Mammalian Cells Engineered To Produce New Steroids

Article abstract of DOI:10.1002/cbic.201800214

Steroids can be difficult to modify through traditional organic synthesis methods, but many enzymes regio- and stereoselectively process a wide variety of steroid substrates. We tested whether steroid-modifying enzymes could make novel steroids from non-native substrates. Numerous genes encoding steroid-modifying enzymes, including some bacterial enzymes, were expressed in mammalian cells by transient transfection and found to be active. We made three unusual steroids by stable expression, in HEK293 cells, of the 7α-hydroxylase CYP7B1, which was selected because of its high native product yield. These cells made 7α,17α-dihydroxypregnenolone and 7β,17α-dihydroxypregnenolone from 17α-hydroxypregnenolone and produced 11α,16α-dihydroxyprogesterone from 16α-hydroxyprogesterone. The last two products were the result of CYP7B1-catalyzed hydroxylation at previously unobserved sites. A Rosetta docking model of CYP7B1 suggested that these substrates’ D-ring hydroxy groups might prevent them from binding in the same way as the native substrates, bringing different carbon atoms close to the active ferryl oxygen atom. This new approach could potentially use other enzymes and substrates to produce many novel steroids for drug candidate testing.

Full text of DOI:10.1002/cbic.201800214

Accepted Article
Title: Mammalian Cells Engineered to Produce Novel Steroids
Authors: Emma S. Spady, Thomas P. Wyche, Nathanael J. Rollins,
Jon Clardy, Jeffrey C. Way, and Pamela A. Silver
This manuscript has been accepted after peer review and appears as an
Accepted Article online prior to editing, proofing, and formal publication
of the final Version of Record (VoR). This work is currently citable by
using the Digital Object Identifier (DOI) given below. The VoR will be
published online in Early View as soon as possible and may be different
to this Accepted Article as a result of editing. Readers should obtain
the VoR from the journal website shown below when it is published
to ensure accuracy of information. The authors are responsible for the
content of this Accepted Article.
To be cited as: ChemBioChem 10.1002/cbic.201800214
Link to VoR: http://dx.doi.org/10.1002/cbic.201800214
A Journal of
www.chembiochem.org

ChemBioChem
10.1002/cbic.201800214
Mammalian Cells Engineered to Produce Novel Steroids
[
a,b,c]
[d]
[a,c]
[d]
Emma S. Spady
, Thomas P. Wyche , Nathanael J. Rollins , Jon Clardy , Jeffrey C.
Way[a,c], and Pamela A. Silver[a,c]
*
[
a] Department of Systems Biology, Harvard Medical School – Boston, MA 02115, United
States.
[
b] Laboratory of Systems Pharmacology, Harvard University – Boston, MA 02115, United
States.
[
0
c] Wyss Institute for Biologically Inspired Engineering, Harvard University – Boston, MA
2115, United States.
[
–
d] Department of Biological Chemistry and Molecular Pharmacology, Harvard Medical School
Boston, MA 02115, United States.
Author Information
Corresponding Author
*
E-mail: pamela_silver@hms.harvard.edu
Current Affiliations
Thomas P. Wyche: Merck & Co., Inc., Cambridge, MA 02141, United States.
1
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Abstract
Steroids can be difficult to modify via traditional organic synthesis methods, but many enzymes
regio- and stereo-selectively process a wide variety of steroid substrates. We tested whether
steroid-modifying enzymes could make novel steroids from non-native substrates. Numerous
genes encoding steroid-modifying enzymes, including some bacterial enzymes, were expressed
in mammalian cells by transient transfection and found to be active. We made three unusual
steroids by stable expression in HEK293 cells of the 7α-hydroxylase CYP7B1, which was
selected because of high native product yield. These cells made 7α,17α-dihydroxypregnenolone
and 7β,17α-dihydroxypregnenolone from 17α-hydroxypregnenolone, and produced 11α,16α-
dihydroxyprogesterone from 16α-hydroxyprogesterone. The latter two products resulted from
previously unobserved CYP7B1 hydroxylation sites. A Rosetta docking model of CYP7B1
suggested that these substrates’ D-ring hydroxylations may prevent them from binding in the
same way as the native substrate, bringing different carbons near the active ferryl oxygen. This
new approach could use other enzymes and substrates to produce many novel steroids for drug
candidate testing.
Keywords: Steroids, Biosynthesis, Synthetic Biology, Cytochromes
2
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Steroids represent an important class of therapeutically useful molecules, but they have
significant dose-limiting side effects. Steroid hormone analogs take advantage of the wide-
ranging effects of their endogenous counterparts, such as immunosuppression and increased bone
density, and are used to treat conditions ranging from arthritis to anemia. Unfortunately, steroid
receptors alter global responses, so steroid drugs have severe side effects. These include diabetes
and osteoporosis for glucocorticoids and masculinizing effects for androgens.[1–3] Steroids are
active at bloodstream concentrations in the nanomolar range and readily cross cell
membranes.[4,5] New combinations of modifications around the central steroid scaffold could
improve binding affinity, pharmacokinetic properties, and receptor specificity.[6]
Steroids must be modified stereo- and regio-specifically to result in useful drugs. Steroid
receptor binding pockets are sensitive to minor modifications; a single hydroxylation or
methylation can cause a steroid to bind a different receptor, dramatically changing its biological
[
6]
effects. Due to its lack of activated carbons, the steroid ring structure is difficult to modify
[
7]
while preserving its integrity. Traditional chemical processes for steroid modifications are
elaborate and frequently less selective than their biological counterparts, to the extent that wild-
type fungi are used as industrial biocatalysts.[7–9]
Biosynthesis can provide an alternative source of novel steroid therapeutics. The enzymes
involved in steroid hormone synthesis modify a large variety of substrates, but still perform
regio- and stereoselective reactions.[10] In principle, these enzymes could accept a wide variety of
synthetic steroid substrates and modify them to create novel products. A single enzymatic step
could replace a long organic synthesis, and thus allow for extensive exploration of the steroid
drug biochemical space. Exogenous expression of steroid enzymes has already been used to
produce complicated steroids intracellularly in yeast and mammalian cells, with glucocorticoids
3
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
being of particular interest.[11,12] However, intentional biosynthesis of chemically novel steroid
products using engineered mammalian cells has not been deeply explored.
Here we report the synthesis of novel steroids by expressing various steroid-modifying
enzymes in mammalian cells and providing them with non-native substrates. We isolated three
steroids by expressing the steroid 7α-hydroxylase CYP7B1, two of which arose from the
enzyme’s previously unknown ability to 11α- and 7β-hydroxylate substrates. This approach
could be used to synthesize a wide variety of steroids for which traditional organic synthesis is
not feasible.
Our strategy for novel steroid synthesis is to express genes encoding steroid-modifying
enzymes in mammalian cells, and then supply the cells with non-native substrates that will result
in novel products (Figure 1). To identify candidates, we expressed a panel of enzymes in
mammalian cells and confirmed their function on native substrates. We then selected one of the
best-performing enzymes, CYP7B1, and chose non-native substrates that would result in novel
products, assuming that the enzyme modifies the substrates selectively. We isolated the products
from the enzyme acting on the alternative substrate, and confirmed the steroids’ structures via
NMR. Notably, we observed the 7α-hydroxylase CYP7B1 performing significant amounts of
1
1α- and 7β-hydroxylation on the alternative substrates. We explained how these unexpected
products may have arisen with a Rosetta docking model for the substrates in the enzyme’s
binding pocket.
Mammalian cells were used in this work because of several unique properties. Firstly,
most of the enzymes tested were human, and cytochrome P450s can be difficult to express in
heterologous hosts.[12,13] Mammalian cells also enable these enzymes to access cholesterol when
required as a substrate. As there is no lipopolysaccharide in this system, any steroids in tissue
4
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
culture supernatant could be tested rapidly on other mammalian cell lines without purification.
While mammalian cells are unlikely to achieve the yields of yeast or bacterial cells, steroids are
very potent drugs, and thus can be studied even in small quantities.[14,15]
Seventeen steroid-modifying enzymes were selected for expression in the human cell line
HEK293. The enzymes were identified based on their gene sequence availability and the diverse
reactions they catalyze (Table S1). Most of these enzymes are cytochrome P450s, and many are
hydroxylases; the group also contains dehydrogenases, a methyltransferase, and a Baeyer-
Villiger monooxygenase. Mammalian steroid-modifying enzymes are generally associated with
the mitochondria or endoplasmic reticulum and may have signal sequences targeting them to
these organelles; these sequences were retained. Bacterial enzymes were expressed without
signal sequences and presumably localized to the cell cytoplasm. The HEK293 cell line was
chosen because of its hardiness and ease of transfection.
Expression constructs encoding each enzyme were transfected into HEK293 cells, and
activity of each enzyme was tested on a native substrate and assayed via LC-MS. Negative
enzyme activity controls consisted of transfecting the same plasmid backbone, but with a
fluorescent reporter protein under the constitutive promoter. Cells were incubated for two days
with a 40 μM solution of a native substrate for the relevant enzyme. The substrate concentration
was chosen because of the low solubility of the less-hydroxylated steroids in media.[16] Lipids
were extracted from the spent cells and media and measured by LC-MS.[17,18] The yield per
milliliter of spent media was determined for each product when a standard was available.
Of the enzymes tested, the human 7α-hydroxylase CYP7B1 and the Pseudomonas
cyclopentadecanone monooxygenase CpdB had the highest yields. CYP7B1 produced 1.4 μg of
-1
7
α-hydroxypregnenolone per mL (2) from a 12.7 μg mL pregnenolone solution (1) (Scheme
5
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
-
1
1
A). CpdB made 2.6 μg of testololactone per mL (10) from an 11.5 μg mL solution of
androstenedione (9) (Scheme S1). Eight other steroid-modifying enzyme plasmids resulted in
significantly more product than the fluorescent protein control when transfected (Table S1).
1
These included two more bacterial enzymes, Δ -KSTD2 and CYP154C3 (Scheme S1). Enzymes
that failed may have done so because of poor gene expression, but rather than troubleshooting,
we focused on those enzymes that worked well immediately. CYP7B1 was pursued further, as it
is a well-studied human enzyme.[19,20] The high 7α-pregnenolone (2) yield implied CYP7B1 had
a good chance of processing alternative substrates into detectable quantities of novel steroids.
The CYP7B1-expressing cells’ yield of 2 improved under different expression
conditions. This ensured that the enzyme’s action on alternative substrates would result in
sufficient product for structure determination. The CYP7B1 electron transfer partner,
cytochrome P450 oxidoreductase (POR), was expressed alongside CYP7B1 to prevent POR
from limiting steroid hydroxylation. A modest 34% increase in 2 yield was achieved by
transfecting CYP7B1 plasmid into stable POR-expressing HEK293 cells (Figure S1A). A
CYP7B1-expressing stable pool was constructed by puromycin selection to enable longer
exposure of cells to substrate. These cells could be plated at low density and grown to
confluency in the same media, allowing for eight days of incubation with substrate instead of 48
hours. HEK293 cells were transfected with the CYP7B1 vector used previously, and were
-1
selected on puromycin. This stable pool improved 2 yield by 229%, to 4.6 μg mL (Figure S1A).
The CYP7B1-expressing stable pool was used in subsequent steroid syntheses, and the POR
stable pool was not used.
Stable CYP7B1-expressing cells were incubated with two alternative substrates that we
predicted would result in novel products. We anticipated that CYP7B1 would convert 17α-
6
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
hydroxypregnenolone (3) to 7α,17α-dihydroxypregnenolone (4) (Scheme 1B). Similarly,
CYP7B1 was expected to hydroxylate 16α-hydroxyprogesterone (7) to produce 7α,16α-
dihydroxyprogesterone (6) (Scheme 1C). The CYP7B1 stable pool cell line was plated at low
density and cultivated in the presence of a 40 μM alternative substrate solution for eight days.
-1
-1
This corresponds to a 13.3 μg mL solution of 3 and a 13.2 μg mL solution of 7. Lipids were
extracted and detected by LC-MS as in the native substrate experiments. The lipids were
compared to those from identically treated negative control stable pool cells, which expressed
GFP instead of an enzyme. The desired hydroxylation product peaks were identified by having
the correct mass and only being present with CYP7B1 expression. This separated the desired
steroids from any hydroxylated isomers due to spontaneous degradation.
Peaks with masses corresponding to hydroxylation of the alternative substrates were
detected exclusively in CYP7B1-expressing cells. CYP7B1 cells provided with alternative
substrate 3 yielded two dihydroxypregnenolone peaks not present in the control (Figure 2A). As
dihydroxypregnenolones lose water during ESI-MS, the product peaks contained m/z =
+
+
3
3
31.2269, 313.2162, and 295.2057, corresponding to [M+H-H
2
O] , [M+H-2H O] , and [M+H-
2
+
H
2
O] ions (Supplementary Results). The later, high signal molecule 4 eluted at 2.7 minutes,
and the earlier, low signal molecule 5 eluted at 2.3 minutes. 0.7 mg of 4 and 0.4 mg of 5 were
-1
-1
subsequently purified, implying a yield of 1.6 μg mL and 0.91 μg mL , respectively.
Analysis of 1D and 2D NMR data (gCOSY, gHSQC, gHMBC, ROESY, and TOCSY)
allowed for structure determination of dihydroxypregnenolones 4 and 5, corresponding to an
expected 7α-hydroxylated product and an unexpected 7β-hydroxylated product (Scheme 1B,
Table S2). COSY interactions between the hydrogen at C-6 (δ 5.57, 5.28) and a hydroxyl-
H
adjacent hydrogen (δ
H
3.8, 3.79) in 4 and 5, respectively, revealed that both products were 7-
7
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
hydroxylated. The C-7 hydroxyl stereochemistry was determined by comparing the isomers to
reference NMR for 7α-hydroxypregnenolone (δC-7 65.20) and 7β-hydroxypregnenolone (δC-7
7
7
3.14).[21,22] Compound 4 (δC-7 65.95) is therefore 7α,17α-dihydroxypregnenolone while 5 (δC-7
3.99) is 7β,17α-dihydroxypregnenolone.
The CYP7B1 stable pool cells provided with alternative substrate 7 exhibited a single
dihydroxyprogesterone product peak (Figure 2B). Dihydroxyprogesterones do not lose water as
easily in ESI-MS, so the product peak was exclusively m/z = 347.2222, corresponding to the
+
[
M+H] ion (Supplementary Results). A single isomer 8 with this m/z eluted at 2.05 minutes, and
was absent in the negative control lipids. 2.2 mg of 8 were eventually purified from tissue
-1
culture, with a yield of 6.9 μg mL .
Compound 8, the single dihydroxyprogesterone product, was determined to be 11α,16α-
dihydroxyprogesterone after analysis of 1D and 2D NMR data (Scheme 1C, Table S2). No other
masses that could correspond to the expected product 6 were detected. The newly hydroxylated
carbon (δ
.24), with which it has a strong COSY correlation. This neighboring carbon has an HMBC with
hydrogens at C-17 (δ 2.58) and C-18 (δ 0.70). This placed the neighboring carbon at C-12, as
C
68.88, δ
H
3.95) was identified by locating its methylene neighbor (δ
C
50.89, δ
H
1.63,
2
H
H
the D-ring hydroxyl had already been assigned and other locations were too far from C-17 and
C-18. Therefore the CYP7B1 hydroxylation was at C-11. The C-11 hydroxyl stereochemistry
was found through ROESY interactions. The hydrogen at C-11 (δ
H
3.95) interacts with
hydrogens at C-18 (δ 0.70) and C-19 (δ 1.35), indicating that it is above the steroid ring plane.
H
H
The hydroxyl at C-11 must face the opposite direction, and thus is an 11α-hydroxyl.
The macromolecular modeling software Rosetta was used to model native substrate 1 and
alternative substrates 3 and 7 binding to CYP7B1 to better understand how the unexpected
8
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
products 5 and 8 could arise. Predictive modeling of cytochrome P450 reaction sites is
difficult;[23] our models instead aim to identify why we observe different reactions for D-ring
hydroxylated substrates than for the native substrate.[20] As there are no CYP7B1 crystal
structures, we used Rosetta to generate structure models before docking the substrates.
Comparative modeling in Rosetta predicts protein structure based on solved structures of
homologs, from which it generates candidate structures.[24] The structure of CYP7A1, which is
4
1% sequence identical to CYP7B1 with 95% coverage, has been solved and served as a starting
point for comparative modeling.[25] Binding modes were generated and those with the lowest
energy were identified by Rosetta’s ΔG analog score, resulting in multiple lowest energy modes
with inseparable energy scores for each substrate. The reactive binding mode was found among
the best predictions by proximity of observed hydroxylation position to the ferryl oxygen (see
Experimental Section). These were analyzed as the best hypotheses for how CYP7B1 catalyzes
7
α-, 7β- and 11α-hydroxylation on the steroids studied (Figure 3).
These structure models suggest that unexpected CYP7B1 activities on 3 and 7 arise from
these substrates shifting out of a tight pocket and tilting relative to the reactive ferryl oxygen.
Cytochrome P450s typically hydroxylate the carbon closest to the ferryl oxygen; the mechanism
does not require any activation of the substrate, though the transient carbanion intermediate
formed favors carbons adjacent to double bonds.[26] As C-7 is adjacent to a double bond in native
substrate 1 and the alternative substrate 3, it is a particularly good hydroxylation site, and thus
observed in their products (Figure 3BC). However, 3’s 17α-hydroxyl does not permit it to
occupy the same binding position as 1 in our model. The hydroxyl forces 3 into a wider channel
to the left in the figure, where the steroid can tilt to result 7β-hydroxylation (Figure 3D). 7’s 16α-
hydroxylation would also result in a steric clash, which implies it will also shift out of 1’s
9
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
binding pocket. The severity of this clash may drive 7 to rotate dramatically, resulting in the
observed 11α-hydroxylation (Figure 3E).
Overall, the hydroxyls’ steric clashes appear to drive steroid positioning more than their
potential for hydrogen bonding. The D-ring hydroxyls of 3 and 7 do not form hydrogen bonds in
any of the model structures, but nevertheless change these substrates’ binding modes. The side
view of CYP7B1’s native substrate, pregnenolone (1), shows its tight fit in a hydrophobic
binding pocket with little room beneath the D-ring (Figure 3F). This fit may hold the substrate
steady to ensure hydroxylation exclusively at position 7α. Functional groups at C-16 and C-17,
such as those in 3 and 7, can disrupt this fit to cause the observed loss of reaction specificity.
We made the three steroids 4, 5, and 8 by expressing CYP7B1 in mammalian cells and
providing it with non-native substrates 3 and 7. 5 and 8 were the result of 7β- and 11α-
hydroxylations, respectively, though CYP7B1 was thought to primarily perform 7α-
hydroxylation. Docking substrates 3 and 7 into the enzyme active site suggests that these
unexpected reactions arise from the substrates’ D-ring hydroxylations pushing them into a tilted
orientation. This work represents a proof of concept in using genetically engineered mammalian
cells to produce novel steroids, and more broadly in using mammalian cells to produce novel
small molecules. Our overall method can take advantage of inexpensive gene synthesis, diverse
steroid-modifying enzymes, and mammalian cells’ unique properties to quickly create steroids
that are compatible with direct biological testing.
Steroids 4 and 5 are novel molecules, but 8 has a straightforward organic synthesis.
Nevertheless, 8’s chemical properties are not published; no NMR structural data were released
when it was initially isolated.[27] 4 and 5 would have been difficult to produce via established
1
0
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
methods. Both have been tentatively identified in mixtures of microbial steroid hydroxylation
products, but no NMR data is available.[28,29] Selective 7β-hydroxylation of pregnenolones is
possible with organic methods, but syntheses that directly favor 7α-hydroxylation do not appear
possible.[30] Nevertheless, C-7 is a relatively easy carbon to chemically oxidize, as it is adjacent
to a double bond.[31] While microbial pregnenolone 7α-hydroxylation has been described, the
organisms make a mixture of products, including non-7-hydroxylated steroids, and the enzymes
have not been identified.[32,21,33] Hence our method for 7-hydroxylation fills a gap in steroid
biosynthetic methods.
We observed unprecedented amounts of 11α-hydroxylated and 7β-hydroxylated products
from CYP7B1. Prior to this work, CYP7B1 only performed 7α-hydroxylation, 6α-hydroxylation,
or 7β-hydroxylation, depending on the substrate.[34] The lattermost product is minor, when
observed at all; it comprised just 1% of the hydroxysteroid product from CYP7B1 action on
DHEA.[35] In comparison, 5 was 36% of the total product isolated from CYP7B1 acting on 3.
1
1α-hydroxylation by CYP7B1 had never been observed until we isolated 8. These reactions
demonstrate that even well-studied cytochrome P450s may process new substrates in unexpected
ways; one cannot assume which isomers these enzymes produce based only on the available
literature.
Our approach reveals that novel steroids can be generated by combining elements from
the human metabolic repertoire. CYP7B1 is a human enzyme; its murine homolog is expressed
primarily in the brain and at low levels in the liver.[36] Its 7α-hydroxylated steroid products are
poorly understood, but improve memory in aged, memory-impaired mice.[34,37] While 7 is not
present in the human body, 3 is an intermediate formed by CYP17A1 in the adrenal cortex and
gonads.[38] However, human CYP17A1 preferentially processes 3 further into
1
1
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
dehydroepiandrosterone.[39] Though the enzyme dissociates from the steroid after hydroxylation,
relatively little 3 can escape into other tissues before CYP17A1 acts again.[40] While CYP7B1
likely does not encounter 3 in nature, in our system the enzyme is capable of processing this
substrate into novel products 4 and 5.
Here, we engineered mammalian cells to produce previously uncharacterized drug-like
small molecules. Numerous protein drugs are already made in mammalian cells, typically due to
species-specific post-translational modifications such as N-linked glycosylation. Mammalian
cells and mammal-based cell-free systems have been engineered to make steroids, but these were
only used to study the biosynthetic pathways of known molecules.[11,41] Engineered steroid
biosynthesis, even in microbes, has been used to make known steroids; novel steroids are
typically detected spontaneously from unmodified cells.[9,12,32,42] Many novel molecules could be
made with engineered cells expressing steroid-modifying enzymes, making more steroids
accessible for biological testing. Our in vivo system could be modified to access all parts of the
steroid ring system, unlike current organic synthesis methods. This work therefore forms a basis
for directed biosynthesis of novel steroids.
1
2
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Experimental Section
Plasmids. The CYP7B1-containing plasmid had a pcDNA3.1 backbone, a CMV promoter
driving intron-free CYP7B1 and puromyin resistance. This pcDNA3.1 backbone from Invitrogen
includes a pUC origin and ampicillin resistance to enable maintenance in Escherichia coli. The
human CYP7B1 isoform 1 cDNA sequence was from the NCBI CCDS database number 6180.1.
The negative control plasmid was identical to the CYP7B1 plasmid, but contained GFP in place
of CYP7B1. This codon-optimized GFP sequence was originally from the pSF-CMV-Ub-daGFP
plasmid (Oxford Genetics). A list of all steroid-modifying enzymes used in similar plasmids can
be found in Table S1. Plasmids were constructed using Gibson assembly to insert IDT gBlocks
containing the gene of interest into PCR-amplified plasmid backbones, which were then
transformed into chemically-competent E. coli K12. Plasmids were prepared using the Qiagen
PlasmidPlus MidiPrep or MaxiPrep kit.
Protein Expression in Mammalian Cells. HEK293 cells (ATCC) were grown in DMEM
(
Invitrogen) with 10% FBS (Gibco) and penicillin-streptomycin (Gibco). Cells were transfected
with 31.5 μg of plasmid DNA per well in 6-well plates using the Lipofectamine 3000 kit
Thermo Fisher). Media was replaced after six hours with fresh media containing 40 μM steroid
(
-1
substrate. For concentrations of each substrate in ng mL , refer to Table S1. Cells were
incubated for 48 hours, after which the negative controls were checked for fluorescence,
indicating acceptable transfection efficiency. Spent media and cells were mixed by scraping, and
the liquid was frozen for storage. Stable pools of HEK293 cells expressing steroid-modifying
genes began with transfection of ~800,000 cells using the lipofectamine 3000 kit, as above. 48
-1
hours after transfection, cells were selected via 1.5 μg mL puromycin (Sigma). After eighteen
-1
days, antibiotic was decreased to 0.5 μg mL puromycin. To modify steroids, stable pool cells
1
3
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
2
were plated at 7,000 cells per cm with 40 μM of the desired substrate. The cells were incubated
for eight days, after which adhered cells were scraped to mix with the media and frozen for
storage.
Steroid Product Isolation. Liquid-liquid separation with a 3:10 methanol : methyl-tert-butyl
ether v/v organic phase removed lipids from the media[17]. The media (2 mL) was extracted twice
with solvent (6.5 mL and 3 mL), vortexing for three minutes each time. The upper layers were
removed, combined, dried down, and resuspended in methanol (100 μL). Lipids were analyzed
with an Agilent 1200 series HPLC system and 6530 qTOF mass spectrometer. Steroids were
separated with a Thermo Scientific Hypersil GOLD C18 column (1.9 μm, 50 x 2.1 mm) in an
acetonitrile gradient in water with formic acid (0.1% v/v)[18]. Known products were compared to
standards for quantification. Please refer to the supporting information for HPLC gradients,
steroid sources, and chromatograms of steroid products with standards.
Unknown Steroid Purification and NMR. Twenty-two T150 flasks of the CYP7B1-expressing
cells were used to process substrate 3, yielding 440 mL of media. Sixteen similar flasks were
used to process substrate 7, resulting in 320 mL of media. Liquid-liquid extraction was
performed with the same proportions as above, with the addition of a sodium sulfate drying step.
The product mixtures from 3 and 7 were resuspended in 3 and 5 mL of methanol, respectively.
An Agilent 1200 semi-preparative HPLC system with a Phenomenex Luna C18 column (5 μm,
2
0
5
50 x 10 mm) was used to purify the steroid products 4, 5, and 8. This resulted in 0.7 mg of 4,
.4 mg of 5, and 2.2 mg of 8. NMR spectra were obtained in CD OD with a Bruker AVANCE
00 MHz spectrometer equipped with a H{ C/ N} cryoprobe and a Bruker AVANCE 500
3
1
13 15
1
3
15
1
13
1
MHz spectrometer equipped with a C/ N{ H} cryoprobe. C and H shifts and assignments
1
4
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
1
3
1
are in Table S2. C, H, HSQC, HMBC, COSY, ROESY, and TOCSY spectra, along with
HPLC methods, can be found in the supporting information.
Rosetta Steroid Docking Models. We generated comparative models for CYP7B1 binding
steroids based on crystallized enzymes with Rosetta[44]. The algorithm selects for structures with
optimal physical interactions, solvent accessible surface area, and bond angles by assigning
structures a Rosetta score; lower Rosetta Energy Unit (REU) values suggest more energetically
favorable conformations[44]. The two solved structures within 30% sequence identity and 10-4
BLAST confidence, and bound to chemically similar substrates, were PDB IDs 3sn5 and
3
v8d[25]. We threaded the sequence of CYP7B1 onto the coordinates of those structures and
combined fragments of each to create ten thousand hybrid models of CYP7B1[24,44]. The heme
was inserted according to the original crystals’ bond angles and obabel-derived partial charges;
this was followed by all-atom optimization[45]. ~2500 reasonable structures were identified by
their Rosetta score falling within fifty REU of the lowest-scoring structure, and were clustered at
3
Å RMSD.
An ensemble of the twenty best-scoring CYP7B1 conformations were used in docking 1,
, and 7, generating ten thousand binding predictions for each steroid. Typical parameters for
3
these RosettaLigand docking runs were used: the steroid was randomly placed within 5 Å of the
binding pocket center, coarsely fit inside, and accommodated by sidechains within 5 Å[24]
.
Energetically feasible binding modes were identified by their overall Rosetta score within 65
REU of the best mode by overall score (to account for variations between the ensemble models),
and their ΔG-binding analog score within 4 REU of the best mode (analogous to the score
contribution of one hydrogen bond). These structures were clustered at 3 Å RMSD between
substrate atoms, and the cluster representative was chosen by the lowest ΔG-binding analog
1
5
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
score. This resulted in approximately seventy energetically comparable structures per steroid. As
it was not possible to determine the best binding mode by energy alone, other information was
incorporated to identify the modes shown. We selected conformations that placed the
hydroxylation site within 3.5 Å of the ferryl oxygen, as these are the only modes capable of
mediating the observed reactions. We avoided conformations that would require unlikely
substrate trajectories into the active site, instead choosing those that resembled the orientations of
substrates in the homologous crystals. Thus we produced a binding mode for each hydroxylation
mode for each substrate, for a total of four models. Please refer to the supporting information for
structure model files and clustering details.
1
6
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Acknowledgements
This work was supported by the Defense Advanced Research Projects Agency HR0011-12-C-
061 and the National Institute of Health P50 GM107618 (Laboratory of Systems
0
Pharmacology) and R01 GM086258 (JC). We are grateful to the ICCB-Longwood Analytical
Chemistry Core facility at Harvard Medical School for use of their mass spectrometer. We would
also like to thank S. Trauger, K. Chatman, and G. Byrd at the Harvard Small Molecule Mass
Spectrometry Core.
Supporting Information Available: Yield tables for all enzymes, descriptions of conditions to
improve yield, supplementary methods, and NMR shifts, assignments, and spectra (PDF).
CYP7B1 substrate binding predictions from Rosetta (ZIP). Intermediate structures used in
CYP7B1 modeling (ZIP). This material is available free of charge via the Internet.
The authors declare no competing financial interest.
1
7
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
References
[
[
1] M. Kadmiel, J. A. Cidlowski, Trends Pharmacol. Sci. 2013, 34, 518–530.
2] E. Sarnes, L. Crofford, M. Watson, G. Dennis, H. Kan, D. Bass, Clin. Ther. 2011, 33,
1
413–1432.
[
[
[
3] B. Haendler, A. Cleve, Mol. Cell. Endocrinol. 2012, 352, 79–91.
4] D. E. Mager, N. Moledina, W. J. Jusko, J. Pharm. Sci. 2003, 92, 1521–1525.
5] P. Keski-Rahkonen, K. Huhtinen, M. Poutanen, S. Auriola, J. Steroid Biochem. Mol. Biol.
2
011, 127, 396–404.
[
[
[
6] P. Huang, V. Chandra, F. Rastinejad, Annu. Rev. Physiol. 2010, 72, 247–272.
7] H. Pellissier, M. Santelli, Org. Prep. Proced. Int. 2001, 33, 1–58.
8] P. Fernandes, A. Cruz, B. Angelova, H. M. Pinheiro, J. M. S. Cabral, Enzyme Microb.
Technol. 2003, 32, 688–705.
[
[
[
9] A. Swizdor, T. Kolek, A. Panek, N. Milecka, Curr. Org. Chem. 2012, 16, 2551–2582.
10] W. D. Nes, Chem. Rev. 2011, 111, 6423–51.
11] P. A. Mathew, J. I. Mason, J. M. Trant, M. R. Waterman, Mol. Cell. Endocrinol. 1990, 73,
7
3–80.
[
12] F. M. Szczebara, C. Chandelier, C. Villeret, A. Masurel, S. Bourot, C. Duport, S.
Blanchard, A. Groisillier, E. Testet, P. Costaglioli, et al., Nat. Biotechnol. 2003, 21, 143–
1
49.
[
[
[
[
[
13] A. Parikh, E. M. Gillam, F. P. Guengerich, Nat. Biotechnol. 1997, 15, 784–788.
14] A. Gerber, M. Milhim, P. Hartz, J. Zapp, R. Bernhardt, Metab. Eng. 2016, 36, 19–27.
15] T. Wriessnegger, H. Pichler, Prog. Lipid Res. 2013, 52, 277–293.
16] P. Kabasakalian, E. Britt, M. D. Yudis, J. Pharm. Sci. 1966, 55, 642–642.
17] V. Matyash, G. Liebisch, T. V Kurzchalia, A. Shevchenko, D. Schwudke, J. Lipid Res.
2
008, 49, 1137–46.
[
[
18] I. Surowiec, M. Koc, H. Antti, P. Wikström, T. Moritz, J. Sep. Sci. 2011, 34, 2650–2658.
19] A. R. Stiles, J. G. McDonald, D. R. Bauman, D. W. Russell, J. Biol. Chem. 2009, 284,
2
8485–28489.
[
[
[
20] A. V. Yantsevich, Y. V. Dichenko, F. Mackenzie, D. V. Mukha, A. V. Baranovsky, A. A.
Gilep, S. A. Usanov, N. V. Strushkevich, FEBS J. 2014, 281, 1700–1713.
21] Y. Wang, D. Sun, Z. Chen, H. Ruan, W. Ge, Biocatal. Biotransformation 2013, 31, 168–
1
74.
22] D. Schmitz, J. Zapp, R. Bernhardt, Microb. Cell Fact. 2014, 13, 81.
1
8
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
[
[
23] L. Olsen, C. Oostenbrink, F. S. Jørgensen, Adv. Drug Deliv. Rev. 2015, 86, 61–71.
24] S. A. Combs, S. L. Deluca, S. H. Deluca, G. H. Lemmon, D. P. Nannemann, E. D.
Nguyen, J. R. Willis, J. H. Sheehan, J. Meiler, Nat. Protoc. 2013, 8, 1277–1298.
[
[
[
[
25] W. Tempel, I. Grabovec, F. MacKenzie, Y. V. Dichenko, S. A. Usanov, A. A. Gilep, H.-
W. Park, N. Strushkevich, J. Lipid Res. 2014, 55, 1925–1932.
26] P. R. Ortiz De Montellano, J. J. De Voss, Substrate Oxidation by Cytochrome P450
Enzymes, 2005.
27] N. S. Wulfson, V. I. Zaretskii, V. L. Sadovskaya, A. V. Semenovsky, W. A. Smit, V. F.
Kucherov, Tetrahedron 1966, 22, 603–614.
28] V. A. Andrushina, A. V. Druzhinina, V. V. Yaderets, T. S. Stitsenko, N. E. Voishvillo,
Appl. Biochem. Microbiol. 2011, 47, 42–48.
[
[
29] O. Schaaf, K. Dettner, J. Steroid Biochem. Mol. Biol. 2000, 75, 187–199.
30] G. O. Schenck, O. A. Neumuller, W. Eisfeld, European J. Org. Chem. 1958, 579, 202–
2
10.
[
[
[
[
[
[
31] E. J. Horn, B. R. Rosen, Y. Chen, J. Tang, K. Chen, M. D. Eastgate, P. S. Baran, Nature
2
016, 533, 77–81.
32] L. H. Shan, H. M. Liu, K. X. Huang, G. F. Dai, C. Cao, R. J. Dong, Bioorganic Med.
Chem. Lett. 2009, 19, 6637–6639.
33] M. I. Choudhary, I. Batool, S. A. A. Shah, S. A. Nawaz, Atta-ur-Rahman, Chem. Pharm.
Bull. (Tokyo). 2005, 53, 1455–1459.
34] K. Rose, A. Allan, S. Gauldie, G. Stapleton, L. Dobbie, K. Dott, C. Martin, L. Wang, E.
Hedlund, J. R. Seckl, et al., J. Biol. Chem. 2001, 276, 23937–23944.
35] K. A. Rose, G. Stapleton, K. Dott, M. P. Kieny, R. Best, M. Schwarz, D. W. Russell, I.
Björkhem, J. Seckl, R. Lathe, Proc. Natl. Acad. Sci. U. S. A. 1997, 94, 4925–4930.
36] G. Stapleton, M. Steel, M. Richardson, J. O. Mason, K. A. Rose, R. G. M. Morris, R.
Lathe, J. Biol. Chem. 1995, 270, 29739–29745.
[
[
37] J. L. W. Yau, J. Noble, M. Graham, J. R. Seckl, J. Neurosci. 2006, 26, 11034–11040.
38] B. C. Chung, J. Picado-Leonard, M. Haniu, M. Bienkowski, P. F. Hall, J. E. Shively, W.
L. Miller, Proc. Natl. Acad. Sci. U. S. A. 1987, 84, 407–11.
[
39] E. M. Petrunak, N. M. DeVore, P. R. Porubsky, E. E. Scott, J. Biol. Chem. 2014, 289,
3
2952–32964.
[
[
[
[
40] E. Gonzalez, F. P. Guengerich, J. Biol. Chem. 2017, 292, 13168–13185.
41] S. Kominami, N. Nishida, S. Takemori, Biochim. Biophys. Acta 1996, 1301, 199–206.
42] C. Duport, R. Spagnoli, E. Degryse, D. Pompon, Nat. Biotechnol. 1998, 16, 186–189.
43] T. R. Kau, F. Schroeder, S. Ramaswamy, C. L. Wojciechowski, J. J. Zhao, T. M. Roberts,
1
9
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
J. Clardy, W. R. Sellers, P. A. Silver, Cancer Cell 2003, 4, 463–476.
44] Y. Song, F. Dimaio, R. Y. R. Wang, D. Kim, C. Miles, T. Brunette, J. Thompson, D.
[
[
Baker, Structure 2013, 21, 1735–1742.
45] N. M. O’Boyle, M. Banck, C. A. James, C. Morley, T. Vandermeersch, G. R. Hutchison,
J. Cheminform. 2011, 3, 1–14.
2
0
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Figures
Figure 1. Overview of our method for novel steroid biosynthesis. Steroid-modifying enzymes
are expressed heterologously in mammalian cells and exposed to a non-native substrate, resulting
in novel steroid product(s).
2
1
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Scheme 1. Steroid-modifying reactions observed from CYP7B1. A) CYP7B1 hydroxylates
native substrate pregnenolone (1) to form 7α-hydroxypregnenolone (2). Yield is in μg of steroid
per mL of media extracted. B) CYP7B1 hydroxylates non-native substrate 17α-
hydroxypregnenolone (3) to form novel product isomers 7α,17α-dihydroxypregnenolone (4) and
7
β,17α-dihydroxypregnenolone (5). C) CYP7B1 hydroxylates non-native substrate 16α-
hydroxyprogesterone (7) to form 11α,16α-dihydroxyprogesterone (8). The anticipated product
α,16α-dihydroxyprogesterone (6) was not observed.
7
2
2
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Figure 2. Extracted ion mass chromatograms comparing products of cells supplied with 3 and 7.
CYP7B1 cells stably expressed that enzyme, while the negative control cells stably expressed
green fluorescent protein. CYP7B1 cell samples were diluted tenfold. A) Only CYP7B1 cells
+
+
exposed to 3 produced peaks corresponding to the [M+H-H
2
O] , [M+H-2H
2
O] , and [M+H-
+
3
H
2
O] ions of 4 and 5. B) Only CYP7B1 cells exposed to 7 produced a single peak
+
corresponding to the [M+H] ion of 8.
2
3
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
Figure 3. Predicted structures of the CYP7B1 active site with steroid substrates 1 (blue), 3
orange), and 7 (green). A) Cartoon representation of holo-CYP7B1 with 1. B-E) Structural
models of CYP7B1 permitting 7α-hydroxylation of 1 (B) and 3 (C), 7β-hydroxylation of 3 (D),
(
2
4
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
and 11α-hydroxylation of 7 (E). The red dashed line highlights the positions of the reactive ferryl
oxygen and the observed hydroxylation site. F) Side view of native substrate 1 in active site.
Yellow dashed lines indicate the proximity of steroid carbons C-16 and C-17 to the edge of the
binding pocket. Distances are in Angstroms.
2
5
This article is protected by copyright. All rights reserved.

ChemBioChem
10.1002/cbic.201800214
2
6
This article is protected by copyright. All rights reserved.