Proc. Nati. Acad. Sci.
USA
Vol. 75, No. 8, pp. 35404547, August 1978
Applied Physical and Mathematical Sciences
Structure determination of crystalline substances by diffraction
methods: Philosophic concepts and their implementation (A Review)*
(phase problem/non-negativity/ordinary structures/biological macromolecules)
JEROME KARLE
Laboratory for the Structure of Matter, Naval Research Laboratory, Washington, D.C. 20375
Contributed by Jerome Karle, June 1, 1978
ABSTRACT Structure determination of single crystals by
diffraction methods is reviewed in terms of the philosophical
and mathematical aspects of the analytical techniques that are
used. A key problem concerns the need to determine the relative
phases of the scattered rays. A major advance in treating this
problem resulted from recognition that useful consequences
ensue from the fact that electron densities in crystals are non-
negative functions. Advantage is also taken of the fact that the
problem is usually greatly overdetermined by the number of
experimental data. For biological macromolecules, the data
available are more limited in range and accuracy and great use
is made of the introduction of suitable heavy atom moieties. The
developments in crystal structure analysis have had a consid-
erable impact on progress in many scientific disciplines.
The determination of the atomic arrangements in crystals by
diffraction techniques affords especially good examples of the
usefulness that is derived from combining physical constraints
with diffraction theory. In applications to single crystals, use
is made of the non-negativity of the electron density distribution
to provide formulas for determining the relative phases of the
scattered rays. The virtue of the concept of non-negativity in
diffraction analysis has been discussed in a previous article (1)
concerning the structures of gaseous and amorphous substances.
In this earlier article, several aspects of diffraction analysis were
discussed. It was pointed out in some detail that diffraction
experiments are generally ambiguous in the sense that non-
physical models can be formulated that satisfy the diffraction
data. The ambiguities are overcome by the introduction of
constraints that are based on physical and chemical consider-
ations. Also discussed in some detail in the previous article (1)
is the concept of bridging, which concerns the adjustment of
theoretical formulas and the treatment of experimental data
to achieve useful results and improve their accuracy. The
ensuing discussion, particularly of single crystals, provides many
illustrations of these many aspects of diffraction analysis.
A characteristic of true crystals is three-dimensional peri-
odicity. Very great numbers and varieties of substances can
agglomerate to form crystals. Monatomic crystals can be formed
from the elements, and complex ones can be formed from
macromolecules such as proteins and viruses. Crystalline ma-
terials provide, as a rule, a large amount of diffraction infor-
mation relative to the number of unknown structural param-
eters to be determined, although this advantageous ratio de-
creases dramatically for macromolecular substances. When the
advantageous ratio prevails, it accounts for the fact that simple
mathematical formulations can provide the solution to a rather
complex analytical problem.
CHARACTERIZATION AND METHOD
The determination of the structures of single crystals has a very
broad range of application. This follows not only from the
propensity that all sorts of materials have to crystallize readily,
but also from developments in techniques for overcoming a key
3540
obstacle to the deduction of atomic arrangements from the
observed diffraction intensity data. The diffraction pattern
taken by the Weissenberg technique and shown in Fig. 1 is
typical. Four numbers are associated with each diffraction
maximum. Three numbers are the Miller indices, which
identify the set of planes in the crystal that is involved in pro-
ducing the particular diffraction maximum, and the fourth
number is a measure of the intensity. The development of a
useful theory for addressing this problem affords a good illus-
tration of the use of physical constraints in structure analysis,
and the development of a suitable procedure provides a
worthwhile opportunity to observe the process of bridging
between mathematical results and practical application. Despite
the complexity of the structures of interest and of the mathe-
matical relationship between the atomic positions and the
diffracted intensities, considerable progress has been made
toward making the direct determination of crystal structures
from the measured intensities a fairly routine operation.
The method for accomplishing this is known as the "direct
method." This terminology implies that the structure is deter-
mined without the use of special information, such as the known
positions of heavy atoms that may be present. The direct
method involves the determination of the phases associated with
the scattered amplitudes directly from the measured intensities.
An intensity is the absolute magnitude squared of the ampli-
tude. Once the phases are known, appropriate Fourier series
that use as coefficients the observed magnitudes of the ampli-
tudes and their corresponding phases can immediately provide
the desired structural information.
A key feature of the attack on the problem of determining
the phases of the diffraction amplitudes was the recognition that
the non-negativity of the electron density in a crystal imposes
a crucial constraint on the system that could possibly lead to a
practical solution of the problem. It is also valuable that the
electron distributions about the individual atoms are known to
a good approximation and that it is possible to transform the
observed x-ray diffractiop intensities to those that would accrue
from essentially point htoms. In point atoms, the electron
density is concentrated in a central point in an atom, rather than
being distributed in space. With the knowledge of the atomic
electron densities at hand, it is easy to show that the problem
is greatly overdetermined by the number of diffracted inten-
sities normally obtained for crystals, except for the very complex
ones such as are formed by proteins and nucleic acids. The
overdeterminacy can inspire the hope that relatively simple
mathematical relationships might be found among the phases
that can have a high probability of being correct. This is, indeed,
what happens, and some of the mathematical details will be
outlined below showing the key role played by the non-nega-
tivity of the electron density.
* By invitation. From time to time, reviews on scientific and techno-
logical matters of broad interest are published in the PROCEED-
INGS.
 Applied Physical and Mathematical Sciences: Karle
-W -
0
-1
FIG. 1. X-ray diffraction pattern of a single crystal by the
Weissenberg technique.
Let us recall that a diffracted ray has an amplitude composed
of a magnitude and a corresponding phase. Measuring instru-
ments such as photographic film or photon counters normally
record only the intensity, which is the square of the magnitude
of the amplitude. The phase is therefore omitted in the re-
cording and apparently lost in the experiment. It is therefore
most remarkable that within the context and the constraints of
the crystal structure problem the phases are recoverable at all,
and that they are, in fact, recoverable from the measured in-
tensities. Strictly speaking, what is recovered is relative phase;
that is, the phase of a scattered amplitude relative to the phase
of other scattered amplitudes. It actually makes no sense to
consider recovering the phase of an individual reflection in
some isolated fashion since, for a number of reasons, this
quantity changes continually throughout an experiment.
However, given some minimal specifications, the relative phases
of the set of reflections are determinate. In phase-determining
procedures, it is necessary to specify the values of some phases
in order to establish an origin in the crystal of interest to which
atomic positions may be referred. The remaining phases that
are evaluated have values that are determined relative to those
that are specified.
The maxima of the electron density distribution p(r) locate
the atomic positions. Since crystal structures possess three-
dimensional periodicity, p(r) may be expressed in terms of a
three-dimensional Fourier series
p(r) = V-1 Fh exp(-2ih * r), [1]
where the coefficients
Fh = IFhI exp(i4h) [2]
are the crystal structure factors associated with the planes la-
beled with the vector h and the h have integer components h,
k, and 1 (the Miller indices) whose values are inversely pro-
portional to the intercepts of crystal planes on the x, y, and z
axes, respectively. The angle qh is the phase associated with Fh,
and r is the position of any point in the unit cell. FhE is the
amplitude of the scattered wave associated with the crystal
plane labeled by h, where E is the electric field vector of the
incident beam. The measured x-ray intensities are proportional
to IFh12.
If the values for the 40h were obtained directly from experi-
ment, the structures could be immediately calculated from Eq.
1. The absence of phase information gives rise to a difficult
Proc. Natl. Acad. Sci. USA 75 (1978) 3541
problem in structure analysis and is generally referred to as the
"phase problem."
The Fourier inversion of Eq. 1 followed by the reduction of
the integral to the sum of contributions from the N discrete
atoms in the unit cell gives, for the Fourier coefficient,
N
E fjh exp(2irih- rj),
IFhI exp(ikh) = J=l [3]
where fjh is the atomic scattering factor (scattering amplitude)
of the jth atom in the unit cell and r1 is its position vector.
To estimate the solvability of the phase problem, Eq. 3 can
be considered to be a system of simultaneous equations since
measurements of the intensities are made for a large number
of h. The unknown quantities are the phases Oh and the atomic
positions rj. The known quantities are the IFhI obtained from
experiment and the fjh, the atomic scattering factors, which
are tabulated. It should also be noted that each of the simulta-
neous equations is actually two equations since both the real part
and the imaginary part of each equation may be set equal
separately. Comparison of the number of unknown quantities
to be determined with the number of independent data avail-
able from the use of an x-ray tube having a copper target in-
dicates that the overdeterminacy could be as great as a factor
of 50 for centrosymmetric crystals and 25 for noncentrosym-
metric ones. In the usual practice somewhat less than this degree
of overdeterminacy is obtained, but the factor is still quite
large.
By multiplying Eq. 3 by its complex conjugate, the phases
are eliminated and we obtain
N N
IFhI2 =
E E
j=1ik=
fZfjk exp[2irih- (rj-rk)]. [4]
This suggests the possibility that the atomic coordinates might
be obtained directly from the measured intensities without the
intermediate use of phases. As a practical matter, it has so far
been found to be generally more feasible to determine struc-
tures by first determining the phases and then computing Eq.
1 than to find the atomic coordinates directly.
The Fourier transform of Eq. 4 is known as the Patterson
function (2, 3),
co
P(r) = EFhi 2 exp(-2-ri h * r). [5]
The maxima of this function represent the interatomic vectors
in a structure. It is evident that the coefficients of Eq. 5 are
directly obtainable from the experimental measurements. The
difficulty with using this function in a general way for structure
determination arises from the lack of resolution that usually
occurs for the N(N - 1) interatomic vectors. There is, however,
a circumstance in which the Patterson function is particularly
useful and has found wide application. This is when a structure
possesses only a few heavy atoms. The interatomic vectors as-
sociated with the heavy atoms are then readily identifiable and
atomic positions for them can then be readily deduced. The
coordinates for the heavy atoms can be used in Eq. 3 to compute
an initial set of phases. There are numerous procedures for
developing a complete structure from this information (4-10).
The Patterson function also becomes somewhat accessible when
it is possible to introduce structural information in the form of
known atomic groupings (11-13). A detailed study of the
properties of the Patterson function has been presented by
Buerger (14).
Despite the limitations on the direct analysis of the Patterson
function, the great overdeterminacy of the problem has moti-
vated the search for alternative methods that could extend the
range of complexity of structures that could be investigated.
3542 Applied Physical and Mathematical Sciences: Karle
The direct method for phase determination has this capability.
Its philosophic basis and mathematical background will now
be outlined.
The usefulness of the non-negativity criterion as described
in the previous article (1) for the electron diffraction of gases
motivated the search for additional applications. The result of
imposing this constraint on the electron density distribution in
crystals is an infinite set of determinantal inequalities of in-
creasing order whose elements are the structure factors (15).
In order of complexity, the first inequality statement is that FOWo
must be non-negative, the second is that IFhj S Fooo, and the
third is a relationship among the structure factors that plays a
primary role in direct crystal structure determination. The
inequality is written
Fooo F -k F-h
V Foo F-h+k I>o 6 Although the Harker-Kasper inequalities did not include in-
| h Fh-k Fooo | [6]
Its important features may be seen by rewriting it in the form
(15)
FkFh-k[
|1
FO O
IFooo F -k 1/2 |Fooo F -h+k 1/2
< Fk FOOOI
| Fh-k Fooo | [7]
The interpretation of this inequality is that the structure factor
Fh is bounded by a circle in the complex plane that has its center
at FkFh-k/Fooo and a radius given by the right side of in-
equality 7 (Fig. 2). When the structure factors Fk and Fh-k have
an unusually large magnitude, inequality 7 becomes quite re-
strictive because the right side becomes rather small. This
permits the conclusion that Oh, the phase of Fh, is approximately
equal to Ok + 4h-k, the phase of FkF h-k. Inequality 7 can be
strengthened by transforming the structure factors F to those
that would be obtained from point atoms. Even with this
strengthening, if it were necessary to be guided solely by the
bounding range of the inequality, the relation
Oh -ok + Oh-k [8]
would be applicable to only the simplest of crystals.
Expression 8 has, however, extensive application. This is so
because inequality 7 has probabilistic implications that, in ef-
fect, confine the bounding range, giving ok + Oh-k as the most
Im
Re
FIG. 2. Circle in complex plane showing interpretation of in-
equality 7, where 6 = FkFh-k/Fooo and r represents the right side of
the inequality. Fh must be on or within the circle. Knowledge of PFhI
further restricts Fh to the line that intersects the circle.
Proc. Natl. Acad. Sci. USA 75 (1978)
probable value of kh. In fact, expression 8 is more probably
correct, the larger the values for the structure factor magnitudes
associated with the phases involved. Such probabilistic features
can be obtained directly in a quantitative fashion by exami-
nation of inequality 7. The quantity FkFh-k/Fooo can be con-
sidered as an expected value for Fh, and a measure of the
variance can be based on the right side of inequality 7. With
these quantities, a probability distribution function for Fh may
be immediately written (16) by use of the central limit theo-
rem.
Various aspects of probability theory have been applied to
the phase problem over the years (17-21), starting soon after
the development of the inequality theory. The probabilistic
characteristics of inequalities had been noted by Gillis (22), who
was working with a set of inequalities derived by Harker and
Kasper (23) somewhat earlier than the determinantal theory.
equality 7, they were stimulating in showing that simple in-
equality relationships could give phase information (24) and
in providing early insight into their probabilistic implications
(22). The determinantal inequalities were derived (15) as a
consequence of the non-negativity of the electron density dis-
tribution. These types of determinantal inequalities were known
to mathematicians working with non-negative Fourier series.
After the determinantal inequalities were obtained, it was noted
that the Harker-Kasper inequalities were contained in them.
Centrosymmetric crystals, i.e., crystals having a center of
symmetry, can have phase values that are only 0 or ir for a
properly chosen origin in the crystal. This means that only a plus
or minus sign attaches to jFhj for such crystals. A simple
probability measure of this in the convenient hyperbolic tangent
form of Woolfson (18) is given by the probability that the
structure factor Fh be plus (i.e., has a phase of zero), P+(h),
given several normalized structure factors Ek and Eh-k,
P+(h) - '/2 + l/2 tanh a3a2-32 I Eh IFk EkE h-k. [9]
The normalized structure factors represent scattering from
essentially point atoms and have the property that the average
value of their magnitudes squared is equal to unity, i.e.,
(IEhj2)h = 1. The quantities an, are defined in terms of the
atomic numbers, Z1,
N
orn = E Z7n [10]
1=1
Certain variance factors that would enhance P+(h), depending
upon how large the I El in Eq. 9 are, have been omitted.
Crystals that lack a center of symmetry can have any value
for a phase. In that case, useful mathematical tools are the ap-
propriate probability distributions (16, 19) and measures of the
variance (16, 21).
The development of a practical procedure of broad appli-
cation from relation 8 and its corresponding probability mea-
sures, the symbolic addition procedure (21), was not immediate.
It required many years and was, in fact, preceded by a proce-
dure for phase determination for centrosymmetric crystals (17)
that afforded experiences that facilitated its development. The
general approach in the symbolic addition procedure involves
initially the stepwise use of expression 8 and the appropriate
corresponding probability measures. Evidently, in order to use
expression 8 certain phase values must be available. They are
derived from certain allowed specifications that, for example,
locate the origin in the crystal and certain others that are
specified by symbols. The stepwise procedure involves the
definition of a large number of phases in terms of the ones that
have been specified, with each step guided by considerations
 Applied Physical and Mathematical Sciences: Karle
of high probability. In a very large number of cases, only a.few
phase values need to be specified symbolically, usually no more
than five and often less. Symbols for phasesiiiciatedwithdl
or imaginary structure factors take on only two values, 0 or X
for the real factors and +7r/2 for the imaginary ones. For
general structure factors, it suffices to limit the phase values to
these four cardinal points.
Many ways exist for evaluating the symbols. Relationships
develop among the symbols as the phase determination pro-
ceeds. In addition, there are several auxiliary formulas available
for helping with the evaluation of symbols. Finally, Fourier
series for the remaining alternative sets of values for the phases
can be computed and examined for suitability as an acceptable
physical answer.
Other procedures have been developed that use many al-
ternative sets of numerical phase specifications instead of the
symbols that can represent these alternatives. The motivation
for this has been the ease with which the programming of a
computer can be performed and, to some extent, the expecta-
tion that certain advantages would accrue from being able to
use phase-determining formulas with numbers rather than
symbols. There is an ambiguity of 2wr that interferes with the
averaging of different symbolic definitions of a phase. It is
possible to use symbols, however, in such a way that the antic-
ipated differential advantages from the use of alternative nu-
FIG. 3. The four cocrystallizing conformers of cyclohexaglycyl. The form at the upper left occurs four times in the unit cell, the one at the
lower left occurs twice, and the other two occur once. [Reprinted with permission from Karle, I. L., Gibson, J. W. & Karle, J. (1970) J. Am. Chem.
Soc. 92, 3755-3760. Copyright by the American Chemical Society.]
Proc. Natl. Acad. Sci. USA 75 (1978) 3543
merical values for phases do not accrue. This is effected by using
the capacity of computers to use and store numerous symbolic
definitions for a phase and to defer the taking of averages until
the symbols are evaluated. In addition, it should be noted that
the capacity of symbols to store phase information can far ex-
ceed the capacity of computing machines to process the indi-
vidual numerical alternatives that the symbols represent.
The most difficult structure determinations are those that
concern essentially equal atom noncentrosymmetric crystals
having many atoms in the asymmetric unit, perhaps 50-100
atoms or more. An asymmetric unit is that part of the unit cell
of a crystal whose composition is unrelated to the symmetry
elements of the crystal. The composition of the entire unit cell
can be generated from the asymmetric unit by application of
symmetry operations appropriate to the crystal. Occasionally,
a phase determination does not succeed in producing an answer.
This can occur because the stepwise procedure is based on
probabilities rather than certainties, and errors can accumulate.
Even this does not present insuperable difficulties, however,
because the stepwisebpath through a phase determination is far
from unique and a fresh start could provide a path to more
accurate phase determination.
A great many structure determinations have been performed
by use of the direct method of phase determination; it has
stimulated research in numerous fields of science because basic
3544 Applied Physical and Mathematical Sciences: Karle
structural information is now much more readily available than
in the past. This is especially true in the fields of organic
chemistry, biological chemistry, and natural products chem-
istry. Structure analysis affords information concerning struc-
tural formula, configuration, and conformation and finds ap-
plication in a variety of investigations concerning, for example,
products of syntheses, biosynthetic pathways, reaction inter-
mediates, rearrangements, reaction mechanisms, ion transport,
and radiation damage to genetic material.
An interesting example of an application is given by the first
substance investigated by the symbolic addition procedure,
cyclohexaglycyl (25). This cyclic peptide crystallizes in space
group P1 with four different conformers in a unit cell con-
taining 196 nonhydrogen atoms, of which 98 are independently
placed (Fig. 3). Not only is the cocrystallization of the four
conformers of inherent interest, but the results of this investi-
gation have been used as models for the prediction of confor-
mation (26, 27).
Efforts to advance the theoretical aspects of crystal structure
determination continue. One path concerns the development
of probabilistic formulas and applications for the higher order
phase invariants, e.g., quartets and quintets. The quartet in-
variants have a long history of development and application,
starting with the Harker-Kasper inequalities (23) and the
sigma-3 formula (17). Recent interest was stimulated by Schenk
and de Jong (28), who suggested a test for selecting a correct set
of phases from among several alternatives. The test was based
on the use of special quartets that, on the basis of indications
from Harker-Kasper inequalities, might be expected to be
negative, i.e., to have a phase of ir. Further developments have
concerned joint probability distributions that include quartets
and their closely related structure factor magnitudes called
neighborhoods (29-31). Such formulas have been derived by
Hauptman (32) and Giacovazzo (33). Similar derivations have
been carried out by Hauptman and Fortier for quintets (34) and
sextets (35). The objective of the introduction of numerous
structure factor magnitudes into the joint distributions is the
expectation that the conclusions from the resulting conditional
probability distributions for the higher order phase invariants
would be more reliable, given the known values of the magni-
tudes from experiment.
An alternative path for theoretical investigation has been the
development of the probabilistic properties of the higher order
determinants, i.e., determinants of the type given by formula
6 but of order 4 and higher. Such studies have been pursued by
Tsoucaris (36) and Karle (16, 37) and bear a relationship to those
that focus on the individual higher order invariants, such as
quartets and quintets, since the determinants contain phase
invariants ranging from triplets to n-tets, where n is the same
order as the determinants. The determinants also contain in-
formation concerning structure factor neighborhoods, and it
is possible to derive special probability distributions for quartets
and higher invariants. This fact, however, is not the primary
motivation for investigating the higher order determinants. The
motivation is derived from the increasingly restrictive bound
that high order determinantal inequalities can place on a phase
or, from another point of view, their closer approach to be-
having like strict equalities. This suggests that advantages would
be derived from being able to use the determinantal probability
distributions and their implications in toto without the re-
duction to special formulas. One aspect of this is the use of a
general form of the maximum determinant rule, an implication
of the determinantal probability distributions, to obtain an
evaluation of the symbolic definitions of phases, as suggested
by Woolfson (38). In a related way, Tsoucaris (36) showed the
value of higher order determinants in selecting the correct set
Proc. Natl. Acad. Sci. USA 75 (1978)
of phases from among alternative ones. The higher order de-
terminants also hold promise for application in the initial stages
of direct phase determination.
What impact the continuing theoretical investigations will
have is a question for the future. The ultimate test is whether
they can significantly facilitate current practice or extend the
range of complexity of structures that can be presently han-
dled.
For additional reading, there are books on x-ray crystallog-
raphy by Zachariasen (39) and by Buerger (40), a recent book
on crystallographic computing techniques edited by Ahmed
et al. (41), an outline of several aspects of phase determination
(42), and review articles (43, 44). Aspects of crystal structure
determination by neutron diffraction have been discussed in
a book by Bacon (45). A book by Vainshtein discusses crystal
structure determination by electron diffraction (46).
MACROMOLECULAR CRYSTALS
Macromolecules often fold up to form globular structures and
can crystallize with three-dimensional periodicity. Some
macromolecules form long, fibrous structures that do not form
globules. The fibers generally agglomerate with two-dimen-
sional periodicity, but the third dimension in the direction of
the fiber axis often exhibits various kinds of disorder. I discuss
'here the structures of large molecules whose crystals possess true
three-dimensional periodicity.
About 100 structure determinations of protein molecules
have been performed to date. They include globins and other
reversible oxygen-transporting molecules, lysozymes, nucleases,
proteases, cytochromes, synthetases, hormones, immunoglob-
ulins, electron transport proteins, and glycolytic enzymes. Some
studies of simple viruses are approaching atomic resolution, and
the structure investigation of tRNA has reached the refinement
stage. The smallest protein contains about 500 nonhydrogen
atoms. This is about as small a molecule that we consider to be
a macromolecule. The viruses currently under investigation
have molecular weights that range from 2 to 9 X 106. They are
composed of numerous protein subunits in association with a
nucleic acid. The problem is somewhat simplified by the
presence of symmetry or near symmetry, so that only part of
the atoms need to be independently located. Nevertheless, the
simplest problem involves the determination of the coordinates
for about 15,000 nonhydrogen atoms.
The development of techniques for determining the struc-
tures of macromolecules has been a major stimulus in the bio-
logical sciences. The results of structure investigations estab-
lished the foundations and have continued to motivate progress
in the field of molecular biology.
Besides the complexity associated with the large number of
atoms that need to be located, macromolecules present addi-
tional problems for the analyst. The main problem is the rela-
tively few data that are available compared to the number of
unknown parameters. The data are limited because of positional
disorder, which occurs in the crystals of these large molecules.
Positional disorder means that the locations of the atoms are not
precisely the same from one unit cell to the other. This has the
effect of damping out the diffraction data at the higher scat-
tering angles. The problems that arise from the limitations in
the range of the data and the general complexity of the mate-
rials have been overcome by special techniques for treating the
phase problem.
The limited data range also affects the accuracy with which
final results may be obtained, and much effort is presently being
expended on structure refinement in order to optimize the
accuracy of the results. The accuracy is often needed for a
 Applied Physical and Mathematical Sciences: Karle
clarification of the relationship of the structure to its func-
tion.
The theoretical basis for the analysis of macromolecular
crystals is derived from the work of Patterson (2, 3) in the
context of its implications regarding the usefulness of the-
presence of a heavy atom in the structure to be determined. As
was pointed out previously, the location of a heavy atom and
the calculation of phases based on the heavy atom affords a
useful first step in structure determination. The situation is
somewhat more complicated for macromolecules. With so
many atoms present, a single heavy atom has too little effect
on the total scattering to sufficiently influence the values of the
phases and thus permit the usual application of the heavy atom
method. One solution would be to try to introduce many heavy
atoms. This makes for other complications, however, such as
the accurate location of all heavy atom substituents, so that
another approach has been followed. It is called the method of
isomorphous substitution and is a more powerful way to use a
heavy atom derivative than afforded by the usual heavy atom
method. The method involves the use of at least two different
crystals that have essentially identical structures except that they
differ with respect to their content of heavy atoms. In appli-
cations to protein structure analysis, one crystal is usually
formed from the native protein and others are made to contain
one or a very few heavy atoms. The usual method for intro-
ducing the heavy atom is to soak the protein crystal in a solution
containing a compound of a heavy atom. The process takes
place by diffusion of the compounds through solvent channels
in the crystal. There are particular sites on the protein molecule
that have a special affinity for certain heavy atom moieties. It
is common to make several heavy atom derivatives and to
combine the results from all the derivatives. This is called the
multiple isomorphous replacement method. In fact, when
applying the method to crystals that lack a center of symmetry,
as do protein crystals, at least two isomorphous derivatives of
the native protein are required, in the absence of additional
information such as is derived from anomalous dispersion, to
avoid an ambiguity in the evaluation of the phases. The analysis
of this aspect of isomorphous replacement was developed by
Bijvoet and coworkers (47, 48).
The isomorphous replacement technique, as applied to
noncentrosymmetric crystals, may be understood by reference
to Fig. 4. The structure factor for the native protein is denoted
by FN, the structure factors for the heavy atom substituents are
denoted by Fy and FZ, and those for the substituted proteins
by FN+Y and FN+Z. They satisfy the equations
FN+ FY = FN+Y [11]
and
FN+ FZ =FN+Z. [12]
In the usual experiment, the information available would be
IFNI, IFN+yI, IFN+zI, Fy, and Fz. FY and F_ are available
since the method involves the initial locating of the heavy atoms
in the unit cell. Eqs. 11 and 12 can be satisfied by two config-
urations, each placed symmetrically about the vectors Fy and
Fz, respectively, 'set at the origin. It is seen in Fig. 4 that only
one configuration for each of the pairs share FN in common.
The ambiguity is thus resolved and the appropriate phase angles
to be associated with the given magnitudes may be measured
from the real axis. Harker (49) suggested an alternative con-
struction for resolving the ambiguity in isomorphous replace-
ment.
Problems arise in the practical application of isomorphous
replacement because errors in measurement and changes in
Proc. Natl. Acad. Sci. USA 75 (1978) 3545
Im
Re
FIG. 4. Construction based on Eqs. 11 and 12 for multiple iso-
morphous replacement showing the relationship among the vectors.
The manner of choosing between alternative values for the phase
angle associated with FN is indicated by the solid lines, which have
FN in common.
structure affect the accuracy with which the phases can be
determined. Multiple isomorphous replacement involving
many different heavy atom derivatives can resolve such
problems and, once good heavy atom derivatives are available,
the practical aspects of the phase determination proceed in a
fairly routine fashion.
The first application of isomorphous replacement to mac-
romolecules was made in the investigation of myoglobin by
Kendrew and collaborators (51) and in the investigation of
hemoglobin by Perutz and his coworkers (52) (Fig. 5). This work
and the particular suitability of proteins for application of the
isomorphous replacement technique have facilitated the ex-
tensive developments that have taken place in the structure
investigations of proteins over the past 15 years. As noted, this
area of research has expanded into the investigation of virus
structures, with considerable progress in several laboratories
during the past few years toward the ultimate goal of atomic
resolution. First steps have also been taken in extending the
application to the investigation of the structures of nucleic acids
with the determination of the three-dimensional structure of
phenylalanine tRNA by Rich, Kim, and coworkers (53) and
Klug and coworkers (54).
Another valuable technique that has found application in
phase determination for macromolecules has also been devel-
oped by Bijvoet and his coworkers (55-57). It makes use of the
anomalous values for atomic scattering factors in the vicinity
of an absorption edge for an atom, referred to as anomalous
dispersion (39). An analysis of the anomalous dispersion tech-
nique as applied to noncentrosymmetric crystals may be found
in a previously cited review article (43). The initial applications
of anomalous dispersion were made by Bijvoet et al. (58) to
solve the problem of the absolute configuration of molecules
whose mirror images are distinct. In this way, they were able
to show that the convention established by Emil Fischer was,
in fact, consistent with what actually occurs in nature.
There is considerable evidence that structures obtained for
protein molecules in the crystalline state are biologically
meaningful. It has been shown, for example, that in many cases
enzymes in the crystalline state are active. The proteins that
3546 Applied Physical and Mathematical Sciences: Karle
FIG. 5. Schematic depiction of the fitting together in space of a
molecular unit of human hemoglobin composed of four protein mol-
ecules, each having a molecular weight of 16,000. The drawing is based
on one by Dickerson and Geis (50).
have been studied maintain the watery environment that they
have in solution by cocrystallizing with large amounts of water.
Some of the water is ordered in the crystal and some is not,
adding to the difficulties in analysis and refinement. Because
of the fairly large scattering factors of hydrogen and deuterium
for neutrons relative to those for other atoms, it is possible to
investigate the water structure in materials by neutron dif-
fraction in some detail. This contrasts with x-ray diffraction,
in which the scattering from hydrogen atoms is quite weak. It
is also possible to design experiments that take advantage of the
fact that the scattering factors for hydrogen and deuterium for
neutrons are negative and positive, respectively. An investi-
gation to this end is the neutron diffraction analysis of the
carbon monoxide derivative of myoglobin by Norvell et al.
(59).
The limited scattering range that is normally obtained for
protein structures places a severe limit on the resolution and
accuracy of the final results unless use is made of additional
information. The first Fourier maps of the electron density
distribution that are obtained from a phase determination for
proteins, in contrast with those for small structures, are usually
poorly resolved and rather inaccurate. Considerable im-
provements are to be expected from the use of mathematical
relationships that can refine sets of phases whose values are
approximately known and especially from the introduction of
chemical information in the form of amino acid sequences, the
structures of amino acid residues, and knowledge of bond dis-
tances, bond angles, and intramolecular interactions. Appli-
cations of these various means for improving the experimental
results have many aspects.
Model-building techniques have been developed by Dia-
mond (60) that make use of chemical sequencing and the
known structure of residues. Phase refinement calculations
based on a variety of phase-determining formulas (16, 61-65)
have been proposed and performed by several workers. Other
refinement techniques that have been developed involve the
use of noncrystallographic symmetry (66), modification of
electron density (67, 68), coordinate refinement with constraints
on the protein chains (69, 70), interactive display, model fitting,
and refinement (71-73), Fourier and least-squares refinement
of coordinates (74), and least-squares refinement of coordinates
coupled with the inclusion of restraints based on general
structural information such as bond distances and bond angles
(75). A major stimulus to the progress in protein structure re-
finement has come from the success of the work of Jensen and
Proc. Nati. Acad. Sci. USA 75 (1978)
Watenpaugh (74, 76). Discussions of several of these topics may
be found in ref. 40.
For readings in protein crystallography there are many ar-
ticles of a review or specialized nature. General reviews have
been presented by Phillips (77), Blundell and Johnson (78), and
Matthews (79). Articles have been written on the preparation
of isomorphous derivatives by Blake (80), on the x-ray crystal-
lography of enzymes by Eisenberg (81), on the molecular ar-
chitecture of oxygen-carrying proteins by Hendrickson (82),
and the evolutionary aspects of protein structures by Dickerson
(83). Noteworthy additions to the literature are a volume on the
structure and function of proteins (84) and a text by Blundell
and Johnson (85).
CONCLUDING REMARKS
The field of structure determination by diffraction methods
has made major advances in the development of mathematical
techniques, data reduction procedures, and analyses that pre-
pare experimental data for application to mathematical theory
and conversely. In these developments, advantage has been
taken of the numerous mathematical and physical constraints
that can be imposed. Considerable benefits have been obtained
from the use of chemical information and from structural in-
formation available from other techniques such as spectroscopy
and microscopy. Mathematical constraints, such as non-nega-
tivity, and their mathematical consequences have led to greatly
enhanced facility and accuracy in analyses. They have made
it possible to effect solutions of the phase problem in crystal
structure analyses. Special experimental techniques such as
isomorphous replacement and their associated theory have
made it possible to develop in depth a heretofore inaccessible
field, the investigation of macromolecular structure and its
correlation with function. It is reasonable to expect that the
progress and ingenuity that has characterized these efforts will
continue. Work is currently under way, for example, that may
one day lead to accurate models of the arrangements of atoms
in amorphous materials or reveal the structures of the complex
organizational features in living cells such as chromosomes and
ribosomes. In view of the significance and broad range of ap-
plication of many of the present studies, the future of these
activities can be viewed with deep interest and anticipation.
1. Karle, J. (1977) Proc. Nati. Acad. Sci. USA 74,4707-4713.
2. Patterson, A. L. (1934) Phys. Rev. 46,372-376.
3. Patterson, A. L. (1935) Z. fur Kristallogr. 90, 517-542.
4. Woolfson, M. M. (1956) Acta Crystallogr. 9,804-810.
5. Bertaut, E. F. (1957) Acta Crystallogr. 10, 670-671.
6. Sim, G. A. (1960) Acta Crystallogr. 13,511-512.
7. Hoppe, W. & Huber, R. (1963) in Crystallography and Crystal
Perfection, ed. Ramachandran, G. N. (Academic, New York),
pp. 61-65.
8. Srinivasan, R. (1966) Acta Crystallogr. 20, 143-144.
9. Karle, J. (1968) Acta Crystallogr. Sect. B 24, 182-186.
10. Sayre, D. (1972) Acta Crystallogr. Sect. A 28, 210-212.
11. Hoppe, W. (1975) Z. Elektrochemie 61, 1076-1083.
12. Nordman, C. E. & Nakatsu, K. (1963) J. Am. Chem. Soc. 85,
353-354.
13. Huber, R. & Hoppe, W. (1965) Chem. Ber. 98,2403-2424.
14. Buerger, M. J. (1959) Vector Space (Wiley, New York).
15. Karle, J. & Hauptman, H. (1950) Acta Crystallogr. 3, 181-
187.
16. Karle, J. (1971) Acta Crystallogr. Sect. B 27, 2063-2065.
17. Hauptman, H. & Karle, J. (1953) American Crystallographic
Association, Monograph No. 3 (Polycrystal Book Service, Pitts-
burgh).
18. Woolfson, M. M. (1954) Acta Crystallogr. 7,61-64.
19. Cochran, W. (1955) Acta Crystallogr. 8,473-478.
 Applied Physical and Mathematical Sciences: Karle
20. Karle, J. & Hauptman, H. (1956) Acta Crystallogr. 9, 635-
651.
21. Karle, J. & Karle, I. L. (1966) Acta Crystallr. 21, 849~591 i
22. Gillis, J. (1948) Acta Crystallogr. 1, 174-179.
23. Harker, D. & Kasper, J. S. (1948) Acta Crystallogr. 1, 70-75.
24. Kasper, J. S., Lucht, C. M. & Harker, D. (1950) Acta Crystallogr.
3, 436-455.
25. Karle, I. L. & Karle, J. (1963) Acta Crystallogr. 16,969-975.
26. Ramachandran, G. N. & Sasisekharan, V. (1968) in Advances in
Protein Chemistry, eds. Anfinsen, C. B., Jr., Anson, M. L., Edsall,
J. T. & Richards, F. M. (Academic, New York), pp. 283-438.
27. Venkatachalam, C. M. (1968) Biopolymers 6, 1425-1436.
28. Schenk, H. & de Jong, J. G. H. (1973) Acta Crystallogr. Sect. A
29,31-34.
29. Hauptman, H. (1977) Acta Crystallogr. Sect. A 33,553-555.
30. Hauptman, H. (1977) Acta Crystallogr. Sect. A 33,568-571.
31. Giacovazzo, C. (1977) Acta Crystallogr. Sect. A 33,933-944.
32. Hauptman, H. (1977) Acta Crystallogr. Sect. A 33,556-564.
33. Giacovazzo, C. (1976) Acta Crystallogr. Sect. A 32,91-99.
34. Hauptman, H. & Fortier, S. (1977) Acta Crystallogr. Sect. A 33,
575-580.
35. Hauptman, H. & Fortier, S. (1977) Acta Crystallogr. Sect. A 33,
697-701.
36. Tsoucaris, G. (1970) Acta Crystallogr. Sect. A 26,492-499.
37. Karle, J. (1978) Proc. Nati. Acad. Sci. USA 75,2545-2548.
38. Woolfson, M. M. (1977) Acta Crystallogr. Sect. A 33, 219-
225.
39. Zachariasen, W. H. (1945) Theory of X-ray Diffraction of
Crystals (Dover, New York).
40. Buerger, M. J. (1962) X-ray Crystallography (Wiley, New
York).
41. Ahmed, F. R., Huml, K. & Sedlacek, B., eds. (1976) Crystallo-
graphic Computing Techniques (Munksgaard, Copenhagen).
42. Ibers, J. A. & Hamilton, W. C., eds. (1974) International Tables
for X-ray Crystallography (Kynoch Press, Birmingham, En-
gland).
43. Karle, J. (1969) in Advances in Chemical Physics, eds. Prigogine,
I. & Rice, S. A. (Interscience, New York), Vol. 16, pp. 131-
222.
44. Robertson, J. M., ed. (1972) Chemical Crystallography (Butter-
worth, London).
45. Bacon, G. E. (1975) Neutron Diffraction (Clarendon, Ox-
ford).
46. Vainshtein, B. K. (1964) Structure Analysis by Electron Dif-
fraction (Pergamon, Oxford).
47. Bokhaven, C., Schoone, J. C. & Bijvoet, J. M. (1951) Acta Crys-
tallogr. 4, 275-280.
48. Bijvoet, J. M. (1952) in Conference on Computing Methods and
the Phase Problem in X-ray Crystal Analysis: Penn State Col-
lege, ed. Pepinsky, R. (X-ray Crystal Analysis Laboratory, De-
partment of Physics, Pennsylvania State College, PA.), pp. 84-
89.
49. Harker, D. (1956) Acta Crystallogr. 9, 1-9.
50. Dickerson, R. E. & Geis, I. (1969) The Structure and Action of
Proteins (W. A. Benjamin, Menlo Park, CA).
51. Kendrew, J. C., Dickerson, R. E., Strandberg, B. E., Hart, R. G.,
Davies, D. R., Phillips, D. C. & Shore, V. C. (1960) Nature 185,
422-427.
52. Perutz, M. F., Rossmann, M. G., Cullis, A. F., Muirhead, H., Will,
G. & North, A. C. T. (1960) Nature 185,416-422.
53. Kim, S. H., Suddath, F. L., Quigley, G. J., McPherson, A., Suss-
man, J. L., Wang, A. H. J., Seeman, N. C. & Rich, A. (1974)
Science 185, 435-440.
Proc. Natl. Acad. Sci. USA 75 (1978) 3547
54. Robertus, J. D., Ladner, J. E., Finch, J. T., Rhodes, D., Brown,
R. S., Clark, B. F. C. & Klug, A. (1974) Nature 250,546-551.
55. i:ijvoet, J. M. (1954) Nature 173,888-891.
56. Peterson, S. W. (1955) Nature 176,395.
57. Bijvoet, J. M. & Peerdeman, A. F. (1956) Acta Crystallogr. 9,
1012-1015.
58. Bijvoet, J. M., Peerdeman, A. F. & van Bommel, A. J. (1951)
Nature 168, 271-272.
59. Norvell, J. C., Nunes, A. C. & Schoenborn, B. P. (1975) Science
190,568-570.
60. Diamond, R. (1966) Acta Crystallogr. 21, 253-266.
61. Weinzerl, J. E., Eisenberg, D. & Dickerson, R. E. (1969) Acta
Crystallogr. Sect. B 25,380-387.
62. Hendrickson, W. A. & Karle, J. (1973) J. Biol. Chem. 248,
3327-3334.
63. Sayre, D. (1974) Acta Crystallogr. Sect. A 30, 180-184.
64. DeRango, C., Mauguen, Y. & Tsoucaris, G. (1975) Acta Crys-
tallogr. Sect. A 31, 227-233.
65. Podjarny, A. D., Yonath, A. & Traub, W. (1976) Acta Crystallogr.
Sect. A 32,281-292.
66. Bricogne, G. (1974) Acta Crystallogr. Sect. A 30, 393-405.
67. Hoppe, W. & Gassmann, J. (1968) Acta Crystallogr. Sect. B 24,
97-107.
68. Collins, D. M. (1975) Acta Crystallogr. Sect. A 31, 388-389.
69. Diamond, R. (1971) Acta Crystallogr. Sect. A 27,436-452.
70. Huber, R., Kukla, D., Bode, W., Schwager, P., Bartels, K.,
Deisenhofer, J. & Steigemann, W. (1974) J. Mol. Biol. 89, 73-
101.
71. Hermans, J. & McQueen, J. E. (1974) Acta Crystallogr. Sect. A
30,730-739.
72. Dodson, E., Isaacs, N. W. & Rollett, J. S. (1976) Acta Crystallogr.
Sect. A 32, 311-315.
73. Ten Eyck, L. F., Weaver, L. W. & Matthews, B. W. (1976) Acta
Crystallogr. Sect. A 32, 349-350.
74. Watenpaugh, K. D., Sieker, L. C., Herriott, J. R. & Jensen, L. H.
(1973) Acta Crystallogr. Sec. B 29,943-956.
75. Konnert, J. H. (1976) Acta Crystallogr. Sect. A 32, 614-617.
76. Jensen, L. H. (1974) Annu. Rev. Biophys. Bioeng. 3,81-93.
77. Phillips, D. C. (1966) in Advanccs in Structure Research by
Diffraction Methods, eds. Brill, R. & Mason, R. (Interscience,
New York), Vol. 2, pp. 75-140.
78. Blundell, T. L. & Johnson, L. N. (1972) in Chemical Crystal-
lography, ed. Robertson, J. M. (Butterworth, London), pp.
199-246.
79. Matthews, B. W. (1977) in The Proteins, eds. Neurath, H. & Hill,
R. L. (Academic, New York), Third Ed., Vol. 3, 403-590.
80. Blake, C. C. F. (1968) Adv. Protein Chem. 23, 59-120.
81. Eisenberg, D. (1970) in The Enzymes, ed. Boyer, P. D. (Aca-
demic, New York), pp. 1-89.
82. Hendrickson, W. A. (1977) Trends in Biochem. Sci. 2, 108-
111.
83. Dickerson, R. E. (1979) in Second Taniguchi International
Symposium in Biophysics on Molecular Evolution and Poly-
morphism (National Institute of Genetics, Mishima, Japan), in
press.
84. Structure and Function of Proteins on the Three-Dimensional
Level Cold Spring Harbor Symposia on Quantitative Biology
(1971) (Cold Spring Harbor Laboratory, Cold Spring Harbor,
NY), Vol. 36.
85. Blundell, T. L. & Johnson, L. N. (1976) Protein Crystallography
(Academic, New York).