Evolutionary selection of enzymatically synthesized

semiconductors from biomimetic

mineralization vesicles
Lukmaan A. Bawazer
a,b,1
, Michi Izumi
b,2
, Dmitriy Kolodin
c,3
, James R. Neilson
a,b,4
, Birgit Schwenzer
b,5
,
and Daniel E. Morse
a,b,c,6
a
Interdepartmental Graduate Program in Biomolecular Science and Engineering,
b
Institute for Collaborative Biotechnologies, California NanoSystems Institute,
and Materials Research Laboratory, and
c
Department of Molecular, Cellular, and Developmental Biology, University of California, Santa Barbara, CA 93106
Edited by David A. Tirrell, California Institute of Technology, Pasadena, CA, and approved April 25, 2012 (received for review October 17, 2011)
The way nature evolves and sculpts materials using proteins inspires
new approaches to materials engineering but is still not completely
understood. Here, we present a cell-free synthetic biological plat-
form to advance studies of biologically synthesized solid-state
materials. This platform is capable of simultaneously exerting many
of the hierarchical levels of control found in natural biomineraliza-
tion, including genetic, chemical, spatial, structural, and morpholog-
ical control, while supporting the evolutionary selection of new
mineralizing proteins and the corresponding genetically encoded
materials that they produce. DNA-directed protein expression and
enzymatic mineralization occur on polystyrene microbeads in water-
in-oil emulsions, yielding synthetic surrogates of biomineralizing
cells that are then screened by ow sorting, with light-scattering
signals used to sort the resulting mineralized composites differen-
tially. We demonstrate the utility of this platform by evolutionarily
selecting newly identied silicateins, biomineralizing enzymes pre-
viously identied from the silica skeleton of a marine sponge, for
enzyme variants capable of synthesizing silicon dioxide (silica) or
titanium dioxide (titania) composites. Mineral composites of in-
termediate strength are preferentially selected to remain intact for
identication during cell sorting, and then to collapse postsorting to
expose the encoding genes for enzymatic DNA amplication. Some
of the newly selected silicatein variants catalyze the formation of
crystalline silicates, whereas the parent silicateins lack this ability.
The demonstrated bioengineered route to previously undescribed
materials introduces in vitro enzyme selection as a viable strategy
for mimicking genetic evolution of materials as it occurs in nature.
directed evolution
|
in vitro compartmentalization
|
DNA shufing
|
metal
oxide nanoparticles
|
self-assembly
A
lthough often treated as conceptually distinct, organic and
inorganic systems have always been intimately connected in
biology. This vital relationship likely originated at reactive min-
eral surfaces, where many believe that lifes rst molecules were
formed (1), and persisted through the emergence of ancient
unicellular organisms able to process or synthesize minerals (2).
The diverse evolutionary products of those initial simple systems,
such as bones, teeth, and shells, serve as inspiration for engineers
interested in developing new and environmentally benign routes
for producing technological materials (38). Bioinspired materi-
als engineering also offers the prospects of both combinatorial
searches for optimizing material performance (9) and genetic
control over biomolecular-mineral interfaces (6, 8, 10, 11). These
prospects and recent research suggest that biology and minerals
are poised to form new intimate connections now in the realm of
human technology. Progress toward this hybrid outcome is high-
lighted by the recent use of semiconductor arrays to sequence a
human genome directly and ultrarapidly (12) and by the fabrica-
tion of high-performing batteries that were genetically encoded
and templated by viruses (11).
Despite these developments, a challenging divide persists be-
tween biological and materials technologies, and the potential
for use of biomacromolecules as tools to control the synthesis of
solid systems, as exemplied in the precision and selectivity of
biomineralization, remains to be fully realized. In contrast to the
wide diversity of biomolecules required for natural biomineraliza-
tion (13), only short stretches of polypeptides (6, 8, 10) or poly-
nucleic acids (14) have been systematically searched by genetic
screening for dened mineral interactions in vitro. In contrast to
load-bearing (4, 15) or other material properties (16, 17) that are
directly selected in nature, laboratory screening has focused on
either simple mineral binding (6, 8, 10) or precipitation (14) as the
primary selection criterion. The challenge of engineering in-
creasingly complex biomolecular behavior is critically dependent
on the development of laboratory platforms in which genetic in-
formation can be expressed and analyzed in new ways. This fact has
been evident in the progress of enzyme engineering (18, 19), in
which emulsion-based cell-free systems have been used as surro-
gates of living cells to overcome the limitations associated with
cellular protein expression (2022). This strategy of in vitro
compartmentalization, as it is called, allows excessive background
activity (from thousands of cellular proteins) to be avoided;
expands opportunities to conduct chemistry that may be toxic or
not well supported in cellular environments; and supports the
screening of large protein libraries [>10
7
variants (2022)] via ow
sorting, a technique by which the articial cells are owed in single
le at high-throughput (>10,000 per second) past one or more
laser-based detectors. Protein variants synthesized in specic cell
surrogates that produce dened uorescence or light-scattering
signals (as an assay for the proteins activity) can be physically
sorted into separate channels, after which the encoding genes can
be isolated for clonal propagation and further study.
Here, we combine lessons learned from silica biomineraliza-
tion in marine sponges (2332) with emulsion-based synthetic
cell surrogates (2022) to select variants of mineralizing enzymes
Author contributions: L.A.B. and D.E.M. designed research; L.A.B., M.I., D.K., and B.S.
performed research; L.A.B., M.I., J.R.N., and D.E.M. analyzed data; and L.A.B. and
D.E.M. wrote the paper.
The authors declare no conict of interest.
This article is a PNAS Direct Submission.
Freely available online through the PNAS open access option.
1
Present address: Faculty of Mathematical and Physical Sciences, School of Chemistry,
University of Leeds, Leeds LS2 9JT, UK.
2
Present address: Pioneer Hi-Bred, Hayward, CA 94545.
3
Present address: Department of Microbiology and Immunobiology, Harvard Medical
School, Boston, MA 02115.
4
Present address: Department of Chemistry, The Johns Hopkins University, Baltimore,
MD 21218.
5
Present address: Pacic Northwest National Laboratory, Richland, WA 99354.
6
To whom correspondence should be addressed. E-mail: d_morse@lifesci.ucsb.edu.
See Author Summary on page 10142 (volume 109, number 26).
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1116958109/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1116958109 PNAS | Published online June 7, 2012 | E1705E1714
B
I
O
C
H
E
M
I
S
T
R
Y
P
N
A
S
P
L
U
S
evolutionarily in vitro, thus advancing capabilities for studying
genetically encoded inorganic materials. The cell-free enzyme
reengineering method we describe is more robust for targeting
a wide range of materials than cellular systems that are susceptible
to poisoning or damage by certain mineral precursors or the
product minerals (24). Silica-synthesizing biomimetic vesicles are
rst investigated with recombinant silicatein , an enzyme that
catalyzes and templates formation of marine sponge biosilica in
vivo. The vesicles are then used evolutionarily to select unique
silicatein variants from a combinatorially altered gene library, with
reactions targeting either silicon dioxide or titanium dioxide syn-
thesis. In the particular version of in vitro compartmentalization
used here (20), the silicatein genes and their product enzymes are
displayed and reacted on the surfaces of polystyrene microbeads.
This allows us to mimic marine sponge skeletal growth in several
ways (Fig. 1), including the use of a silicatein-based scaffold with
an enzymatically active surface as a catalytic template for miner-
alization (20, 23, 25, 30, 32) (Fig. 1A) and the ability to control
reactant species within micrometer-scale compartmentalized
vesicles (26, 28, 31). Because the ampliable genes, their encoded
silicatein-based catalytic template, and the resulting synthesized
minerals all are integrated within the same vesicle-based cell
surrogate, evolutionary selection works on these elements as it
does in living cells. We show that environmentally selected prop-
erties of the mineral and its synthesis system can thus emulate the
evolutionary selection of biominerals as it occurs in nature.
Results
In Vitro Silicatein Expression. Creation of biomimetic vesicles begins
by linking silicatein genes and an antibody directed against a de-
ned peptide tag sequence to polystyrene microbeads (with
a diameter of 0.97 m). The silicatein genes consist of silicatein-
encoding recombinant DNA, whereas the antibody molecules
ultimately act as capture agents for the silicatein proteins that will
be expressed from the DNA. The DNA molecules are anchored to
the beads via their 5-end, leaving the information-encoding length
of nucleic acids accessible to enzymes that can transcribe or rep-
licate the gene. The concentration of DNA in the linkage reaction
is diluted to favor the linking of less than one DNA molecule per
bead. This DNA-to-bead ratio promotes the attachment of unique
DNA molecules (from a mutagenized library of millions of DNA
molecules encoding variant silicateins) to the individual microbe-
ads. This ratio was also used when a homogeneous DNA pop-
ulation encoding parent silicatein was rst used to test the
biomimetic system. Once linked with DNA and coated with anti-
bodies, the beads are immersed in an aqueous solution, including
a suite of biomolecular machinery extracted frombacteria (EcoPro
T7 Mix; Novagen) that enzymatically transcribes the DNA on the
beads to the corresponding mRNA and translates this mRNA to
synthesize multiple copies of the uniquely encoded silicatein pro-
tein (either parental or variant, depending on the DNA sequence).
Genetic modications of the original DNA template ensure that
the expressed silicateins include a unique polypeptide sequence
(i.e., a tag) at their N termini as a handle, and are thus captured by
antitag antibodies that were precoated on the microbead surface. In
a key step ensuring that proteins expressed from a given bead-gene
source become attached to that same bead (instead of diffusing to
bind on another bead), the mixture of DNA-linked beads and
bacterial extracts is rst emulsied in oil, compartmentalizing each
bead-gene to its own oil-bound aqueous vesicle, which serves as
a surrogate of a living cell. The protein expressionreactions yielding
silicatein-coated beads occur simultaneously within the population
of millions of vesicles contained in the emulsion. The emulsion is
then broken by centrifugation, releasing the microbeads that each
contain linked to its surface, its display of silicatein proteins, and
their corresponding encoding gene. Details of this protein expres-
sion method have been described previously by Grifths and Tawk
(20) and Miller et al. (21), and are summarized in Materials and
Methods. In the present study, the presence of recombinant silica-
tein on the bead surfaces was successfully conrmed by ow
cytometry analysis using a silicatein-specic polypeptide (24) as the
uorescent probe (Fig. S1). Antibody-onlycoated beads were used
for all mineralization control reactions in this study.
Biomimetic Mineralization Reactions and Flow-Sorting Analysis. To
form biomimetic mineralization vesicles, the silicatein-displaying
beads (after transcription and translation, as described immediately
above) were reemulsied to form a second unique and distinct
water-in-oil emulsion that included chemical precursors for mineral
synthesis (Fig. 1C), either 100 mM tetraethylorthosilicate [TEOS;
Fig. 1. Overview of the biomimetic mineralization platform used in this study. (A) SEM images of biosilica spicules (Left) and occluded polymerized silicatein
laments (Right) isolated from the marine sponge T. aurantia. [Reproduced with permission from ref. 29 (Copyright, National Academy of Sciences of the
United States)]. (B) TEM images of silicate microstructures formed from biomimetic vesicles, revealing intact bead-silicate composite (Left) and a polystyrene
(PS) microbead scaffold released from the fracture of its silicate shell (Right). (C) Polystyrene microbeads coated with silicatein [by in vitro compartmen-
talization (20)] are reacted with small metal-containing precursors in water-in-oil emulsions, yielding mineral composites such as those shown in B, which are
then isolated from nonmineralized polymer beads by ow sorting, with light scattering used to identify large beads for sorting. (D) Schematic summary of the
gene library used for evolutionary selection experiments. Recombinant genes for two natural isoforms of silicatein were digested and reassembled by DNA
shufing (Materials and Methods) to produce a chimeric library of variant silicateins, which were then screened to identify previously undescribed miner-
alizing variants via the process summarized in C.
E1706 | www.pnas.org/cgi/doi/10.1073/pnas.1116958109 Bawazer et al.
as a precursor for silica (23)] or 100 mM titanium bisammonium
lactatodihydroxide [TiBALDH; as a precursor for titania (30)], with
the emulsions aqueous phase buffered at pH 7.4. After incubation
to allow mineral synthesis (Materials and Methods), the emulsions
were broken and the beads were recovered by centrifugation, as
described previously (20, 21). Beads that support mineralization
from their surfaces and remain intact through centrifugal bead
recovery steps can potentially be detected by ow sorting (using
a Becton Dickinson FACSAria ow sorter) through differential
light-scattering or uorescence signals emanating from the miner-
alized beads. However, the solid oxides targeted in this study do not
exhibit intrinsic photoluminescence. Thus, without functionalizing
these materials with a uorophore, only light scattering can be used
as a trigger for differential bead sorting. In this study, ow sorting
has been used as an assay for mineralization performance. There-
fore, to ensure that light-scattering signals were a reliable indicator
of mineral growth, we tested our biomimetic system using both
uorescence and scattering analysis, combined with EM charac-
terization of the sorted beads, after using initial reaction conditions
that promoted formation of uorescently labeled silica. This was
achieved by using a polypeptide (Si4, MSPHPHPRHHHT) that
has been previously demonstrated to accelerate silica precipitation
fromsilicic acid (33). Fluorescein-labeled Si4 (Flc-Si4) was included
in biomimetic vesicles with silicatein -displaying beads reacted
with TEOS, such that Si4 occluded in the mineralized product
would uorescently label the composite beads.
During ow sorting, beads recovered from an emulsion min-
eralization reaction are hydrodynamically guided past the instru-
ments laser and interrogated at a dened excitation wavelength
(488 nm). A representative population of 100,000 beads is mea-
sured; emission intensities through dened band-pass lters, as well
as light-scattering intensities, are acquired from each bead; and the
results from all 100,000 measured events are plotted by the in-
strument. Showing these plots as 2D histograms as in Fig. 2, where
the intensities through different acquisition channels are shown on
respective axes, provides a useful method for representing the data
and readily identifying subpopulations of interest (34). On gener-
ating these intensity histograms from a control population (i.e.,
reacted beads lacking silicatein on their surfaces), gates are
drawn within the plots to isolate regions of interest. The limits of
each gate are xed arbitrarily and dened by the maximum in-
tensities exhibited by the control beads. Data from the experi-
mental samples are analyzed using the same gates as established for
the control. The instrument calculates the fraction of all (100,000)
events from the control and experimental samples, respectively,
that are binned within each gate. For example, the gate shown in
Fig. 2A, (Fluorescence) was set against 0.2% of the control beads
(i.e., at an intensity limit beneath which 99.8% of the control beads
uoresced; Fig. 2A, Left). By contrast, as indicated in Fig. 2A
(Right), 22.1%of the reacted silicatein bead population fell within
this same gate (Fluorescence), showing that a signicantly higher
amount of Flc-Si4 was associated with the silicatein -generated
Fig. 2. Flow-sorting analysis (from a FACSAria instrument) and TEM characterization of sorted protein-coated beads after reaction with the silica precursor
TEOS in biomimetic vesicles in the presence of Flc-Si4 (main text). (A) 2D histogram maps show the distribution of uorescence emission intensities (integrated
over the indicated band-pass wavelengths) from reacted antibody-coated control beads (Left) or silicatein -coated beads (Right). Percentages of Fluo-
rescence gated beads are shown (green). (B) Light-scattering histograms from the same bead samples as in A. Percentages of Size gated events are shown
(red). (C) Gated subpopulations representing the 5% most highly uorescing beads from the reacted silicatein -coated beads. One thousand beads were
sorted from each subpopulation. (D) Light-scattering distributions from the subpopulations of C. (E) Distribution of average diameters of mineral composite
structures sorted in C, as measured by TEM (representative TEM images are shown in Fig. S2). a.u., arbitrary units.
Bawazer et al. PNAS | Published online June 7, 2012 | E1707
B
I
O
C
H
E
M
I
S
T
R
Y
P
N
A
S
P
L
U
S
reaction products vs. the control products. Similarly, the gate Size
in the light-scattering plots of Fig. 2B was set against the 0.2%most
intensely scattering beads from the control population.
The subpopulations gated in one plot (e.g., Fluorescence) can
be highlighted as intensity distributions in another plot (e.g.,
a light-scattering histogram) by color-coded mapping. In this way,
Fig. 2 A and B shows that the largest beads (gated in Size and
mapped red) were also the most uorescent (gated in Fluores-
cence and mapped green). Further, when four additional gates
were applied that binned intensely uorescent subpopulations
(Fig. 2C), it was seen that intensity increases in light-scattering
signatures correlated positively with uorescence intensities (Fig.
2 Cand D). These results indicated that the largest beads were the
most uorescent as a result of probe incorporation into the en-
zymatically synthesized mineral structures.
Analysis of Sorted Silicatein -Catalyzed Silicate Composites. The
ow sorter can be set to sort beads whose intensities fall within
a specied gate. We ow-sorted beads through the Fluorescence
subgates 1, 2, 3, and 4 of Fig. 2C to examine the re-
lationship between uorescence and scattering within the pop-
ulation further, and to investigate the mineralized materials
produced from this biomimetic system. One thousand beads were
sorted through each subgate into respective collection tubes,
washed, and evaporated onto grids for analysis by transmission
electron microscopy (TEM). Measured sizes of intact silicate
structures conrmed that both uorescence intensity (Fig. 2C) and
light-scattering signatures (Fig. 2D) positively correlated with bead
size (Fig. 2E). TEM images of representative beads sorted through
respective gates (Fig. S2) suggest that mineralization is initiated at
the bead surface and that the resultant pseudospherical composite
morphology (e.g., Fig. 1B) is inuenced by the geometry of the
biomimetic mineralization vesicle. Electron-dispersive spectros-
copy (EDS) of several different mineral-coated beads conrmed
the presence of silicon and oxygen while also revealing a hetero-
geneous distribution of additional metals, including Al, Ca, Cl, K,
Fe, and Na (Fig. S2). The various reagents used for transcription,
translation, and mineralization likely were the source of these ad-
ditional metals. [Reaction components included mineral oil
(Sigma), aqueous buffers, commercially synthesized Flc-Si4 poly-
peptide (SigmaGenosys), commercial streptavidin-coated poly-
styrene beads (Bangs Laboratories), PCR-amplied DNA,
commercial antibody (Roche), the silica precursor TEOS (Sigma),
and a proprietary solution of bacterial extracts for in vitro tran-
scription/translation (EcoPro T7 Mix; Novagen) (Materials and
Methods).] Selected area electron diffraction (SAED) showed that
the mineralized structures were crystalline (Fig. S2); although at-
tributable to the heterogeneous chemistry within these crystalline
products, the patterns could not be unequivocally indexed to known
silica or silicate polymorphs.
The sizes of the conned reaction environments, which vary
among vesicles within the emulsion used for this study (21), can
have a signicant impact on both enzymatic activity (21) and
mineralization (35). Such variable effects, in combination with
the heterogeneous distribution of metal impurities, may con-
tribute to the compositional chemical diversity of the crystalline
phases. TEM and EDS analyses of a sorted fraction of the
TEOS-reacted antibody-only control beads revealed no Si in the
sorted materials and showed only precipitated salts (Fig. S3).
The presence of surface-displayed recombinant silicatein is thus
critical for initiating directed growth of stable, higher order sil-
icate microstructures. The sorted silicatein-containing compo-
sites exhibited crystalline mineral products both immediately
adjacent to and relatively distant from the bead surfaces (Fig.
S2). This suggests that after mineralization is enzymatically ini-
tiated at the bead surface, subsequent mineral growth can occur
through concerted mechanisms involving the TEOS precursor,
the Flc-Si4 polypeptide, the mineral surface, surfactants at the
oil-water vesicle interface, and metal and molecular additives
within the compartmentalized reaction environment. Although
elucidation of the detailed mechanisms controlling mineral for-
mation and growth under these conditions requires further in-
vestigation, this biomimetic system represents a highly tractable
laboratory model relative to living organisms for such studies
(i.e., for investigating general aspects of the concerted genetic,
chemical, spatial, and interfacial mechanisms that operate
during biomineralization).
Silicatein Gene Library Creation. A wide variety of bioengineering
approaches are available for genetically altering recombinant
DNA (1820, 36). We used two approaches that are commonly
utilized for producing diverse gene libraries for evolutionary en-
zyme selection experiments. The rst is error-prone PCR, in which
the encoded amino acid sequence of a length of DNA is randomly
mutated through errors in replication made by DNA polymerases.
The second method is DNA shufing (or molecular sex), in
which two parent genes encoding related proteins with similar
functions and shared sequence identity are randomly recombined
with one another (by enzymatic scission and rejoining), yielding
gene chimeras as progeny. Each progeny DNA molecule has the
same overall length as the parents but randomly recombined
segments of genetic information within its gene sequence (Fig.
1D). The silicateins and from the marine sponge Tethya aur-
antia are compatible with the generation of a shufed library be-
cause they share >70% sequence identity at the gene level (29).
The DNA shufing approach we used, which was a variation of
family shufing with ssDNA as described by Zha et al. (37) (Fig.
S4), also introduced random point mutations into the gene library.
Sequencing of 10 randomly selected variants conrmed that the
library contained silicatein genes exhibiting both gene-to-gene
crossover points and random point mutations (Fig. S5).
Genetic Screening Using Biomimetic Vesicles. The silicatein library
produced as described above yields a diverse gene pool encoding
millions of unique silicatein variants. Partitioning these genes to
different biomimetic mineralization vesicles and screening them
via the mineralization activity of their encoded proteins (as
assayed by ow sorting) allowed us to select previously unde-
scribed functional silicatein enzymes. In vitro expression for dis-
play of silicatein proteins on bead surfaces was conducted as
described above. Here, each set of DNA and corresponding pro-
teins displayed from a given microbead represented a unique sil-
icatein variant, and 2 10
8
genes were applied to beads for
screening in a single experiment (20). In initial screening attempts,
biomimetic mineralization reactions were carried out with TEOS
as the precursor and in the presence of polypeptide Flc-Si4. This
resulted in ow-sorting uorescence and light-scattering gate
enrichments for the library-displaying beads compared with the
control beads (Fig. S6), similar to the effect produced by silicatein
-displaying beads (Fig. 2). TEM analyses showed that the ma-
jority of the reacted control beads remained uncoated by mineral,
whereas the reacted silicatein library-displaying beads were em-
bedded in a matrix of inorganic material (Fig. S6). Also similar to
the results with silicatein , in some product regions, crystalline
materials were observed by SAED (Fig. S6). Collectively, these
results conrmed that the diverse gene pool, when subjected to
evolutionary selection via biomimetic reactions and ow sorting,
included functional (i.e., synthetically active) silicatein variants.
Just as gene replication is a necessary requirement for bi-
ological evolution in nature, in vitro evolutionary selection
requires that successful genes be replicated (amplied) so that
they may be identied and further studied. In bead-based in vitro
compartmentalization (20), gene replication of the DNA at-
tached to the individually selected (ow-sorted) beads is ac-
complished by PCR, in which the genes displayed from the
individual bead surfaces are enzymatically amplied. We found
E1708 | www.pnas.org/cgi/doi/10.1073/pnas.1116958109 Bawazer et al.
PCR amplication of sorted mineralizing genes to be experi-
mentally challenging as a result of the presence of occluding
mineral on the bead surfaces and the very low (femtomolar)
concentration of sorted DNA available for PCR. We attempted
a variety of PCR conditions and bead pretreatments to liberate
the mineral-occluded encoding DNA, including mechanical ag-
itation and silica etching with hydrouoric acid or NaOH. We
also explored various reaction and sorting conditions to mini-
mize mineral synthesis and coverage of the beads (e.g., reduced
reaction times) and to identify surface-exposed genes (e.g.,
sorting based on binding of uorescently labeled cDNA). Using
silica as a model system, we found the most reliable approach to
achieving signicant gate enrichment during ow sorting, while
also promoting successful gene amplication, involved conduct-
ing mineralization in the absence of the uorescently labeled
polypeptide Flc-Si4 to reduce oxide growth subsequent to initial
mineralization at the bead surfaces and then mechanically agi-
tating the sorted beads before PCR. This selection process,
which facilitates use of light scattering alone (instead of uo-
rescence) as a signal for ow sorting, allowed us to select beads
with signicant yet still disruptible mineral oxides without in-
terference from the labeled polypeptide on the mineralization
reactions. Further, these experimental constraints introduced
a unique set of environmental pressures in which the geneti-
cally encoded mineral composites were subjected to mechanical
stresses, thus experiencing at the microscale an analog of the
structural pressures that select for naturally evolved biominerals.
In parallel screening experiments, TEOS was used as the
precursor for silica-targeting reactions and TiBALDH was used
as the mineral precursor for titania-targeting reactions. TEM
analyses conrmed that Si- and Ti-containing materials were
formed under screening reaction conditions (Fig. S6). After re-
covering the beads from emulsion-based biomimetic minerali-
zation reactions and immediately before ow sorting, the beads
were physically agitated by immersion in a sonication bath (5 min)
to disrupt bead agglomerates and background salt precipitates
from control bead surfaces; this treatment (in contrast to Flc-
Si4assisted TEOS reaction conditions) was required to yield
a signicant differential light-scattering signal from the silicatein
library displaying beads relative to control beads. The majority of
the control population then appeared as bead singlets by ow-
sorting analysis (Fig. 3A and Fig. S6). To collect sorted beads for
PCR amplication, the entirety of each reacted silicatein bead
sample was subjected to ow sorting, thereby resulting in circa
600,000 TEOS-reacted and 185,000 TiBALDH-reacted beads,
respectively, sorted through the Size gate (Fig. 3A and Fig. S6).
Gate enrichment and sorting yield were lower in the titania-
targeting selection experiment vs. the silica-targeting selection
experiment. To disrupt the encasing mineral and liberate the
encoding DNA, the sorted beads were subjected to more intense
physical agitation consisting of three cycles of a 5-min sonication
bath, 1 min on a Baxter SP vortexer at maximum speed, and
5 min of centrifugation at 20,800 g. After each centrifugation
step, small particulates became increasingly visible by eye above
the beads along the length of the test tube sidewall; these par-
ticulates were removed along with the supernatant and replaced
with fresh buffer before the subsequent agitation cycle. The ag-
itated and washed beads were nally applied as gene template
carriers to a PCR after dislodged mineral particles were removed
by ltration through 0.22-m lter units (retaining the beads with
a diameter of 1 m). This nal ltration step was critical for
successful gene amplication. Amplied full-length genes (Fig.
3B) were then cloned and sequenced.
Evolved Mineralizing Genotypes. From a sample of 30 sequenced
genes from each selection procedure (targeting silica and titania,
respectively), the majority of genes were found to encode a trun-
cated polypeptide sequence that also was found present in high
numbers in the starting library (SI Text and Fig. S5). However,
within each sampled population, we found one full-length silica-
tein gene (Fig. 4A) different in character from both the parent
silicateins and the representative genes from the starting library
(Fig. S5). We name these variants screened for silica and titania
synthesis, respectively, silicatein X1 and silicatein XT. Each vari-
ant exhibits several unique DNA recombinations and point
mutations. Collectively, almost all (22 of 24) of these changes in
protein sequence involve hydroxyl residues, either directly (Fig.
4A, stars) or indirectly by occurring immediately adjacent to
a hydroxyl moiety (Fig. 4A, asterisks). We consider these changes
likely to be important because hydroxyls have been implicated
previously as surface functional groups capable of guiding mineral
oxide deposition (25, 29, 30, 38).
For Si-O-protein bond formation to occur, pentacoordinated
Si intermediates are required (23, 32, 39). This suggests a bio-
chemical role for hydroxyl-adjacent amino acids, whose impor-
Fig. 3. Mineralization performance of the silicatein library in biomimetic
vesicles. (A) Light-scattering signatures of bead populations reacted with
TEOS, after agitation by immersion in a sonication bath to disrupt bead
agglomeration. Relative population percentages of Size-gated events
(red) and Singlets-gated events (blue) are shown. The control bead pop-
ulations are disrupted into bead singlets (i.e., uncoated and non-
agglomerated polystyrene beads), whereas the silicatein-library bead
population maintains an intensely scattering subpopulation, indicating the
presence of mechanically robust silica-polymer composite structures. Similar
results were obtained from reactions using the titania precursor TiBALDH
(Fig. S6). (B) Schematic illustrations (Upper) and 1% agarose gels of
electrophoresed DNA products from PCR reactions (Lower) show that min-
eralizing enzymes that catalyze synthesis of uniform or extensively ag-
glomerated mineral coatings block gene amplication (E), whereas silicatein
variants that self-assemble and catalyze synthesis of relatively dispersed
oxide nanoparticles meet ow-sorting criteria and are also amenable to
postsorting mechanical disassembly, allowing the encoding gene to be ex-
posed for amplication (S). ref., reference DNA ladder.
Bawazer et al. PNAS | Published online June 7, 2012 | E1709
B
I
O
C
H
E
M
I
S
T
R
Y
P
N
A
S
P
L
U
S
tance is further underscored by a previous study showing that
genetic alterations adjacent to a silicon-interacting hydroxyl
critically affected silica precipitation activity in a cysteine hy-
drolase enzyme (40). Additionally, silicatein surface biochemistry
has been suggested to be critical for its self-assembly into poly-
meric laments in vivo (27, 41). Energy-minimization modeling
using cathepsin L [a structurally closely related homolog of sili-
catein with a known X-ray structure (27, 32)] as the structural
template showed that the majority of point mutations are located
on the solvent-exposed protein surfaces (Fig. S7). The models
further suggest that the active site residues of silicatein (Ser26,
His165, and Asn185) are structurally conserved in silicatein X1
(Fig. 4B) but partially recongured in silicatein XT (Fig. S7).
Silicatein has been proposed to operate through an S
N
2-type
attack by the active site serine-26 on the silicon precursor center,
leading to the release of ethanol and a metal-protein intermediate
(3, 23, 25, 32). Subsequent attack on the metal center by water
releases the hydrolyzed precursor, which can then condense with
other precursor species. In silicatein , the active site histidine-165
hydrogen bonds with the active site serine-26 hydroxyl to increase
the nucleophilicity of the serine oxygen and activate it for metal-
center attack (Fig. 4B). Asparagine-185 is hydrogen-bonded with
His-165 on its opposite site (away from Ser-26) to withdraw
electron density and increase the strength of the imidazole-hy-
droxyl hydrogen bond. In the model of silicatein XT, only Ser-26
appears to be structurally conserved (Fig. S7). Although this
suggests that the catalytic triad of silicatein is energetically
strained in silicatein XT, it is presently unclear whether this
reconguration represents a functionally adjusted active site or
whether a catalytic triad is not necessary for silicatein-mediated
titania formation. The role of active site chemistry in enzymatic
mineralization with different metallorganic substrates is currently
a subject of ongoing investigations (42, 43).
Evolved Mineralized Phenotypes. We overexpressed the selected
variant silicateins and parent silicatein in Escherichia coli, and
puried and refolded the recombinant proteins using column
chromatography (Materials and Methods and Fig. S7). When
freshly refolded silicatein X1, silicatein , or heat-denatured sili-
catein X1 was each reacted with 100 mM TEOS at pH 7.4, no
product was observed from the denatured silicatein X1 reaction.
Insoluble products generated by silicatein and silicatein X1 were
collected by centrifugation, washed, and analyzed by TEM.
Whereas silicatein produced agglomerated, interconnected silica
nanoparticles (Fig. S8), silicatein X1 synthesized relatively dis-
persed silica nanoparticles (Fig. 5A). Analogous results were
obtained with silicatein and silicatein XT from reactions target-
ing titania synthesis from TiBALDH (Fig. 5A and Fig. S8). Under
certain TEOS reaction conditions (static instead of agitated in-
cubation and over an extended incubation time; Materials and
Methods), silicatein X1 produced uniquely ordered silica-protein
composite laments in the form of folded sheets (Fig. 5B and Fig.
S8). In some cases, silicatein X1 or silicatein XT (but not silicatein
) was observed to produce diverse crystalline inorganic products
from these aqueous reactions (Fig. 6 and Fig. S8). These crystals
were larger in size than the amorphous nanoparticles produced
from the same mineralization reactions, and EDS analysis revealed
impurity metals (Al, K, Ca, and/or Fe) to be associated with these
crystalline species (Fig. 6A and Fig. S8). Only C, O, and Si (or Ti)
were found in the amorphous product regions (Fig. 5A). In-
terestingly, some of these mineral crystals appeared to be closely
related to well-known crystalline polymorphs of silica, such as
-quartz, as measured by SAED (Fig. 6B). Synchrotron X-ray
diffraction analysis (44) of the Si-containing products obtained
from reactions with silicatein and silicatein X1 conrmed that
silicatein X1 produces a heterogeneous population of crystals,
whereas the products from silicatein are completely amorphous
(Fig. 6C). The chemical information from EDS suggests the crys-
tals are mixed silicates (Fig. 6A), but the heterogeneous nature of
the products precluded precise structure determination.
Discussion
This study demonstrates multiscale mimicry of skeletal silicica-
tion in marine sponges by integrating several reaction strategies
used in natural biomineralization. During in vivo formation of
spicules (sponge skeletal elements; Fig. 1A), silicatein protein l-
aments are initially assembled intracellularly before being expor-
ted to an extracellular vesicular compartment (26, 28, 31). In both
the in vivo and in vitro processes, silicatein enzymatically hydro-
lyzes precursor molecules to initiate mineralization at the silicatein
lament (or silicatein-bead) surface; mineralization is templated
along the silicatein scaffold, and the mineral grows radially from
the scaffold surface within a shape-constraining compartment as
the templating polymer becomes axially occluded within a spicule
(23, 28) (Fig. 1A) or centrally occluded within a pseudospherical
mineral shell (Fig. 1B). In vivo, additional biomolecules, including
silicatein and other proteins, may guide silica growth within the
vesicle once the lament is covered by mineral (28). The Si4
polypeptide added to the biomimetic mineralization reactions
served to mimic this guided mineral growth while also uo-
rescently labeling the silica product. This labeling, combined with
polydisperse emulsion vesicle sizes (20) (SI Text), was useful for
demonstrating by ow-sorting and TEM analysis that both uo-
Fig. 4. Silicatein sequences and modeled active sites. (A) Multiple protein sequence alignment of silicatein parents ( and ) and progeny (X1 and XT). Stars
indicate sequence changes that directly involve hydroxyl residues; asterisks indicate changes occurring immediately adjacent to hydroxyl residues. Yellow and
blue highlights denote residues structurally conserved in a putative catalytic triad conguration as determined by homology modeling; gray highlights
denote residues in the distorted active site region of silicatein XT (Fig. S7). (B) Homology-modeled active site of silicatein overlaid with that of silicatein X1,
showing the shared putative catalytic triad conguration. Native silicatein has been previously proposed to operate through an S
N
2-type attack by the
active site serine-25/26 on a silicon precursor center, leading to precursor hydrolysis and subsequent polycondensation (23).
E1710 | www.pnas.org/cgi/doi/10.1073/pnas.1116958109 Bawazer et al.
rescence and light-scattering intensities positively correlate with
composite microstructure diameter.
The results from studies of puried, refolded silicateins showed
that the evolutionarily selected silicatein variants acquired several
specic mineralizing phenotypes as a result of the genetic screen,
including the abilities to synthesize dispersed metal or metalloid
oxide nanoparticles, to self-assemble into higher ordered struc-
tures under certain conditions, and to catalyze synthesis of mul-
timetallic oxide crystals from near-neutral aqueous solution at
ambient temperature. The similar ow-sorting and light-scattering
signatures, biochemical mutations, and mineral characteristics
observed from both the silica- and titania-targeting selections
suggest that acquisition of the unique phenotypes was directly
governed by the imposed selection environment. To survive,
a given silicatein gene variant displayed on a bead would need to
(i) code for an enzyme that yields a light-scattering signature (via
mineralization) of sufcient intensity to be included in the ow-
sorting gate (Fig. 3A and Fig. S6) and (ii) be sufciently accessible
for postsorting extraction and polymerase-catalyzed DNA ampli-
cation (Fig. 3B). Self-assembly of the active silicatein enzyme
variant on the bead surface would effectively expand the bead
diameter, thereby providing a larger polymeric template for bio-
mimetic mineralization. Formation of dispersed oxide nano-
particles would then introduce mechanical stabilization while
providing a rough surface, leading to an intense side-scattering
signal for ow sorting. Such a composite architecture would be
susceptible to postsorting mechanical disruption to expose the
encoding DNA (Fig. 3B) yet would be stable enough to resist
presorting physical agitation (Fig. 3A and Fig. S6).
Interestingly, multimetallic silicate crystals have been identied
in natural sponge biosilica (26), suggesting that recombinant sili-
catein activity in these in vitro mineralization vesicles exhibits this
natural activity as well. Our results from genetic screening showed
that this crystal-forming ability could be (re)acquired via evolu-
tionary selection for expression in a range of in vitro environ-
ments. It is unknown whether the crystal-forming ability of the
evolved silicateins was a requirement for selection or a secondary
trait indirectly coselected by the screening reaction environment.
All observed silicatein -catalyzed composites that remained in-
tact through centrifugal recovery and ow sorting also exhibited
a crystalline mineral structure (Fig. S2), providing indirect evi-
dence that crystal formation may promote composite strength.
Completely aqueous (i.e., nonemulsied) mineralization reactions
with bacterially expressed, puried, and refolded enzymes, which
yielded only amorphous silica from parental silicatein (Fig. 6C),
demonstrated that the variant silicatein X1 retained its crystal-
forming behavior as an intrinsic activity (Fig. 6), even under
conditions in which metal additives were scarce and vesicle
interfaces were absent. This conrmed that crystal-forming activity
was uniquely acquired through evolutionary selection from the
biomimetic mineralization vesicles. By random chance alone, it is
unlikely that either variant (silicatein X1 or silicatein XT) would
exhibit the number of hydroxyl-associated changes that we found
as mutations in these two selected protein variants (Fig. 4A). We
thus infer that the observed mutations are mechanistically im-
portant in mineral formation. Collectively, the above ndings
demonstrate that laboratory selection pressures imposed by an
evolutionary reaction environment enabled the detection of
mutationally produced unique or unexpected materials not
formed by conventional chemical syntheses or by reaction with
Fig. 5. (A) TEM image and associated EDS spectrum of mineralization
products from bacterially expressed silicatein X1 reacted with 100 mM TEOS
under stirring conditions (Left) or from bacterially expressed silicatein XT
reacted with 100 mM TiBALDH (Right; also from stirring conditions) (Mate-
rials and Methods), showing that both reactions produce relatively dispersed
nanoparticles on an apparent protein template. (B) TEM images of products
from bacterially expressed silicatein X1 reacted with 100 mM TEOS under
static incubation conditions (Materials and Methods), showing that uniquely
ordered lament and sheet structures are formed. The structures are com-
posites that include silicon, and silicatein yields less ordered structures
under the same conditions (Fig. S8).
Fig. 6. Silicatein variants were evolutionarily selected to catalyze synthesis
of crystalline oxides from buffered aqueous solution, pH 7.4. (A) TEM anal-
ysis with EDS on products from silicatein X1 reacted with TEOS shows that
crystals are formed, which are composed of silicon with additive metals;
analogous results were found from titania-targeting silicatein XT reactions
(Fig. S8). Additional diverse crystals in each system exhibited a variable range
of metal additives in their EDS spectra, whereas the amorphous nano-
particulate product regions (e.g., Fig. 5) showed no additive metals. (B)
Crystal structure of the example presented in A closely resembles that of
-quartz as determined by SAED. d
expmtl
, experimentally measured d-spac-
ings; d
ref
, reference d-spacings for alpha-quartz; % var., % variation be-
tween the experimentally measured and reference d-spacing values. (C)
Synchrotron X-ray diffraction conrms that silicatein X1 uniquely catalyzes
a mixture of crystalline silicates; the heterogeneous nature of the products
precluded precise structure determination.
Bawazer et al. PNAS | Published online June 7, 2012 | E1711
B
I
O
C
H
E
M
I
S
T
R
Y
P
N
A
S
P
L
U
S
unaltered parental protein sequences as obtained directly
from nature.
This study adds to growing evidence that enzymatic routes to
new materials, as exploited in biomineralization, may be effec-
tively harnessed to produce technological materials (3, 23, 24, 32,
40, 43, 45, 46). A typical enzyme, in a properly folded confor-
mation, presents both a chemically catalytic active site and a so-
lution-exposed surface with specic surface chemistry and atomic-
scale topological features. Through genetic evolution, driven by
specic selective conditions, both chemical activity and surface
denitions of an enzyme are tunable. For inorganic materials
synthesis in any context, the dynamic states of both of these
parameters, local chemical changes and molecular-scale inter-
actions at heterogeneous interfaces, are critically inuential in the
nucleation and growth of specic mineral phases. Thus, the es-
sential and genetically tunable features inherent to enzymes make
these functional biomolecules potentially useful agents for medi-
ating and controlling mineral synthesis from aqueous solution.
In this study, enzymes have been evolutionarily reengineered to
synthesize newly identied inorganic materials catalytically. Thus
far, we have produced only a single generation of silicatein progeny
to investigate salient features of the screening process. Our ability
to recover unique functional genes after a single round of selection
was facilitated by our use of a shufed gene library (18, 36). Our
ndings show that enzymes can indeed be engineered to synthesize
new material structures through mutation and evolutionary selec-
tion, but the mineralization reaction environment during screening
must be tightly controlled before more quantitative studies can be
conducted to dene the role of enzymes in achieving specic
crystalline products. The observed problem of contamination is
analogous to similar problems faced in electronic materials fabri-
cation, in which clean-room environments are required for fabri-
cating high-purity, highly controlled semiconducting crystals.
However, although impurity elements pose a concern in elucidating
the enzymatic mechanisms of mineralization, the structural com-
plexity of biominerals in nature suggests that DNA-driven systems
can accommodate and functionally adapt such chemical additives
toward achieving specic new materials properties and perfor-
mance. The in vitro study presented here supports this conclusion
and suggests that we may begin harnessing this exibility for an-
thropogenic materials fabrication.
Conclusion
We have reported the acquisition of previously undescribed en-
zymatic biomineralizing phenotypes through a single generation
of in vitro evolution, including the synthesis of crystalline silicates
from near-neutral aqueous solution. We have thus shown it is
possible to evolve solid-state materials genetically using an ap-
proach that mimics natural biomineral evolution, in which the
ttest materials systems are selected. In principle, the targeted
mineral characteristics could be extended to any material prop-
erty (e.g., photoluminescence, magnetization) for which an ef-
fective selection can be designed. Tunable parameters in the
demonstrated platform for further studies include vesicle size;
vesicle and other mineralization interface chemistries (e.g., sur-
factants, support surfaces); new parental and mutationally altered
proteins encoded by the DNA; mechanisms of genetic mutation;
spatial proximity of the DNA to the encoded proteins; shape,
chemistry, and activities of the polymeric mineral-templating
scaffold; chemistry of the mineralization precursors; and additive
components of interest (e.g., other mineral-binding polypeptides
or aptamers). This simple emulsion-based system thus presents
a multitude of ways to begin investigating mineralization via
DNA-driven genetic algorithms, which have operated over mil-
lions of years of evolution to produce natures biomineral fabri-
cation systems. The close genetic relationship of the silicateins
with the cathepsin family of digestive proteases and the larger
superfamily of hydrolases shows that nature has evolutionarily
recruited enzymes with other functions to control biomineraliza-
tion. Recently, members of this enzyme superfamily (42, 43), as
well as other enzyme families (45, 46) that do not normally cat-
alyze mineral synthesis, have been shown to exhibit catalytic
mineralization activity in vitro, suggesting that a wide variety of
enzymes may be suitable candidates for directed evolution toward
new, specic mineral-synthesis activities. The broad biotechno-
logical capacity to evolve and reengineer enzymes suggests ex-
citing possibilities for further mimicking this natural evolutionary
process, and directing enzyme evolution toward the development
of new solid-state materials. The diverse collection of enzymes in
the natural world and their range of unique chemical activities
could allow, in principle, a correspondingly diverse range of in-
organic materials to be synthesized by this general strategy. As
genetic screening pressures are developed to include functional
material performance for targeted engineering applications, this
approach will begin to allow the same DNA-based evolutionary
processes that have created seashells and skeletons to be har-
nessed to advance human technologies.
Materials and Methods
Preparation of Gene Templates for in Vitro Expression. The starting recombi-
nant T. aurantia silicatein genes used for all methods encoded the mature
218-aa (silicatein ) or 217-aa (silicatein ) polypeptides (29). Parent or
shufed genes were cloned into T7 in vitro expression vector pIVEX 2.6
(Roche) with NotI and SacI sites. Linear templates that included T7 promoter
and terminator regions and the ligated gene were then amplied from this
plasmid by PCR using dual biotin-labeled forward primer LMB21-Long and
reverse primer pIVB1-Long (20). All primer sequences are provided in ref. 20
or in Table S1, and ligation and PCR conditions, as well as conditions for
conrming the presence of recombinant silicatein on the beads by ow
cytometry, are provided in SI Text S1.
Silicatein Gene Library Creation. Silicatein and silicatein genes were PCR-
amplied with primers to produce silicatein genes with one (bottom)
DNA strand that was 5-phosphorylated and silicatein genes that were
similarly labeled but in an inverse manner, with the top strand 5-phos-
phorylated (Fig. S4). Digesting these PCR products with -exonuclease (NEB;
-exonuclease requires 5-phosphorylation of its DNA substrate for proc-
essivity) yielded ssDNA of each parent gene, an intact top strand of silicatein
and an intact bottom strand of silicatein . The ssDNA products were pu-
ried from dsDNA using gel electrophoresis (ssDNA migrates faster than
dsDNA; Fig. S4). Each ssDNA sample was then digested with DNase I at 15 C
for 5 min, and 10- to 80-nt digestion products were excised and puried
from a 10% (wt/vol) PAGE gel. A primerless gene reconstitution reaction was
conducted, followed by a gene recovery PCR using a forward primer for
silicatein and a reverse primer for silicatein . Amplied products showed
a single band at the expected length, and control reactions with mixed
primer sets conrmed the presence at least one -to- crossover point per
shufed gene (Fig. S4).
In Vitro Protein Expression. In vitro compartmentalization for protein ex-
pression was conducted exactly as described previously (20, 21), except the in
vitro expression mixture was supplemented with 40 g/mL BSA in this study.
For control samples subjected to in vitro expression, methionine was with-
held from the in vitro expression mixture. After incubation for protein ex-
pression, the emulsion was broken and beads were washed as described
previously (20), except the carbonate-zinc buffer was omitted in this study.
Instead, after washing the beads with PBS with 0.1% Tween 20 (PBS/T) and
8 mg/mL heparin, the beads were further washed twice more with 10 mM
TrisHCl, pH 7.4, and resuspended in 55 L of the same.
Biomimetic Mineralization Reactions with Flow Sorting. For reactions con-
ducted with the Flc-Si4 polypeptide, 2 L of 1 mg/mL Flc-Si4 peptide (Sigma
Genosys) (33) (in 10 mM Hepes buffer, pH 7.5) was added to the aqueous
phase (giving 57 L of total H
2
O) of the emulsion. For silica-targeting reac-
tions, resuspended beads were rst reemulsied, after which TEOS (Sigma)
was added to 100 mM in the emulsion with stirring by a small magnetic stir
bar immersed in the emulsion at 100 rpm for 3 min. For titania-targeting
reactions, TiBALDH was added to 100 mM in the water phase immediately
before reemulsication. Other than replacement of the aqueous phase (with
TrisHCl-buffered water, pH 7.4, used in place of the in vitro expression
E1712 | www.pnas.org/cgi/doi/10.1073/pnas.1116958109 Bawazer et al.
mixture), all samples were reemulsied with a procedure identical to the
emulsication for protein expression (20). Mineralization and control reac-
tions with silicatein and Flc-Si4 were incubated for 2 h at 16 C (mimicking
the temperature of the sponges natural marine environment). Shufed li-
brary screening reactions from which DNA was amplied were incubated for
12 h at 22 C. Flc-Si4 probe was not added to TiBALDH reactions, nor was it
added to the nal emulsied TEOS reactions conducted for shufed library
screening. Reacted beads were recovered from the emulsions and washed by
centrifugation as described previously (20). After the nal wash, the beads
were resuspended in 100 L of PBS/T. Each sample was diluted 20-fold in PBS/
T and analyzed with a Becton Dickinson FACSAria cell sorter at a rate of
10,000 events per second; uorescence analysis was conducted with a 488-
nm argon ion excitation laser, and emission was measured through 530 15-
nm and 576 13-nm band-pass lters. The forward scatter detector is
a photodiode with a 488 10-nm bandpass lter, and the side scatter de-
tector is a photomultiplier also with a 488 10-nm bandpass lter. Sorting
gates were set as shown in Figs. 2 and 3 and Fig. S6.
TEM Analysis of Sorted Beads. TEM analysis was conducted on either a JEM-
1230 TEM (JEOL) operated at 80 kV with digital images recorded on an
Advantage CCD camera system (AMT) or on an FEI Tecnai G2 Sphera electron
microscope with EDS at an operating voltage of 200 kV. For TEM analysis of
composite structures mineralized within silicatein biomimetic vesicles,
1,000 events were sorted through each applied uorescence subgate (Fig.
2C), washed with 2 mL of ultrapure H
2
O through a 0.45-m vacuumltration
unit (AcroPrep GH Polypro lter plates; Pall), and resuspended in 100 L
ultrapure H
2
O, 20 L of which was dropped onto a copper-formvar TEM grid
and dried at 37 C overnight before imaging. For selected area diffraction
analysis on large mineral structures, the pattern was obtained from near the
outer edge of the structure, where the electron beam could penetrate the
thickness of the material. For TEM analysis of reacted presorted beads from
silicatein library and control biomimetic vesicles, after recovery from emul-
sions, one-fth of the bead solution was washed several times in ethanol by
centrifugation and evaporated onto a copper-formvar TEM grid for imag-
ing; the remainder of the bead solution was used for ow-sorting analysis.
Amplication, Sequencing, and Homology Modeling of Selected Genes. After
gated subpopulations of the reacted library-displaying beads were ow-
sorted, postsorting physical agitation was then conducted as described in
Results, with additional details provided in SI Text. Once agitated, beads
were ltered (SI Text); they were resuspended in a nal volume of 31 L of
1 Pfu buffer (Stratagene) and transferred to a 50-L PCR. The PCR included
primers T7recF and T7recR (0.2 M each), 0.2 mM deoxynucleotide triphos-
phates, and 2.5 units of PfuTurbo (Stratagene). DNA was susceptible to
shearing during the sonication step, causing smearing and low-molecular-
weight bands in gels of the recovered gene products (Fig. 3B). The PCR cy-
cling schedule is provided in SI Text S6.
Products were electrophoresed on a 1% agarose gel, and DNA from the
0.80-kb product band was excised and puried (gel purication; Qiagen) (Fig.
3B). The puried DNA was used as a template in a PCR with primers BFpET
and ARpET to amplify the recovered silicatein genes for cloning into vector
pET151-DTOPO (Invitrogen). Cloning and transfection into E. coli top 10 cells
were conducted according to the manufacturers instructions. After plating
on LB-ampicillinselective media, individual colonies were randomly chosen
for culture, DNA purication (Miniprep; Qiagen) and sequencing (University
of California DNA sequencing facility, Davis, CA).
Homology models (Fig. 4Band Fig. S7) were generated by SWISS-MODEL (47)
and visualized using BALLView v1.3 (www.ball-project.org) (48). Cathepsin L
[Protein Data Bank (PDB) ID code 3HHA, Chain A] was selected as the best
template for homology modeling of silicatein through the Automated
module onthe SWISS-MODEL site, whichcomputationally screens a database of
templates. Once identied, this same structural template was used for model-
ingthe silicateinX1 and silicateinXT structures, using the Alignment module on
the SWISS-MODEL site.
Bacterial Expression, Purication, and Refolding of Silicatein. To investigate
the mineralizing phenotypes of the identied full-length silicatein variants,
genes encoding silicatein X1, silicatein XT, and parent silicatein were
inserted into the pET151-DTOPO plasmid for inducible expression, placing
them under control of the T7 promoter and lacO operator; the plasmid also
encodes a His
6
tag N-terminal to the recombinant protein. Proteins were
overexpressed in and puried from E. coli BL21AI cells, as described in SI Text
S7. Recombinant silicatein was found expressed as inclusion bodies (Fig. S7),
and thus was resolubilized, and then puried and refolded on a 5-mL Ni
HisTrap HP column (GE Healthcare) by gradual removal of denaturing buffer
using an kta purier (Amersham Biosciences) FPLC system (details are
provided in SI Text S7). After eluting the refolded protein from the column,
the buffer was exchanged to 20 mM TrisHCl (pH 8.5), 10 mM NaCl, and 0.1%
glycerol using a HiPrep 26/10 desalting column (GE Healthcare) with the
same kta FPLC system. Protein concentration was determined by absor-
bance at 280 nm using a nanodrop spectrophotometer, with extinction
coefcients determined by the respective amino acid sequence of each
expressed protein using the Protparam tool on the ExPASy Bioinformatics
Resource Portal (http://web.expasy.org/protparam/).
Mineralization Activity of Recombinant Silicateins. To test for mineralization,
50 g/mL silicatein was used in 1.5-mL reactions with 100 mM TEOS or
TiBALDH, or in 5-mL reactions with 1 M TEOS, in 100 mM TrisHCl (pH 7.5) for
12 h at 22 C. Denatured protein was obtained by heating in a water bath at
85 C for 30 min. TEOS reactions that were subject to light physical agitation
while incubating were mixed on a Tomy MT-360 microtube mixer at half-full
speed. TEOS has low miscibility in water; thus, agitation during incubation
constantly refreshed the reaction interface between the two solutions.
TiBALDH is completely miscible in water. At the end of the agitated in-
cubation period, before precipitate recovery by centrifugation, no insoluble
product was observable by eye from the heat-denatured silicatein reactions.
The insoluble product from the silicatein reactions formed a pellet at the
bottom of the test tube. The insoluble product from the silicatein X1 and XT
reactions, although clearly visible by eye, remained as a dispersed suspension
in solution (a qualitative indication that dispersed particles had been
synthesized).
Higher order self-assembled structures from silicatein and silicatein X1
were observed from reaction with 100 mM TEOS after a static incubation
period of 12 h at room temperature, followed by static incubation for 1 wk
at 4 C (Fig. 5B and Fig. S8). To stop the mineralization reactions and recover
precipitated products, reaction mixtures were centrifuged at 20,800 g for
10 min at 4 C and the supernatants were then discarded. For the TEOS
reactions, 1.5 mL of ethanol was added to each reaction before this initial
centrifugation to increase the miscibility of TEOS and create a single-phase
solution. After the rst centrifugation, precipitates were washed ve times
with 1.5 mL of 70% ethanol by centrifugation and decanting, and an aliquot
of the resuspended products was then placed on a TEM copper-formvar grid
and dried at 37 C overnight before analysis with a Tecnai G2 Sphera elec-
tron microscope at an operating voltage of 200 kV. High-resolution syn-
chrotron powder diffraction data were collected using beamline 11-BM at
the Advanced Photon Source, Argonne National Laboratory, using an av-
erage wavelength of 0.13702 (44). A mixture of National Institute of
Standards and Technology standard reference materials, Si (SRM 640c), and
Al
2
O
3
(SRM 676) was used to calibrate the instrument, with the Si lattice
constant calibrating the wavelength for each detector.
ACKNOWLEDGMENTS. We thank the expert beamline staff at 11-BM at the
Advanced Photon Source at Argonne National Laboratory; the laboratory of
P. Daugherty in the Department of Chemical Engineering at the University
of California, Santa Barbara, for assistance with ow sorting; J. Fong,
J. Kunkel, J. Len, G. Mantalas, J. Stuart, and A. Ibish for experimental assis-
tance; past and present members of D.E.M.s laboratory for helpful discus-
sions; and F. Meldrum for careful reading of the manuscript. This work was
supported by US Department of Energy Grant DEFG03-02ER46006. L.A.B.
was partially supported by University of California Systemwide Biotechnol-
ogy Research and Education Program Graduate Research and Education in
Adaptive Bio-Technology Training Grant 2007-19. J.R.N. was partially sup-
ported by a National Science Foundation Graduate Research Fellowship. Use
of the Advanced Photon Source at the Argonne National Laboratory was
supported by the US Department of Energy, Ofce of Science, Ofce of Basic
Energy Sciences, under Contract DE-AC02-06CH11357.
1. Smith JV (1998) Biochemical evolution. I. Polymerization On internal, organophilic
silica surfaces of dealuminated zeolites and feldspars. Proc Natl Acad Sci USA 95:
33703375.
2. Ohmoto H, Kakegawa T, Lowe DR (1993) 3.4-Billion-year-old biogenic pyrites from
Barberton, South Africa: Sulfur isotope evidence. Science 262:555557.
3. Brutchey RL, Morse DE (2008) Silicatein and the translation of its molecular
mechanism of biosilicication into low temperature nanomaterial synthesis. Chem
Rev 108:49154934.
4. Mayer G (2005) Rigid biological systems as models for synthetic composites. Science
310:11441147.
Bawazer et al. PNAS | Published online June 7, 2012 | E1713
B
I
O
C
H
E
M
I
S
T
R
Y
P
N
A
S
P
L
U
S
5. Meldrum FC, Heywood BR, Mann S (1992) Magnetoferritin: In vitro synthesis of
a novel magnetic protein. Science 257:522523.
6. Sarikaya M, Tamerler C, Jen AKY, Schulten K, Baneyx F (2003) Molecular biomimetics:
Nanotechnology through biology. Nat Mater 2:577585.
7. Weiner S, Addadi L (1997) Design strategies in mineralized biological materials. J
Mater Chem 7:689702.
8. Whaley SR, English DS, Hu EL, Barbara PF, Belcher AM (2000) Selection of peptides with
semiconductor bindingspecicity for directednanocrystal assembly. Nature 405:665668.
9. Xiang XD, et al. (1995) A combinatorial approach to materials discovery. Science 268:
17381740.
10. Brown S (1992) Engineered iron oxide-adhesion mutants of the Escherichia coli phage
lambda receptor. Proc Natl Acad Sci USA 89:86518655.
11. Lee YJ, et al. (2009) Fabricating genetically engineered high-power lithium-ion
batteries using multiple virus genes. Science 324:10511055.
12. Rothberg JM, et al. (2011) An integrated semiconductor device enabling non-optical
genome sequencing. Nature 475:348352.
13. Mann S (2001) Biomineralization: Principles and Concepts in Bioinorganic Materials
Chemistry (Oxford Univ Press, UK).
14. Gugliotti LA, Feldheim DL, Eaton BE (2004) RNA-mediated metal-metal bond
formation in the synthesis of hexagonal palladium nanoparticles. Science 304:
850852.
15. Aizenberg J, et al. (2005) Skeleton of Euplectella sp.: Structural hierarchy from the
nanoscale to the macroscale. Science 309:275278.
16. Mann S, Frankel RB, Blakemore RP (1984) Structure, morphology and crystal-growth
of bacterial magnetite. Nature 310:405407.
17. Sundar VC, Yablon AD, Grazul JL, Ilan M, Aizenberg J (2003) Fibre-optical features of
a glass sponge. Nature 424:899900.
18. Arnold FH, Georgiou G, eds (2003) Directed Evolution Library Creation: Methods and
Protocols (Humana Press, Totowa, NJ).
19. Bershtein S, Tawk DS (2008) Advances in laboratory evolution of enzymes. Curr Opin
Chem Biol 12(2):151158.
20. Grifths AD, Tawk DS (2003) Directed evolution of an extremely fast
phosphotriesterase by in vitro compartmentalization. EMBO J 22(1):2435.
21. Miller OJ, et al. (2006) Directed evolution by in vitro compartmentalization. Nat
Methods 3:561570.
22. Tawk DS, Grifths AD (1998) Man-made cell-like compartments for molecular
evolution. Nat Biotechnol 16:652656.
23. Cha JN, et al. (1999) Silicatein laments and subunits from a marine sponge direct the
polymerization of silica and silicones in vitro. Proc Natl Acad Sci USA 96:361365.
24. Curnow P, et al. (2005) Enzymatic synthesis of layered titanium phosphates at low
temperature and neutral pH by cell-surface display of silicatein-alpha. J Am Chem Soc
127:1574915755.
25. Kisailus D, Truong Q, Amemiya Y, Weaver JC, Morse DE (2006) Self-assembled
bifunctional surface mimics an enzymatic and templating protein for the synthesis of
a metal oxide semiconductor. Proc Natl Acad Sci USA 103:56525657.
26. Mugnaioli E, et al. (2009) Crystalline nanorods as possible templates for the synthesis
of amorphous biosilica during spicule formation in Demospongiae. Chembiochem 10:
683689.
27. Murr MM, Morse DE (2005) Fractal intermediates in the self-assembly of silicatein
laments. Proc Natl Acad Sci USA 102:1165711662.
28. Schrder HC, et al. (2007) Apposition of silica lamellae during growth of spicules in
the demosponge Suberites domuncula: Biological/biochemical studies and chemical/
biomimetical conrmation. J Struct Biol 159:325334.
29. Shimizu K, Cha J, Stucky GD, Morse DE (1998) Silicatein alpha: Cathepsin L-like protein
in sponge biosilica. Proc Natl Acad Sci USA 95:62346238.
30. Sumerel JL, et al. (2003) Biocatalytically templated synthesis of titanium dioxide.
Chem Mater 15:48044809.
31. Wang XH, et al. (2011) Evagination of cells controls bio-silica formation and
maturation during spicule formation in sponges. PLoS ONE 6:e20523.
32. Zhou Y, Shimizu K, Cha JN, Stucky GD, Morse DE (1999) Efcient catalysis of
polysiloxane synthesis by silicatein alpha requires specic hydroxy and imidazole
functionalities. Angew Chem Int Ed Engl 38:780782.
33. Naik RR, Brott LL, Clarson SJ, Stone MO (2002) Silica-precipitating peptides isolated
from a combinatorial phage display peptide library. J Nanosci Nanotechnol 2(1):
95100.
34. Mena MA, Treynor TP, Mayo SL, Daugherty PS (2006) Blue uorescent proteins with
enhanced brightness and photostability from a structurally targeted library. Nat
Biotechnol 24:15691571.
35. Stephens CJ, Kim YY, Evans SD, Meldrum FC, Christenson HK (2011) Early stages of
crystallization of calcium carbonate revealed in picoliter droplets. J Am Chem Soc 133:
52105213.
36. Stemmer WPC (1994) Rapid evolution of a protein in vitro by DNA shufing. Nature
370:389391.
37. Zha W, Zhu T, Zhao H (2003) Family shufing with single-stranded DNA. Directed
Evolution Library Creation: Methods and Protocols, eds Arnold FH, Georgiou G
(Humana Press, Totowa, NJ), pp 9197.
38. Hecky RE, Mopper K, Kilham P, Degens ET (1973) Amino-acid and sugar composition
of diatom cell-walls. Marine Biology 19:323331.
39. Tilburey GE, Patwardhan SV, Huang J, Kaplan DL, Perry CC (2007) Are hydroxyl-
containing biomolecules important in biosilicication? A model study. J Phys Chem B
111:46304638.
40. Fairhead M, et al. (2008) Crystal structure and silica condensing activities of silicatein
alpha-cathepsin L chimeras. Chem Commun 17651767.
41. Murr MM, et al. (2009) New pathway for self-assembly and emergent properties.
Nano Today 4(2):116124.
42. Frampton M, Vawda A, Fletcher J, Zelisko PM (2008) Enzyme-mediated sol-gel
processing of alkoxysilanes. Chem Commun 55445546.
43. Smith GP, Baustian KJ, Ackerson CJ, Feldheim DL (2009) Metal oxide formation by
serine and cysteine proteases. J Mater Chem 19:82998306.
44. Wang J, et al. (2008) A dedicated powder diffraction beamline at the advanced
photon source: Commissioning and early operation results. Rev Sci Instrum, 79:
085105.
45. Abbate V, Bassindale AR, Brandstadt KF, Taylor PG (2011) A large scale enzyme screen
in the search for new methods of silicon-oxygen bond formation. J Inorg Biochem
105:268275.
46. Unuma H, Matsushima Y, Kawai T (2011) Enzyme-mediated synthesis of ceramic
materials. Journal of the Ceramic Society of Japan 119:623630.
47. Arnold K, Bordoli L, Kopp J, Schwede T (2006) The SWISS-MODEL Workspace: A web-
based environment for protein structure homology modelling. Bioinformatics 22:
195201.
48. Moll A, Hildebrandt A, Lenhof HP, Kohlbacher O (2005) BALLView: An object-
oriented molecular visualization and modeling framework. Journal of Computer-
Aided Molecular Design 19(11):791800.
E1714 | www.pnas.org/cgi/doi/10.1073/pnas.1116958109 Bawazer et al.
Supporting Information
Bawazer et al. 10.1073/pnas.1116958109
SI Text
S1. Protein Expression by in Vitro Compartmentalization. The ligation
reaction used to insert recombinant silicatein into vector pIVEX
2.6 (Roche) (at NotI and SacI sites) was conducted with T4 DNA
ligase (NEB) in a 25-L ligation solution using 12 ng of the di-
gested insert and 20 ng of the digested vector. One microliter of
the ligation mixture was used as a template in the following PCR
amplication reaction: 2.5 units of PfuUltra (Stratagene); 0.15 M
each primer, LMB21-Long (forward primer, dual biotin-labeled at
the 5 end) and pIVB1-Long (reverse primer) (1) (all primer se-
quences are provided in ref. 20 or in Table S1); and 0.2 mM each
dinucleotide triphosphate (all in 1 PfuUltra buffer). The fol-
lowing thermal cycling schedule was used: 95 C for 2 min; 25
cycles of 95 C for 30 s, 58 C for 30 s, and 72 C for 1 min 30 s;
and 72 C for 6 min. The applied primers ank the T7 promoter
and terminator regions of the vector. The amplied products
appeared as 1.5-kb and 0.9-kb bands on a 1% agarose gel,
corresponding to amplication of the T7 expression region from
the vector with and without the gene insert, respectively. DNA
from the 1.5-kb bands was excised, puried, and used as the
linear T7 expression template for in vitro expression. The linear
DNA constructs encode a nine-residue HA epitope tag fused to
the N terminus of the recombinant protein.
To test protein expression, the linear gene templates encoding
recombinant silicatein were attached to beads and subject to
transcription and translation by in vitro compartmentalization (2).
Streptavidin-coated polystyrene beads (0.97 m, CP01N; Bangs
Laboratories) were coated rst with biotinylated linear gene ex-
pression templates using a 1:3 gene-to bead ratio for the binding
reaction and then with anti-HA antibodies using 3,000 antibody
molecules per bead for the binding reaction. The beads were
emulsied with 500 L of an oil-surfactant mixture and subjected
to in vitro expression exactly as described previously (1, 2).
To assay for recombinant silicatein on the bead surfaces, beads
recovered from the silicatein expression emulsion and control
beads were both incubated with a uorescently labeled, silicatein-
specic polypeptide probe, AntiSil-Flc (3); washed; and ana-
lyzed with a Becton Dickinson FACSAria ow sorter to measure
probe uorescence (Fig. S1). The probe incubation before ow-
sorting analysis included 10 g/mL AntiSil-Flc added to the
beads and mixed under gentle agitation for 20 min. The beads
were then washed four times through 0.22-m ltration units
(retaining the 1-m-diameter beads), with centrifugation con-
ducted at 12 krpm for 1 min for each wash. Wash solutions
consisted of 1 PBS with 0.1% Tween 20 (PBS/T); 1 0.2 M
glycineHCl (pH 2.2); 1 1 M TrisHCl (pH 9.1); and, again, 1
PBS/T. Beads were resuspended in 500 L and analyzed by ow
sorting as described in the legend for Fig. S1.
S2. Transmission Electron Microscopy Analysis of Sorted Silicatein
-Mineralized Beads. The emulsions we used in this study exhibit
polydisperse vesicle sizes. According to Miller et al. (1), who
measured vesicle polydispersity using the same emulsication
conditions we used, the mean droplet diameter within the
emulsion is 23 m and less than 1% of the droplets have a di-
ameter larger than 6 m. Assuming that the silicatein -catalyzed
silicate structures identied and measured by transmission
electron microscopy (TEM) have diameters approximately equal
to the mineralization vesicles in which they were formed (Fig. 2
and Fig. S2), our results are in general agreement with the dis-
tribution of large vesicles reported by Miller et al. (1). Flow
sorting gates 3 and 4 (Fig. 2), which include structures with di-
ameters larger than 6 m, represent 1%of the entire reacted bead
population. Although emulsions with polydisperse vesicle sizes were
useful for investigating the relationship between light scattering and
uorescence signals and corresponding mineral composite sizes, the
low quantity of mineralized materials yielded by ow sorting pre-
cluded a variety of materials analyses beyond TEMcharacterization
(with electron-dispersive spectroscopy and selected area electron
diffraction). Emulsion environments with more tightly controlled
vesicular size distributions to produce a higher yield of mineralized
products will be needed for more detailed studies of the physical
and chemical mechanisms that control mineralization in this system.
The present results demonstrate that biomineralizing enzymes are
critical for producing mineralized composite structures as well as
crystalline mineral products.
S3. TEM Analysis of Mineral Precursor-Reacted Control Beads. The
absence of silicon from beads sorted from the silicatein-minus
control reaction (Fig. S3) is evidence that under the conditions
studied, background autohydrolysis of tetraethylorthosilicate
(TEOS; Sigma) is relatively low, as should be expected at the
reactions buffered near-neutral pH of 7.4. Thus, insufcient si-
licic acid was present for Si4-induced silica precipitation in the
control population. Silica mineral surfaces can participate in
TEOS hydrolysis (4). The Si4 polypeptide likely affects this
process after mineralization is catalytically initialized by silica-
tein at the bead surface; Si4 polypeptide-mediated precipitation
would then consume hydrolyzed species, thus shifting chemical
equilibria to favor TEOS hydrolysis at the new mineral surfaces.
S4. DNA Shufing to Produce a Diverse Silicatein Gene Pool. To
produce a shufed library, selected strands of the parent silicatein
genes were 5-phosphorylated. Silicatein was PCR-amplied
with primers AF-OH and AR-P, and silicatein was PCR-am-
plied with primers BF-P and AR-OH (Fig. S4A). Five micro-
grams of each double-stranded gene was reacted with 10 units of
exonuclease (NEB) at 37 C for 1.5 h to digest the 5-phosphor-
ylated strands (Fig. S4B). For DNase I digestion, 225 ng of each
single-stranded gene was combined in a 50-Lsolution of DNase I
buffer and equilibrated at 15 C for 5 min; 0.6 units of DNase I
(NEB) was then added, and the solution was mixed by pipetting
and incubated for 2 min, followed by immediate enzyme in-
activation at 90 C for 10 min. A smear of 10- to 80-nt products
from each digestion reaction was excised from a 10% (wt/vol)
PAGE gel (Fig. S4C), eluted into ultra-pure H
2
O by soaking
crushed gel slices overnight at 4 C, and further puried using
Centri-Spin 10 silica-gel purication columns (Princeton Sepa-
rations). For preparation of a control reaction, 500 ng of single-
stranded silicatein gene only was digested using the same pro-
cedure. The following primerless gene reconstitution reaction was
conducted: 100 ng of single-stranded gene fragments, 3.5 units of
Expand Hi-Fi polymerase (Roche), 0.24 mM each deoxynucleo-
tide triphosphate, and 2.5 mM MgCl
2
, in a total of 50 L of Ex-
pand Hi-Fi polymerase buffer. The following thermal cycling
schedule was applied: 96 Cfor 90 s; 35 cycles of 94 Cfor 30 s, 65
Cfor 90 s, 62 Cfor 90 s, 59 Cfor 90 s, 56 Cfor 90 s, 53 Cfor 90 s,
50 C for 90 s, 47 C for 90 s, 44 C for 90 s, 41 C for 90 s, 38 C for
90 s, and 72 C for 4 min; 72 C for 7 min; and 4 C thereafter. This
gene reassembly reaction yielded a smear when fragments from
both parent genes were present in the reaction but no products
when only one strand was used (Fig. S4D). This conrmed that
reassembly was initiated by heterogeneous fragment-fragment
hybridization. The long extension times of the reassembly reactions
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 1 of 9
leads to polymerase mistakes that introduce point mutations into
the shufed library. To isolate recombined genes of the appro-
priate length, 0.5 L of the reacted reconstitution solution was
transferred to provide the gene template to a 100-L PCR chi-
meric gene recovery reaction using a forward primer for silicatein
(BF-Rec) and a reverse primer for silicatein (AR-Rec). This
reaction included 6 units of Expand Hi-Fi polymerase, 0.25 mM
each deoxynucleotide triphosphate, 1.75 mM MgCl
2
, and 400 nM
primers, and the following thermal cycling schedule: 94 C for
4 min; 30 cycles of 94 C for 60 s, 58 C for 80 s, and 72 C for 80 s;
and 72 C for 5 min. The reaction yielded a single product band of
the correct molecular weight (Fig. S4E). Although silicatein and
silicatein genes are highly homologous, they are dissimilar in
their priming regions. Thus, to conrm that the gene library in-
cluded at least one -to- cross-over point per gene, the recovered
library was used as a template in four separate PCR reactions
utilizing respective primer sets (-forward, -reverse), (-forward,
-reverse), (-forward, -reverse), and (-forward, -reverse).
Library amplication occurred only with the mixed primer set
(-forward, -reverse) (Fig. S4F), proving the obtained gene
library was chimeric.
S5. Sequenced Sample of the Shufed Silicatein Gene Library. Se-
quencing of randomly selected variants from the shufed silicatein
library showed that a nonfunctional (prematurely truncated) gene
was disproportionally represented in the library (Fig. S5). This
overrepresentation likely resulted from stochastic differential am-
plication during PCRproduction of the assembled library. Alarge
backgroundof nonfunctional genes inthegenepool canbetolerated
as long as the screening method is able to sample the majority of
library diversity, as is the case with the bead-based ow-sorting
approach we used in this study (2). Among the randomly sampled
genes that encoded full-length proteins (Fig. S5), the presence of
multiple cross-over points further suggested that active full-length
silicatein variants were present in the library, because high gene
recombination rates are positively correlated with overall pop-
ulation tness (5, 6). This is especially true when the functional
screen targets an activity that is similar to the activity of the parent
enzymes, which is the case in the present study. We therefore
proceeded with evolutionary selection using the obtained chimeric
library, and we used the background sequence as a marker gene to
check for enrichment of screened gene populations.
S6. Screening the Silicatein Library for Biomimetic Mineralization.
Biomimetic emulsion-based mineralization reactions conducted
with the silicatein shufed gene library conrmed that variants
within the silicatein library could successfully catalyze minerali-
zation in vitro (Fig. S6). After specic subpopulations of bead-
displaying active variants were identied and recovered by ow
sorting (Fig. 1, Fig. 3A, and Fig. S6F), mechanical agitation was
used to expose the encoding DNA bound to polystyrene beads
that had undergone mineralization at their surfaces. First, ow-
sorting collection tubes containing the sorted, mineralized beads
were centrifuged at 4,500 g for 5 min to collect the beads at the
bottom of the tube. The buffer volume (PBS) was adjusted to
1 mL, and the bead mixture was transferred to low-binding
MaxyClear tubes (Axygen). The beads were agitated by sonica-
tion bath, vortexing, and centrifugation as described in Results.
After each agitation cycle, 800 L of supernatant was removed
and replaced with fresh PBS/T. After the last round of physical
agitation, the beads were resuspended in 500 L of PBS/T and
washed through a 0.22-m centrifugal ltration unit (Amicon
ultrafree-MC, PVDF membrane; Millipore). The ltration units
were centrifuged at 7,000 g until dry, and the beads were
further washed twice with 500 L PBS/T and once with 500 L
1 Pfu buffer (Stratagene), resuspended in a nal volume of
31 L 1 Pfu buffer, and transferred to a 50 L PCR reaction,
which included primers T7recF and T7recR (0.2 M each),
0.2 mM deoxynucleotide triphosphates, and 2.5 units Pfu turbo
(Stratagene). The following PCR cycling schedule was used for
the gene recovery and amplication reaction: 95 C for 2 min; 25
cycles of 95 C for 30 s, 50 C for 30 s, and 72 C for 90 s; and
72 C for 6 min. The PCR was centrifuged at 15,000 g for
2 min, 10 L of the supernatant was applied as a template to
a fresh 40 L of PCR solution, and the following cycling schedule
was used in a second PCR: 95 C for 2 min; 45 cycles of 95 C for
30 s, 50 C for 30 s, and 72 C for 90 s; and 72 C for 6 min. If
even a small number of background beads with nonfunctional
genes contaminated the sorted fraction, these had a high likeli-
hood of being PCR-amplied because, in contrast to functional
genes, they would not be occluded by mineral product. Sample-
exhaustive ow sorting was thus performed to increase the sta-
tistical likelihood of liberating and amplifying intact functional
genes from the sorted mineral composites.
S7. Refolding of Bacterially Expressed Silicateins and Homology
Models. For bacterial silicatein expression, 1 L of LB-carbeni-
cillin cultures of BL21-AI cells containing pET151-DTOPO
(Invitrogen)silicatein plasmids (Materials and Methods) was in-
duced with 200 mg of L-arabinose and 1 mM isopropyl -D-1-
thiogalactopyranoside with shaking at 235 rpm with a culture
shaking incubator for aeration, at 37 C for 4 h. Pelleted cells
were frozen overnight at 80 C and then thawed and lysed with
20 mL B-PER detergent reagent (Pierce) in the presence of 20
kilounits of rLysozyme, 0.5 kilounits of benzonase nuclease, and
a protein inhibitor mixture (complete; Roche) with gentle
shaking for 10 min. Soluble and insoluble fractions were sepa-
rated by centrifugation at 20,000 g for 15 min at 4 C. The
insoluble fraction was washed several times by centrifugation
with 125 mL of 10-diluted B-PER reagent; resolubilized in
a buffer of 6 M guanidineHCl, 20 mM TrisHCl (pH 8.0),
10 mM NaCl, and 20 mM -mercaptoethanol; and loaded onto
a Ni HisTrap column (GE Healthcare) that had been pre-
equilibrated with wash buffer [20 mM TrisHCl (pH 8.0), 10 mM
NaCl, 5 mM imidazole, 6 M urea, 15 mM -mercaptoethanol].
The column was washed with 50 mL of wash buffer at a ow rate
of 0.5 mL/min, and the eluant was changed over 150 mL on
a linear gradient to remove urea gradually from the buffer for
protein refolding [nal refolding buffer: 20 mM TrisHCl (pH
8.0), 10 mM NaCl, 5 mM imidazole, 10% glycerol, 20 mM
-mercaptoethanol]. The column was then washed with 50 mL of
nal refolding buffer; the buffer was changed to 50% elution
buffer over a linear gradient of 30 mL and then stepped to 100%
elution buffer [20 mM TrisHCl (pH 8.0), 10 mM NaCl, 500 mM
imidazole, 10% glycerol, 15 mM -mercaptoethanol]; and, -
nally, the column was washed with 20 mL of elution buffer.
Protein exiting the column was monitored for absorption at 254
nm and 280 nm. Peaks were observed in the ow-through during
the rst 20 mL of wash buffer and 20 mL into the elution buffer
gradient (40% elution buffer). Eluted fractions were collected
and checked by electrophoresis on a 10% (wt/vol) poly-
acrylamide denaturing gel, which showed single bands at the
expected molecular weight for each silicatein variant (Fig. S8A).
S8. Mineralized Products from Bacterially Expressed Silicateins. For
formation of protein-mineral sheets and laments, interfacial
effects between water and an undispersed TEOS phase during an
extended reaction incubation period (Materials and Methods) may
have contributed to silicatein-silica composite self-assembly (Fig.
S8B). Silicatein and silicatein X1 differ in their primary protein
sequences (Fig. 4A). Thus, the unique morphologies of the self-
assembled structures produced by the silicatein parent and
progeny indicate that these mineral composite structures are
genetically encoded. Observation of sheet-like to lament-like
morphological transitions apparent in the structures produced by
silicatein X1 (Fig. 5B and Fig. S8B) supports previous studies
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 2 of 9
suggesting that sheet-like intermediates are important in the self-
assembly of native silicatein laments (7, 8). A discussion of
crystalline mineral products observed to be formed through sil-
icatein catalysis (e.g., Fig. S8C) is provided in the main text.
1. Miller OJ, et al. (2006) Directed evolution by in vitro compartmentalization. Nat
Methods 3:561570.
2. Grifths AD, Tawk DS (2003) Directed evolution of an extremely fast
phosphotriesterase by in vitro compartmentalization. EMBO J 22(1):2435.
3. Curnow P, et al. (2005) Enzymatic synthesis of layered titanium phosphates at low
temperature and neutral pH by cell-surface display of silicatein-alpha. J Am Chem Soc
127:1574915755.
4. Brutchey RL, Morse DE (2008) Silicatein and the translation of its molecular mechanism of
biosilicication into low temperature nanomaterial synthesis. Chem Rev 108:49154934.
5. Arnold FH, Georgiou G, eds (2003) Directed Evolution Library Creation: Methods and
Protocols (Humana Press, Totowa, NJ).
6. Stemmer WPC (1994) Rapid evolution of a protein in vitro by DNA shufing. Nature
370:389391.
7. Murr MM, Morse DE (2005) Fractal intermediates in the self-assembly of silicatein
laments. Proc Natl Acad Sci USA 102:1165711662.
8. Murr MM, et al. (2009) New pathway for self-assembly and emergent properties. Nano
Today 4(2):116124.
Fig. S1. Flow cytometry analysis with a silicatein-specic probe (AntiSil-Flc) (3) conrms that silicatein proteins expressed by in vitro compartmentalization
are captured on the surfaces of polystyrene beads. Thirty thousand events from each sample were analyzed to produce the displayed 2D histograms (a de-
scription of the ow-sorting technique is provided in Results; in ow cytometry, this same technique is used but no sorting is conducted.) (A) Flow-cytometry
uorescence maps with 530 15 nm of integrated band-pass intensity on the abscissa and 576 13 nm of integrated band-pass intensity on the ordinate.
Analysis was conducted after the indicated population of polystyrene beads was incubated with AntiSil-Flc and washed to remove nonbound probe. The gate
Fluorescence was set against 0.1% of the most intense background bead uorescence (far left map); the percentage of gated population from each sample is
highlighted in green. (B) Flow-cytometry, light-scattering maps from the same populations as in A, with forward-scattering intensity on the abscissa and side-
scattering intensity on the ordinate. Flow-cytometry detector voltages and intensity windows were adjusted to spread the side-scattering signals from the control
bead population across the displayed intensity range (far left map) to visualize differences in mapped subpopulations between the AntiSil-Flctreated samples
(middle and far right map, respectively) more easily. The Size gate was set against the 1.0% of the most intense background bead-scattering signals (far left
map); the gated percentages are highlighted in red. Correlated mapping shows gated uorescent events (mapped green) are generated from bead singlets only
from the sample subject to silicatein expression (far right map). (C) Polypeptide sequence of AntiSil-Flc (3). a.u., arbitrary units.
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 3 of 9
Fig. S2. (A) TEM images show examples of mineral-composite structures synthesized in silicatein -based biomimetic mineralization vesicles, sorted through
gate 1, 2, 3, or 4 of Fig. 2A, as indicated. Structures shown in Fig. 1B are from gate 2. Where observable, 1-m-diameter polystyrene beads (PS) are
indicated for reference. (Top) Representative microbeads from each sorted subpopulation with selected area electron diffraction (Insets) revealing the
presence of crystalline silicates. (Scale bars = 2 m.) (Middle and Bottom) Mineral growth on bead surfaces is evident in the products from gate 1. Diverse
mineral composite structures are observed at the surfaces of the microbead templates. Some mineral structures appear fractured; fracturing may have occurred
during centrifugation for emulsion breaking and bead washing. (B) Electron-dispersive spectroscopy spectra from seven distinct areas of the polystyrene-oxide
composite structures shown in A and Fig. 1B.
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 4 of 9
Fig. S3. Antibody-coated control beads do not produce crystalline silicates when reacted with TEOS in biomimetic mineralization vesicles. (A) Polystyrene
beads coated only with anti-HA antibodies (i.e., with no silicatein) reacted with TEOS produce more intense ow-sorting, light-scattering signals (Right; gate
Size 2) than unreacted control beads (Left). Gate Size is reproduced from the plots shown in Fig. 2B. One thousand beads from the reacted control pop-
ulation were sorted through the Size 2 gate, washed by centrifugation several times with water, and evaporated onto a TEM grid for imaging. a.u., arbitrary
units. (B) TEM image of recovered, sorted materials. No intact, high-order structures are apparent (compare with Fig. 1B and Fig. S2). (C) Electron-dispersive
spectroscopy spectrum from the recovered, sorted materials shows P, O, Na, Cl, C, and S. No Si is observed.
Fig. S4. Schematic overview of gene shufing protocol used to produce a chimeric silicatein gene library from parent silicateins and of Tethy aurantia. (A)
Opposite DNA strands on silicatein and genes, respectively, are phosphate-labeled by PCR. (B) Phosphate-labeled strands are digested, and ssDNA is re-
covered. (C) ssDNA is digested, and 10- to 80-nt fragments are recovered and puried. (D) Digested fragments are used as a template in a primerless PCR to
reassemble dsDNA, yielding a population of gene chimeras. (E) Reassembly products are used as a template in a PCR with silicatein forward primer and
silicatein reverse primer to amplify the nal shufed gene library of the same length as the silicatein gene. [A side-by-side comparison is shown between the
gel marker lane (Left) and experimental lane (Right), because these DNA samples were run in nonadjacent lanes in the agarose gel.] (F) Final shufed gene
library is ampliable with silicatein forward primer and silicatein reverse primer but not with other silicatein gene primer combinations. This conrms there
must be at least one -to- cross-over point per gene in the shufed library.
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 5 of 9
Fig. S5. Multiple sequence alignment of parent silicateins and with 10 randomly selected genes from the shufed silicatein library. Each full-length
shufed silicatein (L1, L2, and L3) exhibits several gene recombination points and point mutations. Seven of the 10 library genes encode a nonfunctional highly
mutated polypeptide (L4L10) produced during the gene shufing protocol.
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 6 of 9
Fig. S6. Flow-sorting and TEM data from mineralized microbeads containing the chimeric silicatein library, as well as microbeads from control reactions, after
reaction in emulsion-based biomimetic vesicles with TEOS either in the presence (AC) or absence (DF) of uorescently labeled polypeptide Flc-Si4. (A) TEM
images of control beads (Left) and silicatein library-displaying beads (Right). (Scale bars = 0.5 m.) (B) Flow-sorting uorescence plots (Upper) or light-scattering
plots (Lower) from beads shown in A [i.e., beads that were coated with either silicatein library (Right) or anti-HA antibodies only (Left) before emulsied
mineralization reactions]. (C) TEM image and associated selected area diffraction pattern of crystalline product from reacted silicatein library-displaying beads
shown in A and characterized in B (Right). (Scale bar = 0.5 m.) (D) Flow-sorting, light-scattering plots of silicatein-library beads (Right) or antibody-only control
Legend continued on following page
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 7 of 9
beads (Left) after reactions with mineral precursors in absence of polypeptide Flc-Si4. Bead populations were reacted with silica precursor TEOS (Upper) or with
titania precursor titanium bisammonium lactatodihydroxide (TiBALDH) (Lower). (E) TEM images and associated electron-dispersive spectroscopy patterns show
mineral products are formed under reaction conditions that were used for screening, and that led to the scattering signals shown in D. (F) Flow-sorting, light-
scattering plots of TiBALDH-reacted bead populations that yielded scatter plots in D, Lower, shown here after being immersed for 5 min in a sonication bath. For
selection, beads from the Shufed library sample were sorted through the Size gate. Analogous plots of agitated TEOS-reacted beads are provided in Fig. 3A.
Fig. S7. (A) Representative electrophoresed 10% (wt/vol) polyacrylamide gel shows cell lysates from Escherichia coli induced to overexpress silicatein (here,
silicatein ). All assayed silicateins were expressed as inclusion bodies in the insoluble fraction (strong band in the insoluble fraction between the 21.5-kDa and
31.0-kDa marker bands), and they were solubilized and refolded using column chromatography before assaying for activity. (B) Energy-minimized model of
silicatein s catalytic triad region overlaid with residues from silicatein XT that appear in the same region. Representations showing the entire folded
structures of silicatein X1 (C) or silicatein XT (D) are used to highlight the location of point mutations (relative to the parent silicateins) on the folded proteins.
Mutations directly involving hydroxyls are indicated by stars, and mutations occurring immediately adjacent to hydroxyls are indicated by asterisks (Fig. 4A).
Mutations internal to the protein fold are highlighted by space-lling representations showing van der Waals radii.
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 8 of 9
Fig. S8. (A) TEM images of interconnected oxide nanoparticles produced by recombinant silicatein : either silica produced from reaction with TEOS (Left) or
titania produced from reaction with titanium bisammonium lactatodihydroxide (TiBALDH) (Right). (B) TEM images and associated electron-dispersive spec-
troscopy (EDS) spectra of products from bacterially expressed silicateins reacted with 100 mM TEOS silica precursor under static incubation conditions (Materials
and Methods) show that uniquely higher ordered sheet-like and lament-like composite structures are produced by silicatein X1. X1 den indicates silicatein
X1 was heat-denatured before the reaction. (C) Example of a mineralized crystal found as a product from the reaction of silicatein XT with titania precursor
TiBALDH, observed by TEM (image showing lattice fringes; Left) with selected area diffraction (Center) and EDS (Right). Nanoparticle products from the same
reaction (Fig. 5A) exhibited no crystallinity or additive metals (i.e., other than Ti). The challenge of identifying exact materials chemistry and the heterogeneous
nature of the products precluded precise structure determination.
Table S1. List of PCR primers
Primer name [5-Modication]-sequence
AF-OH 5-GCC TAC CCG GAG ACT GTA GAT TGG AGA AC-3
AR-P [Phosphate]-5-GGC GAA TTC CTA GAG AGT GGG GTA GGA GGC ATC ACT AGC AAT CCC ACA TTG ATT GTA C-3
BF-P [Phosphate]-5-CGC GGA CTC CTA CCC AGA GTC GCT GGA CTG GAG AAC A-3
BR-OH 5-CTA CAA GGT TGG GTA GGC GGC ATC AGT TGC AAT ACC ACA TTG GTT G-3
AR-Rec 5-CTG AGT GAG CTC TTA GAG AGT GGG GTA GGA GGC A-3
BF-Rec 5-GTC TGA GCG GCC GCT ACC CAG AGT CGC TGG ACT G-3
T7recF 5-TAA TAC GAC TCA CTA TAG GGG AAT TGT-3
T7recR 5-CTA ATT ATT GCT CAG CGG TGG-3
BFpET 5-CAC CTA CCC AGA GTC GCT GGA CTG-3
ARpET 5-CTA GAG AGT GGG GTA GGA GGC A-3
Bawazer et al. www.pnas.org/cgi/content/short/1116958109 9 of 9