ARTICLE

https://doi.org/10.1038/s41467-019-12876-1 OPEN

Retrofitting metal-organic frameworks

Christian Schneider 1,3, David Bodesheim1,3, Julian Keupp 2, Rochus Schmid 2 & Gregor Kieslich 1*

The post-synthetic installation of linker molecules between open-metal sites (OMSs) and
undercoordinated metal-nodes in a metal-organic framework (MOF) — retrofitting — has
recently been discovered as a powerful tool to manipulate macroscopic properties such as
1234567890():,;

the mechanical robustness and the thermal expansion behavior. So far, the choice of cross
linkers (CLs) that are used in retrofitting experiments is based on qualitative considerations.
Here, we present a low-cost computational framework that provides experimentalists with a
tool for evaluating various CLs for retrofitting a given MOF system with OMSs. After applying
our approach to the prototypical system CL@Cu3BTC2 (BTC = 1,3,5-benzentricarboxylate)
the methodology was expanded to NOTT-100 and NOTT-101 MOFs, identifying several
promising CLs for future CL@NOTT-100 and CL@NOTT-101 retrofitting experiments. The
developed model is easily adaptable to other MOFs with OMSs and is set-up to be used by
experimentalists, providing a guideline for the synthesis of new retrofitted MOFs with
modified physicochemical properties.

1 Department of Chemistry, Technical University of Munich, 85748 Garching, Germany. 2 Computational Materials Chemistry Group, Fakultät für

Chemie und Biochemie, Ruhr-Universität Bochum, Bochum, Germany. 3These authors contributed equally: Christian Schneider, David Bodesheim.
*email: [email protected]

NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications 1

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1

M
etal-organic frameworks (MOFs) combine the variety of robust towards high mechanical pressure21. In a similar fashion,
inorganic coordination chemistry with the large che- Su and co-workers installed linear dicarboxylate CLs in a flexible
mical parameter space of organic chemistry1–3. At the Zr-based MOF to modulate the breathing behavior of the MOF
heart of MOFs is their modular building block principle, which and the sorption properties towards N2 and CO223. In 2014, well
provides experimentalists with the control of structure and before ‘retrofitting’ was introduced as a concept for MOFs, A. A.
functionality through linker topicity, linker chemistry and metal- Talin et al. showed that the redox-active CL TCNQ (7,7,8,8-tet-
node symmetry4,5. In turn, MOFs show a variety of fundamen- racyanoquinodimethane) can bridge two OMSs in Cu3BTC2
tally interesting and technically relevant properties. For instance, (BTC = 1,3,5-benzenetricarboxylate, HKUST-1), which was
one of the recent breakthrough discoveries is the use of MOFs as observed to come with an increase in the electrical conductivity of
working media in highly efficient, non-toxic water recovery the material30. Using the same system, we recently demonstrated
technologies6,7. Other intriguing examples are the application of that retrofitting can be used as a tool to fine-tune the negative
defect-engineered MOFs as Lewis-acid catalysts8,9, the use of thermal expansion of Cu3BTC2, where an allover stiffening of the
MOFs (as precursor) for the synthesis of cathode materials in the material is mainly responsible for reduced negative thermal
oxygen evolution reaction10–12 and synthesis of electrically con- expansion behavior as a function of TCNQ incorporation22.
ducting systems with remaining guest-accessible porosities to Looking at some general experimental considerations, the CL for
name just a few13,14. In the pursuit to further optimize the a retrofitting experiment should be chosen to avoid post-synthetic
physicochemical properties of MOFs for certain applications, linker exchange17, i.e., the binding affinity of the metal-node with
post-synthetic modification (PSM) methods have proved to be an the CL should be weaker than with the linker. Additionally, when
important tool15–17. For instance, a porous but catalytic inactive trying to rationalize the underlying mechanism of CL installation,
MOF can be transformed into a robust heterogeneous catalyst by diffusion limitations are expected to play a role, since diffusion
post-synthetic immobilization of a catalytically active iron com- pathways through the pores are successively blocked with
plex at a functional group of the linker18. Likewise, breathing increasing CL installation. While longer reaction times and
behavior can be introduced in a formally non-flexible MOF relatively high temperatures can overcome such diffusion bar-
through post-synthetic functionalization of the linker mole- riers30,31, vapor phase infiltration at elevated temperatures with
cules19, nicely highlighting the opportunities that come with PSM CLs that exhibit low sublimation temperatures have proved most
methods. Moreover, it has been shown that the properties of acid- suitable in this context13,30,32. It is also interesting to note that
gas degraded zeolitic imidazolate frameworks can be recovered by current literature examples are limited to retrofitting with ditopic
post-synthetic treatment with a fresh linker solution20. Looking at CLs21–23,26,27,31, with bulkier tritopic ligands further increasing
PSM for MOFs from a more fundamental perspective, such concerns related to diffusion limitations. Lastly, it can be observed
methods are only possible due to the combination of the that retrofitting experiments mostly lead to a defective state of the
underlying coordination chemistry, the available functionality of CL@MOF system, with only a partial occupancy of the CL within
the organic back-bone and the guest-accessible porosity of MOFs, the parent MOF network21,22,31. In the big picture, this situation
representing a unique toolbox for experimentalists to fine-tune draws a clear line between the defect chemistry of MOFs and
physicochemical properties. purely inorganic materials, with MOFs exhibiting an additional
In the search for new PSM methods, Yaghi and co-workers degree of freedom for creating highly defective systems.
have recently introduced ‘retrofitting’ as a useful and intuitive Despite the large number of MOFs with OMSs or labile
categorization21. In the most general definition, retrofitting of a monotopic ligands, the examples of retrofitted MOFs are cur-
MOF describes the post-synthetic installation of additional lin- rently limited to only a few systems. To fully understand and
kers between undercoordinated metal nodes or between open unravel the synthetic opportunities that come with retrofitting as
metal sites (OMSs) of a MOF21. Subsequently, retrofitting has PSM, more (systematic) retrofitted MOF series are required.
been discovered as a powerful approach to manipulate macro- When designing a retrofitting experiment in the lab, immediately
scopic physicochemical properties such as the mechanical the question arises, “how can we assess whether a cross linker
robustness, the thermal expansion behavior and responsivity of (CL) is suitable for retrofitting of a given MOF?”. D. S. Sholl and
flexible MOFs to guest adsorption21–23. In a typical retrofitting co-workers approached this question by applying a DFT-based
experiment, a MOF with guest accessible OMSs or labile mono- screening study to identify Cu paddlewheel MOFs that feature
topic ligands is exposed to a molecule with at least two available OMSs for binding of TCNQ33. While this approach is expected to
coordination sites such as nitrile or carboxylate groups. This provide reliable results, a screening of different combinations of
molecule then bridges two OMSs or two undercoordinated metal MOFs and CLs comes at a high computational cost. Likewise, the
nodes, adding an additional connectivity between two different use of computational screening methods based on carefully
metal nodes. In the spirit of retrofitting which originally describes parameterized force-fields requires experience with these com-
the addition of new components to an existing system, e.g., in putational methods, typically limiting the scope for experi-
construction to reinforce the structural stability of historical mentalists for exploiting such screening approaches.
buildings24,25, we refer to these guest molecules as cross linkers Consequently, a robust and adaptive approach is required which
(CLs). It is important to note that in 2016 H. C. Zhou and co- is both easy to use and computationally not demanding. Addi-
workers demonstrated the possibility of installing two linear CLs tionally, as such an approach is most valuable for experi-
with different lengths at two different positions in PCN-700 and mentalists, the underlying methodology should reflect the
referred to this approach as sequential linker installation experimentalists’ chemical intuition and should be usable with a
(SLI);26,27 however, since retrofitting as a concept overarches minimum of experience with computational methods.
several different areas of on-going research and sci- In this work we present an easy-to-use low-cost computational
ence21,22,24,25,28,29, we strongly believe that retrofitting is the framework that allows for evaluating the applicability of a certain
more intuitive and general categorization and is used throughout CL for the use in a retrofitting experiment. We developed a
this manuscript. program named RetroFit that is based on an open-source code
Today, several intriguing examples are known where retro- and can guide experimentalists in the selection of suitable ditopic
fitting was used to manipulate the physicochemical properties of a CLs for their retrofitting experiments. The approach is inspired by
parent MOF. For example, O. Yaghi et al. used 4,4′-biphenyldi- research efforts in the field of molecular modeling in the context of
carboxylate as ditopic CL to render the Zr-based MOF-520 more drug design, where molecular docking of small molecules to

2 NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1 ARTICLE

macromolecular structures is predicted based on their geometry

and interaction potential34,35. Whilst our approach is generally
applicable to MOFs with OMS, we use Cu3BTC2 and a library of
20 CLs (16 different molecules, 4 of them allow for more than one
distinct binding mode) with nitrile groups as a test system
for RetroFit. After identifying suitable CLs, the practicability of
the RetroFit algorithm is shown by synthesizing the most pro-
MIP of test system MOF structure CL structure
mising CL@Cu3BTC2 systems, obtaining a set of three new ret- (DFT) (XRD) (DFT)
rofitted CL@Cu3BTC2 systems with CL = TCNB, DCNB, and
DCNT (TCNB = 1,2,4,5-tetracyanobenzol, DCNB = 1,2-dicyano- RMD, , , E RMM,  RDD, 
benzol and DCNT = 2,5-dicyanothiophene). Finally, we show that
RetroFit can easily be extended to other MOF systems such as
NOTT-100 and NOTT-101, where we identify several promising
CLs for the synthesis of retrofitted systems with the general for- RetroFit program
mula CL@NOTT-100 and CL@NOTT-101.

Results and discussion Position and energy

The RetroFit algorithm. Staying within the realm of ditopic CLs penalty ΔE for CL@MOF
to bridge two OMSs via retrofitting, and raising the question
which CLs are suitable for creating a CL@MOF system, two main
requirements for a CL can be identified: (i) the presence of two Fig. 1 General workflow of the RetroFit program. Three sets of input data,
suitable electron donating groups that can bind to the OMSs of i.e., the energies from single-point DFT calculations of a test system
the parent MOF and (ii) a size and geometry of the CL that brings (orange), the relative position of the OMSs in the MOF (green) and the
these groups into a favorable vicinity of the OMSs for bond relative position of the donor groups of the CL (blue), are used by the
formation. Whereas (i) can easily be fulfilled by choosing CLs RetroFit program (gray) to compute the optimal position of the CL within
with two nitrile, amino or carboxylate groups, (ii) can be the MOF and the corresponding energy penalty (light blue). A workflow
approached by analyzing the CL geometry compared to the with more technical details and a how-to guideline are provided as a part of
spatial arrangement of the OMSs. Combining both requirements the Supplementary data files
in a screening approach is expected to provide a good measure for
estimating if the synthesis of such a CL@MOF system can lead to MOF
a stable host–guest system. Based on these considerations, we
developed the RetroFit program code that allows for evaluating  RMM
the applicability of CLs for the use in retrofitting experiments for M
RMD
M
a given MOF with two OMSs. RetroFit is designed to avoid
RDD
expensive calculations and to circumvent time-consuming trial  D D
and error experiments, in total facilitating the discovery of new 
retrofitted CL@MOF systems. At the core of RetroFit is a mini- CL
mization routine which optimizes the position of a given CL
relative to the OMS positions. The spatial minimization is based
on the binding energy of the CL to the OMSs as reflected in the
model interaction potential (MIP). The MIP itself is used as input 
for RetroFit and is obtained via single point DFT calculations of a
simple test system. A schematic of the required input and output
of RetroFit is shown in Fig. 1, and a detailed workflow including a Fig. 2 Geometric parameters of a general CL@MOF system. The
how-to guide is provided in the supplementary data files. Over- parameters RMM and γ are determined by the spatial arrangements of the
viewing the workflow, the required input is (i) the energies of CL OMSs of the parent MOF (green), while RDD and α are given by the CL
coordination to the specific OMS as a function of bond lengths (blue). θ is computed from α and γ, whereas the parameters RMD and δ
and angles (i.e., the MIP) of the test system which is used in the (orange) are variables and optimized with respect to the underlying MIP.
minimization process, (ii) the structure of the MOF and in par- The two OMSs are related to each other by mirror symmetry
ticular the spatial orientation of the OMSs represented in the
parameters RMM and γ, and (iii) structural information about
the CL that is tested which determines RDD and α. The relative angle as provided by the MIP. The application of RetroFit to
position of the two OMSs and the CL is defined by RMD and various CLs allows for ranking these after their energy penalties
the two angles δ and θ, as illustrated in Fig. 2. Since α, γ, and θ are and further allows for grouping them into three categories (good
interdependent (see Supplementary Note 1), the geometry of the fit, medium fit, poor fit – see Supplementary Note 7). This list can
entire host guest system can be defined by a set of six parameters, guide experimentalists in the identification and prioritization for
i.e., (RMM, RDD, RMD, α, γ, and δ) or (RMM, RDD, RMD, α, θ, and the synthesis of promising CL@MOF systems. Furthermore,
δ), or (RMM, RDD, RMD, α, γ, and θ). The parameters RMM, γ, RDD RetroFit allows to identify favorable spatial arrangements of the
and α are extracted from the structure of the MOF and CL and two electron donating groups (e.g., cis, ortho, etc.) from which
treated as constants in our model, θ only depends on α and γ and further CL structures can be developed.
can be directly computed, whereas RMD, δ are variables that are
optimized during the RetroFit routine. Running through the
RetroFit program for one given CL@MOF combination returns RetroFit for CL@Cu3BTC2. After introducing the general
an energy penalty ΔE which compares the best orientation of the working principle of RetroFit, we applied RetroFit to the well-
CL within the MOF system to the ideal binding distance and established Cu3BTC2 system. Cu3BTC2 was previously used in

NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications 3

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1

retrofitting experiments to obtain TCNQ@Cu3BTC2 and distances fit comparably well such as TCNQ (13, RNN = 4.447 Å,
TCNE@Cu3BTC213,22,32. Notably, TCNQ and TCNE as proto- ΔE = 2.24 kcal mol−1), TCNE (2, RNN = 4.436 Å, ΔE = 2.29 kcal
typical CLs exhibit four nitrile groups that are available for the mol−1) and malononitrile (10, RNN = 4.431 Å, ΔE = 2.32 kcal
coordination to OMSs; however, due to the spatial orientation of mol−1). Looking for trends, nitriles in ortho positions of small
the OMSs in Cu3BTC2 and size dependent restrictions given by cyclic systems as given in tetracyanobutadiene (16) as well as
the CL and MOF combination, only two nitrile groups coordinate geminal nitriles as available in TCNQ show a good fit with
to the MOF while the remaining two point to the center of the Cu3BTC2, while ΔE increases for larger ring systems and nitrile
pore. Therefore, tetranitrile molecules such as TCNQ or TCNE groups in cis position. In contrast, the meta configuration in CLs
are treated as bidentate CLs with two possible configurations with 6-membered rings (20, RNN = 6.893 Å for TCNB) shows a
within RetroFit. Looking at Cu3BTC2 in more detail, Cu3BTC2 relatively high energy penalty of ΔE = 4.815 kcal mol−1 which is
crystallizes in the tbo topology and exhibits two types of large therefore less favorable for retrofitting Cu3BTC2 (see discussion
pores that in principle are accessible for guest molecules. In only in Supplementary Note 7). TCNQ as the prototype CL has a
one of these pores the Cu paddlewheels are oriented with the relatively low value of ΔE = 2.24 kcal mol−1, which is in good
OMS pointing to the center of the pore to allow for retrofitting. In agreement with the results of previous studies30,42,43. Comparing
this pore, 12 OMSs on the axial positions of crystallographically TCNQ to other tested CLs, only CLs based on 3-, and 4-
equivalent Cu paddlewheel metal nodes point to the inside of the membered ring systems show smaller ΔE values. Such molecules
pore. The chemistry of such a Cu OMS shows a large propensity often show flashpoints below 150 °C and are therefore less
for the formation of Cu-N bonds, and in turn CLs with nitrile or suitable for retrofitting experiments which involves vapor phase
amino groups are expected to be suitable CL candidates36–38. infiltration at temperatures typically above 100 °C. In order to
Following the workflow in Fig. 1, a MIP for the system is required evaluate the accuracy of the RetroFit routine, the results are
which is the basis for the calculation of the aforementioned compared with existing computational data from the literature on
energy penalty ΔE for a given CL@MOF system. For Cu3BTC2 TCNQ@Cu3BTC2. A. A. Talin et al. modeled the binding of
the retrofitting experiment is meant to bridge two of the 12 TCNQ in Cu3BTC2 by using DFT (UB3LYP41,44/VTZP level of
available copper OMS which are related to each other by a mirror theory45) and obtained a geometry optimized crystal structure of
plane. Therefore, it is suitable to focus on the MIP of one Cu CL@Cu3BTC2 with CL = TCNQ30. Using their results as a
OMS and subsequently using the mirror symmetry to obtain the benchmark and comparing these with the output values from
interaction potential of the whole system (see Fig. 2). In Cu3BTC2 RetroFit, we indeed observe a good correlation which is
the distance between the Cu OMSs (RCuCu) was extracted from summarized in Supplementary Table 5. The largest deviations
crystallographic data39. The MIP input was obtained by single occur due to CL deformations, which are not accounted for in our
point DFT calculations (B3LYP/TZVPP level of theory, see model. For instance, RNN of TCNQ in the literature structure of
Methods section for details) for a simple Cu(II)formate – acet- TCNQ@Cu3BTC2 is increased by 0.167 Å to RNN’ = 4.614 Å
onitrile complex (CH3CN@Cu2(HCOO)4), where the metal to compared to the free molecule, which results in a RCuN distance
nitrile group distance (RCuN), δ and θ of acetonitrile were varied. which is 0.168 Å shorter and in turn leads to a stronger
A subsequent interpolation of the 3 N dimensional data yielded coordination bond. Taking the distortion of TCNQ into account
the continuous MIP surface that represents one side of the CL- by using RNN‘ and α‘ from the literature as input for RetroFit, a
OMS interaction within the MOF, see Fig. 1 orange box for a significantly lower energy penalty of ΔE = 1.59 kcal mol−1 is
representation of the MIP with fixed θ angle and Supplementary obtained (see Supplementary Table 5). This example shows that
Note 6 for an extensive error analysis of the interpolation. In despite the heuristic approach of RetroFit, the binding situation is
CH3CN@Cu2(HCOO)4 the global energy minimum is reached reasonably represented. It seems, however, that RetroFit slightly
for a linear coordination of acetonitrile (δ = θ = 180°) and a overestimates the energy penalty, whereas in the real
distance of RCuN = 2.2 Å which we define as zero, i.e., ΔE = 0. In CL@Cu3BTC2 system structural deformations can occur that
the following, all energies for the CL@Cu3BTC2 system are further minimize the total energy of the CL@MOF system, i.e., a
expressed as energy difference ΔE from this ideal configuration. slight increase in energy due to unfavorable molecule deformation
Lastly, the geometries of the different CLs are required, which can be overcome through the formation of stronger coordination
were obtained by DFT optimization using the B3LYP hybrid bonds. Therefore, CLs with low energy penalties are expected to
functional and a 6–31 G basis set40,41, see Methods section for lead to a stable CL@Cu3BTC2 system and structural deformations
details. The input of 20 dinitrile-CLs as *.xyz-files are given in can be expected to only play a minor role. Similarly,
the supplementary data files. RetroFit applied to one CL@Cu3BTC2 systems with relatively large energy penalties can
CL@Cu3BTC2 system varies the spatial orientation of the CL with be ruled out, whilst assumptions drawn for CL@MOF combina-
respect to the OMSs within the MOF to minimize ΔE through the tions with intermediate energy penalties are more inconclusive.
parameters RCuN, δ and θ. In turn, the application of RetroFit to For our CL@Cu3BTC2 system, we observe that CL = TCNB (3),
various CLs allows to rank these after their energy penalties. DCNB (5) and DCNT (6) are promising candidates. These
Based on this methodology, we have applied RetroFit to the 20 combine low to medium ΔE values with experimental applic-
different CLs32 which are shown in Fig. 3. ability given by the physical properties of the CLs, and the
To cover a broad range of possible CLs with various geometries synthesis of the corresponding CL@Cu3BTC2 systems is
we tested acyclic (e.g., 7, and 10) and (hetero-)cyclic systems with described in the next paragraph.
3-, 4-, 5-, and 6-membered rings (e.g., 18, 15, 9, and 5), where the
nitrile groups are in geminal, cis (e.g., 1), ortho (e.g., 3) or meta
(e.g., 19) positions, see Fig. 3. By using RetroFit, ΔE for the 20 Proof of concept – experimental validation. After having iden-
tested CL@Cu3BTC2 combinations were calculated and are given tified several promising CLs for retrofitting Cu3BTC2, the synth-
in the energy map shown in Fig. 4 and the bar plot (see below). esis of new CL@Cu3BTC2 materials was attempted. Following the
The global minimum of the energy map ΔEmin, i.e., the molecule results discussed in the previous section, we selected 1,2-dicya-
parameters that suggest the ideal fit, is located at RNN = 5.81 Å nobenzene (5, DCNB), 1,2,4,5-tetracyanobenzene (3, TCNB) and
and α = 60°. Interestingly, the dependence of ΔE on the distance 3,4-dicyanothiophene (6, DCNT) as promising CLs for proof-of-
between the nitrile groups within the CL (RNN) is significantly principle experiments. These CLs show low and intermediate ΔE
stronger than on α. Therefore, molecules with similar RNN values, are commercially available and easy to handle. All of these

4 NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1 ARTICLE

N N N N
N N N N N N
N N
N N
N N

N N S
N N N N
1 2 3 4 5 6 7

N N N N N N N N N N
N N
N N

N N N N O
8 9 10 11 12 13 14
N N N N
N N N N
N N N N N

N
N N N N
N N N N
15 16 17 18 19 20

Fig. 3 Library of potential CLs evaluated with RetroFit. Structural parameters were obtained from geometry optimized structures in the gas phase. The
coordinates can be found in the program package provided in the Supplementary data files

a 180 8 b

7
160
6
140
5
ΔE /kcal mol–1

13
18
19
/°

120 11
12
4
20
10
3
100
17
15
2
16
80 1
7 14 8.9 1
2 6
5
4 3
60
2 3 4 5 6 7 8
RNN/Å

Fig. 4 RetroFit results for Cu3BTC2. a RetroFit map for Cu3BTC2 and the tested dinitrile-CLs. The energy penalty ΔE for the CL parameters RNN and α is
given as color code and increases from blue to red. Values of ΔE, which exceed the color scale bar are set to 8 kcal mol−1 for better visualization. The
smallest ΔE and in turn the best fit is obtained for CLs with 3 and 4-membered rings (15, 16, 17 and 18). A table overviewing the results is given in
Supplementary Note 7 (Supplementary Table 3). b Visualization of the fit of the CLs 5, 12, and 20 with ortho, geminal and meta configuration, respectively.
Cu, C, N, and O atoms are shown in dark blue, black, blue and red. The potential around the OMSs is illustrated as ΔE increasing from blue to red

molecules show sublimation temperatures feasible for vapor phase suggests a primitive space-group, such as Pn-3m, pointing at some
loading experiments that currently seem to be the preferable sort of ordering of the guest molecules within the two chemically
experimental pathway13,22. Following the established solvent-free different pores of Cu3BTC2. Notably, other symmetry-reduction
guest loading procedure13, we prepared CL@Cu3BTC2 com- pathways are possible, but when staying within the 3σ criterion,
pounds with one molar equivalent of DCNB, TCNB and DCNT the Pawley fit with Pn-3m produces the most reasonable fit in all
per Cu3BTC2. Such a loading situation corresponds to two CLs per cases. Attempts of structure solution are challenged by the
large pore, which has been previously determined as the max- expected partial occupancy and complicated ordering mechanism
imum loading for TCNQ13,46. Details on the synthetic procedure of CL molecules within the pores and have not been successful so
and characterization techniques are provided in the Methods far. Additional proof for the presence of the CLs are new signals in
section (see below) and in the references 13 and 22. After infil- the IR spectra that are related to the guest molecules, even though
tration of Cu3BTC2 with the various CLs, powder X-ray diffraction the nitrile bands are less intense when compared to
(PXRD) confirms that the crystallinity of all samples is maintained TCNQ@Cu3BTC2 (Supplementary Note 10). To further underpin
(Fig. 5). An increased intensity of the (111) reflection is observed the accommodation of the CLs inside Cu3BTC2, nitrogen sorption
in all cases, which has previously been attributed to the chemi- isotherms (Fig. 6) and scanning electron microscopy images were
sorption of guest molecules to two neighboring Cu paddlewheels recorded (Supplementary Note 11). From the initial slope of the
within the (111) plane13. Closer investigation of the PXRD pattern type I isotherms comparable BET surface areas of 1170.3 m2 g-1
reveals new reflections which are weak in intensity, see inset in (DCNT), 1151.9 m2 g-1 (DCNB) and 1135.2 m2 g-1 (TCNB) were
Fig. 5. These new reflections are forbidden in the face-centered determined. These values are significantly lower than for pristine
cubic space group of Cu3BTC2 (Fm-3m) and have been previously Cu3BTC2 (1873.8 m2 g-1) but higher than for TCNQ@Cu3BTC2
observed for TCNQ@Cu3BTC213. Indexing of the PXRD pattern (869.9 m2 g-1), which is ascribed to the smaller size of DCNT,

NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications 5

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1

DCNB and TCNB compared to TCNQ. For all new CL@Cu3BTC2 and TCNB@Cu3BTC2, all materials are electrical insulators (see
powders, SEM images confirm the absence of unreacted CLs or Supplementary Note 12). This might be a result of different redox
the formation of amorphous byproducts (see Supplementary potentials of the CLs and the presence of the by-phase in
Note 11). This is different when compared to the case of TCNQ@Cu3BTC2 (a more detailed discussion can be found in
TCNQ@Cu3BTC2, where the formation of CuTCNQ nano-wires Supplementary Note 12). Lastly, we would like to note that tet-
on the surface of the MOF crystals was observed13. Comparing the racyanoethylene (TCNE), which shows an excellent fit according
electrical conductivities of DNCT@Cu3BTC2, DCNB@Cu3BTC2 to RetroFit, was already employed by D’Alessandro and cow-
orkers;32 however, our attempts of the synthesis of TCNE@-
Cu3BTC2 led to a unexpected PXRD pattern that might either
point to a transformation or a relatively drastic change of structure
a
which we could not yet identify (see Supplementary Fig. 19). In
(111)

TCNB@Cu3BTC2 P n3m
summary, following the results from RetroFit, we successfully
Counts/a.u.

rwp = 3.81
synthesized and characterized three new retrofitted CL@Cu3BTC2
systems namely DCNT@Cu3BTC2, DCNB@Cu3BTC2 and
TCNB@Cu3BTC2, highlighting the applicability of the RetroFit
algorithm.
b
DCNB@Cu3BTC2 P n3m
Beyond Cu3BTC2. To show that RetroFit is easily transferable to
Counts/a.u.

rwp = 3.51
other CL@MOF systems, we used RetroFit to screen various
dinitrile-CLs for retrofitting NOTT-100 (Cu2(BPTC), BPTC =
biphenyl-3,3′,5,5′-tetracarboxylate) and NOTT-101 (Cu2(TPTC),
TPTC = [1,1′:4′,1”]terphenyl-3,3”,5,5”-tetracarboxylate) MOFs47.
c Both MOFs crystallize in the nbo topology and are built from Cu
DCNT@Cu3BTC2 P n3m paddlewheel nodes and tetratopic rectangular-shaped linkers.
Counts/a.u.

rwp = 3.73 Since NOTT-100 and NOTT-101 have Cu-based OMSs similarly
to Cu3BTC2, the same MIP can be used as input for RetroFit. In
contrast to Cu3BTC2, the crystal structures of both NOTT sys-
tems exhibit two crystallographically distinct binding sites. We
refer to these sites and corresponding RCuCu and γ values as 3,5-
d position and 3,3′- (NOTT-100) or 3,3”-positions (NOTT-101)
Cu3BTC2 Fm3m which is adapted from the nomenclature of the linker molecules
Counts/a.u.

rwp = 5.35
(see Fig. 7 for a visualization of this situation). In turn, we can
generate two sets of ΔE values, which rank the different CLs after
the two different binding situations. The obtained ΔE values are
listed in Fig. 8 and compared with Cu3BTC2. For the binding site
in the 3,5-position, the RetroFit maps (see Supplementary Note 8)
10 20 30 10 12 14 16
2/° 2/°
resemble the situation previously obtained for Cu3BTC2 (Fig. 4)
due to the similar geometry; however, significant differences can
Fig. 5 PXRD data and Pawley profile fits for CL@Cu3BTC2. Pristine Cu3BTC2 be observed for the 3,3′ and 3,3” binding site. It should be noted
d crystallizes in the face-centered cubic space group Fm-3m. Upon retrofitting that in both NOTT systems, the Cu OMSs in 3,3′ (or 3,3”)
with CL = TCNB a, DCNB b, or DCNT c, new reflections occur (zoom in) position exhibit a small torsion angle. In RetroFit, this torsion
which can be accounted for by using the primitive space-group Pn-3m. angle is neglected for all CLs as the routine translates a 3D
Experimental data, Pawley profile fits and difference curves are shown in blue, multiparameter problem into 2D space. Presumably, this adds
red, and black, respectively. The rwp fitting parameters are provided in the top another offset to the ΔE ranking; however, as a pronounced effect
right corner, respectively. The (111) reflection at 5.8° is indicated in gray of the RNN distance has been found for Cu3BTC2, the torsion
angle is expected to play only a minor role. The results of

a 500 b
1750
1873.8

400
BET surface area/m2 g–1

1500
Vads/cm3 g–1 STP

1250
300
1170.3

1151.9

1000
1135.2

200
750
865.9

Cu3BTC2
1.0DCNT@Cu3BTC2
1.0DCNB@Cu3BTC2
500
100
1.0TCNB@Cu3BTC2
250
1.0TCNQ@Cu3BTC2
0
0
0.0 0.2 0.4 0.6 0.8 1.0 Cu3BTC2 DCNT@ DCNB@ TCNB@ TCNQ@
P /P0 Cu3BTC2 Cu3BTC2 Cu3BTC2 Cu3BTC2

Fig. 6 Porosimetry measurements of CL@Cu3BTC2. Nitrogen adsorption isotherms a and calculated BET surface areas b for samples with CL = DCNT
(orange), DCNB (green), TCNB (red), and pristine Cu3BTC2 (blue). All new materials show a high residual porosity after the retrofitting experiment

6 NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1 ARTICLE

a 180 8 b
7
160
6
140

ΔE / kcal mol–1
5
13
18
/° 120 11 19
4
12
20
10
3
100 17
15
2
16
80 1
2 7 14 8.9 1
4 6
5
3
60
4 5 6 7 8 9
RNN/Å

Fig. 7 RetroFit results for NOTT-101. a RetroFit map for the 3,3″-position and the CL library (Fig. 3). Values of ΔE, which exceed the color scale bar are set
to 8 kcal mol−1 for better visualization. The lowest ΔE for the 3,3″-position is obtained for CL 20 (TCNB, meta position), although CLs with even larger RNN
distances will show better fit. b Visualization of the implications for retrofitting NOTT-101. It is conceivable that two different CLs can be installed in NOTT-
101, e.g., TCNB in the 3,3″-position and TCNE in the 3,5-position. Cu, C, N, and O atoms are shown in dark blue, black, blue, and red. The potential around
the OMSs is illustrated as ΔE increasing from blue to red

7.0
Cu3BTC2 NOTT-100 3,3′-pos.
15.0
NOTT-100 3,5-pos. NOTT-101 3,3″-pos.
6.5
NOTT-101 3,5-pos.
12.5
6.0
ΔE/kcal mol–1

10.0

RNN/Å
5.5
7.5

5.0
5.0

4.5
2.5

0.0 4.0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Cross linker #

Fig. 8 Summary of the RetroFit results. The energy penalties ΔE (bars) of all CL@MOF materials studied in this work are compared. The RNN distance of the
CLs is visualized by the black line. Due to the anisotropy of the linker for the NOTT systems, two different positions, here referred to as 3,5 and 3,3′ (or
3,3”) are possible and have been tested. For bridging two OMSs in a MOF, distinct RNN windows that lead to low ΔE can be identified. Cu3BTC2, NOTT-100
(3,5-position) and NOTT-101 (3,5-position) have very similar ΔE values with an optimum at RNN ≈ 5.1 Å. NOTT-100 (3,3′-position) shows the best fit for
low RNN values, whereas NOTT-101 (3,3”-position) shows the opposite trend

applying RetroFit to NOTT-100 and NOTT-101 are shown in proof-of-principle study, we used RetroFit for screening various
Fig. 8 and compared to Cu3BTC2. In the 3,3′-position of NOTT- CLs for retrofitting experiments of the iconic Cu3BTC2 system.
100, the short RCuCu distance of 7.601 Å in combination with the The synthesis of the most promising CL@Cu3BTC2 candidates
obtuse angle of γ = 145.65° requires molecules with very short was approached in the laboratory, obtaining the three new
RNN distances (see Supplementary Fig. 16). The isoreticular retrofitted systems DNCT@Cu3BTC2, DCNB@Cu3BTC2 and
expansion from NOTT-100 to NOTT-101 results in an increase TCNB@Cu3BTC2. To show the applicability to other systems,
of RCuCu for the 3,3′ (respectively 3,3”) position to 11.607 Å. we applied RetroFit to the well-known NOTT-100 and NOTT-
Therefore, CLs with larger RNN distances, such as TCNB (20, 101 MOFs, testing both possible bidentate bridging modes in
meta), can be accommodated in the structure of NOTT-101 these systems and identifying a few promising CL@NOTT-101
which do not fit into the structure of NOTT-100 or Cu3BTC2. and CL@NOTT-100 candidates. We would like to note that
These results suggest that a system with two distinctly different RetroFit is not aiming at a quantitative description of the
sites such as NOTT-101 can be fitted with two types of CLs, e.g., possible CL@MOF systems but is designed as an easy-to-use
TCNE (12) or malononitrile (10) in the 3,5-position and a TCNB screening tool which can be operated without a profound
(20, meta) across the long side of the terphenyl-linker (3,3”- background in computational chemistry. A detailed how-to
position), providing diffusion limitations can be minimized. guide including the program code and input data for
Notably, TCNB is a candidate CL to bridge both the 3,5-position reproducing the herein presented results are given in the sup-
(ortho coordination) and the 3,3” position (meta coordination), plementary data files. Looking at RetroFit from a more general
making it a particular interesting CL for future retrofitting perspective, the routine translates a complex multiparameter
experiments. 3D optimization problem to a 2D model in which only the
In conclusion we developed a low-cost computational strongest chemical interactions, i.e., the metal-coordination
framework called RetroFit which provides experimentalists bonds are accounted for. With the currently available MIP for
with a tool to identify promising CLs for retrofitting MOFs. In a nitrile-CLs and Cu-based OMS, RetroFit can easily be applied

NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications 7

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1

to all MOFs with Cu-paddlewheel motifs as OMS, such as NU- to the xyz-format using Open Babel (version 2.3.2) and then imported into the
111 (Cu3(5,5′,5”-(benzene-1,3,5-triyltris(buta-1,3-diyne-4,1- RetroFit tool using the Atomic Simulation Environment (ASE) (version 3.16.0) to
compute RNN and α.
diyl))triisophthalate))48, PCN-14 (Cu2(5,5′-(anthracene-9,10- To access energies in the host–guest system consisting of a Cu paddlewheel-
diyl)diisophthalate))49 and NOTT-115 (Cu3(4′,4”‘,4”“‘-nitrilo- based MOF and a nitrile-CL, we optimized a Cu(II) formate paddlewheel and an
tris(([1”“,1”“‘-biphenyl]-3,5-dicarboxylate))))50 to name just a acetonitrile molecule, respectively, and then arranged the two entities that the
few. The next step in the development of RetroFit is the creation nitrile group points towards the OMS of the paddlewheel (CH3CN@Cu2(HCOO)4).
of a library of various MIPs for different metal nodes going By varying RCuN, δ and θ according to the parameters given in Supplementary
Table 2, we obtained an energy of the system for every combination of the three
beyond Cu-based OMS. Likewise, currently 20 dinitrile-based parameters, which allows the generation of a MIP via linear interpolation. In
CLs are contained in the database, which can be expanded to general, the interpolation errors can be improved by a smaller step size but for the
amino-based or carboxylate-based CLs in the future. accuracy demands herein, the chosen step sizes are sufficient (see Supplementary
Finally, it is important to emphasize a few additional points Note 6). The configuration with the lowest energy is defined as 0 kcal mol−1 and all
energies are given as energy differences ΔE. Further, a factor of 2 is applied as the
related to the retrofitting concept. Retrofitting as relatively new host–guest complex involves two coordination bonds of nitrile groups to two
categorization of PSM is an intriguing approach for accessing the OMSs. The single point calculations were done on a DFT level of theory using the
structural dynamics of MOFs. Today only a limited number of TURBOMOLE (V7.1) software package51. The hybrid functional B3LYP41,44 was
retrofitted MOFs exist, a situation that makes it difficult to oversee used with a TZVPP basis set45 and a fine ‘m5’ grid52 for all elements. The
multipole-accelerated53 resolution of the identity approximation54,55 was used for
the full potential of the concept. Most notably, retrofitting as science performance reasons. Grimmes D356 was employed to properly account for
overarching concept is not limited to MOFs. Even in the world of dispersive interactions. The SCF convergence criterion was set to 10-6 Hartree.
covalent organic frameworks (COFs) examples exist that can be Additional details on the single point DFT calculations can be found in
classified under the umbrella of retrofitting, i.e., when metal Supplementary Note 2.
complexes are used to cross-link two-dimensional COF sheets or
when post-synthetic modification of functional groups leads to Synthesis. Cu3BTC2 was synthesized following the literature procedure57. The
crystalline as-synthesized powder was solvent exchanged and activated following
additional framework connectivity28,29. The future progress of the our previously published protocols13. The synthesis of CL@Cu3BTC2 with CL =
retrofitting concept is linked to the discovery of additional 1,2-dicyanobenzene (DCNB), 1,2,4,5-tetracyanobenzene (TCNB), and 3,4-dicya-
CL@MOF systems, which allows to answer open scientific questions nothiophene (DCNT) was performed analogous to the reported procedure for
on potential order phenomena of CLs within the parent MOFs as TCNQ@Cu3BTC213. Therefore, activated Cu3BTC2 (100 mg) and stoichiometric
amounts of DCNB, TCNB, or DCNT were thoroughly mixed to yield mixtures
well as on the influence of CL size and CL-to-OMS bond strength
with x = n(CL)/n(Cu3BTC2) = 1.0. The mixtures were filled into a glass ampule,
on the macroscopic properties. With the synthesis of more which then was evacuated (10−3 mbar) and flame-sealed. The ampules were placed
retrofitted MOF systems, these questions can be approached, to in a convection oven at 180 °C for 72 h. After cooling down, the ampules were
which our RetroFit algorithm adds a powerful tool. Likewise, the opened inside an Ar-filled glovebox and the CL@Cu3BTC2 powders were stored
concept itself can be extended. For instance, the incorporation of until further characterization. For all analytic methods, the powders were handled
under inert conditions (Ar or vacuum) to avoid contamination from the
asymmetric CLs such as amino acids, or CLs that contain functional atmosphere.
backbones such as optically active groups seem to be intriguing
research directions. In these scenarios, diffusion limitations might Powder X-ray diffraction. CL@Cu3BTC2 samples were filled into 0.7 mm bor-
start to play a more significant role but are difficult to estimate osilicate capillaries and sealed. The capillary was mounted onto a PANalytical
without experimental validation. The use of less strong donor CLs Empyrean X-ray diffractometer operated in capillary mode using Cu Kα radiation,
might be a solution to this problem, enabling the equilibrium to be a focusing beam mirror with a 1/8° slit and 0.02 rad Soller slits as the incident beam
optics and a 1/8° anti-scatter slit with 0.02 rad Soller slits and a Ni filter on the
reached at elevated temperatures. Therefore, we strongly believe the diffracted beam side. Diffraction data in the 2θ range of 5–50° with a step size of
retrofitting concept has much to offer in the future, and that 0.013° was collected using a PIXcel 1D detector in scanning line mode. Quanti-
RetroFit can facilitate the discovery of new CL@MOFs, leaving the tative data analysis (Pawley fits) was performed using the Topas Pro v5 software.
territory of trial and error experiments. The fitting parameter, i.e., weighted-profile R-factor (rwp), is provided in the
respective figures.

Methods Porosimetry measurements. Inside of an Ar glovebox, ~60 mg of a sample were

RetroFit algorithm. RetroFit was written in the open source programming lan-
filled into a glass tube and evacuated at room temperature for 3 h at ~10−5 mbar.
guage PythonTM (available at http://www.python.org) and was tested for versions
The exact sample mass was determined and the isotherm was recorded on a
2.7 and 3.7. The algorithm uses three data sets (see Fig. 1) as input, i.e., the relative
Micromeritics 3flex at 77 K in the pressure range between 10−3 and 103 mbar. The
position of the OMSs within the MOF, the structural information of the CL, and a
BET surface area was calculated from the initial slope (0.01 to 0.1 P/P0) of the
MIP representing the interaction between OMS and CL. Whereas the former is
isotherm.
extracted from the MOF crystal structure, the geometry of the CL and the MIP are
obtained by DFT calculations (see below). In the current version of RetroFit, the
structural relation of the OMSs are extracted manually from the MOF crystal Data availability
structure, whereas the geometry of the different CLs is provided as a library of The authors declare that all data supporting the findings of this study are available within
molecular structure files (xyz-files). The continuous MIP is generated by inter- the article and its Supplementary Information or from the corresponding author upon
polation of the input energies from single point DFT calculations of the model reasonable request.
system. A detailed description of the algorithm is provided in Supplementary
Note 3 and the workflow is shown in Supplementary Fig. 10. It should be noted
that herein we release version 1.0 of RetroFit. Future developments of the code as Code availability
well as a library of CLs and MIPs will be deposited on GitHub (https://github.com/ The RetroFit code including all required input files and a How-to-guide is freely available
GKieslich/RetroFit). We believe that the development of RetroFit is a task for on GitHub, see https://github.com/GKieslich/RetroFit and licensed under a MIT license.
computational scientists or users with advanced expertise on computation. On the
other hand, RetroFit is designed to be applied by experimentalists with a minimum
of experience with computational methods. Therefore, future developments of the
Received: 23 July 2019; Accepted: 17 September 2019;
program will be aiming on both to increase the capabilities of the program and to
make RetroFit as user-friendly as possible.

DFT calculations. All molecule geometries were optimized with the Gaussian09
program package. Optimization was performed with DFT with a B3LYP hybrid
functional and a 6–31 G basis set40,41. For the optimization the ‘tight’ convergence References
criterion was used and the Hessian was recalculated after each optimization step. 1. Yaghi, O. M. et al. Reticular synthesis and the design of new materials. Nature
All molecules were symmetry restricted during the optimization process (CLs 2, 3, 423, 705–714 (2003).
4, 12, 13, 16, 17, 19, and 20 were restricted to D2h symmetry while CLs 1, 5–11, 14, 2. Kitagawa, S., Kitaura, R. & Noro, S.-i Functional porous coordination
15, and 18 were restricted to C2v). The resulting Gaussian output file was converted polymers. Angew. Chem. Int. Ed. 43, 2334–2375 (2004).

8 NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1 ARTICLE

3. Lu, W. et al. Tuning the structure and function of metal–organic frameworks 31. Lollar, C. T. et al. Thermodynamically controlled linker installation in flexible
via linker design. Chem. Soc. Rev. 43, 5561–5593 (2014). zirconium metal–organic frameworks. Cryst. Growth Des. 19, 2069–2073
4. Tranchemontagne, D. J., Mendoza-Cortes, J. L., O’Keeffe, M. & Yaghi, O. M. (2019).
Secondary building units, nets and bonding in the chemistry of metal-organic 32. Usov, P. M. et al. Guest–Host Complexes of TCNQ and TCNE with
frameworks. Chem. Soc. Rev. 38, 1257–1283 (2009). Cu3(1,3,5-benzenetricarboxylate)2. J. Phys. Chem. C 121, 26330–26339
5. Guillerm, V. et al. A supermolecular building approach for the design and (2017).
construction of metal–organic frameworks. Chem. Soc. Rev. 43, 6141–6172 33. Nie, X., Kulkarni, A. & Sholl, D. S. Computational prediction of metal organic
(2014). frameworks suitable for molecular infiltration as a route to development of
6. Kim, H. et al. Water harvesting from air with metal-organic frameworks conductive materials. J. Phys. Chem. Lett. 6, 1586–1591 (2015).
powered by natural sunlight. Science 356, 430–434 (2017). 34. Kitchen, D. B., Decornez, H., Furr, J. R. & Bajorath, J. Docking and scoring in
7. Kim, H. et al. Adsorption-based atmospheric water harvesting device for arid virtual screening for drug discovery: methods and applications. Nat. Rev. Drug
climates. Nat. Commun. 9, 1191 (2018). Discov. 3, 935–949 (2004).
8. Liu, J. et al. Applications of metal-organic frameworks in heterogeneous 35. Jorgensen, W. Rusting of the lock and key model for protein-ligand binding.
supramolecular catalysis. Chem. Soc. Rev. 43, 6011–6061 (2014). Science 254, 954–955 (1991).
9. Hall, J. N. & Bollini, P. Structure, characterization, and catalytic properties of 36. Prestipino, C. et al. Local structure of framework Cu (II) in HKUST-1
open-metal sites in metal-organic frameworks. React. Chem. Eng. 4, 207–222 metallorganic framework: spectroscopic characterization upon activation and
(2019). interaction with adsorbates. Chem. Mater. 18, 1337–1346 (2006).
10. Miner, E. M. & Dincă, M. Metal-organic frameworks: evolved oxygen 37. Bordiga, S. et al. Adsorption properties of HKUST-1 toward hydrogen and
evolution catalysts. Nat. Energy 1, 16186 (2016). other small molecules monitored by IR. Phys. Chem. Chem. Phys. 9,
11. Xia, B. Y. et al. A metal–organic framework-derived bifunctional oxygen 2676–2685 (2007).
electrocatalyst. Nat. Energy 1, 15006 (2016). 38. Borfecchia, E. et al. Insights into adsorption of NH3 on HKUST-1
12. Li, W. et al. Unprecedented high oxygen evolution activity of electrocatalysts metal–organic framework: a multitechnique approach. J. Phys. Chem. C 116,
derived from surface-mounted metal–organic frameworks. J. Am. Chem. Soc. 19839–19850 (2012).
141, 5926–5933 (2019). 39. Chui, S. S.-Y., Lo, S. M.-F., Charmant, J. P. H., Orpen, A. G. & Williams, I. D.
13. Schneider, C. et al. High electrical conductivity and high porosity in a A chemically functionalizable nanoporous material [Cu3(TMA)2(H2O)3]n.
guest@MOF material: evidence of TCNQ ordering within Cu3BTC2 Science 283, 1148–1150 (1999).
micropores. Chem. Sci. 9, 7405–7412 (2018). 40. Ditchfield, R., Hehre, W. J. & Pople, J. A. Self‐consistent molecular‐orbital
14. Xie, L. S. et al. Tunable mixed-valence doping toward record electrical methods. IX. An extended gaussian‐type basis for molecular‐orbital studies of
conductivity in a three-dimensional metal–organic framework. J. Am. Chem. organic molecules. J. Chem. Phys. 54, 724–728 (1971).
Soc. 140, 7411–7414 (2018). 41. Lee, C., Yang, W. & Parr, R. G. Development of the Colle-Salvetti correlation-
15. Wang, Z. & Cohen, S. M. Postsynthetic modification of metal–organic energy formula into a functional of the electron density. Phys. Rev. B 37,
frameworks. Chem. Soc. Rev. 38, 1315–1329 (2009). 785–789 (1988).
16. Deria, P. et al. Beyond post-synthesis modification: evolution of metal–organic 42. Chung, Y. G. et al. Computation-ready, experimental metal–organic
frameworks via building block replacement. Chem. Soc. Rev. 43, 5896–5912 frameworks: a tool to enable high-throughput screening of nanoporous
(2014). crystals. Chem. Mater. 26, 6185–6192 (2014).
17. Karagiaridi, O., Bury, W., Mondloch, J. E., Hupp, J. T. & Farha, O. K. Solvent- 43. Hendon, C. H. & Walsh, A. Chemical principles underpinning the
assisted linker exchange: an alternative to the de novo synthesis of unattainable performance of the metal-organic framework HKUST-1. Chem. Sci. 6,
metal–organic frameworks. Angew. Chem. Int. Ed. 53, 4530–4540 (2014). 3674–3683 (2015).
18. Tanabe, K. K. & Cohen, S. M. Engineering a metal–organic framework catalyst 44. Becke, A. D. A new mixing of Hartree–Fock and local density‐functional
by using postsynthetic modification. Angew. Chem. Int. Ed. 48, 7424–7427 theories. J. Chem. Phys. 98, 1372–1377 (1993).
(2009). 45. Weigend, F., Häser, M., Patzelt, H. & Ahlrichs, R. RI-MP2: optimized auxiliary
19. Wang, Z. & Cohen, S. M. Modulating metal−organic frameworks to breathe: a basis sets and demonstration of efficiency. Chem. Phys. Lett. 294, 143–152
postsynthetic covalent modification approach. J. Am. Chem. Soc. 131, (1998).
16675–16677 (2009). 46. Erickson, K. J. et al. Thin film thermoelectric metal–organic framework with
20. Jayachandrababu, K. C., Bhattacharyya, S., Chiang, Y., Sholl, D. S. & Nair, S. high seebeck coefficient and low thermal conductivity. Adv. Mater. 27,
Recovery of acid-gas-degraded zeolitic imidazolate frameworks by solvent- 3453–3459 (2015).
assisted crystal redemption (SACRed). ACS Appl. Mater. Interfaces 9, 47. Lin, X. et al. High capacity hydrogen adsorption in Cu(II) tetracarboxylate
34597–34602 (2017). framework materials: the role of pore size, ligand functionalization, and
21. Kapustin, E. A., Lee, S., Alshammari, A. S. & Yaghi, O. M. Molecular exposed metal sites. J. Am. Chem. Soc. 131, 2159–2171 (2009).
retrofitting adapts a metal–organic framework to extreme pressure. ACS Cent. 48. Farha, O. K. et al. Designing higher surface area metal–organic frameworks:
Sci. 3, 662–667 (2017). are triple bonds better than phenyls? J. Am. Chem. Soc. 134, 9860–9863
22. Schneider, C. et al. Tuning the negative thermal expansion behavior of the (2012).
metal-organic framework Cu3BTC2 by retrofitting. J. Am. Chem. Soc. 141, 49. Ma, S. et al. Metal-organic framework from an anthracene derivative
10504–10509 (2019). containing nanoscopic cages exhibiting high methane uptake. J. Am. Chem.
23. Chen, C. X. et al. Precise modulation of the breathing behavior and pore Soc. 130, 1012–1016 (2008).
surface in Zr‐MOFs by reversible post‐synthetic variable‐spacer installation to 50. Yan, Y. et al. Modifying cage structures in metal–organic polyhedral
fine‐tune the expansion magnitude and sorption properties. Angew. Chem. frameworks for H2 storage. Chem. Eur. J. 17, 11162–11170 (2011).
Int. Ed. 55, 9932–9936 (2016). 51. Ahlrichs, R., Bär, M., Häser, M., Horn, H. & Kölmel, C. Electronic structure
24. Bhattacharya, S., Nayak, S. & Dutta, S. C. A critical review of retrofitting calculations on workstation computers: the program system turbomole. Chem.
methods for unreinforced masonry structures. Int. J. Disaster Risk Reduct. 7, Phys. Lett. 162, 165–169 (1989).
51–67 (2014). 52. Treutler, O. & Ahlrichs, R. Efficient molecular numerical integration schemes.
25. Dadras Eslamlou, S., Masia, M. J., Totoev, Y. Z. & Page, A. W. Effect of J. Chem. Phys. 102, 346–354 (1995).
retrofitting on the structural factors for seismic assessment of unreinforced 53. Sierka, M., Hogekamp, A. & Ahlrichs, R. Fast evaluation of the Coulomb
masonry structures: a review. Aust. J. Struct. Eng. 20, 26–53 (2019). potential for electron densities using multipole accelerated resolution of
26. Yuan, S. et al. Sequential linker installation: precise placement of functional identity approximation. J. Chem. Phys. 118, 9136–9148 (2003).
groups in multivariate metal–organic frameworks. J. Am. Chem. Soc. 137, 54. Eichkorn, K., Treutler, O., Öhm, H., Häser, M. & Ahlrichs, R. Auxiliary basis
3177–3180 (2015). sets to approximate Coulomb potentials. Chem. Phys. Lett. 240, 283–290
27. Yuan, S. et al. Linker installation: engineering pore environment with precisely (1995).
placed functionalities in zirconium MOFs. J. Am. Chem. Soc. 138, 8912–8919 55. Eichkorn, K., Weigend, F., Treutler, O. & Ahlrichs, R. Auxiliary basis sets for
(2016). main row atoms and transition metals and their use to approximate Coulomb
28. Mitra, S. et al. Targeted drug delivery in covalent organic nanosheets (CONs) potentials. Theor. Chem. Acc. 97, 119–124 (1997).
via sequential postsynthetic modification. J. Am. Chem. Soc. 139, 4513–4520 56. Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. A consistent and accurate ab
(2017). initio parametrization of density functional dispersion correction (DFT-D) for
29. Segura, J. L., Royuela, S. & Mar Ramos, M. Post-synthetic modification of the 94 elements H-Pu. J. Chem. Phys. 132, 154104 (2010).
covalent organic frameworks. Chem. Soc. Rev. 48, 3903–3945 (2019). 57. Bhunia, M. K., Hughes, J. T., Fettinger, J. C. & Navrotsky, A.
30. Talin, A. A. et al. Tunable electrical conductivity in metal-organic framework Thermochemistry of paddle wheel MOFs: Cu-HKUST-1 and Zn-HKUST-1.
thin-film devices. Science 343, 66–69 (2014). Langmuir 29, 8140–8145 (2013).

NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications 9

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-12876-1

Acknowledgements Peer review information Nature Communications thanks Matthew Lennox, Rocio
We would like to thank Roland A. Fischer and Christoph Schran for insightful dis- Semino and other, anonymous, reviewers for their contribution to the peer review of this
cussions. Mark D. Allendorf and A. Alec Talin are acknowledged for the support work. Peer reviewer reports are available.
with electrical conductivity measurements and Pia Vervoorts and Dardan Ukaj for
porosimetry measurements. C.S. acknowledges his scholarships from the Studien- Reprints and permission information is available at http://www.nature.com/reprints
stiftung des Deutschen Volkes and from the Fonds der Chemischen Industrie (FCI).
G.K. gratefully thanks the FCI and the DFG SPP1928 (COORNETs) for financial Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
support. published maps and institutional affiliations.

Author contributions
Open Access This article is licensed under a Creative Commons
C.S., D.B. and G.K. developed the concept and algorithm for RetroFit. C.S. synthesized
and characterized the new CL@Cu3BTC2 materials. J.K. and R.S. calculated the model Attribution 4.0 International License, which permits use, sharing,
interaction potential. All authors have contributed to writing of the manuscript with C.S. adaptation, distribution and reproduction in any medium or format, as long as you give
and G.K. as main contributors. appropriate credit to the original author(s) and the source, provide a link to the Creative
Commons license, and indicate if changes were made. The images or other third party
material in this article are included in the article’s Creative Commons license, unless
Competing interests indicated otherwise in a credit line to the material. If material is not included in the
The authors declare no competing interests. article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
Additional information the copyright holder. To view a copy of this license, visit http://creativecommons.org/
Supplementary information is available for this paper at https://doi.org/10.1038/s41467- licenses/by/4.0/.
019-12876-1.
© The Author(s) 2019
Correspondence and requests for materials should be addressed to G.K.

10 NATURE COMMUNICATIONS | (2019)10:4921 | https://doi.org/10.1038/s41467-019-12876-1 | www.nature.com/naturecommunications