1
Real-time visualization of perforin nanopore assembly 1
2
Carl Leung
1,2,‡
, Adrian W. Hodel
1,3,‡
, Amelia J. Brennan
4,‡
, Natalya Lukoyanova
2,‡
, Sharon Tran
4
, 3
Colin M. House
5
, Stephanie C. Kondos
6
, James C. Whisstock
6,7
, Michelle A. Dunstone
6,7,8
, Joseph A. 4
Trapani
5,9
, Ilia Voskoboinik
4,9,
*, Helen R. Saibil
2,
* & Bart W. Hoogenboom
1,3,10,
* 5
6
1
London Centre for Nanotechnology, University College London, London, United Kingdom 7
2
Department of Crystallography/Biological Sciences, Institute of Structural and Molecular Biology, 8
Birkbeck College, London, United Kingdom 9
3
Institute of Structural and Molecular Biology, University College London, London, United Kingdom 10
4
Killer Cell Biology Laboratory, Peter MacCallum Cancer Centre, East Melbourne, Australia 11
5
Cancer Cell Death Laboratory, Peter MacCallum Cancer Centre, East Melbourne, Australia 12
6
Department of Biochemistry and Molecular Biology, Monash University, Clayton, Melbourne, 13
Australia 14
7
The ARC Centre of Excellence in Advanced Molecular Imaging, Monash University, Melbourne, 15
Australia 16
8
Department of Microbiology, Monash University, Melbourne, Australia 17
9
Sir Peter MacCallum Department of Oncology, University of Melbourne, Melbourne, Australia 18
10
Department of Physics and Astronomy, University College London, London, United Kingdom 19
20
‡
These authors contributed equally to this work. 21
*Corresponding authors: ilia.voskoboinik@petermac.org (I.V.); h.saibil@mail.cryst.bbk.ac.uk 22
(H.R.S.); b.hoogenboom@ucl.ac.uk (B.W.H.). 23
24
Perforin is a key protein of the vertebrate immune system. Secreted by cytotoxic lymphocytes as 25
soluble monomers, perforin can self-assemble into oligomeric pores of 10-20 nm inner diameter 26
in the membranes of virus-infected and cancerous cells. These large pores facilitate the entry of 27
pro-apoptopic granzymes, thereby rapidly killing the target cell. To elucidate the pathways of 28
perforin pore assembly, we have carried out real-time atomic force microscopy and electron 29
microscopy studies. Our experiments reveal that the pore assembly proceeds via a membrane-30
bound prepore intermediate state, typically consisting of up to ~8 loosely but irreversibly 31
assembled monomeric subunits. These short oligomers convert to more closely packed 32
membrane nanopore assemblies, which can subsequently recruit additional prepore oligomers to 33
grow the pore size. 34
2
Perforin is a pore-forming protein that is expressed in cytotoxic lymphocytes and that is secreted, 35
together with the pro-apoptotic serine proteases granzymes, into the immune synapse formed between 36
the lymphocyte and a cognate virus-infected or cancerous cell. The loss of perforin expression or 37
function is catastrophic and leads to fatal immune dysregulation
1
. Perforin is part of the Membrane 38
Attack Complex/Perforin/Cholesterol-Dependent Cytolysin (MACPF/CDC) superfamily
2,3
. This 39
superfamily also includes the terminal components of the complement pathway in the immune system, 40
as well as the CDC family of bacterial toxins. The pore-forming mechanism of the superfamily 41
involves the unravelling of two α-helical regions that refold into transmembrane hairpins (TMH1 and 42
TMH2). These create giant β-barrel transmembrane pores up to 30 nm in diameter, formed of up to 50 43
monomers (in the case of CDCs). 44
45
The pathways of CDC assembly have been extensively characterized. CDCs first form differently 46
sized assemblies in a prepore intermediate state on the membrane. The prepore oligomers 47
subsequently undergo a cooperative vertical collapse and insertion of subunits into the membrane, 48
resulting in arc- and ring-shaped pores
4-14
. 49
50
By contrast, the pathways of MACPF assembly remain largely unclear, despite recent advances in the 51
structural characterization of perforin
15
and MAC
16,17
pores. Structurally, MACPF proteins differ from 52
CDCs in forming membrane-spanning β-barrels without vertical subunit collapse: Unlike CDCs, (the 53
structurally characterized) MACPFs have transmembrane hairpins that are sufficiently long to span the 54
membrane without such a collapse. The two-component fungal MACPF protein pleurotolysin can 55
form fully assembled prepore intermediates on the target membrane, in common with the CDCs
18
, but 56
such prepore intermediates have yet to be observed for MAC and perforin. Moreover, binding assays
19
57
and structural data
16,17,20
on MAC components indicate that – unlike the CDCs
10
– the MAC can 58
continue to assemble after insertion into the membrane. This argues for a growing-pore instead of the 59
CDC growing-prepore mechanism; similar suggestions have been made for perforin
21,22
. 60
61
Such mechanistic differences may have important physiological implications. A growing-pore 62
mechanism would explain how membrane insertion of the C5b-C8 initiation complex triggers the 63
recruitment of multiple copies of C9 protein to complete the hetero-oligomeric MAC pore
16,17,19,20
, thus 64
facilitating antimicrobial attack. For perforin, the assembly mechanism must be sufficiently rapid to 65
allow efficient diffusion of granzymes into the target cell within the time frame (< 20 sec) of the 66
membrane-repair response of virus-infected and cancerous target cells to perforin pore formation
23
. In 67
view of the limited understanding of perforin pore assembly and the key role of perforin in the 68
immune system
1
, this study is aimed at elucidating the pathways of perforin self-assembly into 69
membrane-spanning oligomeric pores. 70
71
3
Perforin assemblies on and in the target membrane 72
Atomic force microscopy (AFM) provides direct visualization of pore assembly, as has been 73
demonstrated for CDCs by real-time imaging of membrane pore formation at nanometre 74
resolution
7,10,12-14,24
. When imaged by AFM on a supported lipid bilayer, wild-type (WT) perforin 75
forms a heterogeneous distribution of arc- and ring-shaped assemblies with a variable radius of 76
curvature and a height of 11 nm
15,21
(Fig. 1a). This heterogeneous distribution is at least qualitatively 77
consistent between experiments on supported lipid membranes, on liposomes, and on live cells
15,21,22
, 78
and the observed height is approximately equal to the height of a perforin monomer. When imaged 79
with high aspect-ratio AFM probes, both arc and ring assemblies are found to locally remove and thus 80
fully perforate the membrane in the pore lumen (Fig. 1b-e). Membrane perforation by arc assemblies is 81
in line with earlier observations based on electrophysiology measurements on perforin
21
and based on 82
direct AFM visualization of pore lumens for CDCs
10,12,13
. The formation of arc-shaped pores thus 83
appears to be a widespread feature in MACPF/CDC pore formation
25
. 84
85
To distinguish between different stages of membrane pore formation, a disulphide bond was 86
engineered in perforin to inhibit membrane insertion without affecting the monomer-monomer 87
interaction. Mass spectrometry was used to verify the presence of the additional disulphide bond in the 88
mutant (Supplementary Fig. 1), beyond the 8 disulphide bonds already present in WT perforin. This 89
engineered disulphide bond tethers the TMH1 region such that it is unable to reach the membrane 90
(Fig. 2a), similar to previous constructs for CDCs
6,10
and pleurotolysin
18
. Without the reducing agent 91
dithiothreitol (DTT), the TMH1-lock (A144C-W373C) mutant binds to the membrane in the presence 92
of Ca
2+
, but remains completely inactive (standardized haemolysis
26
< 4%; 100% refers to WT activity 93
in control experiments). Pre-incubation of perforin with DTT led to a decrease in WT activity with 94
increasing DTT concentration, suggesting that – under those conditions, which were avoided hereafter 95
– the stability of the protein was affected by the reduction of the native disulphide bonds. Once 96
TMH1-lock protein was bound to the membrane, however, its activity on post-incubation with DTT 97
was identical to that of WT protein without DTT, for a wide range of protein and DTT concentrations 98
(Supplementary Fig. 2). 99
100
Membrane-bound TMH1-lock perforin is detectable in AFM images by the presence of diffuse 101
features or plateaus above the level of the membrane surface (Fig. 2b). These diffuse, mobile features 102
can be induced to convert into static pores (similar to WT pores, Fig. 1) by the addition of DTT 103
(Supplementary Movie 1). To more easily characterize the diffuse (and presumably not membrane-104
inserted) state by AFM, we restricted its diffusion to membrane domains in phase-separated lipid 105
membranes
27
, making use of the preference of perforin to bind to more loosely packed (here: PC-rich, 106
lower domains in Fig. 1) lipid phases
28
. This caused the membrane bound TMH1-lock perforin to 107
4
appear as featureless plateaus with a height of 11 nm above the membrane (Supplementary Fig. 3), 108
consistent with the height of the perforin monomer
15
. 109
110
Based on these results, we distinguish between mobile proteins that are weakly bound to the 111
membrane surface, and immobile, membrane-inserted pores. In this context, it is worthwhile to note 112
that the membranes used here are too fluid to immobilize surface-bound, non-inserted perforin 113
assemblies, as verified by fluorescence recovery after photobleaching of labelled lipids 114
(Supplementary Fig. 4). This also suggests that immobilization of the membrane-inserted pore 115
assemblies is due to their direct contact, across the membrane, with the mica substrate that supports 116
the lipid bilayer in the AFM experiments. These observations are consistent with previous AFM 117
experiments on CDC prepores, which showed high mobility at minute time scale, except when tightly 118
packed on the membrane
7,10
. 119
120
Characterization of perforin prepores 121
We next set out to determine the oligomeric assembly state of the TMH1-lock perforin on the 122
membrane. To image it at higher spatial resolution, we immobilized the membrane-bound TMH1-lock 123
perforin by glutaraldehyde fixation. In the resulting AFM images (Fig. 2c and Supplementary Fig. 5-124
6), the protein is found to have formed linear and curved prepore assemblies that are markedly shorter 125
and less regularly shaped than WT perforin pores (Fig. 1). This observation was confirmed by 126
negative stain electron microscopy (Fig. 2d). In all these experiments, unfixed membrane-bound 127
prepores could be converted to static pores by exposure to DTT (Fig. 2e-h). 128
129
For a more quantitative analysis of subunit packing and assembly size, we calculated single-particle 130
averages of short segments of perforin prepore and pore arcs and rings, and measured lengths of 131
perforin assemblies from the electron microscopy images. Initially (≤15 min, 37
o
C; as in Fig. 2b-d), 132
TMH1-lock perforin assembles as short, membrane-bound prepore oligomers with loose subunit 133
packing (centre-to-centre spacing ± standard deviation: 3.86 ± 0.13 nm; Fig. 3a). With longer 134
incubation times (50 min, 37
o
C), it evolves into a larger and more densely packed late prepore 135
intermediate (2.89 ± 0.22 nm; Fig. 3b). Upon exposure to DTT, it shows the tight packing (2.55 ± 0.08 136
nm; Fig. 3c) seen in WT perforin pores (2.55 ± 0.09 nm; Fig. 3d). On average, the denser packing 137
coincides with increasing assembly size (Fig. 3e-h) and formation of the pore β-barrel (Fig. 3c, d). 138
After exposure to DTT, the TMH1-lock mutant shows a similar distribution of assembly sizes as the 139
WT protein (Fig. 3g, h), except for the relatively high amount of short (2~6 subunits) assemblies, 140
suggesting slow or incomplete reduction of the disulphide lock. Additional class averages for each 141
condition can be found in Supplementary Fig. 7. 142
143
5
The discovery of loosely packed intermediates raises the question of whether their assembly is 144
reversible. We first noted that the shape of the prepore assembly distributions (Fig. 3e, f) is consistent 145
with the predictions for assembly growth by irreversible monomer association
10
. Next, focussing on 146
conditions under which the early, most loosely packed prepore state prevails (15 min incubation at 37 147
o
C; Figs. 2b-d, 3a, e), we hypothesized that irreversible prepore assembly would impair WT pore 148
formation on co-incubation (and thus on co-assembly) with the TMH1-lock mutant. Such an effect has 149
been previously demonstrated for the CDC suilysin
10
. On the other hand, a reversible assembly 150
process would allow a free exchange of WT and TMH1-lock proteins in an assembly (as, e.g., shown 151
for mutants of the CDC perfringolysin O
29
), leading to the formation of assemblies with sufficient WT 152
content to irreversibly transit into transmembrane pores. 153
154
Experimentally, we found that in non-reducing conditions the addition of TMH1-lock protein indeed 155
prevents co-assembled WT perforin from membrane insertion, in a dose-dependent manner (Fig. 4a, -156
DTT). Perforin pore formation was restored upon subsequent unlocking of the mutant with DTT: This 157
excludes nonspecific aggregation as a cause of the inhibitory effect of TMH1-lock on the WT protein 158
(Fig. 4a, +DTT). In additional control experiments, the WT perforin and the unlocked TMH1 mutant 159
(+DTT) were found to have similar activity when incubated separately (Supplementary Fig. 2). These 160
results were corroborated by in-vitro cell-based experiments, in which added TMH1-lock perforin 161
inhibited the cytolytic activity of (a constant amount of) WT perforin in red blood cell lysis assays 162
(Fig. 4b). In addition, in the context of the physiological immune synapse, cytotoxic activity was 163
significantly lower for primary cytotoxic T lymphocytes that co-expressed both perforin variants, 164
compared to cells expressing only WT perforin (Fig. 4c). 165
166
In the AFM images, we also observe the appearance of discrete assemblies with poorly resolved 167
features that correlate with the fraction of TMH1-lock protein present (Fig. 4a, dotted circles). The 168
localized nature of these assemblies indicates that they contain membrane-inserted subunits or are at 169
least strongly interacting with such membrane-inserted pore subunits. However, their poorly resolved 170
appearance implies that they retain some of the prepore mobility. We tentatively interpret these 171
features as small pore assemblies that have nucleated further growth by association of TMH1-lock 172
prepores to the transmembrane pore assembly. 173
174
Real-time imaging of perforin pore assembly 175
Perforin pore assemblies could thus have a nucleating effect on the recruitment of the early prepores to 176
the growing pore. To validate this interpretation, we visualized pore assembly of WT perforin in real 177
time. For this purpose, we slowed the process by reducing the incubation temperature from 37
o
C to 27 178
o
C. This had the effect of slowing pore formation to tens of minutes, i.e., readily accessible within the 179
time resolution of AFM experiments. At 27
o
C, the early stages of incubation showed WT perforin 180
