Direct PIP2 binding mediates stable oligomer formation of the serotonin transporter

Abstract

The human serotonin transporter (hSERT) mediates uptake of serotonin from the synaptic cleft and thereby terminates serotonergic signalling. We have previously found by single-molecule microscopy that SERT forms stable higher-order oligomers of differing stoichiometry at the plasma membrane of living cells. Here, we report that SERT oligomer assembly at the endoplasmic reticulum (ER) membrane follows a dynamic equilibration process, characterized by rapid exchange of subunits between different oligomers, and by a concentration dependence of the degree of oligomerization. After trafficking to the plasma membrane, however, the SERT stoichiometry is fixed. Stabilization of the oligomeric SERT complexes is mediated by the direct binding to phosphoinositide phosphatidylinositol-4,5-biphosphate (PIP₂ ). The observed spatial decoupling of oligomer formation from the site of oligomer operation provides cells with the ability to define protein quaternary structures independent of protein density at the cell surface.

ARTICLE

Received 19 May 2016 | Accepted 28 Nov 2016 | Published 19 Jan 2017

Andreas Anderluh1, Tina Hofmaier2, Enrico Klotzsch3, Oliver Kudlacek2, Thomas Stockner2, Harald H. Sitte2,* & Gerhard J. Schtz1,*

The human serotonin transporter (hSERT) mediates uptake of serotonin from the synaptic cleft and thereby terminates serotonergic signalling. We have previously found by single-molecule microscopy that SERT forms stable higher-order oligomers of differing stoichiometry at the plasma membrane of living cells. Here, we report that SERToligomer assembly at the endoplasmic reticulum (ER) membrane follows a dynamic equilibration process, characterized by rapid exchange of subunits between different oligomers, and by a concentration dependence of the degree of oligomerization. After trafcking to the plasma membrane, however, the SERT stoichiometry is xed. Stabilization of the oligomeric SERT complexes is mediated by the direct binding to phosphoinositide phosphatidylinositol-4,5-biphosphate (PIP2). The observed spatial decoupling of oligomer formation from the site of oligomer operation provides cells with the ability to dene protein quaternary structures independent of protein density at the cell surface.

DOI: 10.1038/ncomms14089 OPEN

Direct PIP2 binding mediates stable oligomer formation of the serotonin transporter

1 Institute of Applied Physics, TU Wien, Wiedner Hauptstrasse 8-10, Vienna 1040, Austria. 2 Center for Physiology and Pharmacology, Institute of Pharmacology, Medical University Vienna, Waehringerstrasse 13A, Vienna 1090, Austria. 3 EMBL Australia Node in Single Molecule Science, School of Medical Sciences, ARC Centre of Excellence in Advanced Molecular Imaging, University of New South Wales, Sydney, New South Wales 2052, Australia. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed to H.H.S.(email: mailto:[email protected]

Web End [email protected] ) or to G.J.S. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089

The human serotonin transporter (SERT) is a 12-pass transmembrane protein targeted to presynaptic nerve terminals and belongs to the neurotransmitter:

sodium symporter (NSS) or solute carrier 6 (SLC6) family1. These transmembrane proteins mediate the high-afnity uptake of neurotransmitters from the synaptic cleft and are, hence, of pivotal importance for synaptic signal transmission by terminating chemical signal transduction between neurons. SERT is the target for antidepressants like serotonin-selective reuptake inhibitors2 as well as for illicitly used drugs such as amphetamines3; the latter act by reversing the transport direction of SERT, provoking release of serotonin (5-HT) into the extracellular space4.

Biochemical and spectroscopic studies have reported that SERT is present as oligomeric complexes at the plasma membrane59. Likewise, oligomerization of a number of other NSS family members has been described10. The oligomeric state, however, does not seem to be crucial for uptake activity: for example, it was found that oligomerization-decient mutants of the GABA transporter (GAT1) retain unchanged transport activity11. Currently, two possible roles of NSS oligomerization are discussed: (i) oligomerization of correctly folded proteins is necessary to pass the quality control for trafcking from the endoplasmic reticulum (ER)12, in the case of SERT specically by allowing the interaction with SEC24C (refs 13,14). (ii) It has been reported that oligomerization is a prerequisite for the reverse operation of the transporter which affords substrate release15.

Using single-molecule uorescence microscopy, we have previously discovered that SERT forms a broad distribution of assemblies ranging from monomers up to pentamers7. The homo-association at the plasma membrane did not depend on SERT surface density and was stable at least over 10 min. We proposed a model based on kinetic trapping of oligomers at the plasma membrane, subsequent to an equilibration which occurred at an unknown subcellular organelle. The site of equilibration and the mechanism behind kinetic trapping, however, remained unclear.

Some arguments pointed our interest to the negatively charged phospholipid phosphatidylinositol-4,5-biphosphate (PIP2). PIP2 is part of a number of signalling pathways, for example, endo- and exocytosis, cell adhesion, cell motility, phagocytosis or G protein-coupled receptor signalling16. It is a minor phospholipid that is mainly found at the cytoplasmic leaet of the plasma membrane, where it occurs at a surface density of about sPIP B20,00060,000 molecules mm 2 (ref. 17).

Pronounced differences exist in the subcellular localization of PIP2: PIP2 comprises about 1% of total lipid at the plasma membrane17, whereas only trace amounts of PIP2 are present at the ER18. PIP2 binding to transmembrane proteins is frequently observed, for example, for ion channels, where it regulates channel activity by inuencing the open probability19. Likewise, we have recently found that the functional activity of SERT and the dopamine transporter (DAT) was inuenced by PIP2 binding20,21: upon enzymatic depletion of PIP2 or mutation of identied PIP2 binding sites, amphetamine-induced substrate efux was markedly reduced whereas uptake rates were essentially unaffected. Similarly, while the oligomeric conguration does not seem to inuence neurotransmitter uptake11, amphetamine-induced neurotransmitter release has been shown to rely on the quaternary arrangement15.

In the present study, we quantify the degree of SERT oligomerization at different subcellular localizations of Chinese hamster ovary (CHO) cells and analyse the kinetics of protomer turnover. We nd evidence that SERT oligomers are pre-formed at the ER following a dynamic equilibrium model. At the plasma membrane, kinetic trapping arrests the oligomers at the

pre-set stoichiometry. Our data suggest that the different subcellular concentrations of PIP2 mediate the differential

SERT oligomerization behaviour, likely by direct physical connection of SERT protomers.

ResultsSERT oligomerization depends on subcellular localization. First, we addressed if SERT oligomerization differs between the plasma membrane and the ER. Monomeric GFP (mGFP) was inserted at the cytosolic N-terminus of SERT to allow for visualization (mGFPSERT)22. Plasma membrane-localized mGFPSERT was recorded at the bottom cell membrane via total internal reection uorescence (TIRF) microscopy. To retain SERT in the ER, we overexpressed a dominant-negative mutant of Sar1a (Sar1a-T39N)13, a small GTPase that regulates the assembly of COPII vesicles for plasma membrane trafcking. Sar1a-T39N is a GDP-restricted mutant which prevents the formation of COPII in a dominant-negative manner, thereby arresting SERT in the ER. ER-retained mGFPSERT was studied at junctions between ER and the plasma membrane, where single-molecule tracking at high signal-to-noise ratio using TIRF microscopy is feasible23,24.

We used the Thinning out clusters while conserving stoichiometry (TOCCSL) technique previously established in our lab7,2527 to determine the oligomeric state of mGFPSERT. TOCCSL extends uorescence recovery after photobleaching (FRAP) to the level of single-molecule uorescence microscopy (Fig. 1a). Typically, the high density of uorescently labelled proteins results in a homogenously labelled surface, thereby precluding direct single-molecule measurements (Fig. 1a,i). In TOCCSL, a small area of the cell membrane is irreversibly photobleached by a strong laser pulse focused through a rectangular pinhole onto the sample (Fig. 1a,ii,iii). The high laser intensity completely abrogates the uorescence signal of the mGFPs in every SERT molecule within the illuminated area, while retaining full brightness outside this area. During the subsequent recovery phase unbleached molecules and oligomers diffuse into the previously bleached area due to Brownian motion (Fig. 1a,iv). In contrast to FRAP experiments, however, we exploit in TOCCSL the very onset of this recovery process, when individual molecules can be monitored as single, clearly distinguishable uorescent spots (Fig. 1a,v). The brightness of these spots was determined and compared with the brightness of single mGFPSERT molecules recorded under the same conditions and in the same subcellular compartments, yielding the statistical distribution of mGFP SERT oligomers. Note that, similar to uorescence correlation experiments, TOCCSL allows only analysis of the mobile fraction of molecules.

We used this experimental strategy to determine the mean aggregation size of mGFPSERT located at the plasma membrane (Fig. 1b, left) or retained at the ER (Fig. 1b, right) of CHO cells. The differences in the subcellular distributions are apparent in TIRF microscopy; while plasma membrane-localized SERT yields a homogenous intensity distribution over the whole interface of the cell with the glass slide (Fig. 1b, left), we observed the characteristic reticular ERplasma membrane junctions upon Sar1a-T39N overexpression (Fig. 1b, right). The majority of the protein was freely mobile in both cellular compartments, yielding mobile fractions of 828% and 6612% (s.e.m.; nZ10 cells) for plasma membrane and for the ER in FRAP experiments, respectively.

We rst conrmed our recent nding of SERT oligomerization at the plasma membrane: the left panel of Fig. 1b shows the brightness distribution, plotted as a probability density

2 NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089 ARTICLE

a

b

Plasma membrane Endoplasmic recticulum

Mean oligomeric size

TIRF excitation

0.02

0.02

pdf (counts1 )

pdf (counts1 )

0.01

0.01

0 50 100 150

0 50 100 150

Brightness (counts)

Brightness (counts)

c

4

4

Mean oligomeric size

3

3

2

2

1

1

0 400 600 800 1,000

200Mean mGFPSERT density (m2) Mean mGFPSERT density (m2)

0 400 1,000

200

600

800

Figure 1 | Determination of mGFP-SERT oligomer sizes by single molecule brightness analysis. (a) Thinning out clusters while conserving stoichiometry of labeling (TOCCSL). Using a eld stop in the laser beam pathway, a small area of the densely uorescently labelled cell membrane (i) is irreversibly photobleached (ii, iii). During the recovery phase (iv), SERT oligomers diffuse back into the bleached area. At the onset of this process, they can be discriminated as single, well separated uorescent spots (v). (b) The oligomeric state of SERT was evaluated in the plasma membrane (left panel) or the endoplasmic reticulum (right panel) by single-molecule brightness analysis. The brightness distributions (in counts) of mGFPSERT complexes are plotted as pdf. The plots show the distribution of the complexes from the TOCCSL image after the recovery phase (black curves) and the measured brightness of a monomer (red curves); see also Supplementary Fig. 3. A t yields the distribution of oligomeric sizes at the respective organelle. Scale bars, 10 mm.

(c) At the plasma membrane, the mean oligomeric size is independent from the density of SERT (n422 cells per datapoint; plotted protein densities were 2917 (s.e.m.) mm 2, 40231 mm 2 and 84056 mm 2; s.e.m. of the mean oligomeric sizes were smaller than 0.05). In contrast, at the ER higher expression levels correlate with larger oligomeric sizes (n419 cells per datapoint; plotted protein densities are 15313 (s.e.m.) mm 2, 18524 mm 2, 34337 mm 2 and 64336 mm 2; s.e.m. of the mean oligomeric sizes were smaller than 0.05).

function (pdf) obtained from the TOCCSL images. A large spread in the oligomer distribution was observed, and the mean oligomeric size did not depend on SERT surface density (Fig. 1c, left). The results were strikingly different, when we determined SERT oligomerization at the ER membrane. While the overall oligomer distribution remained highly heterogeneous (Fig. 1b, right), we found a pronounced increase of SERT oligomer size with increasing mean SERT density at the ER (Fig. 1c, right).

A second hallmark of dynamic size equilibration would be the exchange of subunits. To discriminate between stable association and rapid subunit exchange, we used the previously established method of repetitive TOCCSL runs7 (Fig. 2a). We performed one run per minute (each consisting of a single bleaching pulse and a single recovery image) over 10 min on the

very same cell. Pooling data from multiple cells provided brightness distributions as a function of time. By this procedure, the amount of active uorophores per cell was substantially reduced to about 50%. In this approach, stable interaction of subunits would reduce the total number of observed spots, but would not alter the brightness distribution (Fig. 2a, scenario i). If the exchange rate of subunits was high, however, bleached subunits would mix with unbleached subunits. Over time, this mixing would increase the number of complexes containing both dark and uorescent subunits, thereby shifting the observable oligomeric distribution towards smaller structures (Fig. 2a, scenario ii). At the plasma membrane, we observed no change in the oligomeric state with increasing number of TOCCSL runs. We thereby conrmed our previous results which indicated stability of SERT oligomers at the minutes

NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089

a

Cytosolic side

Photobleaching

Rapid exchange

No exchange

i ii

b

Plasma membrane

1 2 3 4 5 6 Number of subunits N

Endoplasmic recticulum

1 2 3 4 5 6

c

60

60

50

Start1 min 3 min 10 min

50

Start1 min 3 min 10 min

Distribution [afii9825] N(%)

Distribution [afii9825] N(%)

40

40

30

30

20

20

10

10

0

0

Number of subunits N

Figure 2 | Evaluation of the oligomer stability in the plasma membrane and at the ER. (a) To study the stability of SERToligomers we performed repeated TOCCSL runs on the same cells (1 run per minute over 10 min), and determined the brightness distributions in each run. Two different scenarios can be distinguished: if oligomers were stable over the time course of the experiment, the total number of diffraction-limited spots would be reduced without altering the brightness distribution (left, scenario i). In contrast, if oligomers would exchange subunits during the 10 min, increasing numbers of mixed SERToligomers containing both bleached and non-bleached subunits would be observable, thereby shifting the determined oligomeric distribution towards smaller structures (right, scenario ii). (b) Using the repetitive TOCCSL strategy, we have observed no change in the oligomeric distribution of SERT at the plasma membrane (n 20 cells). Oligomeric distributions are shown at the beginning of the experiment (white bars), after 1 min (dark grey), 3 min

(light grey), and after 10 min (black). (c) At the ER, however, we observed rearrangement of subunits over the timescale of the experiment, as can be seen by the shift of the distributions towards lower oligomer sizes (n 22 cells). Error bars show the s.e.m.

timescale (Fig. 2b). At the ER membrane, however, we found a substantial shift towards oligomers with smaller amounts of active uorophores, indicating rapid exchange of subunits between SERT oligomers (Fig. 2c).

Availability of PIP2 affects SERT oligomerization. Given the described SERTPIP2 interaction20 and the strong difference of PIP2 levels between plasma membrane17 and ER18, we hypothesized that PIP2 impacts on SERT oligomerization. We measured SERT oligomerization at the plasma membrane after depleting PIP2 levels via activation of phospholipase Cg (PLCg)

using the direct PLCg-activator m-3M3FBS (ref. 28). PIP2 depletion had no inuence on the mean mGFPSERT surface density (Supplementary Fig. 1). For low SERT surface densities (25 molecules mm 2), we observed a marked shift of SERT oligomers towards smaller complexes, whereas application of the inert orthologue o-3M3FBS did not elicit any effect (38 molecules mm 2, Fig. 3a). At higher SERT densities, however, oligomers increased in size (Fig. 3b), indicating rapid equilibration of the oligomerization reaction.

A consequence of rapid equilibration would be the continuous exchange of subunits between SERT oligomers. Hence, to test for the stability of oligomers we performed repetitive TOCCSL runs after incubating cells with m-3M3FBS. The quaternary arrangement of SERT in oligomers now showed rapid subunit

exchange (Fig. 3c), which indicated that SERT oligomers were indeed liberated from kinetic trapping. Together, these results show that PIP2 depletion results in equilibration of SERT oligomerization and concomitant subunit exchange at the plasma membrane. Of note, SERT oligomerization at the plasma membrane lacking PIP2 resembles the oligomerization behaviour at the ER membrane (Fig. 2).

PIP2 binds directly to SERT and mediates oligomerization. Next, we investigated whether direct PIP2 binding to SERT sustained oligomer formation or whether PIP2 would exert an indirect effect due to the downstream metabolites IP3 or DAG. Recently, we identied the amino acids K352 and K460 as crucial residues for PIP2 binding to SERT20 (Fig. 4a); mutation of both residues to alanine yielded a substantial decrease of PIP2-induced effects on parachloroamphetamine-induced,

SERT-mediated current and efux. Most importantly, the PIP2SERT interaction was shown to be greatly reduced in pull-down experiments20. We used an mGFP-fusion construct of this mutant to evaluate the effect of reduced PIP2 binding on the oligomeric state. The mGFP-tagged SERTK352AK460A double mutant was efciently trafcked to the plasma membrane and showed similar uptake activity as the wild type20. Single-molecule brightness analysis yielded an oligomeric distribution that differed from wild-type SERT (Fig. 4b): the dominant species

4 NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089 ARTICLE

a

b

4

80

m-3M3FBS o-3M3FBS No treatment

3

Distribution [afii9825] N(%)

Distribution [afii9825] N(%)

60

2

40

20

1

0

1 2 3 4 5 6 7 8

0 200 400 600 800Mean mGFPSERT density (m2)

Number of subunits N

Mean oligomeric size

c

80

Start1 min 3 min 10 min

60

40

20

0

1 2 3 4 5 Number of subunits N

Figure 3 | SERT oligomerization at the plasma membrane depends on PIP2 levels. (a) We enzymatically depleted PIP2 at the plasma membrane via activation of phospholipase Cg (PLCg) by incubating cells for 15 min with the direct PLCg-activator m-3M3FBS (10 mM). This led to a marked shift of

SERT complex sizes towards monomers (dark grey bars). As a negative control, incubation with the inert orthologue o-3M3FBS did not yield any effect (light grey bars) in comparison to the untreated cells (white bars) (n420 cells per experimental condition). SERT surface densities were similar: 2514 (s.e.m.) mm 2 (dark grey bars), 3822 mm 2 (light grey bars), 2917 mm 2 (white bars). (b) PIP2 depletion via m-3M3FBS resulted in marked dependence of mGFPSERToligomerization on mGFPSERT surface density (n420 cells per datapoint; plotted protein densities are 2514 (s.e.m.) mm 2, 4824 mm 2, 8417 mm 2, 18728 mm 2 and 50140 mm 2; s.e.m. of mean oligomeric sizes were smaller than 0.05). (c) To test for the effect of

PIP2 depletion on the stability of the oligomers at the plasma membrane, we performed repetitive TOCCSL runs upon incubating cells with m-3M3FBS (1 mM) (n 19 cells). Now, the SERT complexes showed rapid subunit rearrangement, indicating liberation of SERT oligomers from kinetic trapping. The

SERT surface density was 8921 (s.e.m.) molecules mm 2. Error bars show the s.e.m.

are monomers and dimers, while the fraction of trimers and tetramers was reduced to almost baseline levels. The double mutant showed a pronounced density dependence of its oligomeric assembly, which seemed to saturate at a level of B2.8 transporter molecules per oligomer (Fig. 4c). Repetitive

TOCCSL runs revealed rapid protomer exchange (Fig. 4d). Together, mGFPSERT K352AK460A behaved similar as wild-type SERT after PIP2 depletion.

DiscussionAlthough there is a wealth of data supporting the existence of neurotransmitter transporters in oligomeric quaternary structures10, the nature of the interaction between the subunits has yet not been unravelled. Here, we examined the size and stability of oligomeric complexes of SERT at two different subcellular localizations, the ER membrane and the plasma membrane. We found that dynamic equilibration of SERT oligomers occurs at the ER membrane. After trafcking through the secretory pathway, the pre-formed oligomers undergo kinetic trapping at the plasma membrane. Pre-equilibration of subunit binding at the ER membrane and kinetic trapping of oligomerized protomers at the plasma membrane enables the cell to spatially decouple the oligomerization process from the nal site of oligomer operation.

This appears crucial to render the degree of oligomerization insusceptible to different SERT concentrations at various localizations on the plasma membrane.

Our data indicate that the phosphoinositide PIP2 plays an essential role in this process and that the different PIP2 concentrations of the ER membrane and the plasma membrane are responsible for the pronounced differences. While other phosphoinositides would also be plausible candidates for mediating charge-induced SERT oligomerization, some lines of evidence indicate the specic role of PIP2 in this process.

First, PLCg is a highly specic enzyme for PIP2 hydrolysis, leaving other phosphoinositides virtually unaffected29.

Experiments shown in Fig. 3 hence reveal the specic contribution of PIP2 to SERT oligomerization. Second,

PIP2 is by far the most common phosphoinositide in the plasma membrane16, thereby outcompeting other phosphoinositide species by shear concentration effects. Third, due to rapid dephosphorylation the ER membrane contains phosphatidylinositol (PI) and phosphatidylinositol(4)phosphate (PI(4)P), but virtually no PIP2 or other multi-phosphorylated phosphoinositides16. Figures 1 and 2 thus render a role of PI and PI(4)P in SERT oligomerization unlikely.

In the following, we propose a model how PIP2 may cause kinetic trapping of SERT oligomers at the plasma membrane (Fig. 5), which has three cornerstones:

NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089

a

b

60

mGFPSERTK352AK460A

50

Distribution [afii9825] N (%)

Distribution [afii9825] N (%)

mGFPSERT

40

30

K460

K352

20

10

0 1 2 3 4 5 6 7 8

Number of subunits N

c

d

4

80

Mean oligomeric size

1

3

Start1 min 3 min 10 min

60

2

40

20

0 200 400 600 800

Mean mGFPSERT density (m2)

0 1 2 3 4 5Number of subunits N

Figure 4 | Direct binding of PIP2 to SERT mediates oligomerization. (a) Analysis of the electrostatic eld generated by SERT. The nal structure of a 100 ns simulation of a membrane-inserted SERT is shown as viewed from the cytosole (left) or in side-view (right). SERT surface is shown in white, the electrostatic isosurfaces in red (negative potential) and blue (positive potential). For illustration purposes, a PIP2 molecule was placed into the membrane (in space lled representation) in close proximity to the large positively charged area that includes residue K460. (b) We determined the quaternary assembly of the mutant mGFPSERTK352AK460A at the plasma membrane (white bars). Proteins were expressed at a surface density of 8331 (s.e.m.) molecules mm 2. A distinctive shift to monomers and dimers compared with the wild type (grey bars) was observed (n 23 cells).

(c) A pronounced dependence of mean oligomeric state on mGFPSERTK352AK460A surface density was observed, which saturates around 2.8 transporter molecules per oligomer (n419 cells per datapoint; plotted protein densities are 3511 (s.e.m.) mm 2, 11033 mm 2, 23714 mm 2 and 45235 mm 2; s.e.m. of mean oligomeric sizes were smaller than 0.05). (d) Repetitive TOCCSL runs revealed rapid protomer exchange kinetics for this mutant (n 18 cells). Error bars show the s.e.m.

Endoplasmic reticulum- dissociation/association of oligomers

Plasma membrane - stable oligomers

Figure 5 | A model for PIP2 dependent oligomerization of SERT. High PIP2 concentrations at the plasma membrane (left) saturates PIP2 binding sites on SERT, impeding further oligomerization of the subunits. Also disassembly of the oligomers is efciently prevented: in case of PIP2 unbinding, the vacant position is rapidly re-populated by a new PIP2 molecule before the protomers can separate by diffusion. Together, the two effects lead to the kinetic trapping of the oligomeric state at the plasma membrane. At the ER membrane, however, low PIP2 concentrations lead to coexistence of PIP2-ligated and -unligated

SERT (right), which are capable of mutual binding. Hence, such conditions enable fast equilibration of the oligomerization process, including subunit exchange between SERT oligomers.

6 NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089 ARTICLE

(i) two negatively charged phosphate groups on PIP2 bind electrostatically to positively charged patches on the cytosolic face of two SERT molecules

SERT contains a patch with a strong positive electrostatic potential on the intracellular face which is in contact with the polar headgroups of the membrane20 (Fig. 4b). The cationic patch is generated by basic amino acid residues (including K352 and K460). The inositol sugar ring of PIP2 has two functional

PO4 2 groups in position 4 and 5, each harbouring a negative charge of 2 (ref. 30). These functional PO4 2 groups are

oriented towards opposing ends of the inositol ring and would therefore have the possibility to interact with two separate SERT monomers. Thereby, they would effectively act as an ionic bridge between the two SERT transporters.

(ii) abrogation of PIP2 binding leads to rapid equilibration of the oligomerization process

The oligomerization process of SERT rapidly equilibrates under conditions, where PIP2 does not contribute to the subunit association process. This was inferred from the results of three different approaches: the depletion of PIP2 by enzymatic conversion to IP3 and diacylglycerol by activation of PLCg (Fig. 3); the retention of SERT in the ER, where PIP2 levels are extremely low31 (Figs 1c and 2c); the reduction of PIP2 binding by introduction of two point mutations20 (Fig. 4). All conditions revealed qualitatively similar behaviour, that is, the mean size of SERT oligomers became dependent on SERT surface density, and protomers exchanged rapidly between different oligomers. Intrinsic low afnity proteinprotein interactions between SERT protomers seem to mediate this process, however, contributions of low PIP2 concentrations cannot be ruled out. Of note, equilibration of receptor oligomerization at the ER membrane was recently reported for a GPCR32.

At this stage, we do not know at which exact subcellular location oligomerization is ultimately adjusted; low PIP2 levels in virtually all subcellular membranes except for the plasma membrane suggest that the exchange of subunits remains rapid until the fusion of the cargo vesicles with the plasma membrane.

(iii) elevated PIP2 levels saturate PIP2 binding sites on SERT, thereby stabilizing the oligomer

In principle, PIP2 binding to the polybasic patch at the cytosolic face of SERT, which includes the lysine residues in positions 352 and 460, increases the afnity to other, undecorated SERT protomers. Naturally occurring concentrations of PIP2 at

the plasma membrane (B20,00060,000 molecules mm 2; (ref. 33)), however, lead to saturation of SERT with PIP2,

thereby imposing a charge-based repulsive interaction which precludes further oligomerization.

In consequence, over time most oligomers would disassemble to monomeric SERT until equilibrium is reached, where virtually only PIP2-decorated monomers would be present. Importantly, however, there are mechanisms which strongly slow down this equilibration process (kinetic trapping). In fact, the high PIP2 concentration itself may account for such a mechanism.

Following this line of argumentation, SERT disassembly may require the complete dissociation of one or several PIP2 molecules. The high PIP2 concentrations at the plasma membrane would result in immediate replenishment of the vacant position before separation of the individual protomers.

Hence, according to our model PIP2 levels are not responsible for tuning the oligomeric distribution of SERT, but instead determine the kinetics for reaching the equilibrium of the

oligomerization process: at low PIP2 levels (as present at the ER membrane) equilibration is fast, whereas at high PIP2 levels (as present at the plasma membrane) equilibration is substantially slowed down.

In summary, our data show two important steps in the oligomerization of a transmembrane protein: (i) pre-equilibration of subunit binding at the ER membrane and (ii) kinetic trapping of oligomerization at the plasma membrane. By this, the oligomerization process becomes spatially decoupled from the nal site of oligomer operation. This could be important to make the degree of oligomerization insusceptible to different SERT concentrations at various locations on the plasma membrane.

Methods

mGFPSERT construct. eGFP in the peGFP-C1 vector (Clontech) was converted to mGFP34,35 by mutating alanine 207 of eGFP to lysine using Quik-Change mutagenesis Kit XL from Agilent technologies (Primer: Fw: 50-TAC CTG AGC

ACC CAG TCC AAA CTG AGC AAA GAC CCC AAC-30, rev: 50-GTT GGG GTC TTT GCT CAG TTT GGA CTG GGT GCT CAG GTA-30). cDNAs of WT human SERT and the mutated version SERT-K352AK460A (ref. 20) were cloned tothe mutated vector via HindIII and XhoI restriction sites. For full sequencesee Supplementary Note 1. Functionality of mGFPSERT and mGFPSERT K352AK460A was proven by uptake experiments as described before20,36.

Cell culture. CHO cells were cultured at 37 C and 5% CO2 in Dulbeccos modied Eagles medium (DMEM, PAA Laboratories) supplemented with10% FCS (Invitrogen), penicillin and streptomycin. Cell lines are tested monthly for mycoplasma contamination using MycoAlert (Lonza).

Generation of stable cell lines. A CHO cell line (European Collection of Authenticated Cell Cultures) stably expressing mGFPSERT was generated; transient transfection using the FuGENE 6 transfection kit (Promega) was performed according to the manufacturers instructions. A total of6 mg DNA per 10 cm dish was used. The cells were cultured at 37 C and 5% CO2 in

DMEM F-12 Hams (DMEM F-12 HAMS, PAA Laboratories) supplemented with 10% FCS (Invitrogen), penicillin and streptomycin. After 5 days,800 mg ml 1 of G418 (Sigma-Aldrich) was added as a selection marker for stably transfected cells. After 4 weeks the surviving cells were FACS sorted according to their emission in the 488 nm laser line. Only cells showing clear mGFP expression were used for further incubation. Polyclonal cultures were used to ensure a sufcient range of expression levels for density dependent experiments. Thecells were further cultured in DMEM F-12 HAMS supplemented with 10% FBS, penicillin, streptomycin and 400 mg ml 1 of G418.

Prevention of mGFPSERT trafcking from the ER. To retain mGFPSERTin the ER, a dominant-negative mutant of Sar1a (Sar1a-T39N) was used for co-transfection of CHO cells with mGFPSERT13. A quantity of 990 ng of the expression vector encoding Sar1a-T39N was mixed with 10 ng of mGFPSERT containing plasmid, ensuring that all transfected cells showed retention of mGFPSERT in the ER. For all experiments, imaging was performed 1220 h after the transfection, thereby ensuring similar GFP maturation times.

Coating of glass slides. Proper attachment of the cell lines was ensured by coating the glass slides with bronectin (Invitrogen) as follows: the slides were cleanedin 70% ethanol supplemented with 2% hydrochloric acid for 15 min and washed three times for 5 min in dH2O. 90 ml bronectin (50 mg ml 1 in 1 PBS) was

uniformly distributed on the glass and dried at 50 C. Unbound bronectin was removed by washing the glass slides three times with 1 PBS (PAA Laboratories)

before use.

PIP2 depletion experiments were performed on glass slides coated with poly-D-lysine (PDL, Sigma Aldrich). Cleaned slides were incubated with0.1 mg ml 1 PDL for 1 h at 37 C and washed three times before use.

Microscopy. A 488 nm laser (SAPPHIRE HP, Coherent Inc.) was mode-cleaned using a pinhole and the illumination intensity and timing were adjusted with an acousto-optical modulator (model 1205, Isomet) using a custom written software (Labview, National Instruments). The laser beam was focused onto the back-focal plane of a TIRF objective (NA 1.46, 100 a Plan APOCHROMAT, Zeiss)

mounted on an inverted Zeiss Axiovert 200 microscope. The emission light was ltered using appropriate lter sets for GFP and imaged with a back-illuminated liquid nitrogen cooled CCD camera (Micro Max 1300-PB, Roper Scientic).

To restrict the excitation and photobleaching area an adjustable slit aperture (Zeiss) was used as eld stop.

NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089

All experiments were performed at room temperature. Imaging during all experiments was performed using an objective-type TIRF excitation with an excitation power of B0.50.8 kW cm 2 (determined in epi-conguration)

and stroboscopic illumination with excitation times of 3 ms.

Fluorescence recovery after photobleaching. To determine the mobile fraction of mGFPSERT, an B7 7 mm area of the bottom plasma membrane

or the plasma membraneproximal ER was irreversibly photobleached, and the uorescence recovery over time was monitored (n410 cells). Photobleaching and readout were performed in TIRF conguration. Data were analysed using in-house algorithms implemented in Matlab. The central part of the bleached region was evaluated by integrating all counts and normalizing to the pre-bleach image. For calculation of the mobile fraction a of mGFPSERT the resulting curve was tted with II a 1 exp tt .

TOCCSL. TOCCSL experiments were performed as follows (Supplementary Fig. 2). A pre-bleach image was recorded, which was used for determination of the SERT surface density. After a tpre 50 ms, a conned region of the cell

membrane was photobleached for tbleach 600800 ms with a high laser power

of B 57 kW cm 2. To check for complete bleaching, a post-bleach image was recorded tpost 40 ms after the bleach pulse. Finally, the TOCCSL

image was recorded after an adjustable recovery time of trecovery 1,50012,000 ms.

Images were acquired at low excitation power of B0.50.8 kW cm 2 (all excitation intensities were determined in epiconguration).

Brightness analysis. For single-molecule analysis, images were analysed using in-house algorithms implemented in MATLAB (MathWorks). Individual diffraction-limited signals were selected and tted with a Gaussian intensity prole. The tting routine yielded the single spot brightness B, which was used to determine the oligomeric state of SERT7,26,27,37. The obtained brightness values of each diffraction-limited spot in the TOCCSL image were plotted as a pdf r(B). To obtain the brightness distribution of single mGFP molecules r1(B), cells were extensively photobleached, which reduced the amount of active uorophores to a few molecules per mm2, so that each potential mGFPSERT oligomer containedonly one active uorophore at maximum. By autoconvolution, the monomer brightness distribution was used to calculate the brightness distributions for dimers r2(B), trimers r3(B) and so on. The overall single spot brightness distribution r(B) was then tted by a linear combination of r1(B), r2(B), r3(B) and so on:

r B

XNN1 aN rN B

1

Fitting r(B) with equation (1) yielded the fractions aN of the different numbers of co-diffusing active mGFP molecules (with the number of mGFP molecules; Supplementary Fig. 3). Note that aN is proportional to the oligomeric size of SERT, butdue to incomplete mGFP maturationit slightly underestimates the degree of SERT oligomerization38.

At least 750 datapoints were used for n-mer calculations. Using simulation approaches, this sample size was shown to be sufcient for obtaining statistically signicant results39. To determine error bars, we performed a bootstrapping analysis. Briey, randomly chosen subsamples containing 50% of the data were analysed using equation (1); shown error bars represent the obtained s.d. from 100 repetitions for each oligomeric size divided by 2

p .

Mean oligomeric sizes were determined by Nmean P

N a

N .

Repetitive TOCCSL experiments. To study oligomer stability, a repetitive TOCCSL protocol was applied. One TOCCSL run per minute was performed on the same region of each cell, following the timing protocol shown in Supplementary Fig. 2. Single runs were repeated over 10 min starting from the rst bleach pulse. Both bleaching and image acquisition were done in TIRF mode. The TOCCSL image of each run was used for brightness analysis.

Determination of the SERT surface density. The mean uorescence intensity per mm2 of the bottom plasma membrane was calculated and divided by the mean single-molecule brightness of mGFPSERT. To calculate SERT surface densities at plasma membraneproximal ER, the area fraction of ERPM junctions was determined from super-resolution images23, and the mean intensity of ER-retained SERT was divided by the single-molecule brightness and the respective area fraction.

Enzymatic PIP2 depletion. For activation of phospholipase Cg, cells were incubated for 20 min at 37 C with 25 mM 2,4,6-trimethyl-N-[3-(triuoromethyl) phenyl]benzenesulfonamide (m-3M3FBS, Sigma-Aldrich) in Hanks Balanced Salt Solution (HBSS), with calcium and magnesium (Sigma-Aldrich) supplemented with 2% FCS. m-3M3FBS remained in the imaging buffer during measurements. For the evaluation of the interaction kinetics (repetitive TOCCSL experiments) a concentration of 1 mM m-3M3FBS was used. As a negative control, the inactive ortho-analog o-3M3FBS (Tocris) was used.

Molecular modelling and simulations. The crystal structure of human SERT (PDB ID: 5I6X)40 was used as starting structure. The missing side chains were modelled with MODELLER version 9.15 (ref. 41), creating 100 models usingthe automodel procedure. The best three models, selected according to the DOPE score42, were inserted into a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine containing membrane using the membed procedure43.

The membrane was pre-equilibrated to contain the SERT transporter44. The system was electroneutralized and 150 mM NaCl were added. The environment of SERT was equilibrated while position restraining the transporter. SERT was than released by reducing the position restraints on SERT in four steps, applying 1,000, 100, 10 and 1 kJ mol 1, respectively, each time simulating for 2.5 ns.

Production runs of 100 ns long equilibrium molecular dynamics simulations were carried out with the GROMACS 5.1 MD package45 using the AMBER force eld46 for the protein and the Berger parameters47 for the membrane.

The system was maintained at 310 K while coupling protein, membrane and solvent independently using the v-rescale thermostat48. The pressure was maintained at 1 bar using the weak coupling algorithm, electrostatic interactions were calculated using the smooth particle mesh Ewald method49 with a 1.0 nm cutoff. LennardJones interactions were evaluated applying a 1.0 nm cutoff. Long range corrections for energy and pressure were applied.

Code availability. The Matlab Source code for TOCCSL analysis is available at https://github.com/schuetzgroup/TOCCSL_analysis

Web End =https://github.com/schuetzgroup/TOCCSL_analysis .

Data availability. Data supporting the ndings of this study are available within the article and its Supplementary Information Files and from the corresponding author upon reasonable request. The PDB accession code 5I6X (SERT Structure) was used in this work.

References

1. Saier, M. H., Tran, C. V. & Barabote, R. D. TCDB: the transporter classication database for membrane transport protein analyses and information. Nucleic Acids Res. 34, D181D186 (2006).

2. Schloss, P. & Williams, D. C. The serotonin transporter: a primary target for antidepressant drugs. J. Psychopharmacol. 12, 115121 (1998).

3. Kristensen, A. S. et al. SLC6 neurotransmitter transporters: structure, function, and regulation. Pharmacol. Rev. 63, 585640 (2011).

4. Sitte, H. H. & Freissmuth, M. Amphetamines, new psychoactive drugs and the monoamine transporter cycle. Trends Pharmacol. Sci. 133, 163166

2015:

5. Schmid, J. A. et al. Oligomerization of the human serotonin transporter and of the rat GABA transporter 1 visualized by uorescence resonance energy transfer microscopy in living cells. J. Biol. Chem. 276, 38053810

2001:

6. Just, H., Sitte, H. H., Schmid, J. A., Freissmuth, M. & Kudlacek, O. Identication of an additional interaction domain in transmembrane domains 11 and 12 that supports oligomer formation in the human serotonin transporter. J. Biol. Chem. 279, 66506657 (2004).

7. Anderluh, A. et al. Single molecule analysis reveals coexistence of stable serotonin transporter monomers and oligomers in the live cell plasma membrane. J. Biol. Chem. 289, 43874394 (2014).

8. Kilic, F. & Rudnick, G. Oligomerization of serotonin transporter and its functional consequences. Proc. Natl Acad. Sci. USA 97, 31063111

2000:

9. Jess, U., Betz, H. & Schloss, P. The membrane-bound rat serotonin transporter, SERT1, is an oligomeric protein. FEBS Lett. 394, 4446 (1996).

10. Sitte, H. H., Farhan, H. & Javitch, J. A. Sodium-dependent neurotransmitter transporters: oligomerization as a determinant of transporter function and trafcking. Mol. Interv. 4, 3847 (2004).

11. Scholze, P., Freissmuth, M. & Sitte, H. H. Mutations within an intramembrane leucine heptad repeat disrupt oligomer formation of the Rat GABA transporter1. J. Biol. Chem. 277, 4368243690 (2002).12. Farhan, H., Freissmuth, M. & Sitte, H. H. in Neurotransmitter Transporters Vol. 175 (eds Sitte, H. & Freissmuth, M.) 233249 (Springer, 2006).

13. Farhan, H. et al. Concentrative export from the endoplasmic reticulum of the g-aminobutyric acid transporter 1 requires binding to SEC24D. J. Biol. Chem.

282, 76797689 (2007).14. Sucic, S. et al. Switching the clientele: a lysine residing in the C terminus of the serotonin transporter species its preference for the coat protein complex II component SEC24C. J. Biol. Chem. 288, 53305341 (2013).

15. Seidel, S. et al. Amphetamines take two to tango: an oligomer-based counter-transport model of neurotransmitter transport explores the amphetamine action. Mol. Pharmacol. 67, 140151 (2005).

16. Di Paolo, G. & De Camilli, P. Phosphoinositides in cell regulation and membrane dynamics. Nature 443, 651657 (2006).

17. Hilgemann, D. W. Local PIP2 signals: when, where, and how? Pgers Archiv.Eur. J. Physiol. 455, 5567 (2007).

8 NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14089 ARTICLE

18. Tran, D. et al. Cellular distribution of polyphosphoinositides in rat hepatocytes. Cell. Signal. 5, 565581 (1993).

19. Suh, B.-C. & Hille, B. PIP2 is a necessary cofactor for ion channel function: how and why? Annu. Rev. Biophys. 37, 175195 (2008).

20. Buchmayer, F. et al. Amphetamine actions at the serotonin transporter rely on the availability of phosphatidylinositol-4,5-bisphosphate. Proc. Natl Acad. Sci. USA 110, 1164211647 (2013).

21. Hamilton, P. J. et al. PIP2 regulates psychostimulant behaviors through its interaction with a membrane protein. Nat. Chem. Biol. 10, 582589

2014:

22. Fenollar-Ferrer, C. et al. Structure and regulatory interactions of the cytoplasmic terminal domains of serotonin transporter. Biochemistry 53, 54445460 (2014).

23. Anderluh, A. et al. Tracking single serotonin transporter molecules at the endoplasmic reticulum and plasma membrane. Biophys. J. 106, L33L35 (2014).

24. Nagaya, H. et al. Regulated motion of glycoproteins revealed by direct visualization of a single cargo in the endoplasmic reticulum. J. Cell Biol. 180, 129143 (2008).

25. Moertelmaier, M., Brameshuber, M., Linimeier, M., Schtz, G. J. & Stockinger,H. Thinning out clusters while conserving stoichiometry of labeling. Appl. Phys. Lett. 87, 263903 (2005).26. Brameshuber, M. et al. Imaging of mobile long-lived nanoplatforms in the live cell plasma membrane. J. Biol. Chem. 285, 4176541771 (2010).

27. Madl, J. et al. Resting state orai1 diffuses as homotetramer in the plasma membrane of live mammalian cells. J. Biol. Chem. 285, 4113541142

2010:

28. Bae, Y. S. et al. Identication of a compound that directly stimulates phospholipase C activity. Mol. Pharmacol. 63, 10431050 (2003).

29. Kadamur, G. & Ross, E. M. Mammalian phospholipase C. Annu. Rev. Physiol. 75, 127154 (2013).

30. Kooijman, E. E., King, K. E., Gangoda, M. & Gericke, A. Ionization properties of phosphatidylinositol polyphosphates in mixed model membranes. Biochemistry 48, 93609371 (2009).

31. Stefan, C. J. et al. Osh proteins regulate phosphoinositide metabolism at ER-plasma membrane contact sites. Cell 144, 389401 (2011).

32. Teichmann, A. et al. The Specic monomer/dimer equilibrium of the corticotropin-releasing factor receptor type 1 is established in the endoplasmic reticulum. J. Biol. Chem. 289, 2425024262 (2014).

33. Hilgemann, D. Local PIP2 signals: when, where, and how? Pgers Archiv.Eur. J. Physiol. 455, 5567 (2007).

34. Zacharias, D. A., Violin, J. D., Newton, A. C. & Tsien, R. Y. Partitioning of lipid-modied monomeric GFPs into membrane microdomains of live cells. Science 296, 913916 (2002).

35. Yang, A. et al. A chemical biology route to site-specic authentic protein modications. Science 354, 623626 (2016).

36. Sucic, S. et al. The N terminus of monoamine transporters is a lever required for the action of amphetamines. J. Biol. Chem. 285, 1092410938 (2010).

37. Schmidt, T., Schtz, G. J., Gruber, H. J. & Schindler, H. Local stoichiometries determined by counting individual molecules. Anal. Chem. 68, 43974401 (1996).

38. Ulbrich, M. H. & Isacoff, E. Y. Subunit counting in membrane-bound proteins. Nat. Methods 4, 319321 (2007).

39. Brameshuber, M. & Schutz, G. J. Detection and quantication of biomolecular association in living cells using single-molecule microscopy. Methods Enzymol. 505, 159186 (2012).

40. Coleman, J. A., Green, E. M. & Gouaux, E. X-ray structures and mechanism of the human serotonin transporter. Nature 532, 334339 (2016).

41.ali, A. & Blundell, T. L. Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol. 234, 779815 (1993).

42. Shen, M.-Y. & Sali, A. Statistical potential for assessment and prediction of protein structures. Protein Sci. 15, 25072524 (2006).

43. Wolf, M. G., Hoeing, M., Aponte-Santamara, C., Grubmller, H.& Groenhof, G. g_membed: Efcient insertion of a membrane protein into an equilibrated lipid bilayer with minimal perturbation. J. Comput. Chem. 31, 21692174 (2010).

44. Koban, F. et al. A salt bridge linking the rst intracellular loop with the C terminus facilitates the folding of the serotonin transporter. J. Biol. Chem. 290, 1326313278 (2015).

45. Hess, B., Kutzner, C., van der Spoel, D. & Lindahl, E. GROMACS 4: algorithms for highly efcient, load-balanced, and scalable molecular simulation. J. Chem. Theory Comput. 4, 435447 (2008).

46. Lindorff-Larsen, K. et al. Improved side-chain torsion potentials for the Amber ff99SB protein force eld. Proteins 78, 19501958 (2010).

47. Berger, O., Edholm, O. & Jahnig, F. Molecular dynamics simulations of a uid bilayer of dipalmitoylphosphatidylcholine at full hydration, constant pressure, and constant temperature. Biophys. J. 72, 20022013 (1997).

48. Bussi, G., Donadio, D. & Parrinello, M. Canonical sampling through velocity rescaling. J. Chem. Phys. 126, 014101 (2007).

49. Darden, T., York, D. & Pedersen, L. Particle mesh Ewald: an N.

slog(N) method for Ewald sums in large systems. J. Chem. Phys. 98, 1008910092 (1993).

Acknowledgements

This work was supported by the Austrian Science Fund/FWF (F3506-B20 and W1232 to H.H.S., F3519-B20 to G.J.S. and F3524-B20 to T.S.). E.K. acknowledges funding from in part FEBS Long-term Fellowship and the ARC DECRA Fellowship.

Author contributions

A.A. performed the measurements and data analysis. T.H. and O.K. provided expression vectors and stably expressing cell lines. A.A., T.H., E.K., T.S., H.H.S. and G.J.S. designed the experiments. T.S. performed computational modelling. A.A., G.J.S., H.H.S. andT.S. wrote the manuscript. All authors participated in the discussion of results and contributed to the preparation of the manuscript.

Additional information

Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications

Web End =http://www.nature.com/ http://www.nature.com/naturecommunications

Web End =naturecommunications

Competing nancial interests: The authors declare no competing nancial interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/

Web End =reprintsandpermissions/

How to cite this article: Anderluh, A. et al. Direct PIP2 binding mediates stable oligomer formation of the serotonin transporter. Nat. Commun. 8, 14089doi: 10.1038/ncomms14089 (2017).

Publishers note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afliations.

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

Web End =http://creativecommons.org/licenses/by/4.0/

r The Author(s) 2017

NATURE COMMUNICATIONS | 8:14089 | DOI: 10.1038/ncomms14089 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9