(Received for publication, February 18, 1997, and in revised form, May 7, 1997)
From the Department of Biochemistry and Molecular Biology, Indiana
University School of Medicine, Indianapolis, Indiana 46202 and the
Department of Chemistry, Loras College,
Dubuque, Iowa 52001
The structural determinants of substrate
recognition in the human class IV, or , alcohol dehydrogenase
(ADH) isoenzyme were examined through x-ray crystallography and
site-directed mutagenesis. The crystal structure of
ADH
complexed with NAD+ and acetate was solved to 3-Å
resolution. The human
1
1 and
ADH
isoenzymes share 69% sequence identity and exhibit dramatically different kinetic properties. Differences in the amino acids at positions 57, 116, 141, 309, and 317 create a different topology within
the
substrate-binding pocket, relative to the
1
1 isoenzyme. The nicotinamide ring of
the NAD(H) molecule, in the
structure, appears to be twisted
relative to its position in the
1
1
isoenzyme. In conjunction with movements of Thr-48 and Phe-93, this
twist widens the substrate pocket in the vicinity of the catalytic zinc and may contribute to this isoenzyme's high Km for
small substrates. The presence of Met-57, Met-141, and Phe-309 narrow the middle region of the
substrate pocket and may explain the substantially decreased Km values with increased
chain length of substrates in
ADH. The kinetic properties of a
mutant
enzyme (
309L317A) suggest that widening the middle
region of the substrate pocket increases Km by
weakening the interactions between the enzyme and smaller substrates
while not affecting the binding of longer alcohols, such as hexanol and retinol.
Human alcohol dehydrogenase (ADH)1
isoenzymes are NAD+-dependent, zinc
metalloenzymes that catalyze the reversible oxidation of alcohols to
aldehydes. The ADH system is the major pathway for the metabolism of
beverage ethanol as well as biological important alcohols or aldehydes
like retinol, 3-hydroxysteroids,
-hydroxy fatty acids, and
4-hydroxynonenal (1-3). Each isoenzyme in the ADH family is a dimer
comprised of two 40-kDa subunits. The individual subunits are comprised
of two domains, a catalytic domain and a coenzyme-binding domain (4).
Seven human ADH genes (ADH1-ADH7) have been identified (1,
5). The ADH1-ADH5 genes encode the
,
,
,
, and
subunits, respectively. The protein product of the ADH6
gene has not been identified in vivo. The
subunit is
encoded by ADH7. Polymorphism occurs at both the
ADH2 (
1,
2, and
3) and ADH3 (
1 and
2) loci (6), such that nine distinct human ADH subunits
have been identified. ADH isoenzymes have been assigned to five
distinct classes based on their amino acid sequences as well as their
electrophoretic and enzymatic properties (7). The human
,
,
and
isoenzymes comprise class I, and the
,
,
,
and ADH6 comprise classes II, III, IV, and V, respectively. All ADH
isoenzymes are expressed in the liver except for
ADH, which is
primarily localized in epithelial tissue, such as the stomach mucosa
(8, 9).
The three-dimensional structures of horse and human class I ADHs have
been solved by x-ray crystallography (4, 10-12). Recently, the
structure of human class III ADH was reported (13), as well as the
structure of a cod liver ADH isoenzyme (14). Thus, an increasingly
diverse structural data base exists from which information concerning
the determinants of substrate recognition can be obtained by comparing
the structures and kinetic properties of ADH isoenzymes. Important
amino acids within the substrate-binding site directly affect the
substrate specificity of the human ADH isoenzymes (Table I). For instance, 1
1 ADH, which has a Ser
at residue 48, is the only human isoenzyme able to bind and oxidize
3
-hydroxysteroids (15). Amino acid substitutions within the loops
comprised of residues 55-61 and of residues 113-121 in
ADH
cause these loops to adopt new conformations and contribute to the
enzyme's inability to be saturated with ethanol (13). Mutagenesis
studies on the
1
1 isoenzyme indicate that
residues 93 and 94 account for the increased catalytic efficiency
toward secondary alcohols exhibited by
, relative to
1
1 (16). Residue 116, located at the
entrance to the substrate pocket, also affects the
Km for alcohols by acting as a bottleneck (17).
|
Among the human class I enzymes, 1
1
exhibits the lowest Km for ethanol (18), 120- and
7-fold lower than
and
1
1,
respectively. From ethanol to hexanol, the catalytic efficiency (Vmax/Km) of the
isoenzyme increases 400-fold, while it increases just slightly more
than 2-fold for the
1
1 isoenzyme (18).
Compared with class I
1
1 ADH, class IV
ADH also exhibits very different substrate binding
characteristics. The Km for ethanol exhibited by
ADH is 215-fold higher than that for hexanol (9). In addition,
ADH exhibits the highest catalytic efficiency for the oxidation
of all-trans-retinol to all-trans-retinal among
the known human ADH isoenzymes (19). The production of all-trans-retinal from retinol is thought to be the
rate-controlling step for the production of
all-trans-retinoic acid (20), an important regulator of gene
expression during embryonic development (21). The pathway of retinoic
acid biosynthesis involves retinoid-binding proteins, which may provide
a mechanism to discriminate specific dehydrogenases from nonspecific
dehydrogenases (22). Retinol, in vivo, is bound to the
cellular retinol-binding protein (CRBP). Evidence has been presented
showing that holo-CRBP serves as substrate for microsomal
dehydrogenases (22) and that CRBP may then transfer retinal to the
cytosolic retinal dehydrogenase for oxidation to retinoic acid.
Complete dependence on the CRBP pathway for retinoic acid production
may deny accessibility of retinol to
ADH in vivo.
However, retinoic acid synthesis during embryogenesis was reported to
correlate spatiotemporally with the expression of class IV ADH gene
(23). It was proposed that competitive inhibition by ethanol consumed
during pregnancy can reduce retinoic acid synthesis and may contribute
to the development of fetal alcohol syndrome (24, 25).
In this paper, we examine the structural basis for substrate
recognition in ADH through x-ray crystallography and
site-directed mutagenesis. By comparing the structures of the known
human ADH substrate-binding sites, it may be possible to gain a more
complete understanding of their roles in the metabolism of endogenous
and exogenous alcohols.
The cDNA for
the subunit (5) in M13 was subcloned into the vector pKK223-3
(Pharmacia Biotech Inc.) by site-directed mutagenesis using a
commercial kit (Amersham Corp.) and expressed in Escherichia
coli as described for
1
1 ADH (10).
The lysate was first mixed with DEAE-cellulose (Whatman, Maidstone, UK)
in 50 mM Tris, pH 8.8, at 4 °C, 1 mM
benzamidine, 2 mM dithiothreitol. The unbound proteins were
eluted in a batch procedure and then were buffer exchanged into 7 mM sodium phosphate, pH 6.4, 1 mM DTT using the
Minitan apparatus (Millipore, Bedford, MA) and loaded onto a 5- × 15-cm S-Sepharose column. The protein was eluted with a linear sodium
phosphate gradient from 7 to 65 mM. The enzyme was dialyzed
into 10 mM sodium phosphate, pH 6.4, 1 mM DTT
and applied to a 2.5- × 10-cm Affi-Gel Blue column (Bio-Rad). The enzyme was then eluted with a linear gradient from 10 mM
sodium phosphate, pH 6.4, to 100 mM Tris, pH 8.8, 1 mM DTT. The purified
ADH was dialyzed into 10 mM HEPES, pH 7.0, 1 mM DTT and concentrated with a Microcon 30 concentrator (Amicon, Beverly, MA) before
crystallization. The sitting drop method was employed to crystallize
the protein. Typically 2 µl of an 8 mg/ml
ADH solution was
mixed with 2 µl of the precipitant solution in the drop. The
optimized crystallization conditions for
ADH complexed with
NAD+ were 100 mM cacodylate, pH 6.5, 50-100
mM zinc acetate, 7.5 mM NAD+, and
18% (w/v) polyethylene glycol 6000. The crystals formed as flat
parallelepipeds overnight and grew to maximal size in 1 or 2 more
days.
Single-stranded
cDNA in the M13HinEco1 vector (26) was used as the template for
site-directed mutagenesis. A single oligonucleotide, 45 bases in
length, was used to mutate residues Phe-309 and Cys-317 to Leu and Ala,
respectively. Following identification of the correct mutant clone by
DNA sequencing, the mutant cDNA was subcloned into pKK223-3 and
completely sequenced, prior to expression, to ensure that no unwanted
mutations were present in the cDNA sequence. The mutant enzyme,
309L317A, was expressed and purified using the same procedure as
described for the wild-type enzyme. The kinetic measurements were
evaluated at 25 °C in 100 mM sodium phosphate, pH 7.5, on a Beckman DU-640 spectrophotometer. Enzyme activity was monitored by
following the production of NADH at 340 nm using an extinction
coefficient of 6.22 mM
1 cm
1.
Vmax values were converted to turnover numbers
assuming a molecular mass of 40 kDa per subunit. The
Km values for substrates were determined at a fixed
NAD+ concentration of 2.5 mM, except those for
1-butanol which were determined both at 2.5 mM
NAD+ and by co-variation of NAD+ and 1-butanol.
All kinetic experiments were evaluated using the kinetic programs of
Cleland (27). All reported values are expressed as the means of at
least three separate experiments with their associated standard
deviations.
X-ray diffraction data were collected to 3 Å. Higher resolution data were observed initially (2.6 Å), but severe radiation decay and the inability to flash-cool these crystals prevented collection of the higher resolution data. Four crystals (approximate dimensions, 0.3 × 0.15 × 0.07 mm3) were used to collect the native data set at room temperature on a Rigaku 200HB rotating anode generator equipped with an RAXIS IIC image plate area detector with a crystal-to-detector distance of 145 mm. The data collection statistics are listed in Table II. All crystals exhibited radiation decay and were replaced every 12 h. The diffraction data were indexed, merged, and scaled using the RAXIS IIC data processing software (38).
|
The
structure was solved by molecular replacement using the program package
AMoRe (28) and the data between 15.0 and 4.0 Å. The
1
1 ADH ternary complex dimer with
NAD+ and the inhibitor 4-iodopyrazole (Protein Data Bank
code 1DEH (13)) served as the search model for these calculations. The correlation coefficients for the top two rotation solutions were 25.7 and 19.7, respectively. After the positions for the two dimers in the
asymmetric unit were found, the starting model possessed a correlation
coefficient of 54.7 and an R value of 39.9%. All subsequent
model refinement was performed using the program package X-PLOR
(version 3.1) (29). Rigid-body refinement of the initial model
structure with the data between 8.0 and 3.5 Å brought
Rwork from 40.2 to 37.7% and
Rfree from 40.5 to 37.3%. The atomic positions were refined to 3 Å using the positional refinement protocol in X-PLOR
(30) and an overall temperature factor of 25 Å2. The
resulting structure was inspected using 2Fo
Fc and Fo
Fc maps in CHAIN (31). Amino acid substitutions were
introduced as their positions were identified during refinement.
Additional solvent zinc cations and solvent acetate molecules were
added when strong positive Fo
Fc electron density indicated their presence. In the last refinement procedure, an overall temperature factor for each subunit was refined, and the non-crystallographic symmetry restraints were removed. The final model possesses an average r.m.s.d. of 0.2 Å for the main chain atoms in the four subunits of the asymmetric unit.
C
alignments between
and
1
1
isoenzymes were performed using LSQKAB (32) in CCP4 (1994) suite and
displayed using QUANTA (Molecular Simulations Inc.,
Burlington, MA).
NAD+, grade I and DTT were purchased from Boehringer Mannhein, and PEG 6000 was purchased from Hampton Research; ethanol was purchased from Midwest Grain (Pekin, IL). All other reagents were from Sigma and were of the highest grade available.
The structure of the human class IV,
or , ADH isoenzyme was solved to 3.0 Å by molecular replacement
using the 2.2-Å structure of the class I human
1
1 isoenzyme (12) as the starting model. The final refined structure possesses an Rwork
of 22.5% with an Rfree of 30.5% (Table II).
The stereochemistry of this model was inspected using the program
package PROCHECK (33). The Ramachadram plot showed that 98.6%, or
1471, of the 1492 residues were in the preferred and allowed regions,
and 1.4%, or 21, of the residues were in the generously allowed
region. No non-glycine residues were found in the disallowed region.
Due to the presence of zinc acetate in the mother liquor, 8 solvent
zinc cations and 10 acetate ions were identified as bound to the
enzyme. There was an acetate ion present in the substrate pocket in all
four subunits.
A ADH mutant,
309L317A, was prepared by site-directed mutagenesis. Two residues in
the
isoenzyme, Phe-309 and Cys-317, were mutated to Leu and Ala,
respectively. The choice of these two positions for mutagenesis was
based on their unique characteristics compared with class I enzymes
(Table I). Residue 309 is in the substrate-binding pocket, and residue
317 is behind the nicotinamide ring of NAD+. The substrate
specificity of this mutant was studied and compared with wild-type
ADH (Table III). Mutations at these two residues dramatically increase the Km values toward small
alcohol substrates. For instance, the mutant enzyme exhibits a
Km for ethanol that is 100-fold higher than the
wild-type
ADH. Interestingly, the Km values
for substrates with five or more carbons are less affected. The
Km values for hexanol and retinol are essentially
unaffected by these mutations. As the chain length for straight
alcohols increases, the
Vmax/Km of the mutant enzyme
increases to a greater extent than does the wild-type enzyme. The
Km value for NAD+ was approximately 2 times higher than the wild-type enzyme, whereas the
Ki(NAD+) value obtained from experiments
varying both 1-butanol and NAD+ concentrations was
identical with the wild-type enzyme (0.75 ± 0.03 mM).
|
An alignment
of the C atoms, excluding residues 113-120 and 244-262, in the
dimeric
and
1
1 isoenzymes gives a
r.m.s.d. of 0.60 Å. Alignment of the individual domains within each
subunit yields similar results, with r.m.s.d. values of 0.49 Å for the catalytic domain and 0.61 Å for the coenzyme-binding domain. C
alignment of
ADH with horse and other human ADHs shows that its
structure is most similar to the human and horse liver class I ADH
isoenzymes. Unlike the recently reported structures of the human class
III
isoenzyme (13) and the ADH isoenzyme from cod liver (14),
both of which exhibited semi-open domain structures, the human
isoenzyme exhibits a fully closed conformation of the catalytic and
coenzyme-binding domains when NAD(H) is bound. The alignments reveal
that, relative to
1, there are two major structural
differences in each domain of the
subunit (Fig. 1). In the coenzyme binding domain, the largest difference occurs at the C
terminus of an
-helix comprised of residues 251-258 and the
following turn. This structural change results from the substitutions
of the two Gly residues at positions 260 and 261 in
1
1 by two Asn residues in
ADH. The
main chain conformations of Gly-260 and -261 in the
1
1 isoenzyme are incompatible with the
presence of side chains, thus the main chain path must shift dramatically to accommodate Asn residues at positions 260 and 261. Relative to
1
1 ADH, the C
atoms of
residues 244-262 in
ADH shift, on average, by 1.8 Å, with a
maximum of 4.8 Å at residue 261. The other structural difference in
the coenzyme-binding domain involves residues 297-309, which are
located at the dimer interface and within the substrate-binding pocket
of the neighboring subunit. One structural difference in the catalytic
domain involves residues 17-25 and is not likely to affect enzymatic
activity or subunit interactions. The other structural difference
within the catalytic domain is due to the deletion of residue 117, which shortens the loop at the entrance to the substrate-binding pocket in
ADH. C
alignment between
and
, with each
domain aligned separately and then the results combined, reveals that
in addition to those differences between
and
1
1,
and
differ
substantially at both N and C termini (Fig. 1). Moreover, great
differences exist at residues 55-61 and 112-120. Both regions adopt
new conformations in
ADH and contribute to a larger active site
in
ADH (13).
Our structural comparisons also reveal that there is no evidence that
the interaction between coenzyme and residue 223 is weakened in
ADH, as was suggested by a modeling study (34). In fact, the
hydrogen-bonding distances between adenosine ribose oxygens and
oxygens of Asp-223 are within the range of 2.6-2.7 Å in both
and
1
1 ADHs.
The alcohol binding pocket is an extension of the coenzyme binding
site (4) and is fully formed only after coenzyme binding has occurred.
In ADH, the substrate pocket is a cylinder having dimensions of
approximately 16 by 7 by 6 Å. The substrate specificity for this
enzyme is determined by surface complementarity between the enzyme and
the substrates throughout this cylinder. Changes in the
Km values are related to the effective concentration of the ES complex. Mutations can affect
Km by changing the ratio of productive
versus non-productive encounters with the enzyme. In ADH,
these changes are brought about either through steric exclusion
(preventing productive binding), as seen for the binding of secondary
alcohols to
1
1 ADH (16), or by changing the number of non-productive conformations permitted by altering the
accessible volume of the active site (13). The inner part of the
alcohol site (near the catalytic zinc) includes residues 48 and 93 and
the nicotinamide ring of NAD+.
ADH has a
Km value for ethanol that is 560-fold higher than
1
1 ADH (Table III). One possible cause
for this difference may be the substitution of Cys for Ala-317 near the
nicotinamide ring. To accommodate its longer side chain in
ADH,
the main chain atoms of residue 317 move ~1 Å away from Thr-186,
toward the carboxamide group relative to
1
1 ADH (Fig. 2). To avoid unfavorable contacts with the Cys-317 carbonyl oxygen, the plane of
nicotinamide ring appears to twist in
ADH, relative to its position in the
1
1 isoenzyme (Figs. 2 and
3), This twist creates more space between the nicotinamide ring and the
catalytic zinc in
ADH (Fig. 3). In addition to
these changes, Thr-48 and Phe-93 also shift away from the catalytic
zinc. The distance between the C
atoms of these two residues is 0.9 Å longer in
ADH. Consequently, smaller substrates, such as
ethanol, are not as conformationally constrained in this active site as
in the
1
1 ADH. Thus, a higher
concentration of ethanol is required to produce an equivalent number of
productively bound conformations.
The structural differences near the catalytic zinc in these two
isoenzymes may also explain the weak binding of the inhibitor pyrazole
to ADH (Ki values of 0.60 µM
for
1
1 and 350 µM for
at pH 7.5 (5, 12)). Pyrazole and its 4-substituted derivatives competitively inhibit the binding of alcohol
substrates through the formation of a tight
enzyme·NAD+·inhibitor complex (35), in which pyrazole
nitrogens interact with both zinc and NAD+. We speculate
the bond between the pyrazole nitrogen atom and the C-4 atom on the
nicotinamide ring may be distorted due to the twist of the nicotinamide
ring in
ADH. In fact, if the active sites of the
and
1
1 structures are aligned and the position of 4-iodopyrazole in the
1
1
active site structure is used to examine the geometric constraints on
pyrazole binding to the
structure, the corresponding angle
between the C-4-N-1 and the N-1-N-2 bond is 133°, while it is close
to 120° in the
1
1 structure,
corresponding to a low energy, stable complex. This angular difference
would undoubtedly represent a higher energy conformation and could
account for up to 2.7 kcal/mol of the observed difference (3.8 kcal/mol) using a harmonic potential with a force constant of 0.27 kcal/(mol·degree) (36). In addition, the increased distances between
the N-1 of pyrazole and the C-4 of the nicotinamide ring (by 0.3 Å)
and between residues 48 and 93, where pyrazole is held, could
contribute to lowering the affinity for 4-methylpyrazole.
The middle region of the substrate pocket, which plays an important
role in the interactions with the aliphatic tail of longer substrates,
such as butanol and pentanol, includes residues 57, 141, 294, and 309. Like many other ADH isoenzymes, ADH exhibits Km values for primary straight chain alcohols, which decrease with increasing chain length, whereas the
Vmax values remain relatively constant. Thus,
the catalytic efficiency
(Vmax/Km) increases with
increasing chain length. For example, the
Vmax/Km for hexanol is
138-fold higher than that for ethanol in
ADH (Table III). In
contrast, the Vmax/Km values
for
1
1 ADH vary only 2- to 3-fold for
substrates from ethanol to hexanol. These different characteristics can
be explained by differences in the amino acids within the middle region
of the substrate pocket. The key amino acids within this region are
residues 57, 141, and 309.
1
1 ADH
possesses Leu residues at all these positions, and their side chains do
not appear to create new productive interactions as substrates get
longer (Table III). The presence of Phe at position 309, Met at
position 141, and Met at position 57, in
ADH, narrows the middle
region of the substrate pocket compared with
1
1 (Fig. 3). The twist on the
nicotinamide ring also contributes to a shift in Phe-309, to avoid
unfavorable contact with NAD(H), which further narrows the channel
leading to the catalytic zinc. Although this narrowing does not appear
to directly aid the binding of ethanol, it can explain the decreased
Km values for propanol, butanol, and pentanol in the
isoenzyme relative to
1
1 ADH.
Our modeling indicated that the side chain of residues 57 and 309 would
interact with the substrates at the carbon 4, 5, and 6 positions,
whereas the side chain of residue 141 interacts with carbon 3 and 4. In
the 309L317A mutant, the Km values for propanol,
butanol, and pentanol are increased by 240-, 60-, and 7-fold,
confirming the role of residues 309 and 317 in stabilizing the binding
of these substrates. Moreover, the increase in
Vmax/Km versus the
chain length of the substrates is greater for the mutant than for the
wild-type enzyme. This behavior in the mutant enzyme is due to a much
lower Vmax/Km for small
substrates, since the Vmax/Km
values for hexanol approach those of the wild-type enzyme. The
substitutions in the mutant enzyme thus appear to further widen the
substrate-binding site, relative to
ADH, resulting in a greater
number of permissible non-productive ES complexes. These
changes in the mutant only affect substrates where binding is dependent
on the local topology, such as ethanol, but do not significantly affect
the catalytic efficiency toward hexanol or retinol.
The outer part of the substrate-binding pocket is exposed to solvent
and includes the loop comprised of residues 114-120 and residue 306 from the other subunit within the dimer. The deletion of residue 117 in
ADH shortens the loop comprised of residues 114-120 (Fig. 3)
and widens the entrance to the substrate-binding site. In
1
1 ADH, Leu-116 appears to function as a
door, opening to allow substrates in or out, but then closing to help
keep in bound substrate (17). Consistent with this function, its side chain was found to occupy different conformations in binary and ternary
complexes (10-12). In
ADH, the shift in the position of residue
116 due to the deletion of residue 117 does not permit its side chain
to function in this manner, leaving an open substrate-binding site
(Fig. 3). Widening of the bottleneck at position 116 in
1
1 ADH by mutagenesis dramatically
increased the apparent Km values for primary and
secondary alcohols (17). Consistent with these observations,
ADH
has higher Km values than the
1
1 isoenzyme for straight chain alcohols
and very poor efficiency toward all secondary alcohols (5). The
Km for hexanol and retinol are virtually identical
in the wild-type and mutant enzymes. With the enlarged entrance to the
substrate pocket in the
isoenzyme, the structure of the middle
and inner regions of the substrate pocket would seem to be best suited
for the oxidation of long chain aliphatic alcohols, such as
-hydroxy
fatty acids, farnesyl alcohols, and retinol. To examine long chain
alcohol binding, all-trans-retinol was docked into the
active site using program AUTODOCK (37). The results of this simulation
confirm our previous results based on modeling studies (5), the
-ionone ring of retinol binds at the widened entrance of the
substrate-binding pocket, such that an extended conformation of retinol
can be adopted. Thus, the ability to bind retinol in a more extended
and, presumably, lower energy conformation in
ADH could account
for its higher catalytic efficiency.
The atomic coordinates and structure factors (code 1AGN) have been deposited in the Protein Data Bank, Brookhaven National Laboratory, Upton, NY.