An electrically actuated molecular toggle switch

Abstract

Molecular electronics is considered a promising approach for future nanoelectronic devices. In order that molecular junctions can be used as electrical switches or even memory devices, they need to be actuated between two distinct conductance states in a controlled and reproducible manner by external stimuli. Here we present a tripodal platform with a cantilever arm and a nitrile group at its end that is lifted from the surface. The formation of a coordinative bond between the nitrile nitrogen and the gold tip of a scanning tunnelling microscope can be controlled by both electrical and mechanical means, and leads to a hysteretic switching of the conductance of the junction by more than two orders of magnitude. This toggle switch can be actuated with high reproducibility so that the forces involved in the mechanical deformation of the molecular cantilever can be determined precisely with scanning tunnelling microscopy.

ARTICLE

Received 17 Aug 2016 | Accepted 23 Jan 2017 | Published 9 Mar 2017

Lukas Gerhard1, Kevin Edelmann1,2, Jan Homberg1, Michal Valek1, Safa G. Bahoosh3, Maya Lukas1, Fabian Pauly3, Marcel Mayor1,4,5 & Wulf Wulfhekel1,2

Molecular electronics is considered a promising approach for future nanoelectronic devices. In order that molecular junctions can be used as electrical switches or even memory devices, they need to be actuated between two distinct conductance states in a controlled and reproducible manner by external stimuli. Here we present a tripodal platform with a cantilever arm and a nitrile group at its end that is lifted from the surface. The formation of a coordinative bond between the nitrile nitrogen and the gold tip of a scanning tunnelling microscope can be controlled by both electrical and mechanical means, and leads to a hysteretic switching of the conductance of the junction by more than two orders of magnitude. This toggle switch can be actuated with high reproducibility so that the forces involved in the mechanical deformation of the molecular cantilever can be determined precisely with scanning tunnelling microscopy.

DOI: 10.1038/ncomms14672 OPEN

An electrically actuated molecular toggle switch

1 Institut fr Nanotechnologie, Karlsruhe Institute of Technology, 76344 Eggenstein-Leopoldshafen, Germany. 2 Physikalisches Institut, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany. 3 Department of Physics, University of Konstanz, 78457 Konstanz, Germany. 4 Department of Chemistry, University of Basel, St Johanns-Ring 19, 4056 Basel, Switzerland. 5 Lehn Institute of Functional Materials (LIFM), Sun Yat-Sen University, Guangzhou 510275, China. Correspondence and requests for materials should be addressed to L.G. (email: mailto:[email protected]

Web End [email protected] ) or to F.P. (email: mailto:[email protected]

Web End [email protected] ) or to M.M. (email: mailto:[email protected]

Web End [email protected] ).

NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672

The formation of a molecular junction and thus ultimately the performance of a single-molecule device is determined by quantum mechanics. Thus, junctions of chemically

identical molecules should behave identically allowing for a reliable operation, which is one of the fundamental advantages of molecular electronics1,2. In the past, break-junction experiments largely contributed to the progress of understanding electron transport in single-molecule junctions35. In these measurements, however, the molecular adsorption and the arrangement of the atoms in the metallic electrodes are a priori unknown and often many different nearly degenerate junction geometries exist. This in turn leads to large variations of the electronic properties of the individual junctions and requires a statistical analysis of the data6. Furthermore, break-junction techniques inherently lack the possibility of lateral resolution, which neither allows to directly identify the molecule of interest at the electrode surface nor to deliberately explore variations in bond congurations. Reproducible performance and well-dened characteristics of the molecular junction, however, are prerequisites for both the comprehension of molecular devices and their implementation into future electronic devices7,8. Recently, scanning tunnelling microscope (STM) and atomic force microscope (AFM) set-ups9 emerged as a powerful alternative to break-junction methods. In this approach, the adsorption geometry of the molecules can be determined before contacting a molecule with the tip at specic positions1014. Forces in the range of piconewtons that result from the onset of bond formation can be derived from the frequency shift of the cantilever of an AFM oscillating by some tens to some hundreds of picometres12,14,15.

With a view to future applications, it is necessary to control the properties of single-molecule junctions by external means16 such as mechanical forces17, electric elds1820, currents21,22 or light23,24. In this respect, it is helpful to design the molecule specically such that it binds to the substrate electrode in a well-dened geometry and to lift the functional group of the molecule from the metallic substrate. This can be achieved by three-dimensional molecular carrier platforms that allow for a well-dened arrangement of molecular subunits with respect to the substrate. Tremendous progress in the synthesis of such molecules has been made recently2529.

In this work, we present low-temperature STM measurements and density functional theory (DFT) calculations on tripodal spirobiuorene derivatives (spiro), which are designed to rmly bind to a gold surface via three sulfur atoms at the feet27. They carry, as functional subunit, a freely suspended nitrile group that is well separated from the metallic substrate, freely accessible by the STM tip and mounted on a rigid subunit. In the following, we show that the high reproducibility of the process of contact formation of the nitrile group to a gold electrode allows us to build a fully controllable electromechanical molecular toggle switch and to deduce the forces that are needed to stretch the molecular junction.

ResultsDeposition of spiro A and spiro B molecules. The molecules, entitled spiro A and spiro B, are presented in Fig. 1a,b, and their design comprises a rmly immobilized platform, exposing a rigidly mounted freely suspended cantilever. They were deposited onto clean Au(111) surfaces by a spray deposition method described previously27, followed by mild annealing in ultra-high vacuum (UHV), which promotes the deprotection of the sulfur anchor groups and the proper binding to the gold substrate. Both molecules arrange in ordered islands with a well-dened adsorption geometry (Fig. 1c,d). While the synthesis of

spiro A is already reported27, the synthesis of spiro B is provided in the Methods section. Details of the adsorption geometry are presented in Supplementary Fig. 14 and Supplementary Note 9. Simulated STM images are discussed in Supplementary Figs 12 and 13 and Supplementary Note 8.

Mechanical opening and closing of the molecular junctions. To study the distance-controlled contact formation to the nitrile group, we recorded the current while the STM tip was approached towards the molecule (black curves in Fig. 2a,b) and was retracted back to its initial position (red curves in Fig. 2a,b)30,31. At a certain critical distance zON, the molecular contact closes and the measured conductance jumps to a higher value ( 380 and 11 pm in Fig. 2a,b). Upon retraction, at a larger distance zOFF between tip and sample ( 180 and 4 pm in Fig. 2a,b), the

contact reopens and the conductance jumps back to a lower value. The experimentally observed hysteretic behaviour results from the short-range binding forces between the tip and the nitrile group of the molecule combined with elastic deformations of the molecular bridge, which comprises also surface atoms of the tip and substrate. We manage to extract the forces involved in this purely elastic deformation of the molecular junction as described at the end of the manuscript.

To further investigate the bistable behaviour found in the molecular junctions, we performed DFT calculations of junction opening and closing processes (see the Methods section for details on the applied DFT procedures). In particular, we investigated whether the molecule alone is responsible for the hysteresis in conductancedisplacement curves, as measured in Fig. 2b, or to what extent the electrodes and moleculeelectrode interfaces need to be involved in the ON-to-OFF-state switching. To perform efcient calculations, we assumed that a single Au atom represents the tip. The S atoms of spiro B were assumed to be either completely xed in the plane z 0 or free to move to

describe the role of the substrate. The Au atom was then placed B2 directly above the N atom of a fully relaxed molecule.

Subsequently, it was brought out of and into contact by moving it along the z axis away in steps of 0.1 until the contact broke and the molecule did not feel the tip. Afterwards, the procedure was reversed and the tip approached the molecule until the contact formed again and the Au atom reached its starting position. We computed the change in the AuN bond length as a function of the height of the Au atom above the z 0 plane, as displayed in

Fig. 2c,d. In the case that all S atoms are xed completely, we nd practically no hysteresis (Fig. 2c). But if the S atoms are allowed to move in the xy plane, the legs gain more exibility and a pronounced hysteresis is observed (Fig. 2d). This shows that also the assumed anchoring of the molecule to the substrate plays a crucial role for the junction bistability, or in other words the mechanical properties of the moleculeelectrode interfaces as well as those of the electrodes. Generally speaking, the energy surface at those tip distances, where the hysteresis is observed, must exhibit a double-well potential since both ON and OFF states are stable, of rather similar energy but separated by an energy barrier, thus allowing the state of the system to depend on process history. This means that the elastic energy in the molecular junction must be close to the energy of the AuN bond between the tip and the molecular nitrile group.

At rst sight, it can be seen that the conductance of spiro A is markedly lower than that of spiro B (5 10 6G0 in Fig. 2a

compared to 4 10 4G0 in Fig. 2b, where the conductance is

given in units of G0 2e2/h). To better understand this difference,

we have computed the conductance of spiro A and spiro B with the DFT32 and DFT S methods3335. As described further in

the Methods section, the conductance can be obtained from the

2 NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672 ARTICLE

a

N

b

N

S

S

O

S

O

S

O

O

S

S

O

O

c

d

50 pA

20 pA

40 pA

16 pA

30 pA

12 pA

20 pA

8 pA

10 pA

4 pA

0 pA

0 pA

Figure 1 | Adsorption geometry of the tripodal molecules. (a,b) Chemical structure of spiro A and spiro B as synthesized. (c,d) STM image of ordered layers of spiro A and spiro B recorded in constant height mode at a bias voltage of 1.9 V and 700 mV with the corresponding molecular models superimposed to scale. A red cross indicates a typical position for contact formation. The length of the scale bars is 1 nm.

0

a

b

Conductance G/G 0

106

107

108

Conductance G/G 0

104

105

106

Approach

400

300

200

100

20 15 10 5 0 5 10

z (pm) z (pm)

c d

500

Bond length (pm)

500

Bond length (pm)

450

450

400

400

350

350

300

300

250

250

1.10 1.15 1.20 1.25 1.30 1.35 Tip height (nm)

1.10 1.15 1.20 1.25 1.30 1.35 Tip height (nm)

Figure 2 | Contact formation by approaching the tip towards the sample. (a) Iz curve recorded above the head group of spiro A at a bias of 300 mV. z values are given with respect to the starting position (z 0), and the sweep direction is indicated by arrows. (b) Iz curve recorded above the head

group of spiro B at a bias of 250 mV. (c,d) DFT calculations of the opening and closing of molecular junctions containing spiro B as a result of the tip movement with c, all S atoms xed in their positions, and d, all S atoms free to move in the plane z 0, providing an enhanced mechanical exibility.

Plotted is the distance between the nitrogen atom of the molecular head group and the gold tip atom as a function of the tip height above the surface at z 0.

NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672

DFT

a

b c

g

100

101 102

Transmission

103 104 105

100

101

102

a

b

c

d

e

f

0 2 EEF (eV)

d e f h

DFT+

Transmission

103

104

105

2

a

b

c

d

e

f

2 0 2

Figure 3 | Transport through tripodal molecules on Au(111) in different congurations. (ad) Spiro A. (e,f) Spiro B. (g) Computed transmission curves in DFT and (h) DFT S frameworks are shown for congurations af.

Table 1 | Conductance of the molecular junction.

a b c d e f GDFT 1.6 10 3 2.1 10 3 5.1 10 5 3.9 10 5 5.2 10 3 4.9 10 3

GDFT S 8.0 10 4 5.0 10 4 2.1 10 5 1.4 10 5 2.5 10 3 2.8 10 3

Experiment 5 10 6 4 10 4

Computed conductance of the congurations shown in Fig. 3 using DFT and DFT S methods in comparison with experimental values. Conductances are given in units of G 2e

2/h.

EEF (eV)

energy-dependent transmission function t(E) via equation (4), if phase-coherent transport is assumed. This reduces to GEG0t(EF)

at low temperatures with the Fermi energy EF, and the conductance can thus be estimated from energy-dependent transmission curves. They are shown for different junction geometries of spiro A and B in Fig. 3, and values of the conductance (at vanishing temperature) are listed in Table 1. We observe that the short molecule spiro B shows a larger conductance than spiro A, as expected for off-resonant transport. A rotation of the assumed atomically sharp tip has basically no inuence on transport, but a rotation of the phenyl ring at the head, that may be induced by tipmolecule interactions, leads to a strong suppression of the conductance of spiro A by more than one order of magnitude for perpendicular ring orientation. We observe that the conductance values determined from DFT are consistently a factor of around 24 larger than those of DFT S. This results

from an opening of the highest occupied molecular orbitallowest unoccupied molecular orbital gap that is clearly visible for t(E) in

Fig. 3g,h. DFT S results have been reported to show a good

agreement with experimental conductance values3335, since the DFT S method is constructed to yield improved quasiparticle

energies for the molecule in the metallic junction. For this reason, we will compare below only the DFT S results to the

experiment. Further details of the DFT S calculations can be

found in the Methods section, and Supplementary Note 7 and Supplementary Fig. 11.

To summarize Table 1 and compare to the experiment, values found for spiro B for different tip congurations are centred B2.65 10 3G0, that is, a factor of 7 above the experimental

results. As the terminal phenyl ring of spiro A is freely revolving around the ethynyl connection, we also considered the twisted arrangement of the phenyl ring. In its untwisted conguration, the conductance is B6.5 10 4G0, which is larger than the

experimental value by a factor of 130. A calculation for a twist angle of 90 leads to a value of B1.75 10 5G0 which is, similar

to the case of spiro B, within an order of magnitude from the measurements. In full agreement with the simulated STM images presented in the Supplementary Figs 12 and 13, this indicates that the terminal phenyl ring twists in the presence of the tip.

Typically, such contacting measurements involve plastic deformations of the electrodes or displacements of the molecule, which leads to a widespread range of conductance values of the junction. Robust and reproducible conductance switching, however, is essential with regard to possible applications and requires suitable anchoring groups as is the case for the molecular tripods discussed in this work. On one side, the spiro molecules are rmly attached to the substrate via three sulfur bonds, and on

4 NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672 ARTICLE

a b

5

103

104

105

Conductance G/G 0

4

3

2

1

y(nm)

103

104

105

106

107

800

700

600

500

400

Counts

300

200

100

Conductance G/G0

106

107

Approach

Retract

200 150 50 0

100

0 0 1 2 3 4 5 6 7 x (nm)

z (pm)

c d

103

104

ON

OFF

ON

OFF

Conductance G/G 0

106

107

103

104

105

Conductance G/G 0

106

107

105

200 150 100 50 0

Figure 4 | Reproducibility of contact formation. (a) Conductance measurements with Iz curves recorded at every pixel above spiro B moleculesat a bias of 70 mV. The colour code shows the conductance at the closest distance (z 210 pm) with molecular structure superimposed to scale.

(b) Ten Iz curves recorded on equivalent positions of 10 different molecules marked by red dots in a. (c) Two-dimensional histogram of all 14,000 approach and retraction curves. (d) One-dimensional histogram showing the clearly separated conductance values of the ON and the OFF state including all Iz curves.

0 5,000 10,000Counts

z (pm)

the other side, a nitrile group forms a coordinative bond to the gold tip that can be reproducibly opened and closed. Several thousand conductance curves were recorded on an array of spiro B molecules and all molecular junctions formed resulted in basically the same value of conductance as can be seen in the conductance map presented in Fig. 4a, where all molecular head groups show up in identical colour. Iz curves recorded on equivalent positions of different molecules show identical switching behaviour (Fig. 4b). As can be seen in the corresponding conductance histograms (see Fig. 4c,d, including all 14,000 Iz curves on top of the molecular head group, near to it, on the side of the molecule and above the gold surface), the ONOFF ratio of the switch amounts to about two orders of magnitude with no overlap of the corresponding conductance peaks. The high reproducibility of the nitrilegold contact in the STM experiment goes well beyond break-junction experiments of molecules with similar anchor groups or similar core structures36,37. It allows to unambiguously attribute the remaining variation of the conductance of the ON state to the difference in atomic-scale contact geometries for different tip positions (see lateral mapping of the hysteresis described in detail in Supplementary Fig. 15 and Supplementary Note 10).

Electrical opening and closing of the molecular junctions. The opening and closing of the molecular junction can also be controlled by the applied bias. As shown in Fig. 5a,b, IV curves recorded above the head group exhibit a hysteretic switching between high- and low-conductance curves. The critical voltages that are necessary to open and close the molecular junction strongly depend on the lateral and the vertical position of the tip and range from 1 to 0.5 V and 0.7 to 2 V for closing and

opening the junction, respectively (for details, see Supplementary Figs 16 and 17 and Supplementary Note 11). This behaviour excludes switching mechanisms that are related to a particular energy of the electrons such as inelastic excitations or tunnelling to specic orbitals38,39. However, it can be explained by the electric eld that acts on the dipole moment of the molecular head group as follows: at positive bias voltages applied to the sample, the electric eld points from the sample towards the tip and exerts a torque on the molecular dipole that in turn pushes the molecular head group away from the tip (Figs 5d and 6c). When the molecule is stretched, the elastic energy that counteracts the binding energy in the closed state increases and the contribution of the electrostatic energy becomes crucial. As soon as open and closed congurations of the junction are similar in energy, the electric eld can break the nitrilegold bond and opens the contact between tip and molecule, which leads to a sudden drop in the conductance at a voltage of 1 V for spiro A in Fig. 5a (0.35 V for spiro B in Fig. 5b). Similarly, at negative bias voltages, the electric eld points towards the sample and pulls the head group towards the tip. This closes the tipmolecule contact, and the lone pair of the terminal nitrile nitrogen forms a coordinative bond to the gold tip, which is reected in a sudden increase in the conductance at 0.8 V for spiro A in Fig. 5a

( 0.2 V for spiro B in Fig. 5b). The effect is due to a purely

elastic deformation in contrast to the electric-eld-induced isomerization in azobenzene molecules18 or the recently described oxidationreduction processes of organometallic compounds20.

A series of controlled switching events induced by voltage pulses of 2 V/ 1.5 V with a read-out voltage of 0.5 V is

shown in Fig. 5c, recorded on spiro A. The tip was placed directly over the nitrile group such that the read-out voltage lies within

NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672

a c

20

Closing contact

Opening contact

+2 V

0.49

Current (pA) Current (nA)

0

0.50

107

108

109

20

40

1

1.5 V

2

0 1 Sample voltage (V)

Sample voltage (V)

Sample voltage (V) Conductance G / G

0

b

10

10

870

860

850

5

5

0

0.4 0.2 0.0 0.2 0.4 0 10 20 30 40 50

Time (s)

d

+ + +

p

Height (pm)

840

830

820

810 4 3 2 1 0Electric field (GV m1)

1 2 3 4

+ + +

Figure 5 | Contact formation by application of electric elds. (a,b) IV curves at the position of the head group of spiro A (a) and spiro B (b). The arrows indicate the direction of the voltage ramp. (c) Time traces of applied voltage and resulting current during bistable switching between the two states of spiro A (open and closed junction) by voltage pulses of 130 ms and 2 V/ 1.5 V. Recorded with an integration time of 20 ms. (d) Simulated response

of spiro B to an external electric eld. The z component of the N atom above the plane of the three S atoms at z 0 is shown as a function of the applied

eld. Red circles represent a spiro B molecule with all S atoms xed, while the black squares are for the case, where S atoms can move freely in the xy plane at z 0. Denition of the dipole moment p and the electric eld e as indicated in the molecular models.

the voltage hysteresis, which allows bistability with an ONOFF ratio of the conductance switch of more than two orders of magnitude. Note that every voltage pulse of 130 ms leads to switching and the nal state of the junction solely depends on the polarity of the pulse, which makes this junction an ideal memristor. The short switching times indicate an efcient electric-eld-driven switching mechanism opposed to that of current-induced switching in spin-crossover memristors22. Because the underlying switching mechanism does not involve plastic deformation, similar to the mechanical control presented above, the electric eld allows to switch between two well-dened and distinct conductance values.

To theoretically examine these results, we rst evaluated the dipole moments of benzonitrile as well as of the gas phase relaxations of spiro A and B molecules presented in Supplementary Fig. 8 and Supplementary Note 4. The absolute values of the dipole moments are shown in Table 2, and the angles Fdip between the

surface normal (z direction) and the dipole moment in Supplementary Table 1. We nd that absolute values of the dipole moment of the spiro molecules are similar to those of benzonitrile40. The angle between surface normal and the molecular head group is close to the angle between the surface normal and the dipole moment (for details, see Supplementary Table 1), indicating that the dipole moment of the spiro molecules, indeed, arises mainly from the benzonitrile motive at its head.

With regard to the inuence of electric elds, we performed geometry optimizations of the spiro B molecule in a uniform, static electric eld within DFT. We assume that the three sulfur atoms of the molecule are located in the xy plane at z 0. In

Fig. 5d, we show the z height of the terminal N atom as a function of the electric eld applied along the z direction, similar to the geometry shown on the sides in Fig. 5d. Two cases of constrained relaxation are distinguished. In one case, the three S atoms in the legs are completely xed, and in the other they are just

6 NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672 ARTICLE

Time (s)

a 795 800 805

b

0.8

0.6

Current (pA)

0.4

Energy

0.2

0.0

199 199.5 200 200.5 201 Bias voltage (mV)

c

Time (s) Time (s)

+ + +

686 693 700

181

900 907 914

+[afii9829]

[afii9829]

[p10]

0.8

Current (pA)

p

0.6

0.4

0.2

+

+

0.0

+

+

179 180

Bias voltage (mV)

219 220 221

Bias voltage (mV)

Time (s)

Time (s)

0 2 4 6 8 10 12 14

0 2 4 6 8 10 12 14 Tip position z (pm)

d 0 200 400 600 800 1,000 1,200 1,400 1,600

e 0 25 50 75 100 125

0.8

Current (pA)

6

5

0.6

0.4

Current (pA)

4

3

2

0.2

1

0.0

0

1.0

N ON/(N OFF + N ON)

1.0

N ON/(N OFF + N ON)

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0.0

0.0

3

3

2

2

1

1

E(meV)

0

E(meV)

0

1

1

2

2

3

3

50 100 150 200 250 300 350 Bias voltage (mV)

Figure 6 | Energy differences between the open and closed molecular contact. (a) Statistical conductance switching of spiro A at a bias voltage of 200 mV. (b) Model of a double-well potential for ON and OFF states of the molecular head group. (c) Molecular model for the two states including denitions of the dipole moment p, the electric eld e and the angle F. (d) Statistical switching as a function of applied bias voltage and time. In addition to the enlarged view, shown in a, that corresponds to the grey line in the middle, two more voltage intervals marked in grey are shown in enlarged views (180 and 220 mV). The corresponding population of the ON state NON/(NOFF NON) is indicated by red crosses, the resulting energy difference between

ON and OFF states derived from Boltzmann statistics is shown by black crosses and the red line is a linear t to the data. (e) Statistic switching recorded at 600 mV as a function of time and distance between tip and sample, corresponding population of the ON state (blue crosses) and resulting energy difference between ON and OFF states (black crosses, linear t in blue).

Table 2 | Calculated dipole moments.

Benzonitrile Spiro A Spiro A-rotated ring Spiro B4.33 D 6.45 D 5.77 D 5.06 D

constrained to z 0 and can move freely in the xy plane

otherwise. While the changes in height are somewhat smaller for the molecule with all S atoms xed, as expected due to the less exible geometry, the qualitative features are the same for both: the height of the N atom is reduced for positive elds and increased for negative ones. In perfect consistency with the experimental ndings, this can be interpreted in terms of the

NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672

alignment of the dipole of the nitrile group in the electric eld. We also observe that the displacements of the N as compared to its equilibrium position at zero electric eld are slightly larger at large negative electric eld strength than those at a comparable but positive one. This asymmetry means that it is somewhat easier to stretch the molecule than to compress it.

Deducing forces from statistical conductance switching. With the tip centrally placed above the molecular head group, formation and breaking of the nitrilegold bond leads to switching between two metastable states. However, the short-range bond energy and thus the energy barrier between the two states are drastically reduced, if the tip is placed laterally in front of the nitrile group (Supplementary Fig. 10; Supplementary Note 6). Then, the application of electric elds allows us to ne-tune the energy balance between the two states of the molecular junction such that statistical switching between the two states is observed. Figure 6a shows an example of a current trace, where the junction statistically opens and closes. In addition to the arguments given in the description of the voltage-induced contact formation experiments shown in Fig. 5, the very low currents (o0.1 pA in the OFF state) strongly suggest thermally activated switching and exclude inelastic excitations as the origin of the switching. Furthermore, these low-current densities are unlikely to lead to a considerable increase of the temperature of the molecular junction. The thermal population of the two states in the double-well potential (Fig. 6b) is given by Boltzmann statistics:

NOFF

NON e EOFF EON=kBT; 1

with kB the Boltzmann constant, EOFF and EON the energy of the OFF state, and the ON state, respectively. At a known temperature T, this relation allows us to calculate the difference between the energy levels EOFF EON from the population of the

two states. The energy landscape of our system can be described by a double-well potential formed by a sum of three contributions:

EtotEtip E

dipole

Edeform: 2 Etip is the effective potential for the interaction of the molecular head group with the tip. The energy of an electric dipole p in an electric eld e is given by Edipole e

j j p

j j cosF with

E e

j jU=d, where U is the bias voltage, d the distance between

tip and sample, and F the angle between the dipole and the electric eld (Fig. 6c). Please note that we follow the convention of the dipole moment given by the International Union of Pure and Applied Chemistry (IUPAC) pointing from the negative to the positive charge, while in chemistry often the opposite denition of the direction of a molecular dipole is used. The energy for the elastic deformation of the molecular junction can be approximated to leading order by Edeform 0.5k(z z0)2, with k the effective

stiffness and z0 the height of the relaxed molecule.

Following this model, we rst analyse the energy difference as a function of the applied bias voltage, that is, as a function of the applied electric eld. Figure 6d shows the time trace of the thermally activated statistic switching of the conductance, while the applied voltage was slowly swept from 50 to 350 mV. Three enlarged views are shown at bias voltages around 180, 200 and 220 mV (Fig. 6a,d). Clearly, the population depends on the bias voltage. Averaging the discriminated signal over intervals of 7.5 s results in a smooth curve representing the voltage-dependent population of the ON state (NON/(NOFF NON)), which is then

translated into the energy difference between the ON and the OFF state according to equation (2), taking T 5.2 K from the

thermometer reading. It can be seen that this energy difference scales linearly with the applied voltage as is expected for the energy of a dipole in an electric eld. This clearly shows the role of the

electric eld and indicates that the transition is not caused by the tunnelling current. Furthermore, the slope of a 26 meV V 1

allows to estimate the change in tilt of the dipole moment to B17, when switching between the ON (FON 48) and OFF states

(FOFF 65; for details see Supplementary Note 12). The same

experiment was performed on spiro B, where a change in tilt angle of FOFF FON 7 was found (Supplementary Fig. 18). Although

the precise angle is expected to depend on the tip position, this difference can intuitively be explained by the shorter and therefore more rigid cantilever head group of this variant. In spite of the very low currents involved, we cannot exclude that the molecular temperature differs from the thermometer reading, which would inuence the precise value of the tilt of the dipole.

Second, we extracted the energy difference between the open and closed molecular junction as a function of the distance between tip and sample that allows to deduce the involved force. Figure 6e shows the corresponding time trace of the current for a distance variation of 14 pm for spiro A at constant voltage. As this displacement is negligible in comparison to the overall tipsample distance, the electric eld can be assumed to be constant. Similar to the case of the voltage variation, we observe a continuous transition of the population from the ON state to the OFF state with increasing distance between tip and sample. The counter-intuitive increase in conductance while stretching the molecular junction might be explained by a release of mechanical stress and related changes in the orbital level alignment4143. Strictly speaking, this experiment determines the energy difference between the ON and OFF congurations of the junction as a function of the z position of the tip, which corresponds, following classical mechanics, to the force that is needed to stretch the closed molecular junction. The slope of the tted straight line is 0.9kBT pm 1. Assuming a temperature of

T 5.2 K, this corresponds to 0.4 meV pm 1 which is, in

other units, a force of 64 pN (42 pN for spiro B). In simulations of the stiffness of the molecule, we found values of the same order of magnitude (9.7 pN pm 1 for spiro A and 18.6 pN pm 1 for spiro B, see Supplementary Note 5 and Supplementary Fig. 9).

This leads to a change of the potential energy by 1.5 meV (2.8 meV for spiro B) when compressing or stretching the molecular junction by 7 pm, in agreement with our experimental results. These stiffness values result in the experimentally observed forces when stretching the molecule by 6.6 pm in the case of spiro A (2.3 pm in the case of spiro B). However, the stiffness and the elastic deformation of tip, sample and in particular the bond between the molecule and the tip were not considered in these simulations. The nitrilegold bond is expected to largely inuence the stiffness of the overall molecular junction, as it is the predetermined breaking point. In addition, the bond between molecule and tip is weakened by the voltage applied in the experiments. As the thermally induced opening and closing of the molecular junction does not involve a change of the macroscopic tipsubstrate distance, the force derived in this way does not contain related contributions from van der Waals forces that can amount up to 2 nN (refs 12,14).

DiscussionIn contrast to typical AFM force measurements, only the energy variation related to the stretching of the molecular bridge is involved in our experiment. Thus, the method presented here allows to address exclusively the atomic forces involved in the junction formation provided that current-induced heating can be excluded. Given the low thermal energy of 0.4 meV at a temperature of 5.2 K, minute forces in the range of piconewtons can be extracted with high precision, but larger forces are out of reach due to too few thermally activated transitions.

8 NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications

NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672 ARTICLE

Our results highlight the importance of atomic-scale control of the junction geometry to obtain reliable mechanical and electrical properties in future molecular devices. The very reliable control of the presented molecular toggle switch allowed us to demonstrate new experimental methods that reveal the subtle energy variations during contact formation on the nanoscale and that will be applicable to a wide range of single-molecule junctions.

Methods

Syntheses of the model compounds spiro A and spiro B. Our synthetic strategy used for the preparation of the 2,7,30,60-tetrasubstituted 9,90-spirobiuorene tripodal derivatives spiro A and spiro B is outlined in Supplementary Fig. 1. The rst modular platform based on a rigid 9,90-spirobiuorene with three acetyl protected thiol groups in the positions 2, 30 and 60 exposing a para-cyanophenylethynyl rod in the position 7 (spiro A) was reported recently27. Our synthetic approach is based on a metal-halogen exchange reaction of the 2-iodobiphenyl derivative and its subsequent reaction with 2,7-disubstituted uoren-9-one to afford the carbinol. Further electrophilic cyclization and separation of regioisomers provided the corresponding 2,7,30,60-tetrasubstituted 9,90-spirobiuorene 1 as the key intermediate. This previously prepared spirobiuorene derivative 1 (ref. 27) was also employed for a subsequent synthesis of a shorter derivative spiro B. To introduce the third alkylsulfanyl group, spirobiuorene derivative 1 was treated with 2-(trimethylsilyl)ethanthiol in the presence of Pd2(dba)3, Xantphos and

Hnigs base to provide the desired tripodal spirobiuorene 2 in 75% yield (Supplementary Note 1). Subsequent cyanation of bromo derivative 2 afforded nitrile derivative 3 in 79% yield (Supplementary Note 2). Final transprotectionof the thiol was successfully performed using AgBF4 and acetyl chloride in dichloromethane to obtain the desired tripodal thioacetate spiro B in 95% yield (Supplementary Note 3). The presence of terminal thioacetate groups allows the target spirobiuorene molecules spiro A and spiro B to bind to gold surfaces. The acetyl works as a labile thiol-protecting group, and can mildly and efciently be cleaved by thermal annealing upon deposition of molecules on a gold surface.

All starting materials and reagents were obtained from commercial suppliers and used without further purication. Thin-layer chromatography was performed on Silica gel 60 F254 plates, spots were detected by uorescence quenchingunder ultraviolet light at 254 nm and/or staining with appropriate solutions (anisaldehyde, phosphomolybdic acid, KMnO4). Column chromatography was performed on Silica gel 60 (0.0400.063 mm). All experimental manipulations with anhydrous solvents were carried out in ame-dried glassware under inert atmosphere of argon. Degassed solvents were obtained by three cycles of the freezepumpthaw. Dioxane was dried and distilled from sodium/benzophenone under argon atmosphere. Dichlormethane was dried and distilled from CaH2 under argon atmosphere. S,S0,S00-[7-(4-Cyanophenylethynyl)-9,90-spirobiuorene-2,30,60-triyl] tris(thioacetate) (spiro A)27 and 2-bromo-7-iodo-30,60-bis[2-(trimethylsilyl)ethylsulfanyl]-9,90-spirobiuorene (1) were prepared according to the published procedure27.

All nuclear magnetic resonance (NMR) spectra were recorded at 25 C in CDCl3. 1H NMR (500.16 MHz) spectra were referenced to the solvent residual proton signal (CDCl3, dH 7.24 p.p.m.). 13C NMR (125.78 MHz) spectra with total decoupling of

protons were referenced to the solvent (CDCl3, dC 77.23 p.p.m.). For correct

assignment of both 1H and 13C NMR spectra, the 1H1H correlation spectroscopy (COSY), 13C DEPT-135, heteronuclear single quantum coherence (HSQC) and heteronuclear multiple bond correlation (HMBC) experiments were performed. Electron ionization mass spectra (EI MS) were recorded with a gas chromatography mass spectrometry (GC/MS) instrument (samples were dissolved in diethyl ether, chloroform or introduced directly using direct injection probes (DIP) and direct exposure probes (DEP)), and m/z values are given along with their relative intensities (%) at an ionizing voltage of 70 eV. Infrared spectra were measured in KBr pellets. Analytical samples were dried at 40100 C under reduced pressure (10 2 mbar).

Elemental analyses were obtained using an elemental analyser. NMR spectra of the compounds 2, 3 and spiro B are displayed in Supplementary Figs 27.

Electronic structure calculations. Atomistic simulations are performed to support and help interpret the experimental observations. Electronic structure calculations and related geometry optimizations are carried out by applyingthe quantum chemistry software package TURBOMOLE 6.6 (ref. 44). All the calculations were performed within DFT using the exchange-correlation functional perdew burke-ernzerhof (PBE)4548 and the def-SV(P) basis set4951, which is of split-valence quality with polarization functions on all non-hydrogen atoms.

To analyse the electron transport through the spiro molecules on an Au(111) surface (Fig. 3), we followed the procedure described in refs 52,53. The molecular geometry was rst optimized on the surface, keeping all of the gold atoms xed. Subsequently, the STM tip was connected to the head of the molecule. The Au pyramid, chosen as the STM tip, and the surface represent semi-innite electrodes. They are oriented along the (111) direction, which coincides with the transport direction or z axis.

The theoretical investigation of the quantum transport properties of complex molecules remains challenging because of the large number of atoms involved and the innite, non-periodic geometry of the system. DFT is one of the few ab initio

electronic structure methods that can handle the hybrid metalmoleculemetal contacts. On the other hand, due to self-interaction errors in the standard exchange-correlation functional and missing image charge effects, DFT-based methods have difculties to accurately describe the energy gap and level alignment of molecules at surfaces. This can be improved by adding a self-energy correction, resulting in the DFT S method3335. The main difference in the electronic

structure between DFT and DFT S for a given contact geometry is typically

a pronounced increase of the gap between the highest occupied molecular orbital and lowest unoccupied molecular orbital by several eVs that arise from a (almost symmetric) decrease of the occupied energy levels and increase of the unoccupied energy levels with respect to the Fermi energy EF of the metal electrodes (see Fig. 3 and its discussion). Details regarding our DFT S

implementation can be found in ref. 35, and for more information on our particular DFT S calculations, see Supplementary Note 7 and Supplementary

Fig. 11.Using the electronic structure of DFT and DFT S as input, the transmission

t(E) as a function of energy is calculated using the Landauer-Bttiker formalism expressed in terms of nonequilibrium Greens functions32. In the Landauer-Bttiker approach, the stationary current is given as

I

2e h

Z

1 1

dE tE f E mL

f E mR

: 3

For the linear conductance one obtains

G

dI dV

V0

2e2 h

Z

1 1

dE @Ef

tE: 4

STM experiments. STM measurements were carried out in a home-built STM in UHV at a temperature of 5.2 K. The STM tip was prepared by chemical etching of a tungsten wire and by repeated dipping into the gold surface. The Au(111) single-crystal surface was cleaned by several cycles of Ar sputtering and annealing to 700 K. Spirobiuorene molecules were sprayed onto the surface through a pulse valve at a pressure of B1 mbar. After the deposition, the sample was transferred into the UHV chamber, annealed at 410 K for 1 h and subsequently transferred to the STM. The voltage is applied to the sample.

Data availability. The data that support the ndings of this study are available from the authors on reasonable request, see author contributions for specic data sets.

References

1. Aviram, A. & Ratner, M. A. Molecular rectiers. Chem. Phys. Lett. 29, 277283 (1974).

2. Selzer, Y. & Allara, D. L. Single-molecule electrical junctions. Annu. Rev. Phys. Chem 57, 593623 (2006).

3. Kergueris, C. et al. Electron transport through a metal-molecule-metal junction. Phys. Rev. B 59, 1250512513 (1999).

4. Reichert, J. et al. Driving current through single organic molecules. Phys. Rev. Lett. 88, 176804 (2002).

5. Tao, N. J. Electron transport in molecular junctions. Nat. Nanotechnol. 1, 173181 (2006).

6. Venkataraman, L., Klare, J. E., Nuckolls, C., Hybertsen, M. S. & Steigerwald, M. L. Dependence of single-molecule junction conductance on molecular conformation. Nature 442, 904907 (2006).

7. Nel, N. et al. Controlled contact to a C60 molecule. Phys. Rev. Lett. 98, 065502 (2007).

8. Grill, L. & Moresco, F. Contacting single molecules to metallic electrodes by scanning tunnelling microscope manipulation: model systems for molecular electronics. J. Phys. Condens. Matter 18, S1887S1908 (2006).

9. Butt, H.-J., Cappella, B. & Kappl, M. Force measurements with the atomic force microscope: technique, interpretation and applications. Surf. Sci. Rep. 59, 1152 (2005).

10. Schull, G. et al. Atomic-scale engineering of electrodes for single-molecule contacts. Nat. Nanotechnol. 6, 23 (2011).

11. Mohn, F., Gross, L. & Meyer, G. Measuring the short-range force eld above a single molecule with atomic resolution. Appl. Phys. Lett. 99, 053106 (2011).

12. Hauptmann, N. et al. Force and conductance during contact formation to a C60 molecule. New J. Phys. 14, 073032 (2012).

13. Pawlak, R., Kawai, S., Fremy, S., Glatzel, T. & Meyer, E. Atomic-scale mechanical properties of orientated c60 molecules revealed by noncontact atomic force microscopy. ACS Nano 5, 63496354 (2011).

14. Welker, J. & Giessibl, F. J. Revealing the angular symmetry of chemical bonds by atomic force microscopy. Science 336, 444449 (2012).

15. Albers, B. J. et al. Three-dimensional imaging of short-range chemical forces with picometre resolution. Nat. Nanotechnol. 4, 307310 (2009).

16. Morgenstern, K. Switching individual molecules by light and electrons: From isomerisation to chirality ip. Progr. Surf. Sci. 86, 115161 (2011).

NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14672

17. Quek, S. Y. et al. Mechanically controlled binary conductance switching of a single-molecule junction. Nat. Nanotechnol. 4, 230234 (2009).

18. Alemani, M. et al. Electric eld-induced isomerization of azobenzene by STM.J. Am. Chem. Soc. 128, 1444614447 (2006).19. Wirth, J. et al. Diarylethene molecules on a Ag(111) surface: stability and electron-induced switching. J. Phys. Chem. C 119, 48744883 (2015).

20. Schwarz, F. et al. Field-induced conductance switching by charge-state alternation in organometallic single-molecule junctions. Nat. Nanotechnol. 11, 170176 (2016).

21. Liljeroth, P., Repp, J. & Meyer, G. Current-induced hydrogen tautomerization and conductance switching of naphthalocyanine molecules. Science 317, 12031206 (2007).

22. Miyamachi, T. et al. Robust spin crossover and memristance across a single molecule. Nat. Commun. 3, 938 (2012).

23. Kim, Y. et al. Charge transport characteristics of diarylethene photoswitching single-molecule junctions. Nano Lett. 12, 37363742 (2012).

24. Comstock, M. J. et al. Reversible photomechanical switching of individual engineered molecules at a metallic surface. Phys. Rev. Lett. 99, 038301 (2007).

25. Zhu, S.-E. et al. Self-decoupled porphyrin with a tripodal anchor for molecular-scale electroluminescence. J. Am. Chem. Soc. 135, 1579415800 (2013).26. Kitagawa, T. et al. Rigid molecular tripod with an adamantane framework and thiol legs. Synthesis and observation of an ordered monolayer on Au(111).J. Org. Chem. 71, 13621369 (2006).27. Valek, M. et al. Synthesis of molecular tripods based on a rigid 9,90-

spirobiuorene scaffold. J. Org. Chem. 79, 73427357 (2014).28. Valek, M., Lindner, M. & Mayor, M. Rigid multipodal platforms for metal surfaces. Beilstein J. Nanotechnol. 7, 374405 (2016).

29. Lindner, M. et al. Importance of the anchor group position (para versus meta) in tetraphenylmethane tripods: synthesis and self-assembly features. Chem. Eur.J. 22, 1321813235 (2016).30. Joachim, C., Gimzewski, J. K., Schlittler, R. R. & Chavy, C. Electronic transparence of a single C60 molecule. Phys. Rev. Lett. 74, 21022105 (1995).

31. Fournier, N., Wagner, C., Weiss, C., Temirov, R. & Tautz, F. S. Force-controlled lifting of molecular wires. Phys. Rev. B 84, 035435 (2011).

32. Pauly, F. et al. Cluster-based density-functional approach to quantum transport through molecular and atomic contacts. New J. Phys. 10, 125019 (2008).

33. Quek, S. Y. et al. Amine-gold linked single-molecule circuits: experiment and theory. Nano Lett. 7, 34773482 (2007).

34. Mowbray, D. J., Jones, G. & Thygesen, K. S. Inuence of functional groups on charge transport in molecular junctions. J. Chem. Phys. 128, 111103

2008:

35. Zotti, L. A. et al. Heat dissipation and its relation to thermopower in single-molecule junctions. New J. Phys. 16, 015004 (2014).

36. Mishchenko, A. et al. Single-molecule junctions based on nitrile-terminated biphenyls: a promising new anchoring group. J. Am. Chem. Soc. 133, 184187 (2011).

37. Karimi, M. A. et al. Identication of the current path for a conductive molecular wire on a tripodal platform. Nanoscale 8, 10582 (2016).

38. Wu, S. W., Ogawa, N., Nazin, G. V. & Ho, W. Conductance hysteresis and switching in a single-molecule junction. J. Phys. Chem. C 112, 52415244 (2008).

39. Mohn, F. et al. Reversible bond formation in a gold-atom-organic-molecule complex as a molecular switch. Phys. Rev. Lett. 105, 266102 (2010).

40. Wohlfart, K., Schnell, M., Grabow, J.-U. & Kpper, J. Precise dipole moment and quadrupole coupling constants of benzonitrile. J. Mol. Spectrosc. 247, 119121 (2008).

41. Zotti, L. A. et al. Revealing the role of anchoring groups in the electrical conduction through single-molecule junctions. Small 6, 15291535 (2010).

42. Bruot, C., Hihath, J. & Tao, N. Mechanically controlled molecular orbital alignment in single molecule junctions. Nat. Nanotechnol. 7, 3540 (2012).

43. Reecht, G. et al. Pulling and stretching a molecular wire to tune its conductance. J. Phys. Chem. Lett. 6, 29872992 (2015).

44. TURBOMOLE V6.6 2014, (Universitat Karlsruhe, Forschungszentrum Karlsruhe, TURBOMOLE GmbH, 2014). Available at http://www.turbomole.com.

45. Dirac, P. A. M. Quantum mechanics of many-electron systems. Proc. R. Soc. A 123, 714733 (1929).

46. Slater, J. C. A Simplication of the Hartree-Fock method. Phys. Rev. 81, 385390 (1951).

47. Perdew, J. P. & Wang, Y. Accurate and simple analytic representation of the electron-gas correlation energy. Phys. Rev. B 45, 1324413249 (1992).

48. Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 38653868 (1996).

49. Schafer, A., Horn, H. & Ahlrichs, R. Fully optimized contracted Gaussian basis sets for atoms Li to Kr. J. Chem. Phys. 97, 25712577 (1992).

50. Eichkorn, K., Treutler, O.,hm, H., Haser, M. & Ahlrichs, R. Auxiliary basis sets to approximate Coulomb potentials. Chem. Phys. Lett. 240, 283290 (1995).

51. Eichkorn, K., Weigend, F., Treutler, O. & Ahlrichs, R. Auxiliary basis sets for main row atoms and transition metals and their use to approximate Coulomb potentials. Theor. Chem. Acc. 97, 119124 (1997).

52. Weber, H. et al. Electronic transport through single conjugated molecules. Chem. Phys. 281, 113125 (2002).

53. Brkle, M. et al. Conduction mechanisms in biphenyl-dithiol single-molecule junctions. Phys. Rev. B 85, 075417 (2012).

Acknowledgements

We acknowledge the support by the Research Network Functional Nanostructures of the Baden-Wrttemberg-Stiftung and by the Helmholtz Research Program STN (Science and Technology of Nanosystems). S.G.B. and F.P. gratefully acknowledge nancial support by the German Research Foundation (DFG) through the Collaborative Research Center (SFB) 767 and the Carl-Zeiss Foundation. The simulations were supported by the bwHPC program through the computational resources bwUniCluster and the JUSTUS HPC facility. We thank Magnus Paulsson for inspiring discussions and initial computational modelling. We acknowledge support by Deutsche Forschungsgemeinschaft and Open Access Publishing Fund of Karlsruhe Institute of Technology.

Author contributions

M.L., M.M., L.G. and W.W. conceived and designed the experiments. M.V. and M.M. synthesized the molecules. L.G., J.H. and K.E. carried out STM experiments. S.G.B. and F.P. performed the modelling and ab initio simulations. L.G., J.H. and K.E. analysed the data. L.G., W.W., F.P. and M.M. co-wrote the paper. All authors discussed the results and commented on 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: Gerhard, L. et al. An electrically actuated molecular toggle switch. Nat. Commun. 8, 14672 doi: 10.1038/ncomms14672 (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

10 NATURE COMMUNICATIONS | 8:14672 | DOI: 10.1038/ncomms14672 | http://www.nature.com/naturecommunications

Web End =www.nature.com/naturecommunications