 (Supporting Infor-
mation). The histogram shows that, in 75% of the cases,
the output list contains 10 or fewer structures.
The robust nature of the StrucEluc structure evaluation
strategy is further supported by the fact that in only 2 of
60 test cases (structure 3.8, Figure 3, and structure 4.19,
Figure 4) was the correct structure ranked second by the
dA statistic; in all other cases the dA-ranking placed the
correct structure at the top of the list. This research has
shown that even preliminary ranking of the output struc-
tures by the dF statistic placed the correct structure at the
top of the list in 80% of the test cases. Given the speed
and economy of modern personal computers, the issue of
processing time required for structure determination of a
new natural product is inconsequential. When the extrac-
tion and purification of a new natural product takes months
and sometimes even years, a matter of minutes, hours, or
even a few days to obtain a structural solution on a PC
cannot be considered excessive. Nevertheless, the histo-
gram presented in Figure 5 (Supporting Information)
shows that, for 75% of the test cases, the solution was found
in under one minute; for 90% of the cases the solution time
did not exceed 30 minutes. An analysis of the computation
time with respect to the available data showed that almost
all tasks requiring more than 10 minutes for solution were
solved on the basis of HMBC correlations alone. Clearly,
for the structure determination of complex natural prod-
ucts, utilization of H-H COSY data in conjunction with
HMBC data is highly desirable.
Recent advances in 2D NMR spectroscopy are catalyzing
the widespread use of 1H-15N 2D NMR correlations for the
structure elucidation of natural compounds (see a recent
comprehensive review).65 Further, new pulse sequences are
being developed that differentiate between two- and three-
bond long-range heteronuclear correlations.66 Such precise
knowledge of the correspondence between cross-peaks and
the number of intervening bonds will streamline the
structure generation process and reduce the number of
incorrect structures proposed. Continuing advances in
spectrometer hardware indicate that 2D NMR spectra from
samples as small as about 10 íg will be generally available
in the near future as recent reports indicate.67-69 The
integration of these technological advances with expert
systems such as StrucEluc will have a tremendous impact
on natural products structure determination. The speed at
which new structures can be elucidated, combined with the
ability of chemists to devote their time to less tedious tasks
than structure assignments, will help produce a synergistic
increase in the rate at which new pharmaceutical leads
become available.
Experimental Section
General Experimental Procedures. The software used
for the results reported here was the ACD/Structure Elucida-
tor, StrucEluc, version 5.08, a Windows-based software pro-
gram composed of a number of separate modules.70 The entire
software suite was composed of 1D and 2D NMR data
processing, MS data processing, 1H and 13C NMR chemical
shift prediction and assigned structure databasing tools, and
an integrated chemical structure drawing program. During the
elucidation process the program displayed a number of possible
structures, with comparison of on-screen experimental and
fragment spectra or, in the case of failure, a set of structural
fragments corresponding to portions of the spectrum that were
used to assemble the structure of the unknown compound.
StrucEluc included filters for 1H NMR, IR peaks, mass spectral
(MW) data, and elemental composition and a self-training
system. If the accuracy of spectral calculations for a new class
of compounds was poor, user databases with experimental
chemical shifts could be constructed and utilized during the
elucidation process. Both 1H and 13C databases of assigned
structures with NMR chemical shifts and coupling constants
were available for searching by chemical structure and sub-
structure. The number of entries in each database was
>110 000 for 1H NMR and >111 000 for 13C NMR.
Acknowledgment. The authors thank Dr. Gary E. Martin
of Pharmacia Corporation for the provision of many challeng-
ing data sets to challenge the StrucEluc system discussed in
this publication. We especially appreciate the many stimulat-
ing conversations regarding the application of H-N hetero-
nuclear correlation experiments to the structure elucidation
of natural products. We also thank Dr. Dean Carlson of the
Advanced Chemistry Development Technical Support Group,
Toronto, for applying his exacting and stringent standards of
proofreading to the manuscript. His thorough evaluation of
the science reported here significantly enhanced the quality
of the submission.
Supporting Information Available: Problem distribution his-
tograms (Figures 1-5) providing a graphical summary of the success
rates of StrucEluc for solving the structures for the 60 molecules
examined in this work. This information is available free of charge
via the Internet at http://pubs.acs.org
References and Notes
(1) (a) Elyashberg, M. E.; Martirosian, E. R.; Karasev, Yu. Z.; Thiele,
H.; Somberg, H. Anal. Chim. Acta 1997, 337, 265-286. (b) Elyash-
berg, M. E.; Martirosian, E. R.; Karasev, Yu. Z.; Thiele, H.; Somberg,
H. Anal. Chim. Acta 1997, 348, 443-463.
702 Journal of Natural Products, 2002, Vol. 65, No. 5 Elyashberg et al.

(2) Miyabayashi, N.; Sasaki, S. J. Chem. Inf. Comput. Sci. 1988, 28, 18-
28.
(3) Munk, M. E.; Christie, B. D. Anal. Chim. Acta 1989, 216, 57-68.
(4) Will, M.; Fachinger, W.; Richert, J. R. J. Chem. Inf. Comput. Sci.
1996, 36, 221-227.
(5) Elyashberg, M. E.; Blinov, K. A.; Martirosian, E. R. Lab. Automat.
Inf. Manage. 1999, 34, 15-30.
(6) ) Christie, B. D.; Munk, M. E. J. Am. Chem. Soc. 1991, 113, 3750-
3757.
(7) Munk, M. E.; Velu, V. K.; Madison, M. S.; Robb, E. W.; Baderstscher,
M.; Christie, B. D.; Razinger M. In Recent Advances in Chemical
Information II; Colier, H., Ed.; Royal Society of Chemistry: Cam-
bridge, U.K., 1993; pp 247-263.
(8) Munk, M. E.; Madison, M. S.; Schulz, K.-P.; Korytko A. 13th
Workshop, Nov 13-15, 1998, Bad Durkheim, Germany, 1998.
(9) Munk, M. E. J. Chem. Inf. Comput. Sci. 1998, 38, 997-1009.
(10) Funatsu, K.; Sasaki, S. J. Chem. Inf. Comput. Sci. 1996, 36, 190-
204.
(11) Blinov, K. A.; Elyashberg, M. E.; Molodtsov, S. G.; Williams A. J.;
Martirosian, E. R. Fresenius J. Anal. Chem. 2001, 369, 709-714.
(12) Elyashberg, M. E.; Blinov, K. A.; Martirosian, E. R. European
Conference on Analytical Chemistry (EUROANALYSIS XI), Septem-
ber 3-9, 2000, Lisbon; Book of Abstracts, OC-34.
(13) Peng, C.; Yuan, S.; Zheng, C.; Hui. Y. J. Chem. Inf. Comput. Sci. 1994,
34, 805-813.
(14) Peng, C.; Yuan, S.; Zheng, C.; Hui, Y.; Wu, H.; Ma, K.; Han, X. J.
Chem. Inf. Comput. Sci. 1994, 34, 814-819.
(15) Peng, C.; Yuan, S.; Zheng, C.; Shi, Z.; Wu, H. J. Chem. Inf. Comput.
Sci. 1995, 35, 539-546.
(16) Peng, C.; Bodenhausen, G.; Qiu, S.; Fong, H. H. S.; Farnsworth, N.
R.; Yuan, S.; Zheng, C. Magn. Reson. Chem. 1998, 36, 267-278.
(17) Nuzillard, J.-M.; Massiot, G. Tetrahedron 1991, 47, 3655-3664.
(18) Steinbeck, C. Angew. Chem., Int. Ed. Engl. 1996, 35, 1984-1986.
(19) Lindel, T.; Junker, J.; Koeck, M. Eur. J. Org. Chem. 1999, 3, 573-
577.
(20) Jaspars, M. Nat. Prod. Rep. 1999, 16, 241-248.
(21) Ley, S. V.; Doherty, K.; Massiot, G.; Nuzillard, J.-M. Tetrahedron
1994, 50, 12267-12280.
(22) Nuzillard, J.-M.; Connolly, J. D.; Delaude, C. D.; Richard, B.; Zeches-
Hanrot, M.; Le Men-Olivier, L. Tetrahedron 1999, 55, 11511-11518.
(23) Reynolds, W. F.; McLean, S.; Tay, L.-L.; Yu, M.; Enriquez, R. G.;
Estwick, D. M.; Pascoe, K. O. Magn. Reson. Chem. 1997, 35, 455-
462.
(24) Hadden, C. E.; Angwin, D. T. Magn. Reson. Chem. 2001, 39, 1-8.
(25) ACD SpecManager Program. This program is capable of reading data
presented in practically all formats common in NMR spectroscopy.
http://www.acdlabs.com/products/spec_lab/exp_spectra/
(26) Araya-Maturana, R.; Delgad-Castro, T.; Cardona, W.; Weiss-Lopez,
B. E. Curr. Org. Chem. 2001, 5, 253-263.
(27) Koehn, F. E.; Gunasekera, S. P.; Niel, D. N.; Cross, S. S. Tetrahedron
Lett. 1991, 32, 169-172.
(28) Jimeno, M. L.; Rumbero, A.; Vazquez, P. Magn. Reson. Chem. 1995,
33, 408-411.
(29) Kinouchi, Y.; Ohtsu, H.; Tokuda, H.; Nishino, H.; Matsunaga, S.;
Tanaka, R. J. Nat. Prod. 2000, 63, 817-820.
(30) Leone, P. A.; Redburn, J.; Hooper, J. N. A.; Quinn, R. J. J. Nat. Prod.
2000, 63, 694-697.
(31) Rashid, M. A.; Gustafson, K. R.; Boyd, M. R. J. Nat. Prod. 2000, 63,
531-533.
(32) Capon, R. J.; Miller, M.; Rooney, F. J. Nat. Prod. 2000, 63, 821-824.
(33) Chai, M.-C.; Wang, S.-K.; Dai, C.-F.; Duh, C.-Y. J. Nat. Prod. 2000,
63, 843-844.
(34) Ragasa, C. Y.; Cruz, M. C.; Cula, R.; Rideout, J. J. Nat. Prod. 2000,
63, 509-511.
(35) Zhang, P.; Bierer, D. E. J. Nat. Prod. 2000, 63, 643-645.
(36) Butler, M. S.; Katavic, P. L.; Davis, R. A.; Forster, P. I.; Guymer, G.
P.; Quinn, R. J. J. Nat. Prod. 2000, 63, 688-689.
(37) Qureshi, A.; Faulkner, D. J. J. Nat. Prod. 2000, 63, 841-842.
(38) Macias, M.; Ulloa, M.; Camboa, A.; Mata, R. J. Nat. Prod. 2000, 63,
757-761.
(39) Sitachitta, N.; Williamson, R. T.; Gerwick, W. H. J. Nat. Prod. 2000,
63, 197-200.
(40) Pettit, G. R.; Numata, A.; Cragg, G. M.; Herald, D. L.; Takada, T.;
Iwamoto, C.; Riesen, R.; Schmidt, J. M.; Doubek, D. L.; Goswami, A.
J. Nat. Prod. 2000, 63, 72-78.
(41) Li, C.-J.; Wang, L.-Q.; Chen, S.-N.; Qin, G.-W. J. Nat. Prod. 2000,
63, 1214-1217.
(42) Gonzales, A. G.; Leon, F.; Sanchez-Pinto, L.; Pardon, J. I.; Bermejo,
J. J. Nat. Prod. 2000, 63, 1297-1297.
(43) Gallimore, W. A.; Galario, D. L.; Lacy, C.; Zhu, Y.; Scheuer, P. J. J.
Nat. Prod. 2000, 63, 1022-1026.
(44) Tan, L. T.; Okino, T.; Gerwick, W. H. J. Nat. Prod. 2000, 63, 952-
995.
(45) Nagle, D. C.; Zhou, Y.-D.; Park, P. U.; Paul, V. J.; Rajbhandary, I. J.
Nat. Prod. 2000, 63, 1431-1433.
(46) Wellimgton, K. D.; Cambie, R. C.; Rutledge, P. S.; Bergquist, P. R. J.
Nat. Prod. 2000, 63, 79-85.
(47) Shen, Y. C.; Lo, K.-L.; Chen, C.-Y.; Kuo, Y.-H.; Hung, M.-C. J. Nat.
Prod. 2000, 63, 720-722.
(48) Burgess, E. J.; Larsen, L.; Perry, N. B. J. Nat. Prod. 2000, 63, 537-
539.
(49) (49) Iwamoto, M.; Ohtsu, H.; Matsunaga, S.; Tanaka, R. J. Nat. Prod.
2000, 63, 1381-1383.
(50) Kirsch, G.; Kong, G. M.; Wright, A. D.; Kaminsky, R. J. Nat. Prod.
2000, 63, 825-829.
(51) Torres, Y. R.; Berlinck, R. G. S.; Megalhaes, A.; Schefer, A. B.;
Ferreira, A. G.; Hajdu, E.; Muricy, G. J. Nat. Prod. 2000, 63, 1098-
1105.
(52) Tanaka, R.; Kasubuchi, K.; Kita, S.; Tokuda, H.; Nishino, H.;
Matsunaga, S. J. Nat. Prod. 2000, 63, 99-103.
(53) Ablordeppey, S. Y.; Hufford, C. D.; Borne R. F.; Dwumu-Badu, D.
Planta 1990, 56, 416-417.
(54) Crouch, R. C.; Davies, A.; Spitzer, T. D.; Martin, G. E.; Phoebe, C.
H., Jr.; Sharaf, M. H. M.; Schiff, P. L., Jr.; Tackie, A. N. J. Heterocycl.
Chem. 1995, 32, 1077-1080.
(55) Sharaf, M. H. M.; Schiff, P. L., Jr.; Crouch, R. C.; Davies, A.; Andrews,
C. W.; Martin, G. E.; Phoebe, C. H., Jr.; Tackie, A. N. J. Heterocycl.
Chem. 1995, 32, 1631-1636.
(56) (56) Sharaf, M. H. M.; Schiff, P. L., Jr.; Martin, G. E.; Phoebe, C. H.,
Jr.; Tackie, A. N. J. Heterocycl. Chem. 1996, 33, 239-243.
(57) Sharaf, M. H. M.; Schiff, P. L., Jr.; Minick, D.; Johnson, R. L.; Crouch,
R. C.; Andrews, C. W.; Martin, G. E.; Phoebe, C. H., Jr.; Tackie, A.
N. J. Heterocycl. Chem. 1996, 33, 789-797.
(58) Hadden, C. E.; Duholkc, W. K.; Guido, J. E.; Robins, R. H.; Martin,
G. E.; Sharaf, M. H. M.; Schiff, P. L., Jr. J. Heterocycl. Chem. 1999,
36, 525-531.
(59) Spitzer, T. D.; Crouch, R. C.; Martin, G. E.; Sharaf, M. H. M.; Schiff,
P. L., Jr.; Tackie, A. N.; Boye, G. L. J. Heterocycl. Chem. 1991, 28,
2065-2070.
(60) Tackie, A. N.; Boye, G. L.; Sharaf, M. H. M.; Schiff, P. L.; Crouch, R.
C.; Spitzer, T. D.; Johnson, R. L.; Dunn, J.; Minick, D.; Martin, G. E.
J. Nat. Prod. 1993, 54, 653-670.
(61) Hadden, C. E.; Martin, G. E.; Tackie, A. N.; Schiff, P. L., Jr. J.
Heterocycl. Chem. 1999, 36, 1115-1117.
(62) Martin, G. E.; Hadden, C. E.; Tackie, A. N.; Sharaf, M. H. M.; Schiff,
Jr., P. L. Magn. Reson. Chem. 1999, 37, 529-537.
(63) 2D NMR spectra of strychnine were kindly provided to us by Dr. Gary
E. Martin, Pharmacia, Kalamazoo, MI.
(64) Williams, A. J.; Elyashberg, M. E.; Blinov, K. A.; Molodtsov, S. G.;
Martirosian, E. R.; Martin, G. E.; Hadden, C. E. Magn. Reson. Chem.
(manuscript in preparation).
(65) Martin, G. E.; Hadden, C. E. J. Nat Prod. 2000, 63, 543-585.
(66) Krishnamurthy, V. V.; Russell, D. J.; Hadden, C. E.; Martin, G. E. J.
Magn. Reson. 2000, 146, 232-239.
(67) Martin, G. E.; Guido, J. E.; Robins, R. H.; Sharaf, M. H. M.; Schiff,
P. L., Jr.; Tackie, A. N. J. Nat. Prod. 1998, 61, 555-559.
(68) Russell, D. J.; Hadden, C. E.; Martin, G. E.; Gibson, A. A.; Zens, A.
P.; Carolan, J. L. J. Nat. Prod. 2000, 63, 1047-1049.
(69) Reynolds, W. F.; Yu, M.; Enriquez, R. G. Magn. Reson. Chem. 1997,
35, 614-618.
(70) A detailed explanation of the modules comprising the StrucEluc
program can be found at http://www.acdlabs.com/products/spec_lab/
complex_tasks/str_elucidator/.
NP0103315
Structure Elucidation of Natural Products Journal of Natural Products, 2002, Vol. 65, No. 5 703