Electronic metal-support interaction enhanced oxygen reduction activity and stability of boron carbide supported platinum

Catalysing the reduction of oxygen in acidic media is a standing challenge. Although activity of platinum, the most active metal, can be substantially improved by alloying, alloy stability remains a concern. Here we report that platinum nanoparticles supported on graphite-rich boron carbide show a 50–100% increase in activity in acidic media and improved cycle stability compared to commercial carbon supported platinum nanoparticles. Transmission electron microscopy and x-ray absorption fine structure analysis confirm similar platinum nanoparticle shapes, sizes, lattice parameters, and cluster packing on both supports, while x-ray photoelectron and absorption spectroscopy demonstrate a change in electronic structure. This shows that purely electronic metal-support interactions can significantly improve oxygen reduction activity without inducing shape, alloying or strain effects and without compromising stability. Optimizing the electronic interaction between the catalyst and support is, therefore, a promising approach for advanced electrocatalysts where optimizing the catalytic nanoparticles themselves is constrained by other concerns.

T he oxygen reduction reaction (ORR) is a standing grand challenge in fundamental research in energy, exemplifying a broader class of multi-electron reactions 1 . Oxygen reduction efficiency and long-term catalyst stability are standing fundamental problems that have to be overcome for polymer electrolyte fuel cells 2 to be used in demanding applications such as automobiles and grid-scale energy storage. The relatively large overpotential of the ORR at technically feasible electrodes under practical operating conditions limits conversion efficiency to typically o50% relative to the thermodynamic limit of more than 80% (ref. 3). Challenging cost targets, especially in the automotive industry, provide further motivation to reduce Pt loadings 4 . Finally, the dissolution and/or loss of Pt surface area in the cathode must be greatly reduced to meet automotive and grid-scale storage stability targets.
The work pioneered by Nørskov 5 has provided a rational design approach to improve ORR activity by tailoring the position of the Pt d-band centre, because it determines the stability of reaction intermediates, especial O ads . Significant efforts have successfully focused on improving ORR activity by d-band centre engineering via nanoparticulate Pt alloys supported on high-surface area carbon 2,6-8 . However, the most active alloys 3,9 are unlikely to show long-term stability under cathodic polymer electrolyte fuel cells conditions 10,11 for thermodynamic reasons. At best, one can hope that dealloying is restricted to the surface, leading to Pt skin structures that are, fortuitously, highly active 3,12,13 .
Whether dealloying can be restricted to the surface of the typically 2 nm alloy electrocatalysts over technically relevant lifetimes is an open question. Regardless, stability concerns motivate other design approaches to engineer the Pt d-band characteristics without potentially compromising stability. Recently, research emphasis has shifted to alternative support materials to prevent the loss of catalyst surface area due to support corrosion and dissolution/agglomeration of nanoparticulate Pt (refs 14,15). Alternative supports also provide an opportunity to indirectly improve ORR activity as well, because a change in support electronic structure can influence the catalyst's d-band manifold via electronic equilibration 16 , one aspect of strong metal-support interactions (SMSI).
Utilization of SMSIs is a promising approach to improve both activity and stability of supported Pt catalysts. The SMSI of Pt on oxide supports, particularly titanium dioxide, is well known 17,18 . These Pt/TiO 2 catalysts have previously exhibited higher activities for ORR in relation to Pt/C (ref. 19). The low conductivity of these materials, however, is a challenge 20 . Highly conductive SnO 2 -based supports have, therefore, attracted attention recently 21 , and thin film studies unveiled an intriguing dependence of ORR activity on support orientation and/or termination 22 .
Changing the physical nature of the support from carbons to oxides is a radical step. It potentially impacts catalyst stoichiometry and particle morphology, which in their own right change electrocatalytic properties via size 4 , strain 23 and proximity 24 effects. It is, therefore, nontrivial to experimentally isolate these effects to develop the fundamental understanding needed to rationally exploit SMSIs.
Transition metal carbides have also been widely investigated as promising support materials due to high electrical conductivity, thermal stability and similarities in electronic structure to Pt (ref. 25). However, the low surface area of carbides and liability to oxidation are challenging 26 . Carbides are, nonetheless, attractive to investigate electronic metal-support interactions in isolation, because they are physically closer to carbon than oxides. The material of choice, however, should also show a markedly different electronic structure and resilience against oxidation. Boron carbide (B 4 C) is such a material. It is a refractory compound 26 exhibiting high chemical inertness 27 , semi-conducing band structure and electrochemical stability 28 . The first study of B 4 C as an electrocatalyst support for hydrogen fuel cells was by Grubb & McKee 29 wherein a high resistance to sintering and higher current densities when compared to platinum on carbon black were already observed. More recent work by Lv et al. 30 has also shown promising ORR activity.
Pt nanoparticle supported on graphite-rich B 4 C composites (BC) are compared here with Pt supported on Vulcan XC 72R high-surface area carbon (C) to isolate electronic metal-support interactions. Electrochemical as well as advanced ex situ characterization techniques are used to elucidate the structure-property relationships between catalyst morphology, electronic structure and ORR activity. Transmission electron microscopy (TEM) and extended X-ray absorption fine structure (EXAFS) analysis confirm similar particle size, shape and packing of the clustered nanoparticles on both supports, while X-ray photoelectron spectroscopy (XPS) and X-ray adsorption near edge structure (XANES) analysis show differences in electronic properties due to metal-support interactions. These differences in electronic properties are correlated with enhanced ORR activity: the surface and mass specific oxygen reduction activity of Pt nanoparticles supported on BC under rotating disc electrode (RDE) conditions increases by roughly 40% relative to commercial Pt/C, while the kinetic current at 0.9 V shows a 50-100% improvement. The increased cycle stability of the BC supported catalysts after degradation cycling further promotes the interpretation in terms of increased electronic metal-support interactions on these catalysts. Purely electronic metal-support interactions are, therefore, experimentally isolated demonstrating that optimizing the electronic interaction between the catalyst and support, in conjunction with d-band theory, is a promising approach to designing advanced electrocatalysts without compromising stability under harsh conditions.

Results
Naming scheme. Graphite-rich boron carbide composites (BC) were catalysed with varying amounts of Pt nanoparticles following standard procedures. These Pt/BC catalysts are compared with commercial Pt/C catalysts. It is imperative to keep particle size and dispersion constant between the Pt/BC and benchmark Pt/C catalysts to isolate electronic effects, because these affect ORR activities significantly 4,23-24 . We use a naming convention for all catalysts that follows the nominal platinum loading (that is, wt%) of the commercial benchmark catalysts. The equivalent platinum loading on BC (that is, eq. wt%) designates a catalyst that has the same mass platinum per support surface area (Supplementary Note 1). Support surface area. The BC support Brunauer, Emmett and Teller (BET) surface area was measured to be 80 m 2 g À 1 , in agreement with manufacturer specifications ( Supplementary  Fig. 1). This is considerably lower than seen for high-surface area carbons such as Vulcan XC 72R, which has a surface area of 250-260 m 2 g À 1 . To keep the mass platinum per surface area constant, the actual target platinum loading per mass BC support was, therefore, reduced accordingly (Supplementary Table 1) to maintain a similar Pt coverage per surface area of support and particle size as seen on the commercial Pt/C benchmark catalysts to isolate support-related activity changes.
X-ray diffraction. The X-ray diffraction pattern of BC ( Supplementary Fig. 2) clearly shows peaks indicative of rhombohedral graphite (3R) and hexagonal graphite (2H) in ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15802 addition to rhombohedral B 4 C (R 3m), showing the material to be a composite of graphite and rhombohedral B 4 C. Using Rietveld refinement, the spectrum indicates 58.0±7.7 wt% graphite, 38.7 ± 7.2 wt% B 4 C and 3.0 ± 1.7 wt% B 2 O in the sample. This is in reasonable agreement with the average B:C ratio of 39:61 as obtained from EELS. The high graphite content ensures comparative conductivities between BC and C (Supplementary Note 2). The graphite consists of approximately 32.0 wt% rhombohedral graphite (3R) and 26.0 wt% hexagonal graphite (2H).
The average Pt crystallite size was confirmed by a fit of the Scherrer equation to the Pt(200), Pt(220) and Pt(311) peaks at 46°, 67°and 81°in the X-ray diffraction (XRD) spectra, respectively ( Supplementary Fig. 3). This analysis yielded a volume-weighted average particle size of B3.9 nm for the various loadings, in good agreement with the size measured via TEM (Table 1). XRD routinely overestimates particle size compared to TEM at the relevant length scale, because particle sizes smaller than 2 nm are not identified due to the lower detection limit of XRD (ref. 31).
Transmission electron microscopy. TEM images of the prepared Pt/BC catalysts as well as the as-received commercial Pt/C benchmark catalysts, taken at 270 k times magnification to illustrate particle dispersion and support morphology, are shown in Fig. 1. The Pt loadings on BC were confirmed via ICP-OES analysis to be 10, 20 and 40 eq. wt%, respectively (see Supplementary Table 1 for actual loadings). Pt particle size distribution and dispersion for each catalyst are quantified and contrasted in Table 1, which shows average particle size with s.d. as well as average interparticle distances with s.d.
The BC support particles are homogenous with an average particle size of B40 nm, and Pt nanoparticles are well dispersed on the BC support with average particle sizes ranging from 2.3 to 2.9 nm. Particle size and interparticle distances are very similar for equivalent loadings on both supports at 20 wt% and 40 wt% with little s.d. Non-conformity is seen between the 10 wt% Pt/C and 10 eq wt% Pt/BC catalysts due to the unusually small Pt size on Pt/C. X-ray absorption spectroscopy. The EXAFS was collected for the 20 eq. wt% Pt/BC and 20 wt% Pt/C catalysts. The EXAFS at the Pt L 3 edge gives insights to the supported Pt morphology: the first shell co-ordination number is strongly related to average particle size 32 while the fourth co-ordination number is an indication of particle shape.
The measured and modelled EXAFS spectra are shown in Supplementary Fig. 4. The fitted model, considering up to the fourth neighbour shells, describes the EXAFS data well. The resulting co-ordination numbers (N), distances of neighbouring atoms (R) and the disorder factors (s 2 ) of the catalysts supported on BC and C are listed in Table 2.
The first, second and fourth Pt co-ordination numbers are slightly smaller for Pt supported on BC, while the third is found to be slightly larger. Differences, however, are within experimental error indicating very similar Pt morphologies as further discussed later. Near-neighbour distances also agree very well between the Pt/BC and Pt/C catalysts.
X-ray photoelectron spectroscopy. Figure 2a shows the C 1s and Pt 4f regions of XPS spectra for the Pt/BC and Pt/C catalysts. The XPS spectra are calibrated via alignment of the C(sp 2 ) peak position in the C 1s spectrum to its reference value of 284 eV.
The C 1s region of the Pt/C catalyst is dominated by the C(sp 2 ) peak with a small, broad feature at higher binding energies that has been assigned to C-O bonding structures of carboxyl groups 33 . The C 1 s spectrum of the Pt/BC catalysts is a convolution of contributions from the B 4 C and graphitic carbon within the composite. The C(sp 2 ) peak is assigned to the graphitic carbon, and the peaks at 282.7 and 284.8 eV are attributed to B 4 C. The low energy polymorphs of B 4 C (ref. 34) provide three C environments at the ends and centre of tri-atomic chains and in the polar position within B 11 C icosahedra. The low binding energy peak at 282.7 eV was previously attributed to B 4 C (ref. 35) and is likely due to carbon in low co-ordination (chain centre and/or B 11 C). The C(sp 3 ) peak at 284.8 eV, which is reminiscent of diamond 36 , is also predominantly assigned to B 4 C due to the strong similarity of site environment of tetrahedral carbon at the chain ends in B 4 C and diamond and the lack of evidence for sp 3 hybridized carbon from XRD. This assignment is supported by density functional theory calculations (Supplementary Note 3) that predict a chemical shift of B2.3 eV between the C 1s core electrons in B 11 C and CBC units (Supplementary Table 2), in good agreement with the experimental shift of 2.1 eV. Hence, the relative strength of the C(sp 2 ) and C(sp 3 ) peaks corroborates the phase fractions of B 4 C and graphitic carbon within the composite derived from XRD.
The Pt 4f spectra of the Pt/C catalyst features the usual spin-orbit splitting into a 4f 7/2 peak at 71.0 eV and a 4f 5/2 peak at 74.3 eV. The same splitting is seen for all Pt/BC catalysts. There is, however, a clear shift of both peaks by B0.6 eV to higher binding energies for all Pt loadings on BC to 71.6 and 74.9 eV.
In situ x-ray absorption near edge structure. Figure 2c shows the XANES Pt L 2 and L 3 edges for the 20 wt% Pt/C and 20 eq. wt% Pt/BC catalysts measured under potentiostatic control in electrochemical environment. The increase in the L 2 white line observed for the Pt/BC catalyst indicates an increase in unoccupied d-states above the Fermi level for Pt/BC (ref. 37) in electrochemical environments. Using standard post-processing 38 , d-band occupancy can be estimated to be 63% for Pt/C and 58% for Pt/BC at 0.744 V versus NHE.
Electrochemical characteristics and activity. The calculated electrochemically active surface area (ECSA) and ORR activities are summarized in Table 3. CO stripping voltammetry ( Supplementary Fig. 5) was used to determine the ECSAs assuming a specific charge of 420 mCcm Pt À 2 per CO monolayer adsorbed on the platinum surface. Measurements were taken at room temperature in 0.1 M HClO 4 electrolyte solution. No CO stripping peak was seen on the uncatalysed BC support. ECSAs were confirmed using Cu-UPD 39 .
The specific and kinetic currents reported in Table 3 were obtained from RDE measurements at a rotation speed of 1,600 r.p.m. (Supplementary Fig. 6) during the anodic sweep at 0.9 V versus RHE. Measurements were completed at 20 mVs À 1 to minimize double layer charging artefacts at room temperature in an oxygen saturated 0.1 M HClO 4 electrolyte 4,40 . The ECSA obtained from CO stripping voltammetry was used to calculate surface area specific currents, and the kinetic current was isolated from mass transport limitations using standard post-processing 41 .
The Pt/C catalyst activities fall in line with numerous reported values and compilations of values for ORR activity on these catalysts 4,42-45 . An exception is seen for the 10 wt% Pt/C catalyst, which exhibited large experimental scatter and anomalously small limiting currents during RDE measurements ( Supplementary  Fig. 6a) indicative of various detrimental effects 43 . The 10% Pt/C catalyst is, therefore, not considered a suitable benchmark for our purposes.
The ORR activities for the BC supported catalysts consistently show a more active catalyst with higher mass specific (I mass ), surface area specific (I spec ) and kinetic activities (I k ) for the ORR. Additionally, kinetic activity increases as Pt loading increases on both the Pt/C and Pt/BC catalysts, with the least active ORR catalyst being 10 wt% Pt/C and the most active being 40 eq wt% Pt/BC. These trends are further discussed below.
Further electrochemical characteristics are summarized in Fig. 3. Cyclic voltammograms of the BC supported catalysts are shown in the left diagram of Fig. 3a. In addition to the typical characteristics of supported Pt, a slight coupled reversible reaction at B0.6 V versus RHE is seen for the Pt/BC materials.  *The error associated with distance to neighbouring atoms for the 20 eq wt% Pt/BC and 20 wt% Pt/C is ± 0.00048 Å and ± 0.00056 Å, respectively. Values and confidence intervals obtained from EXAFS fits shown in Supplementary Fig. 4, disorder factors s 2 are given in the fifth column.
The large double layer of the Pt/BC catalysts results in a less well defined H UPD region than seen on the Pt/C catalysts. Therefore, no attempt was made to compare the ECSA as obtainable from the H UPD region with the values given in Table 3. Careful observation of the cyclic votammograms in Fig. 3b, which compares the electrochemical response of Pt/BC with the Pt/C benchmark at 40 wt%, shows slight peak shifts towards higher potentials in the oxide reduction region for Pt/BC. The kinetic ORR results are reiterated in the Tafel plots shown to the right in Fig. 3 for completeness. Figure 3a compares kinetic currents of Pt/BC at various loadings. Again, kinetic activity generally increases with Pt loading over the full potential range although the two lower loadings show comparable kinetic currents at lower voltages. Finally, the Tafel plot of Fig. 3b shows that the 40 eq wt% Pt/BC catalyst exhibits higher kinetic activity for ORR over the measured potential range of 0.85-0.95 V versus RHE than the 40 wt% Pt/C benchmark. In fact, relative activity of Pt/BC versus Pt/C seems to increase further with decreasing potential.
Cycle stability. Catalyst durability was studied for the most active Pt/BC and Pt/C catalysts to investigate the platinum dissolution and platinum agglomeration rate on the different supports. Durability testing consisted of cycling the potential between 0.6 and 1.0 V versus RHE for 6,000 cycles at a scan rate of 50 mVs À 1 and periodically returning to the full voltammogram scan (0.05-1.2 V versus RHE at 20 mVs À 1 ) to compare changes in the H UPD region and oxygen adsorption and desorption region.
The voltammograms of the Pt/BC catalyst show significantly less change after 6,000 cycles compared to Pt/C (Fig. 4a). Changes in the 1.0-1.2 V range for Pt/BC appear to be not purely related to ECSA loss, because less variation is seen in the H UPD region after cycling. Instead, the reduced current response in the oxidative region is in parts attributed to secondary conditioning effects, which could either be the removal of organic impurities from Pt or oxidative decomposition of other trace impurities in solution amongst other things.
Cyclic voltammograms of Pt/C before and after cycling show a significantly altered current response in the oxide as well as in the H UPD region while the double layer region remains invariant after 6,000 cycles, strongly suggesting catalyst degradation and ECSA loss.
The ECSAs shown in Fig. 4b were periodically obtained from the H UPD region. Both catalysts show a relatively stable ECSA for an initial period of B50-100 cycles, which might be due to cleaning and/or roughening effects offsetting initial degradation. A logarithmic loss of ECSA is seen for both catalysts beyond this initial conditioning phase. Focusing on the data from cycle 300   Fig. 7) is compared with ORR activity of the pristine catalysts in Table 4 for one set of catalysts. For consistency, specific and kinetic activities are based on ECSAs measured again after 6,000 cycles by CO stripping (Supplementary Fig. 8). The ECSAs reported in Table 4 agree well with the cyclic H-UPD measurements (Fig. 4), showing a significantly larger loss in surface area for the Pt/C catalyst (23% loss after 6,000 cycles) than for the Pt/BC catalyst (12% loss after 6,000 cycles). Accordingly, mass activity of the cycled Pt/BC catalyst remains with 168A per g Pt B30% higher than seen for  the cycled Pt/C catalyst. Relative loss of mass, specific and kinetic activity of the Pt/BC catalyst are in reasonable agreement with the loss in ECSA for the Pt/BC catalyst. The Pt/C catalyst, however, shows a markedly different behaviour. While mass activity also degrades, the relative loss in mass activity (9% loss) is smaller than the reduction in ECSA (23% loss). Correspondingly, surface specific and kinetic activities increase for reasons discussed later. Nonetheless, kinetic activity of the cycled Pt/C catalyst remains significantly lower than seen for the Pt/BC catalyst after 6,000 cycles.
Platinum nanoparticle size and shape. Platinum was successfully deposited onto BC with the same average particle sizes as on carbon supports (with the exception of 10 wt% Pt/C) and similar particle size distributions (Table 1 and Fig. 1). There is no significant increase in particle size for the higher loadings, indicating that the surface area for both supports is sufficient to accommodate high loadings of nanoparticles. Figure 5 compares the Pt-Pt co-ordination numbers obtained from EXAFS results presented in Table 2 with co-ordination numbers for ideal icosahedra and cubocahedra 46 . The first four Pt-Pt co-ordination numbers are plotted against average number of atoms per particle as obtained from TEM. The error margins are calculated from the TEM particle size distributions and the confidence interval of the modelled parameters as given by Demeter 47 .
The co-ordination numbers N 1y4 for the two supports are within the error margins of the fits and, therefore, suggest very similar Pt particle sizes and shapes on both supports. N 1 is often used to deduce particle sizes 40 . Comparing N 1 with the expectation values for cuboctahedra and icosahedra shows good agreement with the TEM derived particle sizes for both catalysts.
Particle shape has a stronger influence on more distant platinum shells 32 . N 2 and N 3 , however, do not lend themselves to further analysis in our case. The differences between icosahedral and cuboctahedral geometries are too small relative to the precision of our fitted values, which is in part inherently limited by the particle size distribution. For this reason, we have abstained from using the method described by Jentys 48 , which relies on the ratio of N 3 to N 1 to deduce particle size and shape. The analysis of N 4 , however, is instructive due to the strong shape sensitivity of N 4 for icosahedra and cuboctahedra. The N 4 values of both catalysts agree better with a cuboctahedal shape model, suggesting similar exposure of Pt(001) and Pt(111) facets on both supports. Consequently, EXAFS suggests even stronger than TEM that sample averaged particle sizes and shapes are very similar for the Pt/BC and Pt/C catalysts.
Electronic metal-support interaction. As there is no significant change in catalyst size, shape and lattice constant, it is tempting to attribute the significant changes in ORR activity to electronic effects. The Pt 4f spectrum shifts to higher binding energies  10 Figure 5 | Catalyst particle shape. Co-ordination numbers N 1 through N 4 for icosahedra and cuboctahedra with superimposed calculated co-ordination numbers from the EXAFS for the 20 eq wt% Pt/BC and 20 wt% Pt/C against average particle size as suggested by TEM. Error bars indicate s.d. of TEM particle size distribution and confidence intervals from EXAFS fit. Shape models adapted from Glasner and Frenkel 48 .
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15802 ARTICLE relative to the C 1 s core levels for Pt/BC (Fig. 2a). Following Lewara et al. 17 , this might indicate a possible alloy formation with boron or a change in charge transfer between Pt and the support material. There is no evidence of alloy formation in the B 1s XPS (Supplementary Fig. 9). The B 1s peak does not shift, and a model considering only B-B, B-C and B-O bonds is in perfect agreement with the observed spectra. The very similar Pt lattice constants obtained from EXAFS (Table 2) provide further evidence for pure Pt particles. The Pt 4f binding energy shift is, therefore, assigned to purely electronic effects. It is interesting to note that similar electronic effects have been observed for multiwall carbon nanotubes by Ma et al. 49 who find a similar but smaller shift of the Pt 4f core levels. These carbons typically show a work function B0.3-0.5 eV different from graphite 50 . It is, therefore, likely that the measured shift in the Pt 4f core levels are a direct result of changes in the support electronic structure.
Generally, charge is transferred to equilibrate the respective Fermi levels when support and Pt nanoparticle contact. This is qualitatively shown in Fig. 2b. Amongst other factors, the magnitude of the expected charge transfer will depend on the electronic properties of the support material. When changing the support (in this case from C to BC) the extent of this charge transfer changes, resulting in changes in the valence manifolds of the supported catalyst 51 and/or support 52 . This change in electron occupancy can be probed using relative shifts of core levels on either side of the interface 52 .
By aligning the C(sp 2 ) 1s levels for Pt/C and Pt/BC to exclude other effects, a shift in the Pt 4f peak reflects changes in valence state occupancy across the interface as a result of the changing support material, but might also be influenced to some degree by varying final state effects 52 . The reduction of the difference between Pt 4f and C(sp 2 ) 1 s core levels by B0.6 eV shown in Fig. 2, therefore, indicates an increase of Pt 4f and/or decrease of C 1 s core level binding energies due to larger electron transfer into Pt under UHV conditions, strongly suggesting that the improved ORR activity and stability originates from electronic metal-support interactions.
Charge transfer and d-band occupancy. In principle, one should be able to correlate the sign and magnitude of the relative core level shift with the electronic structure of the respective support materials. There is, however, considerable uncertainty around the electronic properties of the materials reported in the literature. For example, the work function of carbons is usually around 5 eV, but varies as a function of surface termination 53 . Further, the band structure of boron carbides is sensitive to the local stoichiometry of samples, and the Fermi level depends on the type/degree of doping. The situation is further complicated by the fact that the support material is a composite containing a significant phase fraction of graphitic carbon although we expect the phase fraction on the surface to be significantly smaller than for the bulk support materials due to the aggressive pre-treatment in nitric acid.
Regardless, stronger electron transfer should occur from B 4 C (typical work function B4-4.5 eV (ref. 54)) to Pt (typical work function B5.5 eV (ref. 55)) than from C (typical work function B5 eV (ref. 53)) to equilibrate the respective Fermi levels as illustrated in Fig. 2b and in agreement with the metal-support interaction suggested by XPS under UHV conditions.
The situation in electrochemical environment, however, is more complex. The in situ XANES spectra shown in Fig. 2c clearly shows an increased hole density in the Pt d-band manifold when particles are supported on BC (that is, more positively charged particles). This apparent contradiction can be resolved by realising that in situ XANES probes a charged interface under potentiostatic control, while ex situ XPS necessarily probes the catalyst under overall neutrality. Because work function and potential of zero charge (pzc) are proportional to each other 56 , the smaller support work function of the BC-based catalysts will cause a negative shift of the pzc and, hence, a more positively charged Pt/BC catalyst if held at the same potential as the Pt/C catalyst in the same aqueous electrolyte. Finally, it is noteworthy that the support-induced decrease in d-band filling seen for Pt/BC is reminiscent of electronic effects seen for highly active Pt-alloys 57 .
Comparison of ORR activity with benchmark data. The trends in ORR kinetic activities (Fig. 3, Table 3) are not only a function of support material. Particle size effects are likewise relevant. The conventional interpretation 58 in terms of changing surface fractions of Pt(100) and Pt(111), which show different activity for ORR 59 , has recently been challenged over the relevant particle size range 40 . However, they are known to significantly impact the results from RDE measurements if the standard protocols and data treatment used here are followed. Figure 6, therefore, compares surface area specific activity of the Pt/C and Pt/BC catalysts with benchmark data from Garsany et al. 44,45 Note that the data from Garsany et al. was corrected for temperature from 30 to 25°C following a suggestion by Neyerlin et al. 60 Table 3). Our results for the benchmark Pt/C catalysts are in good agreement with the literature data attesting to the quality of our films and robustness of experimental procedures. As already mentioned, the 10 wt% Pt/C catalyst does not lend itself as a benchmark for our purposes due to the significantly smaller particle size and considerable experimental scatter.
The Pt/BC catalysts broadly follow a similar particle size dependence but with overall B50% higher surface area specific activity at 0.9 V versus RHE. The increase of surface area specific current density is significantly larger than the expected experimental scatter indicated as grey areas. We, therefore, conclude that purely electronic equilibration between support and catalyst has a bearing on ORR activity of Pt nanoparticles. This is particularly interesting if one considers that the effective charge screening of metals should completely screen the charge redistribution due to Fermi level equilibration between catalyst and support a few atomic distances away from the contact interface [61][62][63] . Hence, one would not directly expect a pronounced electronic effect on large parts of the more than 20 Å thick Pt particles.
Durability. Figure 4 shows the degradation of ECSA over 6,000 cycles for the 40 eq. wt% Pt/BC and the 40 wt% Pt/C catalyst. The voltage range was selected to promote platinum dissolution and/or agglomeration over support corrosion 64 . There is a significantly larger rate of ECSA loss for Pt/C after some initial cycles. The consequences of the logarithmic dependence seen in Fig. 4 are worth highlighting. Due to the 1/3 reduction in loss rate per decade, the Pt/BC catalyst retains more than 90% ECSA for about 2,000 cycles, while the Pt/C catalyst shows a 10% loss in surface area already after about 200 cycles.
Since particle sizes and shapes of both catalysts are initially similar, this cannot be due to a more stable particle size as starting point. Hence, the results suggest that the stronger electronic metal-support interaction between Pt and the BC support leads to a reduction in platinum dissolution and/or agglomeration. The ORR activities of the cycled catalysts reported in Table 4 suggest that mostly Pt agglomeration is inhibited for Pt/BC relative to the Pt/C catalyst. While the relative loss of mass specific ORR activity after 6,000 cycles is in line with the loss of ECSA for Pt/BC, Pt/C shows an increase in surface specific and kinetic activity after cycling. This increase in surface specific activity and simultaneous loss in mass activity is a clear indication of increasing Pt particle sizes on the Pt/C support. CO stripping voltammetry ( Supplementary Fig. 8) provides further evidence for increased Pt agglomeration on the cycled Pt/C. The pre-peak that is observed for both pristine catalysts has been attributed previously to Pt agglomerates 65 . This pre-peak feature increases for the cycled Pt/C catalyst while it disappears for the cycled Pt/BC catalyst, indicating increased particle agglomeration for cycled Pt/C. We have argued above that a stronger dipole field is established at the Pt/BC interface relative to Pt/C, which is corroborated by XPS. This stronger dipole interaction likely reduces the mobility of Pt nanoparticles mitigating agglomeration. A similar effect is seen on carbon supports where the rate of platinum nanoparticle agglomeration is seen to reduce when carbon supports are modified with nitrogen and oxygen containing functional groups that strongly interact with nanoparticles 66,67 .
Likewise, it suggests that Pt nanoparticles can adopt a lower energy state on BC, which, in principle, should mitigate dissolution. The CVs in Fig. 3b support this interpretation. The oxidation/reduction features of Pt/BC are slightly shifted to more positive potentials relative to Pt/C. Hence, the catalyst might be slightly less prone to surface oxidation/reduction and associated anodic and cathodic Pt dissolution with beneficial implications for cycle stability in this potential range.

Methods
Catalyst preparation. As received B 4 C (NaBond Technologies Co., Ltd.) was stirred in conc. nitric acid (Kimix Chemicals and Lab Supplies cc.) at room temperature for 8 h. Organo-metallic chemical deposition was used to deposit platinum onto the treated B 4 C to prepare 10, 20 and 40 eq. wt% Pt/BC, where the eq. wt % was calculated from surface areas calculated from nitrogen physisorption using a TriStar II 3,020 (Micrometrics) and interpreted using BET theory. The catalyst deposition was carried out using platinum acetoacetate (Sigma Aldrich) at 350°C under an argon atmosphere 68,69 . As-received Pt/C catalysts from Alfa Aesar (HiSpec 2000, 3000, 4000) were used as a commercial standard for electrochemical activities.
Ex situ catalyst physical characterization. The prepared support material was characterized by nitrogen physisorption, analysed using BET theory (Micromeritics Instrument Corporation TriStar II), to determine the support surface area. XRD measurements were carried out on a Bruker D8 Advance diffractometer with a Cu Ka radiation source operating at 40 kV to determine the crystalline phases present in the catalyst. Rietveld refinement on the XRD patterns was completed using Bruker AXS TOPAS software, Version 4.1. TEM was performed on a Tecnai G2 electron microscope operating at 200 kV to determine particle sizes and a Tecnai F20 with a FEG at 200 kV to obtain images shown in Fig. 1. Elemental analyses were carried out by using a Cs probe-corrected JEOL ARM200F (cold-FEG) TEM/STEM operated at 200 kV and equipped with a Gatan GIF spectrometer. EELS data were acquired with an energy dispersion of 0.25 eV per channel. XPS was carried out by Nexus XPS service at Newcastle University using a Thermo Scientific K-Alpha instrument with an Al Ka X-ray source. The supported catalysts were characterized using TEM for Pt particle size using an average of at least 100 particle diameters sampled across three images for each catalyst; particle distances were obtained by measuring the closest neighbouring platinum particle distance for 100 particles across three TEM images for each catalyst. XRD was used to verify the Pt crystallite size, XPS to investigate the binding energies, and ICP-OES (Varian 730-ES) to confirm the Pt loading on the support. Ex situ advanced surface characterisations were completed using X-ray adsorption spectroscopy on the B18 line at the Diamond Light Source to investigate the morphology of Pt on the surface of the support. The Pt L 2 and L 3 edges (13,273 and 11,564 eV) were measured on beamline B18 at Diamond Light Source which operated with a ring energy 3 GeV and at a current of 300 mA. The monochromator used Si(111) crystals operating in Quick EXAFS (QEXAFS) mode. A total of three spectra were averaged for each sample. The measurements were collected using the ionization chambers in transmission mode at 298 K. Calibration of the monochromator was carried out at both edges using Pt foils. XAS was collected for the 20 wt% Pt/C and 20 eq. wt% Pt/BC catalysts only. These catalysts were prepared using boron nitride to bind pellets of the catalysts, the pellets were then placed under a H 2 atmosphere to reduce surface oxides. The absorption spectra were modelled with Demeter by Bruce Ravel using Ifeffit by Matt Newville 17 to solve the EXAFS equation for the first four Pt shell co-ordination numbers as described by Frenkel et al. 49 The ink formulation for the 20 eq wt% Pt/BC catalysts used for XANES consisted of 228 mg catalyst, 3.42 ml 5 wt% Nafion solution, 7.6 ml ethanol (Sigma Aldrich); a catalyst layer of 91.46 mg was coated onto a Teflon coated carbon base layer. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15802 ARTICLE Catalyst electrochemical characterization. Catalyst inks were prepared for the electrochemical characterization techniques. The Pt/BC catalyst ink consisted of 2.5 mg of the catalyst, 0.5 ml of ethanol (Sigma Aldrich) and 25 ml of 5 wt% Nafion solution. The mixture was sonicated for 30 min and 5 ml was placed on a glassy carbon electrode. The Pt/C catalyst ink consisted of 5 mg of the catalyst, 5 ml of water (Milli-Q 18.2 mOcm), 1.5 ml isopropanol (Kimix) and 25 ml of 5 wt% Nafion solution. The mixture was sonicated for 30 min and 10 ml was placed on a glassy carbon electrode. Prior to this, the glassy carbon electrode was polished to a mirror finish using 0.1 and 0.05 mm Al 2 O 3 polish (Buehler). The reference electrode used was Hg/HgSO 4 with a platinum counter electrode. Cyclic voltammetry was conducted by cycling the potential between 0.05-1.2 V versus RHE with a scan rate of 50 mVs À 1 for 50 cycles at room temperature in an oxygen-free 0.1 M HClO 4 electrolyte solution to electrochemically clean the catalyst surface. The scan rate was then changed to 20 mVs À 1 for 5 cycles and the potential cycled between 0.05 and 1.2 V versus RHE at room temperature in an oxygen-free 0.1 M HClO 4 electrolyte solution. CO stripping was conducted by purging the electrolyte solution with CO for 20 min while holding the potential at 0.1 V versus RHE, argon was then bubbled for 20 min to rid the solution of CO, after which the potential was cycled between 0.05 and 1.2 V versus RHE. The ORR activity was measured by RDE experiments in an O 2 saturated 0.1 M HClO 4 electrolyte (70% Suprapur-Merck) using the anodic scan between 0.05 and 1.2 V versus RHE, with a scan rate of 20 mVs À 1 and rotation rate of 1,600 r.p.m. at room temperature. All scans were normalized for ECSA and corrected for ohmic drop measured by impedance spectroscopy and capacitive current in an oxygen-free electrolyte. Error margins were obtained from 2 to 4 repeats for each data point. The catalyst degradation studies were completed by cycling between 0.6 and 1.0 V versus RHE with a scan rate of 50 mVs À 1 in 0.1 M HClO 4 electrolyte solution for 6,000 cycles at room temperature, and periodically reverted to a cyclic voltammogram between 0.05 and 1.2 V versus RHE with a scan rate of 20 mVs À 1 to investigate the changes to the H UPD region. The reference electrode used was Hg/HgSO 4 with a gold counter electrode. The H UPD region was used for changes in ECSA in this instance. ORR activity and ECSA were reassessed after cycling using RDE and CO stripping following the procedures described above.
Data availability. All data supporting this study are openly available from the University of Southampton repository at http://dx.doi.org/10.5258/SOTON/D0067.