Applying Computer-Assisted Structure Elucidation Algorithms for the Purpose of Structure
Validation: Revisiting the NMR Assignments of Hexacyclinol
A. J. Williams,*,† M. E. Elyashberg,‡ K. A. Blinov,‡ D. C. Lankin,§ G. E. Martin,⊥ W. F. Reynolds,| J. A. Porco, Jr.,3
C. A. Singleton,3," and S. Su3,`
ChemZoo, 904 Tamaras Circle, Wake Forest, North Carolina 27587, AdVanced Chemistry DeVelopment, Moscow Department, 6 Akademik
BakuleV Street, Moscow, 117513, Russian Federation, Department of Medicinal Chemistry and Pharmacognosy, College of Pharmacy,
UniVersity of Illinois at Chicago, Chicago, Illinois 60612-7231, Schering-Plough Research Institute, Rapid Structure Characterization
Laboratory, Pharmaceutical Sciences, Summit, New Jersey 07901, Department of Chemistry, UniVersity of Toronto, 80 St George St., Toronto,
Ontario M5S 3H6, Canada, and Department of Chemistry and Center for Chemical Methodology and Library DeVelopment, Boston UniVersity,
24 Cummington Street, Boston, Massachusetts 02215
ReceiVed October 3, 2007
Computer-assisted structure elucidation (CASE) using a combination of 1D and 2D NMR data has been available for
a number of years. These algorithms can be considered as “logic machines” capable of deriving all plausible structures
from a set of structural constraints or “axioms”, defined by the spectroscopic data and associated chemical information
or prior knowledge. CASE programs allow the spectroscopist not only to determine structures from spectroscopic data
but also to study the dependence of the proposed structure on changes to the set of axioms. In this article, we describe
the application of the ACD/Structure Elucidator expert system to help resolve the conflict between two different
hypothetical hexacyclinol structures derived by different researchers from the NMR spectra of this complex natural
product. It has been shown that the combination of algorithms for both structure elucidation and structure validation
delivered by the expert system enables the identification of the most probable structure as well as the associated chemical
shift assignments.
In recent years, computer-assisted structure elucidation (CASE)
has offered an additional option to scientists challenged by difficult
chemical structure elucidation problems. The reasons for considering
CASE methods should be obvious: there may be significant time
savings when applying algorithmic approaches; the subjective bias
of a scientist can be reduced relative to an algorithm; the thought
process associated with the analysis of a large and diverse data
collection made up of spectrum-structural information used to
elucidate a chemical structure offers a significant logical and
combinatorial challenge. The nature of this process was already
revealed in the pioneering works published in the late 1960s and
1970s.1–4 When scientists attempt to solve structural problems, they
logically draw definitive structural conclusions from a set of
spectroscopic data and a priori knowledge assembled into a set of
initial axioms. It has been shown5 that the definition of a spectrum-
structural problem is equivalent to formulating a specific axiomatic
theory.
The solution of a structural problem using 2D NMR spectroscopy
data can be divided into four main stages:
1. Prepare the experimental data to create spectrum-structure
axioms that serve as the basis of the structure elucidation process.
This stage encompasses both raw data processing and peak picking.
1D and 2D peak tables are produced as an output from this stage.
2. Create axioms and hypotheses on the basis of NMR peak
tables. The information contained in the peak tables is considered
as true and consistent. The following axioms are the most common:
(a) If the hydrogen atom H(i) shows an HSQC (HMQC) correlation
with the carbon atom C(i), then the atom H(i) is attached to the
atom C(i); (b) if the hydrogen atom H(i) attached to the carbon
atom C(i) shows a COSY correlation with the hydrogen atom H(j)
attached to the carbon atom C(j), then carbons C(i) and C(j) make
up a chemical bond in a molecule; (c) if the hydrogen atom H(i)
shows an HMBC correlation to a carbon atom C(j), then the distance
between those atoms in a molecule is 2 or 3 bonds; (d) if the
hydrogen atom H(i) shows a NOESY (ROESY) correlation with
the hydrogen atom H(j), then the distance through space between
atoms H(i) and H(j) is less than 5 Å. Additional structural constraints
are produced on the basis of chemical knowledge including
information regarding sample origin and associated structural
knowledge as well as Bredt’s rule, etc.
3. The logical inference of all direct structural conclusions, if
one is possible, comes from following the set of axioms outlined
above. These conclusions form a chemical structure file where each
structure within the file has to satisfy all the axioms applied (i.e.,
be entirely consistent with the experimental data) and has to contain
the assigned experimental chemical shifts.
4. The verification of all structural hypotheses proceeds in order
to choose the most probable solutions consistent with the data. For
this purpose the prediction of NMR spectra associated with the
candidate structures in combination with specific chemical consid-
erations is applied and hypothetical structures are rank ordered based
on the agreement between the experimental data and the algorith-
mically generated structures.
If all initial axioms used to deduce the structural formula of an
unknown contain no contradictions, then the structural file produced
will be valid6 and will contain the actual structure. It is evident
that if at least one of the axioms in the set is false or conflicts with
the other axiom(s), then the correct structure will not be determined.
A contradictory system of axioms simply produces no structure.
Research has shown that the information obtained from 2D NMR
spectra can frequently be fuzzy, contradictory, and incomplete.7–10
These problems primarily arise due to the presence of long-range
2D NMR correlations that span more than the typical number of
bonds allowed by the axioms outlined above (so-called nonstandard
correlations, NSC8) or severe overlap in the NMR spectra. Overlap
* To whom correspondence should be addressed. Tel: +919.341 8375.
Fax: +919 300-5321. E-mail: antony.williams@chemspider.com.
† ChemZoo.
‡ Advanced Chemistry Development, Moscow Department.
§ University of Illinois at Chicago.
⊥ Schering-Plough Research Institute.
| University of Toronto.
3 Boston University.
" Current address: Waters Corporation, 34 Maple St., Milford, MA,
01757.
` Current address: The Scripps Research Institute, 10550 North Torrey
Pines Rd., La Jolla, CA 92037.
J. Nat. Prod. 2008, 71, 581–588 581
10.1021/np070557t CCC: $40.75  2008 American Chemical Society and American Society of Pharmacognosy
Published on Web 02/08/2008

leads to equivocal signal assignments in the 2D spectra and
consequently to ambiguities in the spectroscopic interpretations.
These possibilities mean that the initial axioms may fail (for
instance, the hydrogen atom H(i) in the real molecule is attached
not to the carbon atom C(i) but to another carbon atom) and some
axioms can be violated (for instance, the topological distance
between two HMBC correlated atoms H(i) and C(j) is longer than
three bonds (3JCH)).
To overcome these difficulties, a human expert employs all of
their knowledge, experience, and intuition. One common approach
is the suggestion of a structure for the unknown using a series of
similar molecules as a basis that have assigned chemical shifts in
both 1H and 13C NMR spectra. The comparison of the molecules
with the proposed structure and its NMR spectra allows chemical
shift assignment that appears to be consistent with the suggested
structure. The correctness of the assignments is validated by
checking the consistency of the chemical shifts with the topological
distances between the intervening atoms: an assignment is consid-
ered acceptable if all distances are in agreement with the postulated
axioms. The problem is reduced to either confirming or refuting
the proposed structure. Commonly, some fragments of the unknown
compound can be inferred by the chemist through an understanding
of the origin of the sample (reactants, plant genus, etc.), and these
fragments can also be utilized in the elucidation process.
Data analysis can vary from simple to very complex depending
on the nature of the problem. For complex problems chemists can
arrive at a structure for the unknown compound that is incorrect.
Our experience7–10 shows that CASE methods offer the possibility
of dramatically accelerating the process of structure elucidation and
validation, as well as increasing the reliability of the derived
structure.
Expert systems now exist that are capable of modeling all stages
of the structure elucidation process via mathematical algorithms
implemented in the corresponding software programs. To date the
most advanced CASE system is ACD/Structure Elucidator (StrucE-
luc), which has been described in a large number of publications.7,10,11
The StrucEluc system mimics the sequence of steps performed by
spectroscopists during the process of structure elucidation. It should
be noted that inside the program the structure elucidation process
is combined with structure verification in such a way that the
structure verification procedure can be carried out in an independent
manner. StrucEluc is supplied with an extensive and branched
knowledge base, sophisticated algorithms for inferring plausible
structures from an initial set of axioms, and multiple algorithms
for the accurate and rapid prediction of 1H, 13C, 15N, 19F, and 31P
nuclei chemical shifts. The efficiency of this system has been amply
demonstrated by solving hundreds of structural problems.9 More-
over, the application of StrucEluc allowed authors12 to recognize
the structure of a complex natural product that had previously
remained unsolvable by highly qualified spectroscopists.
In this article we will demonstrate how the StrucEluc system
can be applied to the validation of different hypothetical structures
derived by different researchers from the same initial experimental
data. For this aim, two important capabilities of StrucEluc were
utilized: (a) the system makes it possible to follow the impact of
how changes in the initial axioms influence the inferred structures;
(b) the application of spectrum prediction for different proposed
structures allows comparison of different structural hypotheses and
allows the choice of the most probable structure(s).
This study was initiated by a discussion in the literature and on
Web site blogs regarding the complex chemical structure of a newly
identified natural product, hexacyclinol.13–16 The essence of the
problem will be explained in the next section.
The History of Hexacyclinol and Its Various Structural
Forms
Hexacyclinol was first described by Gräfe and co-workers in
2002.13 The compound was isolated in Siberia from basidiospores
collected from Panus rudis strain HKI 0254. The proposed structure
(Figure 1a) contained a reactive epoxyketone and a highly strained
endoperoxide moiety.
The total synthesis of the structure proposed by Gräfe13 was first
reported by La Clair in 2006.14 Shortly after this work, the structure
of hexacyclinol was revisited by Rychnovsky based on calculated
13C NMR chemical shift correlations.15 To determine the structure
of hexacyclinol, highly oxygenated and unsaturated molecules were
compared with predictive NMR calculations. When evaluating the
chemical shift difference between calculated and experimental
hexacyclinol, it was found that the Gräfe13 structure of hexacyclinol
had an unusually high deviation, with an average of 6.8 ppm
difference, and five carbon chemical shifts with more than a 10
ppm difference from the calculated structure. Using GIAO NMR
predictions, the structure of hexacyclinol was revised from the one
proposed by Gräfe13 to the structure shown in Figure 1b.15
Following this, the revised hexacyclinol structure was indeed
synthesized by Porco and co-workers.16 The analytical data were
identical to that reported for natural hexacyclinol for 1H and 13C
NMR and optical rotation studies. In addition, an X-ray crystal
structure was obtained, providing unequivocal structural confirma-
tion. These data allow for a thorough assessment of the power of
CASE systems in elucidating heretofore unknown, complex natural
products.
Structure Comparison Using ACD/Labs Methods
The initial question identified for investigation was assuming
that the NMR data obtained by Gräfe et al.13 are appropriately
recorded and accurately reported. The NMR data presented in the
original publication are shown in Table 1. These data can be
considered as the initial set of spectroscopic-structural axioms
defining the given problem.
The molecular formula, C23H28O7, and the spectroscopic infor-
mation from the 1H, 13C, COSY, HMQC, and HMBC data sets
were manually input into the StrucEluc program. A molecular
connectivity diagram (MCD)7–11 was created using the criteria that
chemical bonds between oxygen atoms would be allowed, but triple
bonds are forbidden (see Supporting Information Figure A). Using
the most reliable multiplicities of the 1H NMR signals presented
in Table 1 (where the measured coupling constants coincide for
intervening nuclei), the following numbers of attached hydrogen
atoms were set for specific carbon nuclei: 18.6 (0), 26.1(0), 120.7
(1), 75.8 (1), 47.8 (2), 139.6 (1), 40.9(1), 26.6(0), 24.7(0), 49.1(0).
This additional structural information significantly reduces the
number of plausible structures. Note that the atom properties and
the number of attached hydrogen atoms displayed in the MCD
complement the system of axioms given in Table 1.
According to the methodology of the StrucEluc application the
MCD was checked for the presence of contradictions using the usual
Figure 1. Two different hexacyclinol structures proposed by Gräfe
et al.13 (a) and by Rychnovsky15 (b).
582 Journal of Natural Products, 2008, Vol. 71, No. 4 Williams et al.

assumptions,8 and the program reported that the analyzed data were
consistent. As has been reported elsewhere,10 the most appropriate
structure generation fuzzy mode uses m ) 0–15, a ) 16, with
generation stopped at m ) mg. m is the number of connectivities
to be lengthened or removed and a is the number of bonds to be
added to the connectivity length (a ) 16 represents the removal of
connectivities), mg is the m value at which the resultant structural
file is not empty. It is evident that if m ) 0 (when the data contain
no NSCs), then strict structure generation will be carried out. The
fuzzy structure generation procedure was initiated utilizing structural
filtering via spectroscopic libraries, the permanent BADLIST
(containing fragments that are unlikely in organic chemistry), and
the checking of chemical structures by Bredt’s rule. Four-membered
rings were forbidden since these are relatively rare in natural
products. These constraints accelerate the process of structure
generation.
As a result of structure generation, 5425 molecules were
generated at mg ) 1 (denoted as k ) 5425). The structure generation
process took tg ) 56 s. 13C NMR chemical shift prediction was
carried out for all generated structures by both the incremental and
the neural net approaches using ACD\CNMR Predictor.17 The
calculations consumed 66 s. The resulting file was reduced to 4961
structures after removal of duplicates (k ) 5425f4961). Structures
containing the O-OH group were filtered out since this fragment
was deemed to be unlikely in natural products. Excluding hydro-
peroxide moieties reduced the k value to 3097. According to the
methodology described earlier,7,11 structures were ranked in order
of the increasing deviation between the experimental and calculated
chemical shifts of the predicted structures. Four structures at the
top of the file ranked by the deviation of spectra calculated via
increments dI(C) are shown in Figure 2.
Figure 2 shows the most probable structure when ranking is
performed by dI(C). Figure 3 displays the COSY and HMBC
connectivities of this structure. The next three structures are very
similar to Tanimoto structure similarity coefficients of 0.99, 0.99,
and 0.97, and two of them are distinguished as preferable using
other calculation approaches. Structure 3 in Figure 2 is distinguished
as the best structure overall by both the 13C HOSE and 1H neural
net calculations as well as by the complex match factor. Structure
4 was ranked as the best structure by the 13C neural net approach
(the upper right-hand corner of the structure boxes in the quadrants
displayed in Figure 2 are labeled appropriately). The deviations
calculated for Gräfe’s structure are relatively large, but when the
complexity of the structure is taken into account, they are certainly
acceptable.
All of these structures contain only one NSC (a ) 1) between
the carbon resonances at 47.8 ppm and 53.1 ppm. The presence of
at least one NSC is frequently observed for organic molecules. Since
the structures are both very similar and very complex, the applied
methods of chemical shift prediction cannot reliably prove or refute
any of the structures. The results obtained show that the structure
suggested by Gräfe et al.13 from the data displayed in Table 1 was
also inferred by the CASE algorithms as one of the most probable.
The location of the structure proposed by Gräfe13 in the CASE
output file should be considered as a consequence of the total set
of initial axioms and assumptions used by Gräfe, including the
allowance of the presence of one NSC and O-O bond in the target
structure.
The number of NSCs in a structure could be very large, as has
been reported in our earlier work.10 It is interesting to identify what
type or types of structure(s) are generated if we allow the presence
of up to four NSCs in the analyzed structures as well as forbidding
the peroxide group, as was suggested for Rychnovsky’s structure.15
Setting a larger number of possible NSCs is too time-consuming
for this structural problem. Fuzzy structure generation was per-
formed in the secure mode {m ) 4, step by step, a ) 16}, implying
that all m values between 0 and 4 will be taken into account.10 To
prevent the production of too large an output file, incremental 13C
NMR spectrum prediction was performed during structure genera-
tion and only structures that had dI(13C) value less than 4 ppm were
saved to disk. The results obtained were as follows: 290 141
molecules were generated and 503 molecules were stored, removal
of duplicates gave k ) 503f386, and the process was performed
in a generation time of tg ) 16 min 42 s. All of the structures in
the output file contained two epoxy-containing cycles as represented
in Rychnovsky’s structure. The “best” structure (1) distinguished
by the minimum sum of deviations D(sum) ) dA(C) + dI(C) +
dNet(C) when the structures were ranked by the complex match
factor dA(Σ) ) dA(C) + dA(H) is shown below (structure 1).
Four nonstandard COSY correlations are shown by the two-headed
arrows, and one of them between the proton bearing carbon atoms
resonating at 40.40 and 61.00 ppm was unusually long, a 6JHH coupling.
There is no convincing explanation of why this coupling would exist.
Comparison between the deviations determined for Gräfe’s structure
(Figure 3) and the structure 1 are given in Table 2.
Table 1. NMR Data Utilized by Gräfe et al.13 for Structure Elucidation
position δC δH COSY HMBC
1 18.6 q 1.77 s 142.2, 120.7
2 142.2 s
3 26.1 q 1,72 s 142.2, 120.7
4 120.7 d 4.82 d, 10.1 H-5
5 75.8 d 5.46 d, 10.1 H-4 60.5, 202.9
6 60.5 s
7 202.9 s
8 53.1 d 3.23 d, br, 3.5 H-9, H-10 202.9, 54.5
9 54.5 d 3.64 m H-8, H-10, H-13 47.8
10 47.8 d 2.74, dd, 5.2, 7.8 H-9, H-11 54.5, 60.5
11 71.5 d 4.99 dd,5.2 br H-10, H-12
12 40.4 d 3.55 m H-11, H-13 61.0, 71.5, 72.7
13 72.7 d 3.8 dd, 9.5, 1.5, 2.54 br (OH) H-12, H-9 40.4, 54.5, 47.8, 53.1
14 61.0 d 3.51 dd, 2.9, 0.5 H-12, H-15 40.4
15 53.2 d 3.29 d, 3.2 H-14 132.5, 192.8
16 192.8 s
17 132.5 s
18 139.6 d 6.73 dd 5.3, 2.4 (allyl) H-19 192.8, 40.9
19 40.9 d 3.59 d, 5.3 H-18 139.6
20 77.3 s
21 26.6 q 1.26 s 77.3, 40.9
22 24.7 q 1.15 s 77.3, 40.9
23 49.1 q 3.02 s 77.3
NMR Assignments of Hexacyclinol Journal of Natural Products, 2008, Vol. 71, No. 4 583

The table shows that, on one hand, the deviations found for
structure 1 are markedly smaller than Gräfe’s structure, but on the
other hand four NSCs including one corresponding to a 6JHH
coupling appear in the structure. It is to be expected that further
refining of the possible structure by increasing the m value during
fuzzy structure generation would result in further violation of the
initial axioms declaring that the lengths of the COSY and HMBC
correlations should be standard. The results obtained suggest that
a revision of the initial data displayed in Table 1 may be necessary.
As mentioned earlier, Rychnovsky15 suggested the hexacyclinol
structure differing from that suggested by Gräfe’s and offered
chemical shift assignments for this newly suggested structure using
the spectroscopic data published by Gräfe et al.13 Deviations based
on NMR predictions were calculated for both structures suggested
by Rychnovsky and Gräfe. The results are listed in Table 3.
All deviations calculated for the structure suggested by Rych-
novsky are slightly smaller than those for the structure suggested
by Gräfe. However, checking Rychnovsky’s structure relative to
the 2D NMR data shows the presence of 12 NSCs with unusually
long topological distances between the intervening nuclei. Using a
designation introduced previously,10 the set of NSCs can be
symbolized as follows:
m ) COSY{1a(5) + 2a(4) + 2a(3) + 2a(2)} + HMBC {2a(5)
+ 2a(4) + 1a(3)} ) 12
where na(p) denotes that there are n connectivities that exceed by
p bonds the standard length assumed for a given 2D NMR technique
(see Table 4). Rychnovsky’s structure with the corresponding NSCs
Figure 2. The four structures at the beginning of the CASE structure elucidation output file are ranked by the deviation of incrementally
calculated spectra dI(C). The inscriptions accompanying the structures list the type of spectrum prediction method for which a given structure
was ranked in first position.
584 Journal of Natural Products, 2008, Vol. 71, No. 4 Williams et al.

is shown below with the nonstandard connectivities marked by bold
lines (structure 2 (COSY) and 2a (HMBC)).
Analysis of the spectroscopic data led Reynolds18 to conclude
that the structure and NMR data could be made more consistent if
the 1H NMR chemical shifts at 3.55 and 3.59 ppm were exchanged.
This exchange only slightly reduced the deviations, while the
number of NSCs and their lengths still remained fairly large (a )
2–4). Up to this point in the present study attempts were made to
resolve the problem assuming that only the 2D NMR data of Gräfe13
were available and that the structure suggested by Rychnovsky15
has chemical shift assignments based on these data.
The investigation led to the following conclusions: all possibilities
to apply the 2D NMR data published by Gräfe for clarifying the
conflict between the two suggested structures were exhausted, and
it was impossible to either prove or refute either of them. It became
clear that either the structure suggested by Rychnovsky was
incorrect or that the chemical shift assignments suggested in his
work should be reexamined using new 2D NMR data. Rych-
novsky’s structure for hexacyclinol has been unambiguously
confirmed by X-ray analysis,16 so the challenge remains in aligning
the chemical shift assignments of the structure in agreement with
all of the NMR data.
NMR Confirmation of the Structure of Hexacyclinol
In order to determine appropriate chemical shift assignments for
hexacyclinol, 1D and 2D NMR spectra of the compound synthesized
by Porco et al.16 were acquired under conditions described in the
Figure 3. Structure of hexacyclinol proposed by Gräfe and the 13C
NMR chemical shift assignment suggested in his work.13 The
COSY (two-headed arrows) and HMBC (unidirectional arrows)
connectivities are shown.
Table 2. Comparison of Deviations Calculated for Structure 1
and for Gräfe’s Structure (Figure 3)
structure 1 Gräfe’s structure13
dA(13C) 3.808 dA(13C) 5.071
dI(13C) 2.873 dI(13C) 3.808
dN(13C) 2.564 dN(13C) 4.130
dA(1H) 0.412 dA(1H) 0.434
dI(1H) 0.420 dI(1H) 0.395
dN(1H) 0.356 dN(1H) 0.433
dA(Σ) 6.126 dA(Σ) 8.462
Table 3. Comparison of Deviations for Structures Suggested
by Gräfe and Rychnovsky Using the Data Listed in Table 1
Gräfe’s structure13 Rychnovsky’s structure15
dA(13C) 5.071 dA(13C) 5.027
dI(13C) 3.808 dI(13C) 3.699
dN(13C) 4.130 dN(13C) 4.018
dA(1H) 0.434 dA(1H)0.343
dI(1H) 0.395 dI(1H) 0.297
dN(1H) 0.433 dN(1H) 0.289
dA(Σ) 8.462 dA(Σ) 8.457
Table 4. Nonstandard Connectivities Present in the Structure
Suggested by Rychnovsky (marked in bold)
COSY HMBC
40.9f61.007JHH 40.9f61.0 6JCH
40.9f72.7 6JHH 40.9f72.7 5JCH
40.9f71.5 5JHH 40.9f71.5 4JCH
40.4f139.6 4JHH 54.5f72.7 5JCH
72.7f54.5 6JHH 53.1f72.7 6JCH
54.5f47.8 4JHH
53.1f47.8 5JHH
NMR Assignments of Hexacyclinol Journal of Natural Products, 2008, Vol. 71, No. 4 585

Experimental Section. Positions and numbers of resonances local-
ized in the strongly overlapped areas of 1H NMR were refined using
the 1H NMR spectrum registered at the frequency of 900 MHz.
Using ACD/SpecManager19 we automatically determined peak
positions, multiplicities, and coupling constants (when it was
possible) in the 1H NMR spectrum. With these data, assignments
of series of peaks observed in 2D NMR spectra were revised. The
problem was reformulated in the following manner: once the
structure is determined unambiguously, determine those chemical
shift assignments that remove the extremely long-range correlations.
Generally speaking, the criterion based on the minimization of the
number of NSCs and their length is of a heuristic nature, so this
criterion should be added to the system of axioms used for
confirmation of the structure of hexacyclinol.
To assist the process of understanding, we will denote new
experimental chemical shifts in the 1H and 13C NMR spectra using
an italic font. The consideration of nonstandard connectivities
presented in structure 2(2a) shows that exchange of pairs of carbon
atom chemical shifts, specifically exchanging the pair 40.90(40.84)
and 40.40(40.32) ppm, immediately removes the nonstandard
lengths of three of the unusually long HMBC connectivities:
40.90–71.50 (40.84–71.51), 40.90–72.70 (40.84–72.64), and
40.90–61.00 (40.84–60.94) ppm. This exchange of assignments also
adjusts the proton–proton connectivities in the COSY spectrum
between the protons attached to the following pairs of carbons:
40.90–71.50 (40.84–71.51), 40.90–72.70 (40.84–72.64), 40.90–61.00
(40.84–60.94), and 40.4–139.6 (40.32–139.70) ppm to acceptable
lengths (3-4JHH). To edit the COSY (6JHH) and HMBC (5JCH)
connectivities 72.70–54.50 (72.64–54.48), we took into account that
four 1H NMR resonances, 3.53, 3.61, 3.63, and 3.66 ppm, are
observed in a very narrow spectroscopic interval (see Supporting
Information, Figure B, where an expansion of the 1H NMR spectrum
obtained at 900 MHz is displayed) related to several equivalent
possibilities for HSQC peak assignment may be expected. The 1H
NMR chemical shifts determined from this spectrum are very close
to those found from the HSQC spectrum. Assuming that 72.64(3.84)
and 40.84(3.63) ppm are newly assigned HSQC assignments, then
both the noted COSY and HMBC connectivities will assume
standard lengths. The 4JHH COSY connectivity 54.50–47.80
(54.48–47.71) ppm is adjusted to a standard length when the COSY
correlation 3.66–2.76 ppm is replaced by the correlation 3.63–2.76
ppm. The COSY peak 3.23–2.74 (53.10–47.8) ppm shown in Table
1 might be assigned to a cross-peak 3.20–2.76 ppm in the
experimental spectrum, but careful analysis of the spectrum shows
that this peak is absent from the 2D NMR data and should be
considered as an error. According to Table 1 the HMBC cross-
peak at 54.50–72.70 (54.48–72.64) ppm is related to two peaks:
3.64–72.70 (3.66–72.64) ppm and 3.80–54.50 (3.84–54.48) ppm.
Exchanging the assignments between 3.66 ppm and the nearby shift
of 3.63 ppm leads to a 2JCH correlation length. The HMBC peak
3.84–54.48 ppm is not observed in the experimental spectrum. The
remaining long (6JCH) HMBC connectivity 53.10–72.70
(53.09–72.64) ppm corresponds to the cross-peak at 3.80–53.1
(3.84–53.09) ppm and is transformed into a standard correlation
when the 13C NMR chemical shift 53.09 ppm is exchanged with
the peak at 53.21 ppm.
The final chemical shift assignment of the Rychnovsky15
structure is given in Table 5 (see ref 21), where NMR data obtained
in this work are shown in comparison with Gräfe’s data.13 The 13C
and 1H NMR chemical shifts determined in this work differ only
very slightly from those determined in the original work.13 The
near coincidence of the chemical shifts provides us with the basis
to conclude that the compound identified by Gräfe et al.13 had a
structure consistent with that suggested by Rychnovsky.15 This
conclusion is further confirmed by the coincidence of multiplicities
and coupling constants in the 1H NMR spectra obtained by Gräfe
et al.13 and in this report.
The 13C and 1H NMR chemical shift assignments are displayed
in structure 3 below:
The COSY correlations accompanied by the peak multiplicities
and the measured coupling constants are shown in structure 3a:
When the new NMR data presented in Table 5 were entered
into the StrucEluc program, only three structures were produced
in 0.095 s. After selection of the best structure, the program placed
structure 3 at the first position and with the same 1H and 13C NMR
assignments shown. Table 6 lists the deviations calculated for
hexacyclinol with both the old (see Table 3) and the new
assignments.
It is evident that the deviations calculated by all methods are
smaller for the structure for which chemical shifts were assigned
in this work. Note that all 2D NMR correlations in the structure 3
are now of standard length.
Experimental Section
The 900 MHz NMR data were obtained on a Bruker AVANCE-
900 NMR spectrometer, which was equipped with a Bruker 5 mm
TCI cryoprobe. A synthetic sample of (+)-hexacyclinol (∼1.2 mg)
was dissolved in ∼120 uL of CDCl3, contained in a 3 mm o.d.
NMR sample tube (∼3 cm sample height) and centered in the
receiver coil of the cryoprobe. 1H NMR spectra were acquired with
64K data points and zero-filled to 256K data points prior to Fourier
transformation. All 2D experiments (gCOSY, gHSQC, gHMBC)
employed gradient-enhanced pulse sequence versions and were
acquired with proton detection in f2. The gCOSY data was acquired
586 Journal of Natural Products, 2008, Vol. 71, No. 4 Williams et al.

with 2K points in f2 with 256 increments in f1. The f1 increments
were linear predicted to 2K, and both f1 and f2 were zero-filled to
4K points to yield a 4K × 4K data set after Fourier transformation.
The gHSQC and gHMBC experiments were acquired with 2K
points in f2 and 128 increments in f1 (gHSQC) and 256 increments
in f1 (gHMBC) and subsequently linear predicted to 1K increments
and 2K increments in f1, zero-filled to 4K × 2K (gHSQC) and 4k
× 4k (gHMBC), respectively, prior to Fourier transformation. The
frequency range of the carbon-13 sweep widths for the gHSQC
and gHMBC experiments was 180 and 220 MHz, respectively.
Conclusions
It is rather an uncommon situation when different research groups
publish different chemical structures for newly separated or
synthesized organic molecules, especially when 2D NMR data are
used for the structure elucidation. This is more likely to happen if
there is severe overlap in the NMR spectra since this eventuality
leads to ambiguous interpretation of the data. With severe overlap
in the proton spectrum, incorrect chemical shift assignments can
carry through the homonuclear and heteronuclear HSQC (HMQC)
spectra. If two researchers employ slightly different initial data,
then it is possible that they can arrive at different structures. In
this article, we have demonstrated that resolving contradiction(s)
between two or more proposed structures the capabilities within
the expert system StrucEluc7–10 can be extremely advantageous
since the system is capable of both generating and validating
different structural solutions derived from the data.
The system was applied to the selection of the most appropriate
structure from two suggested variants of the natural product
hexacyclinol. This compound was first separated and structurally
characterized by Gräfe et al.13 Subsequently the structure of
hexacyclinol was revised15,16 and an alternative structure was
confirmed via total synthesis. Combining the procedures of both
Table 5. Comparison of 1D and 2D NMR Data Presented in Ref 13 with the Data Obtained in the Current Work (with those
chemical shifts marked in italics)
# δC δH COSY HMBC
1 18.60 q 1.77 s 142.20, 120.70
18.50 1.79 4.85 142.26; 120.67; 26.25
2 142.20 s
142.26
3 26.10 q 1,72 s 142.20, 120.70
26.25 1.73 4.85 142.26; 120.67; 18.50
4 120.70 d 4.82 d, 10.1 5.46
120.67 4.85 m 5.49; 1.7, 1.79 75.74; 26.25; 18.50; 60.40
5 75.80 d 5.46 d, 10.1 4.82 60.50, 202.90
75.74 5.49 d, 9.04 4.85 60.40; 202.85; 142.26; 40.32; 120.67; 71.51
6 60.50 s
60.40
7 202.90 s
202.85
8 53.10 d 3.23 d, br, 3.5 3.64, 2.74 202.90, 54.50
53.09 3.25 d,3.67 3.66 202.85; 60.40
9 54.50 d 3.64 m 3.23, 2.74, 3.80 47.8
54.48 3.66 t, 3.53 3.25, 5.01
10 47.80 d 2.74, dd, 5.2, 7.8 3.64, 4.99 54.50, 60.50
47.71 2.76 dd, 10.31, 5.23 5.01; 3.63 54.48; 60.40; 40.32; 202.85; 71.51; 75.74
11 71.50 d 4.99 dd,5.2 br 2.74, 3.55
71.51 5.01 dd 5.10, 3.67 2.76, 3.66 53.09; 75.74; 60.40
12 40.40 d 3.55 m 4.99, 3.80 61.00, 71.50, 72.70
40.32 3.61 dd, 5.37, 3,11 6.76 60.40; 77.08; 139.70; 47.71; 202.85; 132.44; 24.69; 26.61
13 72.70 d 3.80 dd, 9.5, 1.5, 2.54 br (OH) 3.55, 3.64 40.40, 54.50, 47.80, 53.10
72.64 3.84 d, 9.61 2.33 47.71; 53.21; 60.94
14 61.00 d 3.51 dd, 2.9, 0.5 3.55, 3.29 40.40
60.94 3.53 d 3.4 3.32 40.84; 53.21; 72.65
15 53.20 d 3.29 d, 3.2 3.51 132.50, 192.80
53.21 3.32 d, 3.4 3.53 132.44; 192.75; 60.94
16 192.80 s
192.75
17 132.50 s
132.44
18 139.60 d 6.73 dd 5.3, 2.4 (allyl) 3.59 192.80, 40.90
139.70 6.76 dd, 5.65, 2.54 3.61 192.75; 40.32; 60.40
19 40.90 d 3.59 d, 5.3 6.73 139.60
40.84 3.63 tt, 9.9, 2.7 2.76 139.70; 72.64; 132.44
20 77.30 s
77.08
21 26.60 q 1.26 s 77.30, 40.90
26.61 1.28 77.08; 40.32; 24.69
22 24.70 q 1.15 s 77.30, 40.90
24.69 1.17 77.08; 40.32; 26.61
23 49.10 q 3.02 s 77.30
49.10 3.04 77.08
OH 3.84 72.64; 40.84
Table 6. Comparison of the Deviations Calculated for the
Rychnovsky Structure with both Old and New Chemical Shift
Assignments (all entries are in ppm)
old assignment new assignment
dA(13C) 5.03 dA(13C) 4.46
dI(13C) 3.70 dI(13C) 3.07
dN(13C) 4.02 dN(13C) 2.60
dA(1H) 0.34 dA(1H) 0.33
dI(1H) 0.30 dI(1H) 0.29
dN(1H) 0.29 dN(1H) 0.28
dA(Σ) 8.46 dA(Σ) 5.40
NMR Assignments of Hexacyclinol Journal of Natural Products, 2008, Vol. 71, No. 4 587

structure generation and spectrum prediction incorporated into the
expert system we checked both structural hypotheses and indepen-
dently confirmed the revised structure.
In order to help in a fresh analysis of the NMR data, we
reacquired 1D and 2D NMR spectra at 900 MHz for a synthetic
sample as described elsewhere.16 With these data we have been
able to reassign both the 1H and 13C chemical shifts. The
reassignment was governed by the heuristic requirement of
eliminating unusually long COSY and HMBC correlations. As a
result, a number of the chemical shift pairs with very similar
chemical shifts were permuted and all 2D NMR connectivities were
converted to standard lengths. Simultaneously, the chemical shift
deviations calculated by all available prediction methods were
smaller than those suggested by Rychnovsky, thereby endorsing
the new assignment. The number of exchanged chemical shifts is
actually quite large, and the criterion of minimizing the number of
nonstandard correlations is heuristic in nature and leads to the
question, is the suggested assignment correct? Turning to the works
of Sigmund Freud20 we quote: “An attribute of scientific thinking
is the possibility to be content with an approximation to the truth
and to continue creative work in spite of the absence of final
confirmation”. We have no further data available21 to us to assist
in checking the suggested assignment, so we accept the conclusions
as being accurate and appropriate.
Acknowledgment. The purchase of the 900 MHz NMR spectrometer
and construction of the UIC Center for Structural Biology were funded
by NIH Grant GM068944.
Supporting Information Available: A graphic indicating the nature
of a molecular connectivity diagram generated by the Structure
Elucidator software program during the computer-assisted structure
elucidation process. An expansion of the 900 MHz 1H NMR spectrum
of hexacyclinol is shown as evidence of the nature of peak overlap in
the spectrum as discussed in the text. This material is available free of
charge via the Internet at http://pubs.acs.org.
References and Notes
(1) Lederberg, J.; Sutherland, G. L.; Buchanan, B. G.; Feigenbaum, E. A.;
Robertson, A. V.; Duffield, A. M.; Djerassi, C. J. Am. Chem. Soc.
1968, 91, 2973–2976.
(2) Elyashberg, M. E.; Gribov, L. A. Zh. Prikl. Spectrosc. 1968, 8, 296–
300.
(3) Sasaki, S. I.; Abe, H.; Ouki, T.; Sakamoto, M.; Ochiai, S. Anal. Chem.
1968, 40, 2220–2223.
(4) Nelson, D. B.; Munk, M. E.; Gasli, K. B.; Horald, D.L. J. Org. Chem.
1969, 34, 3800–3805.
(5) Elyashberg, M. E.; Gribov, L. A.; Serov, V. V. Molecular Spectral
Analysis and Computer (in Russian); Nauka: Moscow, 1980.
(6) Elyashberg, M. E.; Martirosian, E. R.; Karasev; Yu., Z.; Thiele, H.;
Somberg, H. Anal. Chim. Acta 1997, 348, 443–463.
(7) Elyashberg, M. E.; Blinov, K. A.; Molodtsov, S. G; Williams, A. J.;
Martin, G. E. J. Chem. Inf. Comput. Sci. 2004, 44, 771–792.
(8) Molodtsov, S. G.; Elyashberg, M. E.; Blinov, K. A.; Williams, A. J.;
Martin, G. E.; Lefebvre, B. J. Chem. Inf. Comput. Sci. 2004, 44, 1737–
1751.
(9) Elyashberg, M. E.; Blinov, K. A.; Molodtsov, S.; G; Williams, A. J.;
Martin, G. E. J. Chem. Inf. Model. 2006, 46, 1643–1656.
(10) Elyashberg, M. E.; Blinov, K. A.; Molodtsov, S.; G; Williams, A. J.;
Martin, G. E. J. Chem. Inf. Model. 2007, 47, 1053–1066.
(11) Blinov, K. A.; Carlson, D.; Elyashberg, M. E.; Martin, G. E.;
Martirosian, E. R.; Molodtsov, S. G.; Williams, A. J. Magn. Reson.
Chem. 2003, 41, 359–372.
(12) Blinov, K. A.; Elyashberg, M. E.; Martirosian, E. R.; Molodtsov, S. G.;
Williams, A. J.; Sharaf, M. M. H.; Schiff, P. L. Jr.; Crouch, R. C.;
Martin, G. E.; Hadden, C. E.; Guido, J. E.; Mills, K. A. Magn. Reson.
Chem. 2003, 41, 577–584.
(13) Schlegel, B.; Härtl, A.; Dahse, H.-M.; Gollmick, F. A.; Gräfe, U.;
Dörfelt, H.; Kappes, B. J. Antibiot. 2002, 55, 814–817.
(14) La Clair, J. J. Angew. Chem., Int. Ed. 2006, 45, 2769–2773.
(15) Rychnovsky, S. D. Org. Lett. 2006, 8, 2895–2898.
(16) Porco, J. A., Jr; Su, S.; Lei, X.; Bardhan, S.; Rychnovsky, S. D. Angew.
Chem., Int. Ed. 2006, 45, 1–4.
(17) ACD/CNMR Predictor v.10.0; Advanced Chemistry Development Inc.:
110 Yonge St., Toronto, Ontario, Canada M5C 1T4.
(18) Reynolds, W. F. The Hexacyclinol Controversy; What Went Wrong
With The Original Structure Determination?, Poster #34, SMASH
NMR Meeting., Burlington, , 2006.
(19) ACD/SpecManager v.10.0; Advanced Chemistry Development Inc.:
110 Yonge St., Toronto, Ontario, Canada M5C 1T4.
(20) Freud, S. New Introductory Lectures on Psycho-analysis; London,
1933.
(21) The current assignments given in Table 5 are consistent with the
reassignments reported in the hexacyclinol total synthesis report. See
the Supporting Information associated with ref 16 for further details.
NP070557T
588 Journal of Natural Products, 2008, Vol. 71, No. 4 Williams et al.