Electron penetration triggering interface activity of Pt-graphene for CO oxidation at room temperature

Achieving CO oxidation at room temperature is significant for gas purification but still challenging nowadays. Pt promoted by 3d transition metals (TMs) is a promising candidate for this reaction, but TMs are prone to be deeply oxidized in an oxygen-rich atmosphere, leading to low activity. Herein we report a unique structure design of graphene-isolated Pt from CoNi nanoparticles (PtǀCoNi) for efficiently catalytic CO oxidation in an oxygen-rich atmosphere. CoNi alloy is protected by ultrathin graphene shell from oxidation and therefore modulates the electronic property of Pt-graphene interface via electron penetration effect. This catalyst can achieve near 100% CO conversion at room temperature, while there are limited conversions over Pt/C and Pt/CoNiOx catalysts. Experiments and theoretical calculations indicate that CO will saturate Pt sites, but O2 can adsorb at the Pt-graphene interface without competing with CO, which facilitate the O2 activation and the subsequent surface reaction. This graphene-isolated system is distinct from the classical metal-metal oxide interface for catalysis, and it provides a new thought for the design of heterogeneous catalysts.

1. Page 3, Figure 1: Have the authors prepared Pt/(Co, Ni)/C samples, and how does their catalytic performance and spectroscopic characteristics compare to the Pt/CoNi@C materials? I would be interested to know how the d-hole count of Pt from XAS would look on a sample intentionally prepared to have Pt in contact with Co and/or Ni (e.g, perhaps bimetallic PtCo nanoparticles supported on C), as one would expect this might be different from the samples in this paper that have the C layer in between Pt and Co or Pt and Ni. This could provide further evidence that the authors proposed structure is present throughout the bulk of the sample (and not just in the limited regions observed by TEM), and if the spectroscopic data for PtCo particles on C is distinct from those reported in the manuscript, it would provide evidence that Co does not escape from the C shell and alloy with Pt over time. I will note that this is addressed to some extent on page 8, wherein PtCoNiOx is less active than the fresh samples, which the authors propose results from a lack of free electrons in the CoNiOx for donation to Pt.
2. Page 3: The authors mention "it is interesting to find that Pt NPs are selectively deposited on the surface of graphene with CoNi NPs inside but rarely on the surface of the concomitant CNT" and refer the reader to TEM images. The TEM images will only represent a minority of each sample, is there any additional quantitative surface characterization the authors can provide that Pt NPs are selectively deposited near the CoNi NPs? Perhaps IR spectra of adsorbed CO (or Pt XPS) would show a different peak center for Pt/C compared to Pt near encapsulated CoNi? The XANES data show, on average, that Pt/CoNi@C has different edge energy than Pt/C, but these data for Pt/CoNi@C could reflect a mixture of Pt particles distant from CoNi particles and Pt particles adsorbed at CoNi. Figure 1: is the Pt4 model realistic given that the Pt NPs are 1-2 nm in size and likely comprised of >4 Pt atoms? This is my strongest criticism of the paper in multiple areas -whether or not this model is representative of the samples. Were other models with larger numbers of Pt atoms considered, and were the results similar to those reported here? The effect of the number of Pt atoms on the calculated parameters should be included in the article or the SI. Additionally, the authors model a single Pt4 particle on each CoNi cluster, but from TEM (e.g., Figure 1d) it appears plausible that multiple Pt particles are attached to each encapsulated CoNi cluster. What is the effect of inclusion of multiple Pt particles on the calculated parameters? I would imagine either a larger number of atoms per particle, or a larger number of Pt particles per CoNi cluster, would disperse the electronic effects to a greater extent, perhaps minimizing the magnitude of the difference relative to C without CoNi encapsulated. These calculated parameters include Bader charge and Pt binding energy (page 5, Figure 1h), PDOS (Figure 1j), DFT calculations ( Figure 3). Figure 1h: is Pt binding and Bader charge at an empty C sphere an appropriate comparison for these characteristics when Pt binds to C-encapsulated CoNi particles? When not binding to an encapsulated CoNi, Pt would realistically be bound to an extended C surface, so perhaps a filled C sphere would be a better comparison for these calculations than a hollow C sphere (assuming there are not hollow C spheres in the C supports used by the authors).

5.
Page 6: what leads to the deactivation over 24 h, and why can the material be regenerated in H2? Have the authors performed XPS (or TEM) on the catalyst after 24 h reaction, and how does this compare to that for the fresh catalysts? TEM after exposure to reaction conditions at higher temperatures is reported in Figure 2c, but I did not see the TEM for the materials after 24 h exposure to reaction conditions at 298 K.
6. Page 6: what was observed for Pt during the NAP-XPS experiments? It would benefit the reader if the Pt XPS data during these treatments were included in Figure 2. Furthermore, I would note that these CO and O2 pressures are much lower than those used for catalysis, so the effect of extended exposure to 1 kPa CO and 20 kPa O2 on the oxidation states of Pt, Co, and Ni at 298 K is unclear from the data reported in Figure 2b. Could the authors perform operando Pt, Co, and/or Ni XAS to monitor the oxidation states of these elements at reaction conditions? 7. Page 7, Figure 2: for the caption of figure 2e, it would help if the treatment conditions were included in the caption (gas pressures, treatment durations, flowrates per mass catalyst, etc.). 8. Page 9: Figure 2e shows that there is O on the samples after reaction at 298 K, that cannot be removed even after reduction at 513 K. The authors propose this O reflects activation of O2 on the graphene surface. Is this signal O2, or O adatoms? In either case, the fact that these are not removed during after reduction at 240 C suggests that these may be structural or strongly bound O atoms, that may not be participating in catalysis. Are these present on the CoNi@C and C materials in the absence of Pt? It was unclear to me if these O are part of the same reservoir of O that is ultimately incorporated in CO2, or if these O are spectators/unreactive. In the caption of Figure S5, the authors mention that O2 can be activated on the surface of CoNi@C -do the authors mean O2 is dissociated on this surface? If so, what evidence is provided that O2 is dissociated into O atoms in the absence of Pt? Perhaps the LEIS data provide this evidence, if the peak for O is characteristic of O atoms and not O2.
9. Page 10 - Figure 3c -it was unclear what the model is that was used for Figure 3c -the Pt particle is much larger than the Pt4 model used for DFT, Bader charge, and Pt adsorption energy calculations. This model may be more representative of the Pt particles present on the samples in this study.
10. SI comments: Experimental section -a variety of details are missing that could inhibit efforts to reproduce the work. a. Line 28: Please add the supplier for the SiO2 spheres, or the procedure for synthesizing these. b. Line 29: For the equimolar Co and Ni solution what was the concentration used? c. Lines 30-32: For the reduction in H2 and CH4 to form CoNi, what flow rate per g catalyst was used, and what was the temperature ramp rate? d. Lines 34 and 40: what temperature and pressure were used for the vacuum drying steps? e. Line 38: how long was the ultrasonic oscillation performed? f. Line 57-58: additional details about the reaction chamber for LEIS should be added, or a reference to the design added. g. Line 60: what was the inner diameter and material used for the reactor? How was the reactor temperature measured (thermocouple inside the reactor or on the external surface?) h. Line 61: what was the particle size of the catalyst? Was the catalyst sieved before loading the reactor? i. Line 63: Can the authors add references or description of the "methane converter," and mention which GC column was used for separation? j. Line 64: What were the purities of the H2, CO, O2, and He used in the study? 11. Figure S3: It appears the catalyst would continue deactivating at 298 K if the reaction were run for >24 h. If possible, the authors should add an example of the reaction run for 72 h (or longer) without intermediate reduction steps, and then characterize the sample (XAS, TEM, XPS) afterwards to try to explain the cause of the deactivation. If there is no change in the CO pressure in the feed, why would the reaction rate decrease due to CO poisoning as a function of time on stream? In the caption of Figure S3: The last sentence of this caption should have a reference added, or additional description added.

Reviewer #2 (Remarks to the Author):
This is a study to understand the effect of graphene encapsulated transition metals (TM) on the activity of a supported metal NP-graphene interface. The authors conclude that the TM/graphene/TM configuration enhances the catalytic activity by the electron penetration effect from the CoNi alloy to Pt NP. The DFT calculations are used to support such results. I have major reservations. 1. To model the Pt-graphene interface, the authors employed a Pt4 cluster model for the Pt nanoparticle (NP). Under this model, O2 shows exclusive adsorption with Pt4/CoNi@C compared to Pt4/C. However, the finite-size effect in the simulation of NPs is critical and well-known to yield wrong binding energetics, especially for smaller sized clusters (1-5 atoms) (J. Phys. Chem. Lett. 2013, 4, 222−226 & Journal of Catalysis 223 (2004. Experimental Pt NP is about 2 nm, but Pt4 cluster used here to model it is ~0.5 nm. I strongly suspect the artifactual high binding affinity of this 4-atom cluster (of the order of > 1 eV according to the above refs). 2. The CoNi NP inside the carbon is also modelled by 55-atom cluster (1.2 nm in diameter) while experimental CoNi-NP is about 5 nm (~4500 atoms). Electronic structure of the model 55-atom cluster (1.2 nm) used here and realistic 4500-atom cluster (5 nm) are expected to be, again, quite different, and quite possibly a higher reactivity of smaller clusters may have contributed to the calculated stronger binding energetics claimed here. 3. If the electron penetration is important in Pt4/CoNi@C, the local configuration of CoNi nanoparticle can be critical to affect the catalytic activity also. How did the authors determine the configuration of 55-atom CoNi alloy cluster (aside from the previous comment on the potentially artificial finite size effect of 55-atom cluster), and how the calculated energetics vary with different CoNi configurations? 4. In Fig.3, the adsorption energies are calculated without any CO coverages, but the reaction barriers are calculated with 4 pre-adsorbed CO. Why? The presence of pre-adsorbed CO molecules could affect greatly the adsorption energetics of CO, O2, and the activation barriers altogether. One has to systematically show the coverage dependent energetics for CO binding, O2 binding, and CO oxidation.
With all Pt atoms saturated with CO, O2 could be adsorbed at the interface also. 5. To really show the effect of CoNi alloy inside the graphene, reaction profiles using Pt NP/CoNi@C should be compared with those using Pt NP/C. Reviewer #3 (Remarks to the Author): The manuscript reports enhanced CO oxidation reactivity of graphene-isolated Pt nanoparticles from CoNi alloy. The authors interpreted the enhanced reactivity, i.e. near 100 % CO conversion at room temperature, as modified Pt electronic structures due to electron penetration effect, i.e. delivering electrons from CoNi nanoparticles to Pt nanoparticle. With several state-of-art modern experimental tools and DFT calculation, the authors tried to deliver the messages that graphene-meditated electrons between Pt and CoNi nanoparticles play the major roles in the enhanced activity. While the DFT calculations shows energy states with detailed intermediate steps with convincing arguments, I cannot completely agree with the analysis of experimental results, which need to be clarified before the publication.
First, as a verification step, XANES was employed to estimate the d-hole of Pt nanoparticles. From the quality of figure, it is rather difficult to judge how the background subtraction was made for the calculation of d-hole. The slight change of XANES background normalization on higher photon energy side can easily change the estimated number of d-hole. It would be good to show the detailed figures in supplemental information for the clarification.
Second, the change of work function is suggested based on the DOS calculation. The change of work function can be probed with the use of NAP-XPS. The kinetic energy of gas phase signal can provide the modification of work function of the system. I wonder if the authors had a chance to look over this aspect from their results.
Lastly, but most importantly, according to the result of NAP-XPS, the graphene started to fail to protect CoNi nanoparticles from its oxidation at 150 C. To me, this shows that graphene is not fully attached to CoNi. Many previous reports showed that graphene on metal substrates can stay up to several hundred degrees of Celsius. The failure of protecting CoNi nanoparticles at this low temperature possibly indicate that there can be many defects on graphene layers. In fact, this graphene defects can easily trigger the activation of O2 adsorption/dissociation. The role of graphene defects as active sites have been repeatedly reported and found to be very important for the study of graphene. I assume that most of the findings in this manuscript can be originated from the defect of graphene layers, instead of Pt-graphene interface.

Response to the reviewers' comments
We are grateful for the reviewers' in-depth comments and suggestions, which have helped us a lot in improving the quality of this manuscript. The revised parts are highlighted in yellow in the files of "Revised Manuscript" and "Revised Supplementary Information".  Co and/or Ni (e.g, perhaps bimetallic PtCo nanoparticles supported on C), as one would expect this might be different from the samples in this paper that have the C layer in between Pt and Co or Pt and Ni. This could provide further evidence that the authors proposed structure is present throughout the bulk of the sample (and not just in the limited regions observed by TEM), and if the spectroscopic data for PtCo particles on C is distinct from those reported in the manuscript, it would provide evidence that Co does not escape from the C shell and alloy with Pt over time. I will note that this is addressed to some extent on page 8, wherein PtCoNiO x is less active than the fresh samples, which the authors propose results from a lack of free electrons in the CoNiO x for donation to Pt.

Comments
Author reply: According to your suggestions, Pt-CoNiO x /CB and Pt-CoNiO x /CNT catalysts were intentionally prepared for comparison. First, CoNiO x was precipitated on carbon supports in a form of layered double hydroxides with metals loading of 10 wt%. The procedure is the same as that for CoNiO x /SiO 2 described in the section of Methods. Then, Pt NPs (1-2 nm) were deposited on CoNiO x /CB and CoNiO x /CNT with a Pt loading of 4 wt%. The procedure is the same as that for Pt/CoNi@NC described in the section of Methods. This two-step approach can create exposed Pt-CoNiO x interfaces for a better comparison, because such interfaces but not PtCoNi alloy would be more likely formed for Pt/CoNi@NC catalyst if CoNi can escape from its graphene shell during CO oxidation reaction. It can be seen that the catalytic activity is enhanced after introducing CoNiO x onto both carbon supports and quite close for Pt-CoNiO x /CB and Pt-CoNiO x /CNT. The minor support effect indicated that most of Pt NPs should be in close contact with CoNiO x on both catalysts. The formed Pt-CoNiO x interfaces presented a much lower activity of CO oxidation than the fresh Pt/CoNi@NC catalyst (Fig. R1). It implies that the activity would decrease if Co or Ni escapes from the graphene shell during reaction, and this has been confirmed by the experimental design shown in Fig. 2d. However, the activity of the broken Pt/CoNi@NC catalyst (Fig. 2d) is a little lower than that of Pt-CoNiO x /CNT (Fig. R1), which may be due to a lower number of Pt-CoNiO x interfaces for the former, because there are still some Pt NPs deposited on the concomitant CNT in Pt/CoNi@NC. Thus, it is better to do the comparison of XANES spectra between the fresh and broken Pt/CoNi@NC catalysts.

Fig. R1
Temperature-dependence CO conversion in CO oxidation reaction over the pre-reduced catalysts. 1% CO and 20% O 2 in He (1 bar). Space velocity: 60000 mL·g -1 ·h -1 . can support our inference that the oxidation of CoNi cannot enhance the catalytic activity significantly (the 3rd run in Fig. 2d) since CoNiO x has no more free electrons to modulate the catalytic properties of the outer graphene and Pt NPs (Page 8). Thus, we would like to add Fig.   R2 as Supplementary Fig. 12 in the SI. We also added Fig. R1 as Supplementary Fig. 11  in-situ characterization for CO oxidation reaction, especially in our electron-mediated situation.
We had tried the CO-adsorbed IR testings over Pt/CoNi@NC and other catalysts in either DRIFTS or transmission mode even at a low temperature (-40 °C), but no signal of adsorbed CO can be observed after slightly vacuuming (Fig. R3a) or simply purging with Ar (Fig. R3b). It should be due to the strong absorption of IR on such a black material, which is a common phenomenon for carbon-based catalysts. Additionally, the authors model a single Pt 4 particle on each CoNi cluster, but from TEM (e.g.,  Author Reply: According to your suggestions, we appended two systems to investigate the size effect of Pt cluster. One is to increase the number of Pt atoms from 4 to 9 (Fig. R4), and the other is to build Pt nano-strip (81 atoms) on the graphene layer with or without CoNi underneath (Fig.   R5). The results of Bader charge analysis and the binding energies between Pt and graphene are listed in Table R1. For Pt 9 , it possesses a negative charge on CoNi@C but a positive charge on C.
For Pt nano-strip, it also possesses a negative charge on graphene/CoNi but a positive charge on graphene. Both trends are the same as for Pt 4 models used in our manuscript. The binding energy between Pt and C also shows the same trend for those models, which can be enhanced after introducing CoNi NPs. The lower binding energy of Pt 9 on C than that of Pt 4 on C is due to the increase of cohesive energy of Pt cluster for a larger cluster. The higher binding energy of Pt nano-strip models than others is due to the larger number of Pt atoms. The PDOS analysis for those systems also gives the same trend of increasing the Fermi level of C after introducing CoNi NPs, as shown in Fig. R6.     (Table R1) and the PDOS of C (Fig. R6), the overall trend 8 does not change for two Pt 4 clusters comparing with one Pt 4 cluster. It is reasonable in consideration of that the electron penetration from CoNi to Pt through graphene should mainly occur at the local interface. We appended all the models and calculation results here to the SI and quoted them in the manuscript on Page 5. Author Reply: According to your suggestions, we compared the electronic structure of Pt 4 cluster on an empty C sphere with that on a filled C sphere (Fig. R8). The difference of Bader charge of Pt 4 cluster between the two models is about 0.006 e -, which is neglectable. The reason is that the van der Waals interaction between the carbon layers is too weak to affect the strong bonding interaction between Pt and C.

Fig. R8
Structure of Pt 4 cluster on filled C sphere. The inner carbon sphere is colored in brown.

Question 5. Page 6: what leads to the deactivation over 24 h, and why can the material be regenerated in H 2 ? Have the authors performed XPS (or TEM) on the catalyst after 24 h reaction,
and how does this compare to that for the fresh catalysts? TEM after exposure to reaction conditions at higher temperatures is reported in Figure 2c, but I did not see the TEM for the materials after 24 h exposure to reaction conditions at 298 K.
Author Reply: According to your suggestions, we supplemented the TEM and XPS characterizations on the sample of Pt/CoNi@NC after reacting at 25 °C for 24 h. As shown in Fig 2011,133,[4498][4499][4500][4501][4502][4503][4504][4505][4506][4507][4508][4509][4510][4511][4512][4513][4514][4515][4516][4517], and this CO poisoning mechanism were also used to explain the deactivation of CO oxidation over one Pd catalyst (Angew. Chem. Int. Ed. 2015, 54, 15823-15826). However, a possible O 2 -induced deactivation can still not be excluded in our case, as the electron-rich Pt atoms especially around the interfaces may be more susceptible to oxidation under such a high partial pressure of O 2 and the oxidized Pt will present a lower activity for CO oxidation than the metallic Pt (J. Phys. Chem. C 2016, 120, 17996-18004). Actually, we have got a preliminary experimental hint for this possibility, that is, the pre-reduced Pt/CoNi@NC catalyst after purging with 20% O 2 /Ar at room temperature for a while exhibited a lower activity for CO oxidation. This phenomenon is unusual for CO oxidation over Pt-based catalysts (e.g. Pt/CNT).
A thorough kinetic study is undergoing to investigate the deactivation mechanism. In any case, the decrease in the number of active sites caused by either CO poisoning or O 2 oxidation can be easily restored by a simple H 2 reduction. We added Fig. R9 as Supplementary Fig. 8 in the SI and quoted it in the manuscript on Page 6. Thank you for this useful suggestion.   We are sorry that we cannot make the operando XAS experiments due to the restrictions of platform facilities and booking time, but a quasi-in-situ XPS testing was performed over Pt/CoNi@NC to observe the change of valence states of both metals (Pt, Co, and Ni) and light elements (C, O, and N). Same to the LEIS setup, we connected the in-situ reaction chamber next to the UHV-XPS chamber. The in-situ reaction chamber allows the reduction and reaction of catalysts with the same parameters of temperature and pressure as for the actual testings. Firstly, the sample was pretreated in a flow of H 2 at 240 °C for 2 h. Then, the reaction was performed under a flow of mixed gas (1% CO and 20% O 2 in He) at room temperature for 72 h. After each 24 h, the in-situ reaction chamber was evacuated to UHV, and then the sample was transferred into the UHV-XPS chamber for detection. As shown in Fig. R11, there is no significant change for Co, Ni, Pt, and N signals, but the O signal gradually increases after each reaction stage (Table   R2) and the peak position of 531.5 eV corresponds to the C=O species, indicating the activation of O 2 on the graphene surface.  suggestion, here we performed the LEIS testings over CoNi@NC and CB without loading Pt. To avoid some potential influence of CO, synthetic air was used to flow through the samples at room temperature for 2 h between two reduction processes (H 2 , 240 °C, and 2 h). As seen from Fig.   R12, the O signal of CoNi@NC increased after exposing in air flow and cannot be efficiently removed by such a mild reduction, which is consistent with the results of Pt/CoNi@NC shown in Fig. 2e. In contrast, the O signal of CB did not change too much after these treatments (Fig. R12).
As a conclusion, O 2 can be activated on CoNi@NC but not on CB at room temperature, which can further confirm that the graphene turns to be more active when encapsulating CoNi NPs. We added Fig. R12 as Supplementary Fig. 13 to further support this conclusion on Page 9. Author Reply: The model in Fig. 3c is a schematic diagram and was not used in our calculations.
This model is indeed more representative, but the computational cost is too high to even perform static calculations, let alone the reaction-related calculations shown in Fig. 3b. It contains more than 1000 carbon atoms and 1000 metal atoms, which is very challenging for the DFT method.
However, as mentioned above, we constructed a similar Pt nano-strip model (Fig. R5)  h. Catalysts were sieved through a 200-mesh sieve so that the particle size is within 80 μm.
i. The methane converter is a micro high-temperature furnace (330 °C) with Ni-based catalyst that can fully convert CO and CO 2 into CH 4 for FID detection. Instead of TCD, FID was used to achieve a higher sensitivity for GC analysis. Thus, the methane converter is necessary considering that CO and CO 2 cannot be directly detected by FID. The GC column used here is a packed column (TDX-01).
j. The purity of H 2 and Ar is 99.999%. We didn't use the pure gases of CO, O 2 and He but a mixed gas of 1% CO, 20% O 2 , and 79% He as the reaction gas for this work.
Question 11. Figure S3: Author Reply: According to your suggestion, we performed the 72-h stability testings over Pt/CoNi@NC at 25 °C under two space velocities (Fig. R13). The initial CO conversion can also reach near 100% under a shorter residence time (the higher space velocity), and the ratio of CO conversion between the two space velocities tends to approach 2 in the final stage. It means that the coverages of CO and O 2 have not reach the steady state at the initial stages with high CO conversions for both curves, indicating a CO poisoning process occurred. Also from Fig. R13, we can observe two deactivation rates with 36 h as dividing line, indicating a different deactivation mechanism existed. As stated in the reply to your Question 5, a possible O 2 -induced deactivation may co-exist. At present, by only judging from the TEM and XPS data ( Fig. R9 and Fig. R11) the deactivation mechanism is still unclear. A more sensitive operando XAS (in plan) may provide more information and an in-depth kinetic study is more necessary and undergoing for this purpose, which hope to form another independent work. Here for the description in the caption of Supplementary Fig.7 (Figure S3 in previous SI), it was modified as "the slow deactivation should be due to either the gradual CO poison on Pt NPs or the possible oxidation of Pt NPs, which is still under investigation".

Reviewer #2
Comments: This is a study to understand the effect of graphene encapsulated transition metals  Author Reply: Thank you for your insightful comment. Smaller sized clusters usually present higher adsorption energy, which is a very common phenomenon for DFT calculations. We also noticed the relatively high adsorption energies for both O 2 and CO on our model than on the Pt slab model reported elsewhere (e.g. Science 2010, 328, 1141-1144). Thus, we appended a larger Pt nano-strip model (Fig. R14) to simulate the interface structure in our real catalyst. It consists of 81 Pt atoms, 110 Co atoms, 110 Ni atoms and 280 carbon atoms. The electronic structure analysis (Table R3 and Fig. R15) indicates that, despite the quantitative values are different, the main trend and conclusion are the same for Pt 4 cluster model and Pt nano-strip model. For instance, the charge density of Pt is increased and the work function of graphene is decreased when introducing CoNi into each model. Furthermore, the binding strength for Pt on graphene/CoNi is also stronger than that on graphene, which can be directly observed from the side view of the structures (Fig. R14).  on Pt or at the Pt-graphene interface (Fig. R16 or Supplementary Fig. 17). Comparing with Pt 4 cluster models (Table R4), the adsorption of CO on Pt nano-strip models become slightly weaker but are still stronger than those reported in literatures (e.g. -1.64 eV for CO on Pt(111) from the above mentioned paper), which is mainly because the adsorption in the literatures took place at close-packed sites whereas edge sites were used here. However, the main conclusion of CoNienhanced CO adsorption drawn from the P 4 cluster model is still valid for the Pt nano-strip model.   Table   R5, the charge of Pt cluster can also gain electrons (0.06 e -) from the reversed CoNi. CO adsorption is slightly weakened by 0.11 eV, whereas O 2 adsorption at interface is enhanced from -0.37 to -0.75 eV. The data further validate the enhancement of electron penetration effect on the reaction, and the trend becomes more favorable. It is reasonable to predict that the average effect falls in between these two configurations if we take more configurations into account. cluster (Table R6). However, the adsorption of the 5th CO on 4CO-adsorbed Pt 4 (around -1.6 eV) is still stronger than that of O 2 on a clean Pt 4 (around -1 eV, Fig. 3a), indicating that O 2 cannot compete with CO for adsorption on Pt site no matter with the coverage of CO. That is why we start the energy profile form 4CO-adsorbed Pt 4 (model I in Fig. 3b), and the first step is the adsorption of the 5th CO on it to reach the saturated coverage. After that, we found that O 2 can still adsorb at the Pt-graphene interface of Pt 4 /CoNi@C (model III, ΔE = -0.21 eV) but cannot adsorb at the interface of Pt 4 /C, as we described on Page 11 of the manuscript. For the influence of CO coverage on the reaction barrier, we calculated the combination step between adsorbed CO and O on a clean Pt 4 /CoNi@C, which gave a slightly lower barrier of 0.47 eV relative to that of 0.63 eV on the CO-saturated Pt 4 /CoNi@C (Fig. 3b). We added Fig. R17 as Supplementary Fig.   15 in the SI and quoted it in the manuscript on Page 9.  is not possible since no local minimum exists for this configuration. Therefore, the subsequent reaction is almost impossible through this mechanism. We proposed that the reaction for Pt 4 /C follows the classical mechanism, where O 2 adsorption has to compete with CO adsorption on Pt site. This will result in low activity at room temperature for Pt/C catalyst. On the other hand, as shown in Fig. 3b of the manuscript, the free energy of O 2 adsorption for Pt 4 /C model is even higher than the maximum barrier of the whole reaction profile for Pt 4 /CoNi@C model, which can already illustrate the significant role of the encapsulated CoNi. Author Reply: Thank you for your suggestion. Fig. R18 shows the original XAS spectra of Pt/CoNi@NC, Pt/CoNi@C, Pt/CNT, Pt/CB, Pt foil, and PtO 2 . Their edge positions are exactly the same. Thus, during processing the data, E 0 was firstly calibrated to 11564.0 eV with the same energy shift of -0.1 eV for all the spectra. Also from Fig. R18, the absorption strength is weaker for the four catalysts samples than Pt foil and PtO 2 due to their low Pt loadings. Thus, the preedge and post-edge ranges were initially determined based on the spectra of Pt foil and PtO 2 and then applied to others to ensure the same ranges for all the samples. The pre-edge range we used is from -180 to -80 eV relative to E 0 , and the post-edge range is from +150 to +800 eV relative to E 0 . From Fig. R19, we can see that these ranges can fit all the samples well. The spectrum of Pt/CoNi@C has a slight bend at the high energy region probably due to the background drift. We have added Fig. R19 as Supplementary Fig. 3 in the SI according to your suggestion. Actually, if we use a pre-edge range of -150 to -50 eV relative to E 0 , the XANES spectra (Fig. R20a) are almost unchanged comparing with Fig. 1f, and so do the estimated d-hole counts. If we change the post-edge upper limit from +800 to either +750 or +850 eV, only the spectrum of Pt/CoNi@C fluctuates a lot (Fig. R20, b and c), but this fluctuation does not change the calculated d-hole counts (1.6 ± 0.02) of this sample too much. Pt/CoNi@NC. We added it as Supplementary Fig. 6 in the SI to support our DFT calculation on Page 6 of the manuscript.  Author Reply: Thank you for your concern about the possible defects of graphene in our catalyst.

Comments: The manuscript reports enhanced CO oxidation reactivity of graphene-isolated
We would like to answer it from the following three aspects.
Firstly, the graphenes of CoNi@NC and Pt/CoNi@NC can be proven to be intact by the acid leaching experiments, based on the fact that only one carbon vacancy can create a hole with a diameter of 2.1 Å which can allow Co 2+ or Ni 2+ (whose diameters are around 1.4 Å) be leached out. It should be noted that, a long-term acid leaching process had been carried out to obtain the CoNi@NC and CoNi@C samples according to our previous works (Angew. Chem. Int. Ed. 2015, 54, 2100-2104Energy Environ. Sci. 2016, 9, 123-129;Energy Environ. Sci. 2020, 13, 119-126), during which the not fully encapsulated metals can be removed, thereby leave some hollow graphene cages as seen from the TEM images (Fig. 1a, Supplementary Fig. 1, and Supplementary   Fig. 2a-2b). Here for your question, we did a further acid leaching process on the obtained CoNi@NC and Pt/CoNi@NC samples in a 10 wt% HCl aqueous solution at room temperature for 12 h, and the Co and Ni residuals determined by ICP-OES in the filtrate were negligible for both. It means that the graphene layer is intact to protect metals from acid leaching, and this is why CoNi@NC can also be used as a HER electrocatalyst in H 2 SO 4 electrolyte (Angew. Chem. Int. Ed. 2015, 54, 2100-2104).
Secondly, the relatively low oxidation temperature (150 °C) of graphene in our catalyst is due to the unique electron penetration effect of this kind of catalyst (Angew. Chem. Int. Ed. 2020, 59, 15294-15297). The encapsulated metals can donate electrons to the outer graphene and then 29 decrease the local work function of graphene as we originally found in our previous work (Angew. Chem. Int. Ed. 2013, 52, 371-375), thus activating the inert graphene for chemical reactions. It should also be noted that this 150 °C is the onset temperature for oxidation, which is rational compared with those reported in literatures. For example, for a flat graphene on an inert SiO 2 /Si substrate, O 2 oxidation at 200 to 300 °C can create strong hole doping in graphene (Nano Lett. 2008, 8, 1965-1970. Furthermore, the oxidation of graphene is highly sensitive to substrates (J. Phys. Chem. Lett. 2016, 7, 867-873;Nanoscale 2016, 8, 11494-11502). Zhang et al.
found that graphene remains inert on SiO 2 and h-BN but becomes increasingly weak against oxidation on metal substrates because of enhanced charge transfer and chemical interaction between them (J. Phys. Chem. Lett. 2016, 7, 867-873), similar to the electron penetration effect in our work. They experimentally confirmed that graphene can be easily oxidized on fully covered Cu foil at 270 °C (not the onset temperature) in air, and their DFT calculations implied that Co and Ni foils can promote the oxidation of graphene to a larger extent comparing with Cu foil. CoNi-alloy NPs used in our case should be more active that the single-metal foils. Thus, we can say that the 150 °C for the oxidation of the graphene over 5-nm CoNi NPs is rational and reasonable. Actually, the intact graphene has already protected the CoNi NPs from oxidation below 150 °C, because our NAP-XPS results indicate that the small CoNi NPs will be easily oxidized at near room temperature without the protection of graphene ( Supplementary Fig. 10).
Thirdly, we would like to discuss the role of graphene defect in CO oxidation. Indeed, graphene defect can activate O 2 as we also found before (Chem. Commun. 2011, 47, 10016-10018). Thus, we prepared Pt NPs on defect-rich graphene from graphene oxide according to our previous work (J. Catal. 2019, 377, 524-533), and then performed CO oxidation over this Pt/rGO catalyst. As shown in Fig. R22, we can see that its activity is between those over Pt/CNT and Pt/CB, indicating that the defect alone cannot effectively activate O 2 at room temperature and the underneath CoNi NPs are still the key to the high room-temperature activity. If one assumes that the role of graphene defect is to expose CoNi for directly activating O 2 , actually we have confirmed that CoNi NPs will be oxidized at room temperature once exposed to O 2 and then loss the effectiveness to promote the activity of CO oxidation (Fig. 2d). In conclusion, as we stated on Page 8 of the manuscript, the intact graphene layer is very important for the keep of CoNi in the metallic state and thus for the effectiveness of the electron penetration effect on the roomtemperature activity of CO oxidation. Reviewer #2 (Remarks to the Author): Overall, the authors have addressed most of the comments quite thoroughly with additional DFT calculations on enlarged nano-strip models and different CO coverage effects. The nano-strip models have shown consistency in results with the smaller original graphene cage models. The CO coverage effect on interfacial O2 adsorption has shown to weaken O2 binding energies within a reasonable range of <0.2 eV.
However, there is still some clarification that could be made regarding the authors' response to Question 3. For the suggested reversed-CoNi@C model, the interfacial O2 adsorption strengthened by a significant amount (0.38 eV), while the Bader charge of Pt4 is not much different (only 0.02). Therefore, it seems that other factors may be affecting the interfacial O2 adsorption other than solely the electron penetration to Pt4 (otherwise the Bader charge should have a meaningful difference in value to account for the 0.38 eV difference in binding energy). As the interfacial O2 adsorption involves chemisorption of O atoms at both Pt and C, perhaps the electronic structure properties of the C atoms in proximity of the binding site (or directly binding to an O atom) or looking into O2 binding on CoNi@C without Pt clusters may help to more fully understand why interfacial O2 adsorption occurs on the system studied.
Reviewer #3 (Remarks to the Author): The authors answered my previous questions with detailed explanation. So, I don't have any further questions in regard to those. However, I found the other question in the response letter. In the answer to question 6 of reviewer #1, the author showed the AP-XPS spectra of Pt 4f 7/2 in Fig. R10, claiming no significant change in Pt element. I strongly disagree with this argument. In Fig. R10, the peak of Pt 4f moves to the lower binding direction when the temperature reaches to 100C. The estimated peak position of Pt 4f is ~70.8 eV, which corresponds to the metallic states of bulk Pt 4f.
(Please, find the figure attached.) After 100C, the spectra shift back to higher binding direction, possibly indicating the oxidation of Pt atom under oxygen rich CO oxidation condition. So, now it makes me think about the validity of XPS analysis and its interpretation.
First of all, from the first look, the line shape of Pt at T=25C shows the presence of oxidation. It is well known that oxidized Pt shows the increased intensity at higher binding energy side.

Response to the reviewers' comments
We are grateful for the reviewers' helpful comments and suggestions. The revised part is highlighted in yellow in the file of "Revised Supplementary Information". Similar to the small difference in the charge of Pt, the Bader charge difference of the carbon sites 2 between the two models is also very small (-0.004 e -). The 2p density of states (DOS) of the carbon sites also have no obvious difference. Thus, the activation of the carbon surface and Ptgraphene interface is not much affected by the local structure of the encapsulated metal. However, the adsorption of O 2 at the interface may induce local structural changes in the vicinity of the adsorption site depending on the element type, component, and size of the encapsulated metal, as found in our previous work (Angew. Chem. Int. Ed. 2013, 52, 371;Energy Environ. Sci. 2016, 9, 123;Nano Energy 2018, 52, 494). This could lead to fluctuated adsorption energies at the interface though the electronic structures of the adsorption sites are similar in the two models. Author Reply: Thank you for your comment on the NAP-XPS spectra of Pt 4f 7/2 ( Figure S9).

Comments
Firstly, we would like to point out that the peaks displayed in the previous Figure S9 are too broad to observe the peak position accurately. Here we present a screenshot of the normalized display of these original spectra in CasaXPS software ( Figure R1), from which we can see that the low-energy edges of them are overlapped and their peak positions fall into the range of 71.3 ±0.1 eV, which is within the allowed error range of XPS test. Thus, it is reasonable to describe that there is no obvious change for the signal of Pt 4f 7/2 during the in-situ reaction from 25 to 210 °C (Page 6 of the manuscript). For a clear observation, we replaced the previous chart in Figure S9 with the current Figure R1 (Page 10 of SI).  First of all, I am not questioning about the calibration procedure. The calibration procedure with Ag is a very well-known standard procedure and I do believe that authors follow the standard procedure properly. Nonetheless, whenever the XPS spectra are measured, Fermi level is also checked as an another internal reference. In metal, the Fermi level is the position of zero binding energy. I wonder if the authors have measured the Fermi level also.
Second, I don't agree with author's suggestion of using Figure R1 copied from CasaXPS software screen. In the Figure R1, authors made the reference point on binding energy of 73 eV and compared the highest peak positions of Pt 4f. However, this is not how XPS spectra should be compared. Any experienced XPS users will be able to tell that this procedure is not correct. The background are not matching in both high and low binding energy side. Of course, I understand the difficulty of matching background of XPS spectra obtained from nanoparticle system due to many complex XPS effect. For this reason, I prefer the previous figure that authors used for first revision. In previous plot, I assumed the authors applied the simple normalization on raw data. You can see more clear evolution of Pt 4f peak as a function of temperature in previous plot. (See the attachment.) Third, when I mentioned about the possible oxidation state of Pt 4f, I did NOT mean that the presence of PtO or PtO2 oxide. Pt is a noble metal, which does not form stable oxide at RT temperature. What I meant from previous comments was the presence of oxide peak, PtOx, under room temperature. If the sample is exposed to air for any brief time, the contamination of water from atmosphere can generate metastable Pt oxide. Considering the high surface areas of nanoparticles, the formation of surface oxide is certainly possible. So, I suspected the high background of Pt 4f as a sign of PtOx. Of course, as reviewer pointed out, the final state effect can interfere with this and generate high background. I will get back to this discussion later.
Fourth, I would like to discuss about the binding energy position of Pt 4f. The author claimed that the metallic state of Pt 4f in Pt (111)  To summary, I agree most of the authors' claims in response letter, except the pure metallic states of Pt based on XPS analysis in Figure R1. Author claimed that there is no alloy formation between Pt and CoNi alloys. However, the peak position at 71.3 eV at RT and 25 C and 50 C can be interpreted as either interfacial oxide or alloy formation. Several previous results, e.g. RSC Adv., 2016, 6, 71501-71506, NATURE COMMUNICATIONS | (2019)10:1743, Pt 4f peak moves to the higher binding direction as Pt forms nanoparticle alloys. Another possibility is that Pt nanoparticles are heavily influenced by interfacial graphene and also CoNi, resulting in broad metallic states, without alloy formation. I would like to see the solid proof of metallic state of Pt, other than XPS.  Author Reply: Thank you for your helpful comments. We didn't finely scan the valance band spectra during the in-situ NAP-XPS testings over our catalyst, but we measured the survey spectra from 0 to 1000 eV, from which we can roughly see that there is almost no significant change for the Fermi level at 0 eV during the testings ( Figure R1). Figure R1. Survey spectra from 0 to 20 eV during in-situ NAP-XPS testings over Pt/CoNi@NC Second, I don't agree with author's suggestion of using Figure R1 copied from CasaXPS software screen. In the Figure R1, authors made the reference point on binding energy of 73 eV and compared the highest peak positions of Pt 4f. However, this is not how XPS spectra should be compared. Any experienced XPS users will be able to tell that this procedure is not correct. Author Reply: Thank you for the detailed explanation. We adopted the previous figure as Figure   S9 in SI according to your suggestion. Author Reply: We totally agree with your assumption that the surface atoms of Pt NPs can be partially oxidized to PtO x when exposing to air. During our in-situ XPS testings, there can also in-situ form some metastable PtO x surface species in the presence of O 2 and CO. However, after the in-situ H 2 reduction (without exposing to air or reactant gases), the peak position of Pt 4f 7/2 is still at around 71.3 eV ( Figure S8 and Figure S9), indicating that the amount of the possibly formed PtO x species may not be sufficient enough to cause a significant shift of the binding energy of Pt 4f, and the Pt NPs should be mainly in the metallic state, at least for the bulk phase. Author Reply: Thank you for sharing these literatures with us. We noticed that 71.32 eV was found for pure Pt NPs (RSC Adv. 2016, 6, 71501-71506) and 71.2 eV was found for Pt in commercial Pt/C catalyst (Nat. Commun. 2019, 10, 1743. The latter paper also stated that the 71.2 eV for Pt/C is an indicative of Pt 0 , but their focused catalyst is not PtCo alloy but singleatom Pt on Co0.85Se alloy. For the PtNi alloy in the former paper, the Pt 4f peak moves to the lower binding direction (negative shift) when forming alloy. It seems that the shift direction of Pt 4f core level is metal-dependent, because Pt presented a positive shift in PtCo alloy as you suggested but had almost no shift in PtCu alloy (Surf. Interface Anal. 2000, 30, 475-478).
Nevertheless, PtCo or PtNi alloy will not be formed in our case, because Pt NPs (Chem. Mater. 2000, 12, 1622-1627 and CoNi@NC were separately prepared by different methods and then mixed together in ethanol at room temperature to obtain the Pt/CoNi@NC catalyst (details see Methods in manuscript). Thus, according to the two literatures you shared and more others (Nat.