ARTICLE doi:10.

1038/nature14953

Structure and mechanism of an active

lipid-linked oligosaccharide flippase
Camilo Perez1, Sabina Gerber1{, Jérémy Boilevin2, Monika Bucher1, Tamis Darbre2, Markus Aebi3,
Jean-Louis Reymond2 & Kaspar P. Locher1

The flipping of membrane-embedded lipids containing large, polar head groups is slow and energetically unfavourable,
and is therefore catalysed by flippases, the mechanisms of which are unknown. A prominent example of a flipping
reaction is the translocation of lipid-linked oligosaccharides that serve as donors in N-linked protein glycosylation. In
Campylobacter jejuni, this process is catalysed by the ABC transporter PglK. Here we present a mechanism of
PglK-catalysed lipid-linked oligosaccharide flipping based on crystal structures in distinct states, a newly devised in
vitro flipping assay, and in vivo studies. PglK can adopt inward- and outward-facing conformations in vitro, but only
outward-facing states are required for flipping. While the pyrophosphate-oligosaccharide head group of lipid-linked
oligosaccharides enters the translocation cavity and interacts with positively charged side chains, the lipidic polyprenyl
tail binds and activates the transporter but remains exposed to the lipid bilayer during the reaction. The proposed
mechanism is distinct from the classical alternating-access model applied to other transporters.

The translocation of lipids across the bilayer, termed flipping, is not hydrolysed to the monophosphate by undecaprenyl-pyrophosph-
only essential for maintaining lipid asymmetry in membranes, but atase22 (Extended Data Fig. 2).
also underpins various processes including signalling, vesicle forma- To understand the mechanism of ATP-driven LLO flipping, we
tion in secretory and endocytic pathways, the regulation of membrane have developed an in vitro assay and determined the crystal structures
protein activity, and asparagine-linked protein glycosylation1–5. In of C. jejuni PglK in distinct states. Our results suggest a mechanism in
animals, trans-bilayer lipid asymmetry affects processes such as blood which only the outward-facing conformations are relevant for flip-
coagulation, macrophage recognition6 and apoptosis7. In bacteria, ping, thereby allowing the passage of the oligosaccharide moiety
precursors of cell wall components are coupled to lipid carriers and through the transporter while the polyprenyl tail remains partially
exported by flippases, which is essential for cell wall assembly8–10. embedded in the lipid bilayer but attached to the surface of PglK.
Flipping reactions are catalysed by passive or active transporter This newly proposed mechanism is distinct from the alternating-
proteins11. These include energy-independent scramblases that access model that underlies the mechanism of most primary and
randomize the orientation of lipids across the bilayer12,13, proton- or secondary transporters studied so far.
sodium-driven secondary antiporters14, and ATP-driven, vectorial
transporters that actively translocate specific lipids. The latter include PglK in vitro LLO flipping and ATPase activity
P4-ATPases15 and ATP-binding cassette (ABC) transporters11,16. Flipping of polyprenyl-linked oligosaccharides has previously been
Despite extensive research, remarkably little is known about the investigated using radiolabelled natural lipids23,24. By contrast, we
structure and mechanism of active, flippase-catalysed lipid transloca- exploited the fact that PglK can process LLOs with shortened oligosac-
tion. Until now, crystal structures have only been reported for the charides both in vivo and in vitro20,25,26. We therefore generated pro-
MsbA protein16,17, the bacterial flippase of lipid A-core, but no mech- teoliposomes by co-reconstituting PglK and a trisaccharide LLO
anism could be deduced from the structural data. In addition to the (tLLO), which contained a reducing-end bacillosamine and two
scarcity of structures, there are few reliable in vitro assays that allow N-acetylgalactosamine (GalNAc) moieties (Fig. 1a and Extended
the study of native substrate flipping. Data Fig. 1). To determine the rate of ATP-driven, PglK-dependent
A prominent substrate of flippases in eukaryotes is the lipid-linked- tLLO flipping, the amount of tLLO molecules in the outer, accessible
oligosaccharide (LLO) Man5GlcNAc2-PP-dolichol, which is translo- leaflet of the proteoliposomes was quantified by radiolabelling using
cated from the cytosolic to the luminal side of the endoplasmic purified glycosyltransferase PglH27–29, an enzyme of the C. jejuni
reticulum membrane, where the oligosaccharide is transferred to N-glycosylation machinery that was shown to attach up to three
asparagine residues of nascent polypeptide chains in protein GalNAc residues to the non-reducing end of tLLO (Extended Data
N-glycosylation1,2. In the homologous bacterial N-glycosylation path- Fig. 2). We used excess UDP-GalNAc, thereby forcing three GalNAc
way, a chemically similar LLO (GlcGalNAc5Bac-PP-undecaprenyl; molecules to be added to each tLLO molecule, allowing accurate deter-
Extended Data Fig. 1) is flipped in the human pathogen C. jejuni18,19. mination of tLLO flipping to the liposome lumen (Fig. 1b and Extended
The responsible flippase, PglK, is a homodimeric ABC transporter Data Fig. 9a). The conversion of tLLO to hexa-saccharide LLO was
and an essential component of the bacterial protein N-glycosylation confirmed by transferring the resulting oligosaccharides to fluores-
machinery20. The translocated LLO is handed over to the oligosac- cently labelled acceptor peptides using purified oligosaccharyltransfer-
charyltransferase PglB, which catalyses oligosaccharide transfer to ase PglB, followed by SDS–PAGE analysis26,30,31. PglK-catalysed in vitro
acceptor proteins21. Undecaprenyl-pyrophosophate is subsequently tLLO flipping was strictly dependent on ATP hydrolysis. A PglK
1
Institute of Molecular Biology and Biophysics, ETH Zürich, CH-8093 Zürich, Switzerland. 2Department of Chemistry and Biochemistry, University of Berne, CH-3012 Berne, Switzerland. 3Institute of
Microbiology, ETH Zürich, CH-8093 Zürich, Switzerland. {Present address: GlycoVaxyn AG, Grabenstrasse 3, 8952 Schlieren, Switzerland.

2 7 AU G U S T 2 0 1 5 | VO L 5 2 4 | N AT U R E | 4 3 3
G2015 Macmillan Publishers Limited. All rights reserved
 RESEARCH ARTICLE

a b
WT –ATP Liposomes (tLLO)
E510Q +ADP
P P P
P P P
3. PglH 100

Normalized flipping (%)

1. ATP
PP PP PP
2. ADP [3H]-UDP-GalNAc
c 90

PP PP P
P * PP P
P
80

P
70

***
P
60

0 10 20 30 40
ATP ADP+Pi Time (min)

Figure 1 | In vitro LLO flipping assay. a, Schematic of designed assay. acetylgalactosamine. PglK activity is inhibited by the addition of excess ADP,
Proteoliposomes contain PglK homodimer (subunits in orange and grey) and after which proteoliposomes are incubated with purified PglH and 3H-UDP-
tLLO, both in 50in:50out orientations (see Methods and Extended Data Fig. 9c, GalNAc to label and quantitate tLLO remaining in the external leaflet. b, tLLO
d). Owing to the external addition of ATP, only inside-out transporters flipping of wild-type (WT) PglK and the E510Q mutant. Controls include
catalyse tLLO flipping. The undecaprenyl tail of tLLO is shown in purple. empty liposomes containing only tLLO (liposomes), the constant presence of 4
Red hexagon denotes di-N-acetylbacillosamine; yellow square denotes mM ADP (1ADP), and the absence of ATP (2ATP). Error bars denote s.d.
N-acetylgalactosamine; and yellow square with an asterisk denotes 3H-N- (n 5 3).

mutant containing a glutamine residue in the Walker-B motif Structures of PglK in distinct states
(E510Q) exhibited an almost 100-fold decrease in the flipping rate, We determined three crystal structures of the PglK mutant E510Q
which correlated with strongly reduced in vivo activity (Extended at resolutions ranging from 2.9 Å to 5.9 Å (Fig. 2a, Extended
Data Fig. 3a–c). The ATPase activity of PglK in the presence of Data Fig. 4 and Extended Data Table 1). Two structures of the
LLO follows Michaelis–Menten kinetics, with a Km value of apo protein were determined at higher resolution and revealed
0.46 6 0.08 mM (mean 6 s.d.), matching the Km of ATP-dependent distinct inward-facing conformations (designated apo-inward-1
tLLO flipping of 0.36 6 0.03 mM (Extended Data Fig. 3d). This cor- and -2), whereas the third was of ADP-bound PglK, resulting
roborates the interpretation that ATP hydrolysis is required and in an unprecedented, outward-occluded conformation. Given the
driving PglK-catalysed LLO flipping. As was observed with other limited resolution of the latter, model building was helped by a
ABC transporters, binding of native substrate can stimulate the previous finding35 that transmembrane helices of ABC exporters
ATPase activity. PglK-catalysed ATP hydrolysis is stimulated ,2.5- move in pairs during structural transitions. We also confirmed the
fold by full-length LLO and tLLO both in detergent and in liposomes, register of the resulting model by collecting anomalous diffraction
demonstrating that the terminal GalNAc and the branching glucose data of Hg-soaked crystals and of a selenomethionine derivative,
residues of native LLO are not relevant for ATPase stimulation or which allowed us to position cysteine and methionine residues
flipping (Extended Data Fig. 3e–g). Notably, although the hydrolysis (Extended Data Fig. 5).
and flipping rates are almost 100-fold lower than those of wild-type The fold of PglK is similar to that of other ABC exporters, first
PglK, the ATPase activity of the E510Q mutant is also stimulated by revealed by the structure of the bacterial ABC transporter Sav1866
LLO, suggesting that the interaction with LLO is not affected by the (ref. 36) (Extended Data Fig. 6a). However, PglK contains an addi-
E510Q mutation. Unlike remotely related multi-drug ABC transpor- tional, short a-helix (termed external helix ‘EH’) at the periplasmic
ters32,33, the stimulation of the ATPase activity of PglK is highly membrane boundary and parallel to the plane of the membrane (see
specific. Commonly used drugs such as verapamil, Hoechst33342, below). The apo-inward-1 and outward-occluded structures were
rhodamine 6G and acrilflavin had no stimulating effect (Extended obtained from PglK, crystallized using lauryl maltose neopentyl glycol
Data Fig. 3f). The addition of lipid A, known to stimulate MsbA34, (LMNG) as a detergent, whereas that of apo-inward 2 was obtained
was equally unable to stimulate the ATPase activity of PglK. using N-dodecyl-b-D-maltopyranoside (DDM) (Extended Data Table 1).

a b
EH EH EH EH EH EH
Out

TM5

ICL2

TM4 TM6
In

40°

44 Å 30 Å
ADP ADP

ADP-bound
Apo-inward-1 Apo-inward-2
outward-occluded

Figure 2 | Structures of PglK in distinct conformations. a, PglK subunits are in ribbon representation. The same helices of a superimposed PglK subunit in
in orange and grey. EH denotes the external helix. b, Conformational changes the apo-inward-1 state (light yellow) and in the outward-occluded state (dark
associated with the conversion of inward- and outward-facing states. A single orange) are shown as ribbons. ICL, intracellular loop.
PglK subunit is shown in light orange, with transmembrane helices 4–6 shown

4 3 4 | N AT U R E | VO L 5 2 4 | 2 7 AU G U S T 2 0 1 5
G2015 Macmillan Publishers Limited. All rights reserved
 ARTICLE RESEARCH

Crystal contacts in the three crystal forms are distinct, as are the PglK (Fig. 3a). Given its unusual location and surface, we investigated
crystallization conditions (Extended Data Fig. 7). This finding sug- the functional relevance of the EH by studying PglK mutants in vivo
gests that the type of detergent and the lattice contacts, rather than and in vitro. We generated a deletion mutant (termed DEH) in which
intrinsic properties or dynamics of the protein, define the degree of the entire helix (residues S46–P67) was replaced by a flexible linker
nucleotide-binding domain (NBD) separation in the inward-facing containing eight residues. In addition, we generated a triple mutant
conformations. By contrast, the NBD arrangement in the outward- converting prominently located tyrosine residues into alanines
occluded state resembles a ‘closed sandwich dimer’ observed in several (Y50A/Y56A/Y63A, termed EHm1) and a double mutant of posi-
nucleotide-bound structures of ABC transporters17,36 and in various tively charged residues involved in stabilizing interactions (R53A/
isolated ABC domains37,38. A comparison of the three PglK structures K55A, termed EHm2). When reconstituted in liposomes, the three
revealed that the main conformational changes involve a scissor-like EH mutants showed unaltered basal ATPase activity compared to the
motion, with transmembrane helices TM4 and TM5 moving as a rigid wild type. However, no stimulation of ATPase activity by LLO was
body towards TM6 and tilting by 16u (apo-inward-1 versus apo- observed, suggesting that the mutations had abolished the role of EH
inward-2) or by 40u (apo-inward-1 versus outward-occluded) in linking LLO binding to ATPase stimulation (Fig. 3b). In detergent,
(Fig. 2b and Extended Data Fig. 6b). At the periplasmic surface, the the basal ATPase levels of the two EH mutants EHm1 and EHm2 were
outward-occluded conformation of PglK differs from the fully out- higher than that of wild type, whereas that of DEH was slightly lower.
ward-facing Sav1866 structure36 in that only a small opening from the Notably, however, ATPase stimulation by LLO was absent for all three
central translocation pathway exists. Outward-occluded PglK thus EH mutants in detergent as it was in liposomes. In agreement with the
reflects an intermediate conformation between the occluded state of lack of LLO-stimulated ATPase activity, the in vitro flipping rate of
the antibacterial peptide ABC exporter McjD39 and the fully outward- the DEH mutant was too low to be accurately determined, whereas
open Sav1866 (ref. 36) (Extended Data Fig. 6C). those of EHm1 and EHm2 were found to be between 15- and 33-fold
lower than wild type (Fig. 3c and Extended Data Fig. 3b).
Role of external helix EH in PglK–LLO interaction Furthermore, the DEH mutant had completely lost in vivo activity,
The loop connecting transmembrane helices TM1 and TM2 of PglK whereas the mutants EHm1 and EHm2 revealed only slightly lower in
contains the external helix EH (Figs 2a and 3a), the orientation of vivo output compared to wild type (Fig. 3d). It is important to note
which relative to the transmembrane domain is defined by several that reductions of the in vivo output of the biosynthetic glycosylation
hydrophobic, hydrophilic and cation-p interactions (Fig. 3a). As a pathway as a consequence of a PglK mutation may not directly
consequence, two hydrophobic grooves are formed near the external correlate with the reduction of in vitro flipping rates because the
membrane boundary, one between each EH and TM1 and TM2 of rate-limiting step of the in vivo process is unknown. A similar phe-
nomenon was previously observed for mutants of PglB that exhibited
reductions of in vitro glycosylation rates by two orders of magnitude
a or more, but showing only marginally lower in vivo output21,26,30,31.
R53
D47
K55 We speculated that the hydrophobic grooves formed near the EH
(Fig. 3a) might contribute to the binding of LLO to PglK. Such an
Y50 Y63 interaction could anchor the lipid tail of LLO and favour the position-
T43
Y56 ing of the pyrophosphate moiety for entering a translocation cavity
and thus initiate the flipping reaction. Because no structure of PglK
TM2
TM1 with bound LLO could be determined, we investigated PglK activity
using synthetic compounds representing structurally truncated ver-
sions of native LLO (Fig. 3e and Extended Data Fig. 1). Notably, none
b Detergent Liposomes c of the truncated LLOs stimulated the ATPase activity of PglK recon-
Normalized flipping

stituted in proteoliposomes, suggesting that the shorter polyprenyl

ATPase (nmol Pi

160 1.0
mg–1 min–1)

120 0.8
tails (maximum of 20 carbon atoms in geranyl-geranyl-PP-GlcNac)
(min–1)

80 0.6
40 0.4
0.2
may prevent their proper recognition by PglK in a lipid bilayer
0
0 (Extended Data Fig. 8). By contrast, the ATPase activity of PglK in
+

W +

EH 1+

2+
–

m –

m –

m –
EH H+

EHEH+
ΔE T+

ΔE T+
ΔE –

Δ –
W –

W –
EH m1

EH m2

EH 1

EH m2
EHm1

2
H

H
T

m
W

2
H
T

detergent was stimulated by farnesyl-pyrophosphate (FPP), but also

m

m
W

ΔE
EH

EH

d e by farnesyl-PP-GlcNAc and geranyl-geranyl-PP-GlcNac. Unlike in

1

2
m

m
H

Detergent Liposomes lipid bilayers, the stimulation of the ATPase activity of PglK in deter-
T

EH

EH
ΔE
W

150
gent may be due to the simultaneous binding of several truncated LLO
ATPase (nmol Pi
glycan

mg–1 min–1)
Anti-

100 molecules. Ubiquinone, a molecule with a polyprenyl tail similar to

50 that of native LLO but containing a distinct head-group (Fig. 3e and
0 Extended Data Fig. 1), was unable to stimulate the ATPase activity of
Anti-
PglK

PglK both in detergent and in liposomes. This demonstrated that both

+ G FPP + T
G Gl FPP

b i lcN c
in c
We
+G P + T–
G Gl PP

b i lcN c
in c
e
+U G A
qu A
on

+ U P G NA
qu A
on
W

P P cN

F
P c

the polyprenyl tail and the pyrophosphate moieties are essential for a
P
+F

productive, ATPase-stimulating interaction of LLO with PglK. When

+

Figure 3 | Structural and biochemical characterization of the external helix using native LLO, we found positive cooperativity of ATPase stimu-
EH. a, Left, hydrophobic groove (green) formed by the EH and lation (Km(LLO) 5 8.2 6 0.8 mM; Hill coefficient (nH) 5 1.9 6 0.4,
transmembrane helices 1 and 2 (TM1 and TM2). Right, close-up of the EH Extended Data Fig. 3h), which correlates with the presence of two
with functionally relevant residues indicated. b, ATPase activity in the presence regions on the surface of PglK that can interact with LLO molecules.
(1) or absence (2) of LLO of wild-type PglK and EH mutants in detergent
(LMNG) or proteoliposomes. c, In vitro tLLO flipping rate of PglK variants. Functions of inward- and outward-facing cavities
Bars indicate initial rates. d, In vivo LLO flipping by PglK variants expressed in
The inward-facing PglK structures revealed a large cavity that is access-
Escherichia coli SCM6 cells. e, ATPase activity of PglK in detergent or
proteoliposomes in the presence of diverse lipids and glycolipids: FPP (90 mM ible to the cytoplasmic side and to the inner leaflet of the bilayer
farnesyl-pyrophosphate), FPPGlcNAc (90 mM farnesyl-PP-GlcNAc), (Fig. 4a). Analogous cavities have been observed in nucleotide-free,
GGPPGlcNAc (90 mM geranyl-geranyl-PP-GlcNAc), and ubiquinone (90 mM). inward-facing states of other ABC transporters17,33,40, where they have
DEH, EH replaced by GSSGSSGS linker; EHm1, Y50A/Y56A/Y63A; EHm2, been interpreted to reflect substrate-binding pockets. Accordingly, one
R53A/K55A. Error bars denote s.d. (n 5 3). might speculate that during LLO flipping, the 55-carbon polyprenyl tail
2 7 AU G U S T 2 0 1 5 | VO L 5 2 4 | N AT U R E | 4 3 5
G2015 Macmillan Publishers Limited. All rights reserved
 RESEARCH ARTICLE

a ADP-bound b side of the membrane (Fig. 4a, b). In both states, the outward-facing
Apo-inward-1 LLO outward-occluded TM3 TM1
cavity spans almost the entire membrane, and whereas the outward-
occluded conformation would not provide sufficient space to contain
R86′ R302′
R302 R86 the pyrophosphate and oligosaccharide moieties of LLO, the fully
R260′ R260 outward-facing conformation could. We identified multiple arginine
R309 R309′ residues (R86, R260, R302 and R309) that lined the interior of the
outward-facing cavity, accounting for a total of eight positive charges
(Fig. 4b). Owing to the scissoring motion required to convert ABC
exporters from inward- to outward-facing states, these arginines are
buried and/or form salt-bridges in the inward-facing conformations
c Detergent Liposomes but become accessible when PglK is in an outward-facing or outward-
occluded state. Because arginines might serve as binding partners of
ATPase (nmol Pi

160
mg–1 min–1)

120
80
the pyrophosphate moiety of LLO, we generated individual alanine
40 mutants (R86A, R260A, R302A and R309A) and a fourfold mutant
0
(‘tetra’) with all charges removed. Whereas the single mutants main-
W –
R8 T+
R 8 6 A–
R2 6A+
26 –
+
30 –
+
30 –
Te A+
S2 4F Te a–
F/ 97 +
97 –
W+
W –
R8 T+
R 8 6 A–
R2 6A+
26 0A–
+
30 –
+
30 –
+
S2 94F Te ra–
F/ 29 +
97 –
+
T

R 60A

R 02A

R 09A

V2 W

R 02A

R 09A

V2 7W
R3 0 A

R3 2 A

94 /V2 tra

R3 0 A

R3 2 A

9A

9 4 / V tra

W
tr
W

tained a certain level of flipping activity both in vivo and in vitro, the

et
6

T
R

fourfold mutant was unable to catalyse LLO flipping (Fig. 4d, e).
9
S2

S2
d e Furthermore, whereas the single mutants retained some stimulation
7W
6A 2 9
Normalized flipping

R8 F / V

1.0 of ATPase activity by LLO (albeit with a lower basal ATPase rate), the
0A

R3 A
Te A
02

09
94

0.8 tra
26
R3
S2
(min–1)

ATPase activity of the fourfold mutant was no longer stimulated

R

0.6
glycan

0.4
Anti-

0.2
0
(Fig. 4c), suggesting an apparent loss of interaction between LLO
and mutant PglK.
T

tra

7W
A

0A

2A

9A
W

86

Te
26

30

30

9
R

Anti-
PglK
V2
R

F/
94

Flipping mechanism
S2

Figure 4 | Analysis of translocation pathway. a, Comparison of cavities The accumulated biochemical and structural evidence allow us to
(green) in inward-facing and outward-occluded states of PglK, with native propose a mechanism of PglK-catalysed LLO flipping, shown in
LLO shown as space-filling model for size reference. b, Side view of fully Fig. 5. The two conformations of PglK that are relevant for activity
outward-facing homology model of PglK based on Sav1866 structure36. Cavity- are the outward-occluded (ADP-bound) and outward-facing (ATP-
exposed, positively charged side chains are shown as blue sticks and labelled. bound) ones. Given millimolar concentrations of ADP and ATP in
c, ATPase activity of wild-type PglK and mutants in detergent (LMNG) and the cell, an apo-state of the transporter, represented by an inward-
in proteoliposomes, in the presence (1) or absence (2) of native LLO.
d, In vitro tLLO flipping rate of PglK variants. Bars indicate initial rates. e, In
facing structure, is probably transient and sparsely populated. It is
vivo LLO flipping of PglK variants expressed in E. coli SCM6 cells. Tetra, indeed conceivable that fast exchange of ATP for ADP would not
R86A/R260A/R302A/R309A. Error bars denote s.d. (n 5 3). allow the ATP-binding sites of homodimeric PglK protein to be nuc-
leotide-free at the same time. We propose that while the polyprenyl
enters the inward-facing cavity from the lipid bilayer to fold (‘curl up’) tail of LLO interacts with the surface of PglK including and around
into the very bottom of the cavity, where the surface is hydrophobic the EH, the pyrophosphate moiety is directly transferred into the
outward-facing cavity of the ATP-bound state, where it can form
(Fig. 4a and Extended Data Fig. 6d). Although difficult to reconcile with
electrostatic interactions with exposed arginine residues. The oligo-
the essential nature of the EH in facilitating LLO interactions (see
saccharide then follows the pyrophosphate moiety but does not
above), we tested this hypothesis and reduced the volume of the hydro-
require any specific interactions with PglK during the translocation
phobic section of the inward-facing cavity by ,60% (ref. 41) by intro-
process, which could explain the low specificity of the flipping reac-
ducing four bulky residues, two in each PglK subunit (S294F/V297W).
tion with respect to the type and number of saccharides. This hypo-
The resulting mutant retained wild-type in vivo and in vitro flipping
thesis is consistent with the observation that PglK can flip distinct
rates and unaltered ATPase stimulation (Fig. 4c–e), arguing against the
LLOs that are normally transported by the multidrug/oligosacchar-
involvement of this cavity in polyprenyl binding and suggesting that
idyl-lipid/polysaccharide (MOP)-type flippase Wzx20. It can also sug-
the polyprenyl tail remains exposed to the lipid bilayer during flipping. gest how related ABC-type flippases could translocate long, linear
We next considered whether the pyrophosphate-oligosaccharide polysaccharides during O-antigen generation42: as the elongated
head group could be transferred to the inward-facing cavity during oligosaccharide chain is imagined to slide directly into an outward-
flipping. However, this seems highly unlikely for the following reason: facing translocation pathway and leave on the external side, no lim-
as a consequence of the domain-swapped architecture of ABC expor- itation on chain length may exist. In our model, the driving force of
ters, TM4–TM5 of each PglK subunit cross over and bind the NBD of the flipping reaction of PglK has two components. First, as ATP
the opposite subunit. The cavity formed by the inward-facing state is hydrolysis causes the NBDs to be pushed apart for product release,
thus defined by the separation of TM4 and TM6 (Fig. 2b). If the transmission of this motion to the TMDs may cause pressure to be
pyrophosphate and oligosaccharide moieties of LLO were to enter exerted on the substrate in the translocation pathway. This is remin-
this cavity, the LLO molecule would be ‘pinched off’ during the trans- iscent of the peristaltic component proposed for the mechanism of
ition of PglK to outward-facing, as the gap between TM4 and TM6 the vitamin B12 ABC importer43. Second, once the pyrophosphate
closes and a distinct gap (between TM1 and TM3) opens to the out- and oligosaccharide moieties have diffused out of the translocation
side. This would trap the LLO molecule and produce a massive steric pathway, PglK will probably adopt an outward-occluded conforma-
clash, preventing substrate release. The above considerations lead to tion as observed in our ADP-bound structure, thereby effectively
the inevitable conclusion that during the flipping cycle of PglK, the closing the door behind the substrate. We propose that the energy
inward-facing cavity does not contribute to LLO binding. gained by the hydrolysis of ATP is primarily used to break the inter-
The outward-facing cavity is distinct in chemical nature from the action of the pyrophosphate moiety with the essential arginine resi-
inward-facing one. Exploiting previous findings35, we modelled a fully dues lining the outward-facing cavity of PglK. It is conceivable that
outward-facing state of PglK based on the structure of Sav1866 (ref. pyrophosphate-containing LLO accumulated on the periplasmic side
36). The resulting model features a similar cavity as that of our out- might act as an inhibitor of the flipping reaction, which might con-
ward-occluded structure, but with a wider opening to the periplasmic tribute to the slowing of the in vitro flipping rate with time. In vivo, a
4 3 6 | N AT U R E | VO L 5 2 4 | 2 7 AU G U S T 2 0 1 5
G2015 Macmillan Publishers Limited. All rights reserved
 ARTICLE RESEARCH

Out Out
2ADP
R86′ R302′ E 2ATP 2ADP
R302 R86 A
R309 R260
R260′ R309′
2ATP 2Pi
P
In In P P
Basal P
ATPase
activity

ADP ADP ATP ATP

5 1 2

LLO
outside leaflet

D B
LLO
inside leaflet

P 4 3
P

P
C P

2Pi

ADP ADP ATP ATP

Figure 5 | Proposed LLO flipping mechanism including five states (circled 1 and 4 are based on the outward-occluded structure, states 2 and 3 are
numbers). PglK is shown as a cylinder model, with subunits coloured grey modelled outward-open conformations. The proposed molecular events
and orange, and the EH coloured red. Functionally important arginine residues indicated by arrows are: A, polyprenyl tail of LLO interacting with PglK region
are shown as blue sticks and labelled. The polyprenyl tail of LLO is shown in including and around EH, ATP replacing ADP; B, pyrophosphate-
purple; phosphate groups are indicated by circled ‘P’. Red hexagon denotes di- oligosaccharide head group entering outward-facing cavity; C, ATP hydrolysis,
N-acetylbacillosamine; yellow square denotes N-acetylgalactosamine; and LLO head-group release on external side of the membrane, release of inorganic
blue circle denotes glucose. State 5 reflects the apo-inward-1 structure, states phosphate; D, polyprenyl tail release; E, futile ATPase cycle.

similar problem might occur if undecaprenyl-pyrophosphate were to Online Content Methods, along with any additional Extended Data display items
and Source Data, are available in the online version of the paper; references unique
accumulate after glycan transfer by PglB21, but is probably alleviated to these sections appear only in the online paper.
by the catalysed hydrolysis of undecaprenyl-pyrophosphate to the
monophosphate species22 (Extended Data Fig. 2). Received 20 February; accepted 15 July 2015.
The LLO flipping mechanism deduced for PglK has a resemblance Published online 12 August 2015.
to the ‘credit card swipe’ model proposed for the flipping of phospho-
1. Burda, P. & Aebi, M. The dolichol pathway of N-linked glycosylation. Biochim.
lipids by P4-ATPases flippases44–46. Two key differences are the pro- Biophys. Acta 1426, 239–257 (1999).
posed recognition of the polyprenyl tail on the PglK surface by means 2. Helenius, J. et al. Translocation of lipid-linked oligosaccharides across the ER
of the EH, and the requirement of a long, sufficiently wide transloca- membrane requires Rft1 protein. Nature 415, 447–450 (2002).
3. Sprong, H., van der Sluijs, P. & van Meer, G. How proteins move lipids and lipids
tion pathway that contains the pyrophosphate and oligosaccharide move proteins. Nature Rev. Mol. Cell Biol. 2, 504–513 (2001).
moieties during flipping. Our mechanism is distinct from that pro- 4. Sebastian, T. T., Baldridge, R. D., Xu, P. & Graham, T. R. Phospholipid flippases:
posed for MsbA function, which suggested that the entire lipid A-core building asymmetric membranes and transport vesicles. Biochim. Biophys. Acta
1821, 1068–1077 (2012).
may enter the inward-facing, nucleotide-free state of the trans- 5. Hankins, H. M., Baldridge, R. D., Xu, P. & Graham, T. R. Role of flippases,
porter17. More generally, our mechanistic proposal is also distinct scramblases and transfer proteins in phosphatidylserine subcellular distribution.
from the canonical alternating-access model used to explain the Traffic 16, 35–47 (2015).
6. Krahling, S., Callahan, M. K., Williamson, P. & Schlegel, R. A. Exposure of
mechanism of transporters of small, hydrophilic substrates. phosphatidylserine is a general feature in the phagocytosis of apoptotic
lymphocytes by macrophages. Cell Death Differ. 6, 183–189 (1999).
Conclusions 7. Balasubramanian, K. & Schroit, A. J. Aminophospholipid asymmetry: A matter of
life and death. Annu. Rev. Physiol. 65, 701–734 (2003).
Our results define the structural and mechanistic basis of lipid-linked 8. Cuthbertson, L., Kos, V. & Whitfield, C. ABC transporters involved in export of cell
oligosaccharide flipping by an ABC transporter. The data suggest a surface glycoconjugates. Microbiol. Mol. Biol. Rev. 74, 341–362 (2010).
flipping mechanism suitable for the translocation of a large head 9. Cuthbertson, L., Kimber, M. S. & Whitfield, C. Substrate binding by a bacterial ABC
transporter involved in polysaccharide export. Proc. Natl Acad. Sci. USA 104,
group containing a pyrophosphate moiety and a linear polysacchar- 19529–19534 (2007).
ide, both of which are shielded from the lipid bilayer in a translocation 10. Clarke, B. R., Cuthbertson, L. & Whitfield, C. Nonreducing terminal modifications
cavity. We predict that the proposed direct jump into the outward- determine the chain length of polymannose O antigens of Escherichia coli and
couple chain termination to polymer export via an ATP-binding cassette
facing conformation and the putative interaction of the lipid tail with transporter. J. Biol. Chem. 279, 35709–35718 (2004).
the membrane-facing surface of the transporter may be a feature of 11. Sharom, F. J. Flipping and flopping–lipids on the move. IUBMB Life 63, 736–746
other ABC-type LLO flippases and possibly of MOP-type transporters (2011).
involved in bacterial cell wall assembly or N-glycan flipping in eukar- 12. Kodigepalli, K. M., Bowers, K., Sharp, A. & Nanjundan, M. Roles and regulation of
phospholipid scramblases. FEBS Lett. 589, 3–14 (2015).
yotes14,20. Understanding the molecular mechanism of LLO flipping 13. Brunner, J. D., Lim, N. K., Schenck, S., Duerst, A. & Dutzler, R. X-ray structure of a
may open new ways to modify the glycosylation machinery in bac- calcium-activated TMEM16 lipid scramblase. Nature 516, 207–212 (2014).
teria, which could be exploited for the generation of new glyco- 14. Hvorup, R. N. et al. The multidrug/oligosaccharidyl-lipid/polysaccharide (MOP)
exporter superfamily. Eur. J. Biochem. 270, 799–813 (2003).
conjugates or glycoproteins that may have applications in diagnosis 15. Lopez-Marques, R. L., Theorin, L., Palmgren, M. G. & Pomorski, T. G. P4-ATPases:
or therapeutics47,48. lipid flippases in cell membranes. Pflugers Arch. 466, 1227–1240 (2014).

2 7 AU G U S T 2 0 1 5 | VO L 5 2 4 | N AT U R E | 4 3 7
G2015 Macmillan Publishers Limited. All rights reserved
 RESEARCH ARTICLE

16. Eckford, P. D. & Sharom, F. J. The reconstituted Escherichia coli MsbA protein 36. Dawson, R. J. & Locher, K. P. Structure of a bacterial multidrug ABC transporter.
displays lipid flippase activity. Biochem. J. 429, 195–203 (2010). Nature 443, 180–185 (2006).
17. Ward, A., Reyes, C. L., Yu, J., Roth, C. B. & Chang, G. Flexibility in the ABC transporter 37. Zaitseva, J. et al. A structural analysis of asymmetry required for catalytic activity of
MsbA: Alternating access with a twist. Proc. Natl Acad. Sci. USA 104, 19005–19010 an ABC-ATPase domain dimer. EMBO J. 25, 3432–3443 (2006).
(2007). 38. Smith, P. C. et al. ATP binding to the motor domain from an ABC transporter
18. Wacker, M. et al. N-linked glycosylation in Campylobacter jejuni and its functional drives formation of a nucleotide sandwich dimer. Mol. Cell 10, 139–149
transfer into E. coli. Science 298, 1790–1793 (2002). (2002).
19. Young, N. M. et al. Structure of the N-linked glycan present on multiple 39. Choudhury, H. G. et al. Structure of an antibacterial peptide ATP-binding cassette
glycoproteins in the Gram-negative bacterium, Campylobacter jejuni. J. Biol. Chem. transporter in a novel outward occluded state. Proc. Natl Acad. Sci. USA 111,
277, 42530–42539 (2002). 9145–9150 (2014).
20. Alaimo, C. et al. Two distinct but interchangeable mechanisms for flipping of lipid- 40. Shintre, C. A. et al. Structures of ABCB10, a human ATP-binding cassette
linked oligosaccharides. EMBO J. 25, 967–976 (2006). transporter in apo- and nucleotide-bound states. Proc. Natl Acad. Sci. USA 110,
21. Lizak, C., Gerber, S., Numao, S., Aebi, M. & Locher, K. P. X-ray structure of a bacterial 9710–9715 (2013).
oligosaccharyltransferase. Nature 474, 350–355 (2011). 41. Voss, N. R. & Gerstein, M. 3V: cavity, channel and cleft volume calculator and
22. Tatar, L. D., Marolda, C. L., Polischuk, A. N., van Leeuwen, D. & Valvano, M. A. An extractor. Nucleic Acids Res. 38, W555–W562 (2010).
Escherichia coli undecaprenyl-pyrophosphate phosphatase implicated in 42. Whitfield, C. & Trent, M. S. Biosynthesis and export of bacterial
undecaprenyl phosphate recycling. Microbiology 153, 2518–2529 (2007). lipopolysaccharides. Annu. Rev. Biochem. 83, 99–128 (2014).
23. Sanyal, S. & Menon, A. K. Specific transbilayer translocation of dolichol-linked 43. Korkhov, V. M., Mireku, S. A. & Locher, K. P. Structure of AMP-PNP-bound vitamin
oligosaccharides by an endoplasmic reticulum flippase. Proc. Natl Acad. Sci. USA B12 transporter BtuCD-F. Nature 490, 367–372 (2012).
106, 767–772 (2009). 44. Vestergaard, A. L. et al. Critical roles of isoleucine-364 and adjacent residues in a
24. Sanyal, S. & Menon, A. K. Stereoselective transbilayer translocation of mannosyl hydrophobic gate control of phospholipid transport by the mammalian P4-
phosphoryl dolichol by an endoplasmic reticulum flippase. Proc. Natl Acad. Sci. ATPase ATP8A2. Proc. Natl Acad. Sci. USA 111, E1334–E1343 (2014).
USA 107, 11289–11294 (2010). 45. Stone, A. & Williamson, P. Outside of the box: recent news about phospholipid
25. Linton, D. et al. Functional analysis of the Campylobacter jejuni N-linked protein translocation by P4 ATPases. J. Chem. Biol. 5, 131–136 (2012).
glycosylation pathway. Mol. Microbiol. 55, 1695–1703 (2005).
46. Pomorski, T. & Menon, A. K. Lipid flippases and their biological functions. Cell. Mol.
26. Lizak, C. et al. A catalytically essential motif in external loop 5 of the bacterial
Life Sci. 63, 2908–2921 (2006).
oligosaccharyltransferase PglB. J. Biol. Chem. 289, 735–746 (2014).
47. Hug, I. & Feldman, M. F. Analogies and homologies in lipopolysaccharide and
27. Troutman, J. M. & Imperiali, B. Campylobacter jejuni PglH is a single active site
glycoprotein biosynthesis in bacteria. Glycobiology 21, 138–151 (2011).
processive polymerase that utilizes product inhibition to limit sequential glycosyl
transfer reactions. Biochemistry 48, 2807–2816 (2009). 48. Jones, C. Vaccines based on the cell surface carbohydrates of pathogenic bacteria.
28. Abeijon, C. & Hirschberg, C. B. Topography of initiation of N-glycosylation An. Acad. Bras. Cienc. 77, 293–324 (2005).
reactions. J. Biol. Chem. 265, 14691–14695 (1990). Acknowledgements We thank the staff scientists at the PX beamline of the Swiss Light
29. Hanover, J. A. & Lennarz, W. J. The topological orientation of N,N9- Source for help with data collection, and M. Napiorkowska and A.Ramirez for assistance
diacetylchitobiosylpyrophosphoryldolichol in artificial and natural membranes. with PglB assays. This work was supported by the Swiss National Science Foundation
J. Biol. Chem. 254, 9237–9246 (1979). (SNF 31003A–146191 to K.P.L. and Transglyco Sinergia program to M.A., J.-L.R. and
30. Gerber, S. et al. Mechanism of bacterial oligosaccharyltransferase: in vitro K.P.L.). C.P. acknowledges support from the ETH postdoctoral fellowship program.
quantification of sequon binding and catalysis. J. Biol. Chem. 288, 8849–8861
(2013). Author Contributions C.P. determined the structures of PglK, established the in vitro
31. Lizak, C. et al. Unexpected reactivity and mechanism of carboxamide activation in flipping assay, and performed in vivo flipping studies. S.G. crystallized PglK in the
bacterial N-linked protein glycosylation. Nature Commun. 4, 2627 (2013). apo-inward-2 state, M.B. assisted in expression and purification of PglK. J.B., T.D. and
32. Siarheyeva, A. & Sharom, F. J. The ABC transporter MsbA interacts with lipid A and J.-L.R. synthesized LLO analogues. K.P.L., S.G. and C.P. conceived the project. K.P.L.,
amphipathic drugs at different sites. Biochem. J. 419, 317–328 (2009). M.A., and C.P. analysed the data. K.P.L. and C.P. wrote the manuscript.
33. Jin, M. S., Oldham, M. L., Zhang, Q. & Chen, J. Crystal structure of the multidrug
transporter P-glycoprotein from Caenorhabditis elegans. Nature 490, 566–569 Author Information Atomic coordinates and structure factors were deposited with the
(2012). RCSB Protein Data Bank (PDB) under accessions 5C78 (apo-inward-1), 5C76
34. Raetz, C. R., Reynolds, C. M., Trent, M. S. & Bishop, R. E. Lipid A modification (apo-inward-2) and 5C73 (outward-occluded). Reprints and permissions information
systems in gram-negative bacteria. Annu. Rev. Biochem. 76, 295–329 (2007). is available at www.nature.com/reprints. The authors declare no competing financial
35. Lee, J. Y., Yang, J. G., Zhitnitsky, D., Lewinson, O. & Rees, D. C. Structural basis for interests. Readers are welcome to comment on the online version of the paper.
heavy metal detoxification by an Atm1-type ABC exporter. Science 343, Correspondence and requests for materials should be addressed to K.P.L.
1133–1136 (2014). ([email protected]).

4 3 8 | N AT U R E | VO L 5 2 4 | 2 7 AU G U S T 2 0 1 5
G2015 Macmillan Publishers Limited. All rights reserved
 ARTICLE RESEARCH

METHODS non-crystallographic symmetry averaging and solvent flatting. Electron density

No statistical methods were used to predetermine sample size. Experiments were maps were calculated using CCP4 programs56. The protein model was built using
not randomized, and the investigators were not blinded to allocation during Coot57, and refined using Phenix52. X-ray data and refinement statistics are given
experiments and outcome assessment. in Extended Data Table 1.
PglK expression and purification. The gene encoding C. jejuni PglK was cloned The apo-inward-2 structure was solved by molecular replacement using a partial
into a modified pET-19b vector (Novagen) with a His10 affinity tag fused to N model from the apo-inward-1 structure as a search model using the program
terminus. PglK was overexpressed in E. coli BL21-Gold (DE3) (Stratagene) cells in a Phaser58, and refinement was performed using Phenix52 combined with manual
30 l fermentor (Infors). Cells were grown at 37 uC in modified Terrific Broth (TB) building in Coot57. The structure of the outward-occluded state was solved by a
medium supplemented with 1% glucose (w/v) to A600 nm of 10.0 before expression combination of molecular replacement using a partial model from the apo-inward-
was induced by the addition of 0.2 mM isopropyl b-D-1-thiogalactopyranoside 1 structure, and a single wavelength anomalous dispersion (SAD) data set from a
(IPTG) for 1 h. All following steps were performed at 4 uC unless specified selenomethionine derivative of PglK, as well as data from a derivative after labelling
differently. Cells were collected by centrifugation, re-suspended in 50 mM cysteine and histidine residues with EMP and PtCl4, respectively, in order to
Tris-HCl, pH 8.0; 500 mM NaCl; 7 mM b-mercaptoethanol; 0.5 mM phenyl- confirm the polypeptide chain register (Extended Data Fig. 5a–c). X-ray data
methanesulfonylfluoride (PMSF). They were disrupted in a M-110L microflui- and refinement statistics are given in Extended Data Table 1.
dizer (Microfluidics) at 15,000 p.s.i. chamber pressure. Membranes were PglH expression and purification. The gene encoding C. jejuni PglH was cloned
pelleted by ultracentrifugation at 100,000g for 0.5 h. PglK was solubilized in into a modified pET-19b vector (Novagen) with a N-terminal His10 affinity tag
50 mM Tris-HCl, pH 8.0; 500 mM NaCl; 20 mM imidazole; 15% glycerol (v/v); fused to PglH. PglH was overexpressed in E. coli BL21-Gold (DE3) (Stratagene)
7 mM b-mercaptoethanol; 1% DDM (w/v) (Anatrace); 1% C12E8 anapoe cells in modified TB medium supplemented containing 1% glucose (w/v). Cells
(Anatrace) by stirring for 2 h. The supernatant was loaded onto a NiNTA were grown at 37 uC to A600 nm of 3.0 before the culture was induced by the
superflow affinity column (Qiagen), washed once with the same buffer but addition of 0.5 mM IPTG, followed by a transfer to 18 uC for 16 h. All following
containing 10% glycerol (v/v), 50 mM imidazole and 0.016–0.03% DDM, and steps were performed at 4 uC unless specified differently. Cells were collected by
then washed a second time with the same buffer containing 0.016–0.03% LMNG centrifugation, re-suspended in 50 mM Tris-HCl, pH 8.0, 200 mM NaCl, 20 mM
(Affymetrix). Elution was performed in the same buffer but containing 200 mM imidazole and 0.5 mM PMSF. Cells were disrupted in a M-110L microfluidizer
NaCl and 200 mM imidazole. The protein was further purified by size exclusion (Microfluidics) at 15,000 p.s.i. chamber pressure followed by the addition of 1%
chromatography (Superdex 200 10/300 GL, GE Healthcare) and peak fractions Triton X-100 (w/v)27. After centrifugation, the supernatant was loaded onto a
were pooled and concentrated to 8–10 mg ml21 in an Amicon Ultra-15 con- NiNTA superflow affinity column (Qiagen), washed once with the same buffer
centrator (Millipore) with a molecular mass cut-off of 100 kDa. but containing 50 mM imidazole. Elution was performed using a buffer contain-
Seleno-methionine derivative production. PglK was overexpressed in E. coli ing 50 mM Tris-HCl, pH 8.0 and 500 mM imidazole. The protein was desalted
BL21 (DE3) RIPL (Stratagene) cells as described above with minor modifications. into 50 mM Tris-HCl, pH 8.0 and 200 mM NaCl.
In brief, cells in TB-glucose media were grown at 37 uC to A600 nm of 0.5–1.0. The LLO and tLLO extraction. Isolation of LLOs was performed as described prev-
cells were then used to inoculate a culture of M9 media supplemented with iously30. In brief, LLOs were extracted from E. coli SCM6 cells carrying a C. jejuni
vitamin B1 hydrochloride. Cells were then grown until A600 nm of 0.5 and used pglBmut cluster, containing an inactivated pglB gene (for LLO extraction) or a
to inoculate 2 l M9 media supplemented with vitamin B1 hydrochloride. Cells pglHmut:pglBmut cluster, containing an additionally inactivated pglH gene (for
were grown overnight at 37 uC until A600 nm of ,0.9. At that point a cocktail of tLLO extraction). Extraction was performed using a mixture of chloroform:-
amino acids (lysine, threonine, phenylalanine, leucine, isoleucine and valine, all at methanol:water (10:10:3) for LLO extraction or methanol:chloroform (1:2) for
100 mg l21) including selenomethionine (200 mg l21) was added to the culture. tLLO extraction. Extracts were dried in a rotary evaporator and reconstituted in a
After 30 min, expression was induced with 0.2 mM IPTG for 90 min. Cells were buffer containing 10 mM Tris, pH 8.0, 150 mM NaCl and 1% Triton X-100 (w/v).
collected immediately thereafter. The concentration of reconstituted LLOs was determined by titrating various
Native crystals. For crystallization of the outward-occluded state PglK was incu- amounts of LLOs against a constant amount of acceptor peptide in an in vitro
bated for 0.5 h with 5–10 mM ADP and 5–10 mM MgCl2. PglK was crystallized by glycosylation assay as described before59.
vapour diffusion in sitting drops or hanging drops at 20 uC using reservoirs Reconstitution of PglK and variants in preoteoliposomes. Liposomes (20 mg
containing 100 mM Tris-HCl, pH 8.2; 150 mM Mg-acetate; 25% PEG400 for lipid ml21) from a mixture 3:1 (w:w) of E. coli polar lipids and L-a-phosphatidilcho-
the apo-inward-1 state crystallization; 50 mM glycine-NaOH, pH 9.4, 300 mM line (Avanti Polar Lipids) were prepared by extrusion through polycarbonate filters
KCl, 50 mM MgCl2, 21% PEG600 for the apo-inward-2 state crystallization; and (400 nm pore size) and diluted in 10 mM Tris-HCl, pH 8.0, 150 mM NaCl and 2
50 mM Na/K phosphate, pH 7.5, 250 mM NaCl, 3.5 M ammonium sulphate for mM b-mercaptoethanol. After saturation with Triton X-100, the liposomes were
the outward-occluded state crystallization. The protein to reservoir volume ratio mixed with purified protein at a lipid/protein ratio of 30–50:1 (w/w) and ,4 mM
was 2:1–1:1. Crystals typically appeared after 3–4 days and matured to full size tLLO. BioBeads were then added to remove detergent. Finally, proteoliposomes
within 2 weeks. Crystals were cryoprotected by gently increasing the cryoprotec- containing a final concentration of 20 mg ml21 lipids, 6.2–10.1 mM PglK and 20 mM
tant concentration in the drops (up to 30% PEG400 for apo-inward-1 and -2, and tLLO were centrifuged and washed before being frozen in liquid nitrogen and
up to 3.0 M ammonium sulphate for outward-occluded) and directly flash frozen stored at 280 uC.
by immersion in liquid nitrogen before data collection. Determination of PglK orientation in liposomes. A fully functional mutant of
Heavy-metal derivatives. Native crystals were soaked for 1 day in 1 mM ethyl PglK containing none of the native cysteines but with an engineered cysteine at
mercury phosphate (EMP) or 1 mM potassium tetrachloroplatinate before back- each NBD (C269L/C352S/C386S/C549L/S544C) was reconstituted in proteolipo-
soaking and flash-freezing by immersion in liquid nitrogen. somes. After incubation with the thiol-reactive fluorescent dye Alexa Fluor 488
Data collection. Crystals belonged to the space groups P1 (apo-inward-1), P21 C5 maleimide, samples were analysed by SDS–PAGE and a fluorescence gel
(apo-inward-2) and P43212 (outward-occluded). Native data were collected at scanner. The ratio of PglK with NBDs facing outwards was calculated after
beamline X06SA at the Swiss Light Source (Villigen). EMP and PtCl4 derivative comparison of the band intensities obtained from non-disrupted and disrupted
data sets (Extended Data Table 1) were collected at the same station. Data were proteoliposomes (0.3% Triton X-100). From this analysis it was determined that
processed and merged with XDS49 and anisotropic scaling/ellipsoid truncation 51.8 6 2.8% of PglK molecules are oriented with NBDs facing outwards.
was performed50. In vitro tLLO flipping assay. PglK proteoliposomes diluted in 10 mM Tris-HCl,
To improve the usable resolution and quality of the resulting electron density pH 8.0, 150 mM NaCl and 2 mM b-mercaptoethanol were extruded through
maps, we used the Karplus CC* (Pearson correlation coefficient)-based data cut- polycarbonate filters (400-nm pore size) and incubated with 5 mM MgCl2 and
off approach51 (Extended Data Table 1). The resolution limit was set taking into 2 mM ATP to initiate the tLLO flipping reaction. To stop the translocation
account a CC1/2. ,40% based on data merging statistics and a CC* analysis reaction, samples were diluted into a buffer containing 4 mM ADP. Labelling
against unmerged intensities in the Phenix package52, satisfying Karplus CC* of non-flipped tLLO remnant in the external leaflet of proteoliposomes was
against CCwork and CCfree criteria, as well as Rfree of the highest resolution shell achieved by incubation with purified glycosyltransferase PglH in the presence
against the refined structure being less than or equal to ,50% (Extended Data of 50 mM 3H-UDP-GalNAc and 5 mM MnCl2. To stop the labelling reaction, the
Table 1). samples were diluted into a cold stop buffer and filtered using a Multiscreen
Structure determination. Experimental phases of the apo-inward-1 structure vacuum manifold (MSFBN6B filter plate, Millipore). Radioactivity trapped on
were determined using multiple isomorphous replacement with anomalous scat- the filters was determined using a gamma counter (CobraII Auto-Gamma,
tering (MIRAS). Mercury positions were found using SHELX53 and refined using Packard). Nonlinear fitting of data and initial velocities determination were per-
SHARP54. Solvent flattening was performed using Solomon55 and the resulting formed using GraphPad Prism 5 and the following equation: Y 5 (Y0 2 plateau) 3
phases were combined with a higher-resolution native data set, followed by e(2K3 X) 1 plateau Initial rates were calculated as the derivative at time 5 0.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH ARTICLE

Product analysis and determination of tLLO orientation in liposomes. glycosylated 3D5. Immunodetection of PglK was performed with anti-His-HRP
Glycosylation of tLLO by purified PglH was achieved after incubation of non- serum (Sigma).
disrupted or disrupted proteoliposomes (0.3% Triton X-100) by addition of 3.5– Mutagenesis. PglK mutants were generated by the QuickChange method or by
50 mM UDP-GalNAc, 2 mM MnCl2 and 0.2 mg ml21 PglH. The reaction was using gBlocks gene fragments (Integrated DNA technologies). The resulting
stopped by heat shock (90 uC for 10 min). The products were then analysed by in plasmids of all constructs were validated by DNA sequencing (Microsynth).
vitro glycosylation of a fluorescently labelled substrate peptide (DQNAT sequon) PglK variants were cloned into pMLBAD as above and used for in vivo flipping
catalysed by PglB30,31. In brief, the reaction products were incubated in a mixture assays.
containing 10 mM MES, pH 6.5, 100 mM NaCl, 10 mM MnCl2, 3% glycerol (v/v), Truncated LLOs synthesis. Farnesyl- and geranyl-geranyl-PP-GlcNAc were syn-
1% Triton X-100 (w/v), 0.5 mM fluorescently labelled acceptor peptide and 1 mM thesized in seven steps from commercially available lipids and sugar according to
purified PglB. The mixture was incubated at 30 uC and stopped by addition of SDS- the previously published strategy with slight modifications61. As the procedure
containing gel loading buffer. Samples were analysed by Tricine–SDS–PAGE in involves the formation of highly acid sensitive phosphorylated intermediates,
mini-gels (8 3 8 cm) consisting of a 16% resolving gel with 6 M urea, a 10% spacer flash column chromatography were performed on silica gel basified with an
gel, and a 4% stacking gel30,31. Fluorescent bands for glycopeptide were visualized ammonium hydroxide solution to avoid product degradation during purification.
using a Typhoon Trio Plus imager (GE Healthcare) with excitation at 488 nm and a The final LLOs were obtained from the protected precursor by deacetylation
526-nm SP emission filter. The amount of formed glycopeptide was determined using an ammonium hydroxide solution in methanol for 6 h. After freeze-drying,
from band intensities of fluorescence gel scans (ImageJ). 16 mg of farnesyl-PP-GlcNAc and 42 mg of geranyl-geranyl-PP-GlcNAc (5%
ATPase assays. ATP hydrolysis reactions were performed as described prev- overall yield from the lipid) were respectively isolated as final compounds and
iously, using a modified molybdate-based colorimetric method43. Protein con- were shown to be pure by 1H, 13C, 31P NMR as well as negative-mode ESI-HRMS.
centrations in the assays ranged from 0.1 mg ml21 to 0.3 mg ml21. All reactions
49. Kabsch, W. Xds. Acta Crystallogr. D 66, 125–132 (2010).
were performed in the presence of 2 mM ATP (or a range of ATP concentrations 50. Strong, M. et al. Toward the structural genomics of complexes: crystal structure of
for Km determination) and 5 mM MgCl2. ATPase rates were determined using a PE/PPE protein complex from Mycobacterium tuberculosis. Proc. Natl Acad. Sci.
linear regression. Nonlinear regression and statistical analysis was performed USA 103, 8060–8065 (2006).
using GraphPad Prism 5. 51. Karplus, P. A. & Diederichs, K. Linking crystallographic model and data quality.
In vivo LLO flipping assay. To analyse the flipping activity of PglK, we used an Science 336, 1030–1033 (2012).
52. Adams, P. D. et al. PHENIX: a comprehensive Python-based system for
assay similar to that previously described to determine in vivo PglB glycosylation macromolecular structure solution. Acta Crystallogr. D 66, 213–221 (2010).
activity30. PglK and variants were cloned into a pMLBAD plasmid60 with an 53. Sheldrick, G. M. A short history of SHELX. Acta Crystallogr. A 64, 112–122 (2008).
N-terminal His10 tag fused to it. This plasmid was transformed into E. coli cells 54. Bricogne, G., Vonrhein, C., Flensburg, C., Schiltz, M. & Paciorek, W. Generation,
defective for O-antigen biosynthesis (SCM6), carrying the plasmids pCL21 (ref. representation and flow of phase information in structure determination: recent
59) and pACYC-wlaB::kan25. pCL21 encodes for the expression of the single- developments in and around SHARP 2.0. Acta Crystallogr. D. 59, 2023–2030
(2003).
chain Fv fragment of 3D5 carrying a DQNAT glycosylation site in the linker 55. Abrahams, J. P. & Leslie, A. G. Methods used in the structure determination of
region and a C-terminal Myc tag fused to 3D5. pACYC-wlaB::kan codes for the bovine mitochondrial F1 ATPase. Acta Crystallogr. D 52, 30–42 (1996).
biosynthesis of the C. jejuni LLO with a knockout of the C. jejuni pglK gene. A 5 ml 56. Collaborative Computational Project, Number 4. The CCP4 suite: programs for
pre-culture was inoculated from a single clone and grown overnight at 37 uC in LB protein crystallography. Acta Crystallogr. D 50, 760–763 (1994).
medium. The main culture was inoculated to an optical density A600 nm of 0.05 in 57. Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of
Coot. Acta Crystallogr. D 66, 486–501 (2010).
10 ml LB medium and grown at 37 uC to A600 nm of 0.5. The culture was induced
58. McCoy, A. J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674
by addition of arabinose to 0.1% (w/v) and grown for 4 h at 24 uC. For extraction (2007).
of periplasmic proteins, an equivalent of 1 ml culture volume with an A600 nm of 59. Lizak, C., Fan, Y. Y., Weber, T. C. & Aebi, M. N-Linked glycosylation of antibody
2 was collected by centrifugation, re-suspended in 100 ml extraction buffer, fragments in Escherichia coli. Bioconjug. Chem. 22, 488–496 (2011).
consisting of 30 mM Tris-HCl, pH 8.5, 20% (w/v) sucrose, 1 mM EDTA and 60. Lefebre, M. D. & Valvano, M. A. Construction and evaluation of plasmid vectors
1 mg ml21 lysozyme (Sigma) and incubated for 1 h at 4 uC. A final centrifugation optimized for constitutive and regulated gene expression in Burkholderia cepacia
complex isolates. Appl. Environ. Microbiol. 68, 5956–5964 (2002).
step yielded periplasmic proteins in the supernatant. Glycosylation of 3D5 and 61. Liu, F. et al. Rationally designed short polyisoprenol-linked PglB substrates for
expression of PglK were analysed by immunoblot following SDS–PAGE. engineered polypeptide and protein N-glycosylation. J. Am. Chem. Soc. 136,
Immunodetection was performed with anti-glycan serum hR6 (ref. 20) to observe 566–569 (2014).

G2015 Macmillan Publishers Limited. All rights reserved

 ARTICLE RESEARCH

Extended Data Figure 1 | Structures of LLOs and chemical analogues used in this study.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH ARTICLE

Extended Data Figure 2 | LLO synthesis and N-glycosylation in C. jejuni. tail is shown in purple. A periplasmic polypeptide chain is shown in blue. PglH
Red hexagon denotes di-N-acetylbacillosamine; yellow square denotes (red) is the glycosyltransferace used in the in vitro tLLO flipping assays.
N-acetylgalactosamine; and blue circle denotes glucose. The LLO polyprenyl Und-PP-ase, undecaprenyl pyrophosphatase.

G2015 Macmillan Publishers Limited. All rights reserved

 ARTICLE RESEARCH

Extended Data Figure 3 | Biochemical characterization of wild-type and represents the initial LLO flipping rate in proteoliposomes at distinct ATP
mutant PglK. a, In vitro tLLO flipping rates of PglK wild-type and E510Q concentrations. Cpm, counts per million. e, ATPase activity in the presence (1)
mutant. Bars indicate initial velocities. b, In vitro tLLO flipping of PglK mutants or absence (2) of native LLO of wild-type PglK and E510Q mutant in detergent
E510Q, EHm1 and EHm2. 1ADP, assay in the presence of 4 mM ADP; 2ATP, (LMNG) or proteoliposomes. f, ATPase activity of PglK in detergent in the
assay in the absence of ATP. Liposomes, empty liposomes containing only presence of LLO (20 mM), tLLO (20 mM), diverse drugs (50 mM) and lipid A (20
tLLO. ADP alone does not cause tLLO to disappear from the external liposomes mM). g, Determination of Km values of ATP hydrolysis in the presence (1) or
leaflet. c, In vivo LLO flipping of wild-type PglK and E510Q mutant expressed absence (2) of LLO (20 mM). Stimulation results in higher Vmax, while the Km
in E. coli SCM6 cells containing the C. jejuni pgl operon30. E.V., empty vector, for ATP remains almost unaltered in the presence (Km(1LLO) 5 0.54 6 0.03
N.I.C., non-induced cells. Anti-glycan refers to HR6 antibody recognizing the mM) or absence (Km(2LLO) 5 0.36 6 0.04 mM) of native LLO.
N-glycan of a substrate protein (see Methods), whereas anti-PglK is used to h, Determination of Km value of ATPase stimulation by native LLO in detergent
monitor PglK expression level. d, Determination of Km values of ATP (rates are normalized against the basal activity in absence of LLO). Error bars
hydrolysis and ATP-driven in vitro tLLO flipping. The black curve represents denote s.d. (n 5 3).
the ATPase rate of PglK at distinct ATP concentrations. The blue curve

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH ARTICLE

Extended Data Figure 4 | Electron density maps. a, b, Stereo views (cross- complete PglK dimer of the structure at 5.9 Å and close-up of the NBDs
eyed) of the 2Fo 2 Fc electron density maps for the complete PglK dimer of the showing the Fo 2 Fc electron density map for the bound ADP molecules.
structures at 2.9 and 3.9 Å, respectively. c, Stereo view of the non- 2Fo 2 Fc and NCS maps are shown at 1.0s level. Fo 2 Fc maps are shown
crystallographic symmetry (NCS)-averaged electron density map for the at 3.0s level.

G2015 Macmillan Publishers Limited. All rights reserved

 ARTICLE RESEARCH

Extended Data Figure 5 | Validation of side-chain register of outward- levels are between 4.0 to 5.0s. In a, anomalous density was observed for 9 out of
occluded PglK model. a–c, Anomalous electron density maps define 10 selenomethionines of PglK.
selenomethionine (a), cysteine-bound mercury (b) and PtCl4 sites (c). Contour

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH ARTICLE

Extended Data Figure 6 | Structural features of PglK. a, Ribbon diagram of occluded state, and the ABC exporter Sav1866 (PDB code 2HYD) in outward-
one PglK subunit depicting the secondary structure arrangement, based on the open state. Transmembrane helices TM1 and TM3 (purple) define the extent of
Sav1866 nomenclature36. b, Conformational changes of TM4 and TM5 the external opening. Subunits in each dimer are shown in orange and grey.
visualized after superposition of subunits of apo-inward-1 (light orange) and d, Side and cytoplasmic view cut-off of PglK apo-inward-1 structure and
apo-inward-2 (dark orange) structures. c, Structures of the antibacterial peptide vacuum electrostatic surface representation showing the internal cavity.
ABC exporter McjD in occluded state (PDB code 4PLO), PglK in outward-

G2015 Macmillan Publishers Limited. All rights reserved

 ARTICLE RESEARCH

Extended Data Figure 7 | PglK crystal packing. a–c, The main crystal contacts of apo-inward-1 (2.9 Å resolution) (a), apo-inward-2 (3.9 Å resolution) (b), and
outward-occluded (5.9 Å resolution) (c) states.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH ARTICLE

Extended Data Figure 8 | Putative interactions of native and synthetic LLO di-N-acetylbacillosamine; yellow square denotes N-acetylgalactosamine; blue
analogues with PglK in apo-inward and outward-occluded states. circle denotes glucose; and blue square denotes N-acetylglucosamine. The LLO
a, b, Native LLO (a) and synthetic LLO analogues (b). Red hexagon denotes polyprenyl tail is shown in purple.

G2015 Macmillan Publishers Limited. All rights reserved

 ARTICLE RESEARCH

Extended Data Figure 9 | Product analysis and control experiments of (104.2 6 10.8%) relative to the amount at t 5 0. ADP alone does not cause a
PglK-catalysed in vitro flipping assay. a, Product analysis of tLLO decrease in the concentration of tLLO in the external liposomes leaflet, ruling
glycosylation catalysed by PglH. The reaction products were analysed by in out a potential sequestration of tLLO in the outward-occludded state of PglK.
vitro glycosylation of a fluorescently labelled substrate peptide (DQNAT c, Determination of tLLO orientation in proteoliposomes. Disrupted liposomes
sequon) catalysed by PglB as reported earlier30,31. Depending on the presence (0.3% Triton X-100) and non-disrupted proteoliposomes were incubated with
and size of the N-glycan, peptides show a different mobility after Tricine–SDS– PglH in the presence of excess GalNAc. The amount of glycopeptide was
PAGE. Bands were visualized using a fluorescence gel scan (488 nm excitation determined from band intensities of fluorescence gel scans. 48.2 6 7.5% of
and 526 nm emission). Lane 1, product of the deglycosylation reaction of tLLO is located in the outer leaflet of the bilayer. d, Determination of PglK
hexasaccharide LLO catalysed by a-N-acetylgalactosaminidase, which removes orientation in proteoliposomes. The fully functional mutant PglK(C269L/
terminal GalNAc molecules. This demonstrates the purity of the LLO used C352S/C386S/C549L/S544C) was reconstituted in proteoliposomes and
with respect to GalNAc. Lane 2, tLLO used in biochemical assays. Lane 3, labelled with negatively charged Alexa Fluor 488 C5 maleimide. The
products of tLLO glycosylation when the tLLO:GalNAc molar ratio is 2:1. Lane fluorescence of non-disrupted and disrupted proteoliposomes was compared.
4, products of tLLO glycosylation when the tLLO:GalNAc molar ratio is 1:1. 51.8 6 2.8% of the PglK molecules are oriented with NBDs facing outwards.
Lane 5, products of tLLO glycosylation when the tLLO:GalNAc molar ratio is Red hexagon denotes di-N-acetylbacillosamine; yellow square denotes
1:10. b, PglK proteoliposomes incubated in the presence of 4 mM ADP. N-acetylgalactosamine.
The level of tLLO in the external leaflet after t 5 40 min remains unchanged

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH ARTICLE

Extended Data Table 1 | X-ray data collection and refinement statistics

Values in brackets are before anisotropic truncation; values in parentheses are for the last resolution shell. Rmerge 5 ShklSi | Ii(hkl) 2 ,I(hkl). | /ShklSi Ii(hkl)
* Before density modification.
{ After density modification.

G2015 Macmillan Publishers Limited. All rights reserved