Introduction

C60 fullerene is the most abundant member of the so-called fullerene family1,2,3,4. However, monomeric C60-based endohedral metallofullerenes (i.e., fullerenes with metal atom(s) encapsulated), referred to hereafter as M@C60, have not been obtained in macroscopic and pure form to date. Most previous studies on metallofullerenes reported during the past 25 years have focused on higher fullerenes than C60 such as C80, C82, and C842, even though the presence of M@C60 has been confirmed by mass spectrometry of supersonic cluster beams in the gas phase3 as well as laser-vaporized processed soot4. Theoretical calculations also suggested the stability and possible structure of M@C605,6,7. An exception is [Li@C60]+(SbCl6) salt, which was produced by Li-ion bombardment with C60 onto a target plate in vacuum8, 9. Neutral Li+@C60∙− radicals have been obtained by the electrochemical reduction of cationic Li+@C60; however, Li@C60 oligomerizes10 and has a dimerized form in the crystal11.

The inability to extract and purify M@C60 is due to their very small highest occupied molecular orbital-lowest unoccupied molecular orbital (HOMO-LUMO) gaps predicted, which lead to high chemical reactivity toward surrounding soot, air, moisture, and various organic solvents. There have been, however, several attempts12,13,14,15 to try and extract and purify M@C60 by, for example, vacuum sublimation from soot followed by liquid chromatographic separation13, 14. Unfortunately, none of these efforts led to the final macroscopic preparation of pure and isolated M@C60. For this reason, M@C60 are often referred to as the missing metallofullerenes2, 16, 17.

One of the main reasons for the long-standing interest in M@C60 is due to their expected novel electronic and magnetic behavior at low temperatures such as superconductivity, because the electronic band structures near the Fermi levels of M3+@C603− solids resemble those of superconducting alkali-doped C60 fullerides such as (K3)3+C603− 18. If this is the case, the presence of heavy atoms like lanthanoid elements inside C60 might enhance the superconducting transition temperatures19. One way to stabilize such species and open the HOMO-LUMO gap is through chemical functionalization. Here we report the macroscopic preparation of purified (>99.9%) trifluoromethylated Gd@C60 and La@C60, stable as Gd@C60(CF3)3,5 and La@C60(CF3)3,5, respectively, and their structural determination, and electronic and magnetic properties.

Results

Enhanced stabilization of Gd@C60 and La@C60

To stabilize Gd@C60 and La@C60, we have employed in situ trifluoromethylation during the arc-discharge synthesis of the metallofullerenes developed in the present laboratory (see Supplementary Discussion 1 and Supplementary Figure 1)11, 19. For example, the in situ trifluoromethylation of Gd@C60 generates Gd@C60(CF3)n (n = 1–6) together with other sizes of trifluoromethylated Gd-metallofullerenes, Gd@C2m(CF3)n (2m ≥ 70) (see Supplementary Figures 2 and 3 for La). The high-temperature arc-discharge induces evaporation of polytetrafluoroethylene (PTFE) placed near the arc zone to produce CF3 radicals11. We found trifluoromethyl-derivatized metallofullerenes, Gd@C60(CF3)n and La@C60(CF3)n (n = 3, 5), were formed efficiently. Trifluoromethyl derivatives of Gd@C60 and La@C60 are totally soluble and stable in toluene and carbon disulfide, which enabled us to perform high-performance liquid chromatography (HPLC) purification and thus subsequent characterization. As discussed in later sections, the stability of these trifluoromethyl derivatives, Gd@C60(CF3)3,5 and La@C60(CF3)3,5, can be attributed to their closed-shell electronic structures, leading to wider HOMO-LUMO gaps than those of pristine Gd@C60 and La@C60, respectively. With the current separation/isolation protocol (see Supplementary Figure 4), one may, for example, obtain ca. 1.0–2.0 mg of purified (>99.9%) Gd@C60(CF3)5 within 24 h from the initial o-xylene extraction of the soot containing Gd@C2m(CF3)n obtained by 10 arc-discharge syntheses (see Supplementary Figures 58). The absorption onsets of the ultraviolet-visible-near-infrared (UV-Vis-NIR) absorption spectra of CS2 solution of Gd@C60(CF3)5 and La@C60(CF3)5 suggest the presence of enlarged HOMO-LUMO gaps of ca. 1.2 eV approaching that of C60 (1.6 eV) (see Supplementary Figure 9). Interestingly, Gd@C60(CF3) and Gd@C60(CF3)2 have not been solvent-extracted because of their much smaller HOMO-LUMO gaps even though mass spectrometric analysis indicates the presence of these metallofullerenes in raw soot.

Because of the very high reactivity of M@C60, it has not been self-evident that the metal is encapsulated in the conventional truncated icosahedral Ih-C60. The C60 might have a so-called non-isolated pentagon rule (non-IPR) structure20, 21, where two (or three) pentagons are fused with each other, since non-IPR fullerenes have frequently been observed in metallofullerenes during the past decade2. To answer this question, the molecular and crystal structures of Gd@C60(CF3)5 (structural isomers I and II) and La@C60(CF3)5 (I) have been determined by synchrotron radiation single-crystal X-ray diffraction (XRD) at SPring-8 (see Supplementary Tables 15).

Synchrotron single-crystal XRD measurements

The X-ray results clearly show that the C60 cage structures of Gd@C60(CF3)5 (I, II) and La@C60(CF3)5 (I) are very similar to that of the conventional empty C60(Ih). The five CF3 groups of isomer I are attached to carbon atoms numbered as 6, 9, 12, 15, and 53 in a Schlegel diagram shown in Fig. 1a. In isomer II the five CF3 groups are attached to carbon atoms numbered as 6, 9, 12, 15, and 36. In both isomers, four of the five CF3 groups are attached to carbon atoms 6, 9, 12, and 15 bonded to carbon atoms 5, 1, 2, and 3 on a pentagon. Figure 1b–e shows the molecular structures of Gd@C60(CF3)5(I) and (II), respectively. Very interestingly, the remaining CF3 group is attached not to carbon atom 18 but to carbon atom 53 (isomer I) or 36 (isomer II) located far from the 1-2-3-4-5 pentagon. This asymmetric attachment of five CF3 groups results in C1 symmetry. If the fifth CF3 group had been attached to carbon atom 18, neighbouring carbon atom 4 on the pentagon, the molecule would have had five-fold symmetry. Instead of the fifth CF3 group, a Gd (La) atom inside the C60 cage is located near carbon atoms 18 and 4. The distances between the Gd atom and carbon atoms 18 and 4 are ~2.38 Å in both isomers (see Supplementary Tables 25).

Fig. 1
figure 1

Gd@C60(CF3)5 (I) and (II). a Schlegel diagram of C60. CF3 groups attached to Gd@C60(CF3)5 (I) and (II) are shown as red circles. b, c Molecular structure of Gd@C60(CF3)5 (I). d, e Molecular structure of Gd@C60(CF3)5 (II). Front views facing the bond between carbon atoms 4 and 18 (b, d) and side views (c, e) are shown. The thermal ellipsoids in (be) are drawn at 50% probability level. f, g Molecular packing in the Gd@C60(CF3)5 (I) crystal viewed along the a- and c-axis, respectively. Pseudo-hexagonal close-packed layers in the bc-plane are stacked along the a-axis. h Steric and electrostatic stacking of the Gd@C60(CF3)5 (I) molecules with an electric dipole moment (P) along the a-axis, which form a close-packed array

The selective location of the Gd (La) atom is explained in terms of the elongation of C–C bond lengths by the attachment of CF3 groups. The conventional C60(Ih) molecule consists of two kinds of C–C bonds: short 6:6 bonds fusing two hexagons with a bond length of 1.39 Å and longer 6:5 bonds fusing a hexagon and a pentagon with a bond length of 1.45 Å. As expected, the C–C bonds around the carbon atoms with CF3 groups attached (6, 9, 12, 15, and either 53 or 36) have sp3 character, resulting in the elongation of the bonds. The 6:6 bond lengths between 6-5, 9-1, 12-2, and 15-3 carbon atoms are ~1.51 Å in both of the isomers (see Supplementary Tables 2 and 5). The 6:6 bond lengths between 18 and 4 atoms (~1.46 Å) are also longer than the normal 6:6 bond length (1.39 Å). This suggests that the Gd is interacting with the pz-orbitals of atoms 18 and 4, resulting in the breaking of the π-bond and elongation of the bond length. At the same time, there is a general attraction for the Gd toward the area of the C60 surface that has been functionalized, hence this location beneath two carbon atoms rather than staying at the center of the neighboring hexagon.

Gd@C60(CF3)5 (I) and La@C60(CF3)5 (I) form pure crystals that contain neither solvent nor ligand molecules. Figure 1f, g shows the molecular packing of the Gd@C60(CF3)5 (I) crystal. The centrosymmetric crystal with a space group of P21/c contains two chiral isomers of Gd@C60(CF3)5 (I) with the same number of molecules. Most of the endohedral metallofullerene crystals so far reported contain either solvent or ligand molecules2. Gd@C60(CF3)5 (II) has been obtained only as a co-crystal with Ni(OEP) (OEP: octaethylporphyrin). The crystals of Gd@C60(CF3)5 (I) and La@C60(CF3)5 (I) are the first example of pure crystals of M@C60 derivative. The crystal structure determined possesses a pseudo-hexagonal close-packed layer of the molecules in the bc-plane (Fig. 1f) and stacking of the layers along the a-axis (Fig. 1g). The stacking of the molecules is caused by intermolecular steric and electrostatic interactions. The molecule has an electric dipole moment (P) because of the electron-withdrawing CF3 groups asymmetrically attached to the C60 cage and the presence of the off-centered Gd (La) atom inside. The stacking of the turtle-like molecules with an electric dipole moment can provide an electrostatically stable close-packed allay along the a-axis as shown in Fig. 1h (see Supplementary Figures 1011). The detailed crystallographic data are summarized in Supplementary Tables 16.

We then carried out 19F-nuclear magnetic resonance (NMR) measurements of La@C60(CF3)n (see Supplementary Discussion 3 and Supplementary Figure 14). The spectra of the two La@C60(CF3)5 isomers consist of five equal-intensity peaks, which are four quartets (or multiplets) due to the through-space coupling of the four grouped CF3 units and the other singlet due to fifth more isolated CF3 unit22, 23. In contrast, the 19F-NMR spectrum of La@C60(CF3)3 only showed a singlet peak at −75.5 ppm because the three trifluoromethyl groups are geometrically equivalent, indicating C3 symmetry of La@C60(CF3)3.

Discussion

To obtain information on the growth mechanism, structure and electronic properties of Gd@C60(CF3)5, we perform spin unrestricted density functional calculations under the local spin density approximation24,25,26, applying relativistic pseudopotentials27. Wavefunctions are constructed from 38/90/28 independent Gaussian-based functions for C/Gd/F, respectively, up to l = 3 angular momentum, with an electron temperature of kBT = 0.04 eV for level occupation and 300 Ha cut-off plane-waves for the charge density. To explore the addition pathways for CF3 functionalization of Gd@C60(CF3)n, n = 0-5, we assume that CF3 adds to the most thermodynamically favored site and is then immobile, which is justified by a calculated high surface migration barrier of CF3 (n = 1) of 59 kcal mol−1. Taking the most stable nth isomer as a starting point for n + 1 stepwise addition allows us to reduce the number of isomers considered to a manageable several hundreds. Where two stable isomers are closer in energy than 2.3 kcal/mol of each other, we have used both as starting points for subsequent addition, allowing us to trace bifurcations in the addition pathway.

The Gd atom in Gd@C60 is initially covalently bound beneath hexagon (3-4-15-16-17-18) with 2.41 Å Gd–C bond lengths. The most stable CF3 addition pathway involves sequential addition to back bonds of the 1-2-3-4-5 pentagon, i.e., addition at sites 6, 9, 12, and 15 in that order shown in top line of Fig. 2a. Instead of completing the pentagon at site 18, which is excluded due to the presence of Gd below this last pentagon back bond with ~2.35 Å Gd–C bond lengths, the fifth CF3 group adds behind the neighboring pentagon at sites 53 (isomer I) or 36 (isomer II). Adding a sixth CF3 group is endothermic compared to ½(CF3)2, confirming that n = 5 is the end point in the addition sequence in agreement with the experimental observation. Interestingly, isomer II matches a La@C60(CF3)5 structure proposed previously7, but the intermediate n = 4 structure proposed there is different to the structure we find here and can only be linked to n = 3 and 5 with additional surface CF3 migration.

Fig. 2
figure 2

Calculated functionalization sequences and electronic properties of Gd@C60(CF3)n. a Calculated preferential addition sequences for CF3 to Gd@C60. Previous addition sites marked in blue, new addition site in red. Pink shading indicates location of the encapsulated Gd atom. Numbers indicate energy release in kcal mol−1 according to the reaction Gd@C60(CF3)n − 1 + ½ (CF3)2 → Gd@C60(CF3)n. All pathways are shown where isomers are within 2.3 kcal mol−1 of the most stable. Fullerenes where next addition is endothermic have black outlines and represent stable end points of a reaction pathway. b Calculated Kohn-Sham eigenvalues (eV) for (left) C60, (center) Gd@C60, and (right) Gd@C60(CF3)3, showing CF3 functionalization results in a closed-shell system, reopening the HOMO-LUMO gap to nearly that of C60. c Spatial distribution of calculated Kohn-Sham eigenstates showing highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) and their neighboring states around the Fermi level for the most stable Gd@C60(CF3)5(I) isomer. The states are delocalized across the fullerene cage with very limited CF3 character

After n = 1 addition at site 6 there is a second stable isomer for n = 2, where Gd maintains its hexagonal site and CF3 adds to site 36 (lower line of Fig. 2a). The pathway then bifurcates once again, the energetically favored route continuing the 9, 12, 15 addition sequence to isomer II. However, an alternative n = 3 addition at site 13 results instead in a C3 symmetry Gd@C60(CF3)3 isomer. Further CF3 addition to this is endothermic, showing this to be a third stable end point in the addition sequence. This mapping process thus shows remarkable agreement with the current X-ray and 19F-NMR experimental results, successfully predicting the three observed Gd@C60(CF3)n species, i.e., Gd@C60(CF3)5 (I), Gd@C60(CF3)5 (II) and Gd@C60(CF3)3.

These three stable isomers have multiplicity 7/2, associated with the Gd atom. They are closed-shell systems (unlike many of the less-stable isomers), restoring the gap between HOMO and LUMO to ~2/3 that of C60 (Fig. 2b) with the largest gap for Gd@C60(CF3)3. The unpaired Gd f-states are deep lying with the HOMO and LUMO localized primarily over the fullerene cage with minor CF3 character (Fig. 2c). Gd plays an interesting dual role when selecting CF3 addition sites, acting as donor and localizing surface charge, while also affecting surface chemistry through strong hybridization with the carbon 2p states. This suggests different addition sequences are likely for non-covalently bound endohedral +3 oxidation species. As seen above, the agreement between the X-ray and the theoretical results for the presence of the stable isomers of Gd@C60(CF3)3,5 strongly suggests that CF3 groups are sequentially added to Gd@C60 during the arc-discharge synthesis. The energy levels, HOMO-LUMO and spin-up and -down charge density for Gd@C60(CF3)3 and Gd@C60(CF3)5 (I,II) are shown in Supplementary Figures 15 and 16, respectively.

Literature calculations for M@C60(CF3)3 using comparable methods give a calculated HOMO-LUMO gap of 0.81 eV for Y@C60(CF3)316, and ~0.7 eV for La@C60(CF3)37, at first sight in discrepancy with our significantly larger calculated gap for Gd@C60(CF3)3 of 1.4 eV. This is because the literature calculations consider a M@C60(CF3)3 isomer with the CF3 groups arranged around a shared pentagon. Our calculated HOMO-LUMO gap for this Gd@C60(CF3)3 isomer is indeed 0.78 eV, close to these values. However, we do not predict this structure to be stable in isolation but simply an intermediate structure en-route to the formation of isomer I and isomer II Gd@C60(CF3)5 structures. The Gd@C60(CF3)3 isomer we predict to be a stable end point is instead the C3 symmetry isomer shown in Fig. 2, with CF3 groups arranged symmetrically around a central hexagon. This arrangement of CF3 groups gives the much wider calculated HOMO-LUMO gap of 1.4 eV.

The magnetic properties of Gd@C60(CF3)3 metallofullerene was investigated by SQUID measurements (see Supplementary Discussion 5). The HOMO-LUMO gap of Gd@C60(CF3)3 (1.4 eV) is larger than that of Gd@C60(CF3)5 (1.2 eV) suitable for solvent extraction and HPLC isolation. Magnetization curves of solid Gd@C60(CF3)3 indicate no hysteresis as shown in Fig. 3, where the ordinate is normalized by experimentally determined saturation magnetization Ms of 41 108 emu ∙ G mol−1. Except for the magnetization curve at 2 K, other data points are on the same curvature trace. When magnetic moments thermally fluctuate without strong interaction between the magnetic moments, magnetization curves are fitted by the following Brillouin function BJ(x)

$$M = M_{\mathrm{s}}B_J\left( x \right)$$
(1)
$$M_{\mathrm{S}} = NgJ\mu _{\mathrm{B}}$$
(2)
$$B_J\left( x \right) = \frac{{2J + 1}}{{2J}}{\mathrm{coth}}\left( {\frac{{2J + 1}}{{2J}}x} \right) - \frac{1}{{2J}}{\mathrm{coth}}\left( {\frac{1}{{2J}}x} \right)$$
(3)
$$x = \frac{{gJ\mu _{\mathrm{B}}H}}{{k_{\mathrm{B}}T}}$$
(4)

where N is the spin concentration, g the g-factor, J the total quantum number, μB the Bohr magneton, and kB the Boltzmann constant. From the curve fitting, we found that the total quantum number J = 7/2 is needed to explain the experimental results by setting g = 2. This gives an effective magnetic moment μeff \(\left( { = g\sqrt {J\left( {J + 1} \right)} \mu _{\mathrm{B}}} \right)\) of 7.94 μB per Gd@C60(CF3)3 molecule. This experimental analysis supports a trivalent Gd3+ state by transferring three electrons to the C60 cage (Gd3+@C603−) and these electrons are then used to make chemical bonds with three CF3 on C60. Mulliken charge analysis shows that most of the transferred charge is indeed spread around the C60 cage beneath the CF3 groups, on the central pentagon, and on the neighbors of the Gd. The magnetic moment is thus essentially located on the encaged Gd3+ i on (4f7, J = 7/2 with orbital quantum number L = 0, i.e., Gd3+ has spin 7 μB) in agreement with the theoretical result shown in Fig. 2b. According to the previous studies on M3+@C823− (M = Gd, Dy, Ho, Er) metallofullerenes28,29,30,31,32, the magnetic moments observed are significantly small compared to the expected moments of free M3+ ions.

Fig. 3
figure 3

Normalized magnetization curves of Gd@C60(CF3)3. a Whole trace of magnetization taken at 273, 100, 20, 10, 4.2, and 2 K, and b is an expansion around the origin of abscissa. Ms value determined experimentally is 41108 emu · G mol−1. Solid lines are the calculated magnetization curves using Brillouin function with J = 8/2, 7/2, and 6/2. Red open circles are for the data taken at 2 K. Only this data set did not fit with the same curvature

This reduction of the magnetic moments was theoretically explained by antiferromagnetic coupling between the trivalent endohedral metals and free spins on the carbon cage33. In the present Gd (La)-encapsulating C60 fullerides, since free spins on C60 are very small by forming the chemical bonds with CF3 on the cage, this guarantees an almost ideal magnetic moment of Gd3+ ion. Theoretically, a trace counter-spin can be seen localized on the C60 cage between the Gd and CF3 groups at 0 K (Fig. 2c), presumably too weak to be detectable by SQUID but partially responsible for the weak antiferromagnetic coupling observed at low temperature. This was also confirmed by the analysis of temperature dependence of molar magnetic susceptibility as explained in Supplementary Discussion 5.

To further investigate the electronic structures of Gd@C60(CF3)5 (I,II), we also performed electron spin resonance (ESR) measurements in CS2 frozen solution (see Supplementary Discussion 6 and Supplementary Table 7). The spectra recorded by an X-band continuous wave ESR spectrometer at 4 K and by a W-band at 20 K are shown in the upper and lower panels of Fig. 4, respectively, where the observed spectra with fine structure in red are compared with simulated ones in blue34, 35. The parameters shown in Supplementary Table 7 were determined so that both of the observed spectra by the X- and W-band measurements should simultaneously be reproduced by the simulation. In both cases the spin quantum number of S = 7/2 was obtained, consistent with the theoretical and magnetic measurement results described above. The zero-field splitting parameters, D = 0.162 cm−1 and D = 0.193 cm−1, were obtained for Gd@C60(CF3)5 (I) and Gd@C60(CF3)5 (II), respectively, comparable to those for S = 7/2 states of Gd@C8234 and Eu@C8235. The lift of the x and y degeneracy of E = 0.012 cm−1 for Gd@C60(CF3)5 (II), which is bigger than that (E = 0.000 cm−1) of Gd@C60(CF3)5 (I), suggests that the position of the Gd3+ ions in the two isomers differs slightly, since the E parameters originate basically in the spin-orbit coupling of the Gd3+ ions. Our calculations indeed indicate that the Gd3+ ion of isomer (I) lies more symmetrically beneath the C–C bond (so more symmetrically with respect to the four CF3 groups around the pentagon) than that of isomer (II). The observed ESR parameters are summarized in Supplementary Table 7.

Fig. 4
figure 4

Confirmation of ESR parameters for Gd@C60(CF3)5(I) and Gd@C60(CF3)5(II). a, c ESR spectra (red lines) of Gd@C60(CF3)5(I) and Gd@C60(CF3)5(II) recorded by X-band (9 GHz) spectrometer in CS2 at 4 K, b, d W-band (90 GHz) spectrometer in CS2 at 20 K. ESR parameters were determined so that both observed spectra in ad should be simultaneously reproduced by simulation (blue lines)

In summary, the missing metallofullerenes, Gd@C60 and La@C60, in the trifluoromethylated forms of Gd@C60(CF3)3,5 and La@C60(CF3)5, have been obtained in pure bulk forms. In situ trifluoromethylation widens the HOMO-LUMO gaps of Gd@C60 and La@C60, enabling isolation of the metallofullerenes and the subsequent single-crystal synchrotron XRD. The crystals exhibit a pseudo-hexagonal close-packed layer of molecules in the bc-plane and stacking of the layers along the a-axis (a turtle-like stacked structure). The temperature dependence of the magnetic susceptibility of Gd@C60(CF3)3 is well fitted by the Curie-Weiss law. The magnetic moment is located on the encaged Gd3+ i on (4f7, J = 7/2 with orbital quantum number L = 0), with very small magnetic moment on the cage in agreement with the theoretical results. The intact forms of Gd@C60 and La@C60 might exhibit superconductivity as the electronic structures resemble those of superconducting alkali-doped C60 fullerides. The preparation and isolation of intact Gd@C60 fulleride is now underway in the present laboratory.

Methods

Synthesis, purification, and crystallization

A cross-sectional view of the DC arc-discharge chamber is illustrated in Supplementary Figure 1, where PTFE rods (40 g) are placed near the discharge area. Graphite rods (100 g) impregnated with Gd (La) (0.8 mol%, Toyo Tanso Co.LTD) was used as the anode. A pure graphite rod (Toyo Tanso Co.LTD) was used as the cathode. Arc-discharge was performed at a DC current of 500 A in a flowing He atmosphere with a pressure of 7–9 kPa. During arc-discharge because of the high-temperature around the arc zone, PTFE was decomposed and evaporated to produce CF3 radicals. Normally, 50–70 g of raw soot was obtained per discharge. Gd-metallofullerenes and empty fullerenes were extracted from the raw soot with o-xylene.

Extraction and separation

The rapid separation of the metallofullerenes from empty fullerenes was carried out by the TiCl4 Lewis acid method developed in the present laboratory. To a 500 mL CS2 solution of the crude mixture of Gd-fullerenes and empty fullerenes, ca. 5 mL of TiCl4 was added. Metallofullerenes were reacted immediately and insoluble complexes were precipitated out. After mixing for 5 min, the precipitate was collected on a PTFE membrane filter and washed with 10–20 mL of CS2 to separate from the empty fullerenes solution. Deionized water was passed through the filter to decompose the complex of metallofullerene/TiCl4, and then washed with acetone to eliminate extra water. Finally, CS2 was passed through the filter to collect desired Gd-metallofullerenes as a solution as shown in Supplementary Figure 2 (equivalent for La in Supplementary Figure 3).

Multi-stage HPLC purification

HPLC purification was conducted by using a JAI (Japan Analytical Industry Co. LTD.) recycling preparative HPLC LC-9104HS. The overall separation and isolation scheme of Gd@C60(CF3)n is shown in Supplementary Figure 4. For the identification of metallofullerenes, mass spectrometric analysis was performed on Shimadzu MALDI-TOF-MS Spectrometer. Vis/NIR absorption spectra of metallofullerenes in CS2 were recorded by a Jasco V-570 spectrophotometer. Three isomers of Gd@C60(CF3)n were isolated from the mixture by the multi-stage HPLC method developed in the present laboratory. Two kinds of columns were used alternatively with toluene eluent for the isolation, i.e., Buckyprep column (20 mm diameter × 250 mm, Nacalai Tesque Inc.) and Buckyprep-M column (20 mm diameter × 250 mm, Nacalai Tesque Inc.). The initial (the first stage) HPLC purification was performed with Buckyprep-M. Gd@C60(CF3)n (n = 3, 5) was obtained in fraction 1. The HPLC chromatograms are shown in Supplementary Figure 5 (equivalent for La in Supplementary Figure 6). Identification and isolation of the metallofullerenes were checked by MALDI mass spectroscopy as shown in Supplementary Figures 7 and 8. UV-Vis-NIR absorption spectra of the metallofullerenes exhibit characteristic features and also provide an estimate on their HOMO-LUMO gaps judging from the absorption onsets. The spectral features between Gd- and La-metallofullerenes are almost the same with each isomer as shown in Supplementary Figure 9.

Synchrotron single-crystal X-ray structure analysis

Single-crystal XRD measurements of Gd@C60(CF3)5 (I), La@C60(CF3)5 (I), Gd@C60(CF3)5 (II), and Gd@C60(CF3)3 were performed at SPring-8 BL02B1. The crystallographic data and various bond lengths for Gd@C60(CF3)5 (I,II), Gd@C60(CF3)3, and La@C60(CF3)5 (I) are summarized in Supplementary Tables 1-6. The crystal structures for Gd@C60(CF3)5 (I), La@C60(CF3)5 (I), and Gd@C60(CF3)5 (II) are shown in Supplementary Figures 1012, respectively. The charge density surface of Gd@C60(CF3)3 is shown in Supplementary Figure 13.

Gd@C60(CF3)5 (I) and La@C60(CF3)5 (I): Single crystals of Gd@C60(CF3)5 (I) and La@C60(CF3)5 (I) were obtained from CS2 solution by vapor diffusion. Results of the XRD measurement are summarized in Supplementary Table 1. The crystal structures were determined by using SIR and SHELX with good reliable factors. Gd@C60(CF3)5 (I) and La@C60(CF3)5 (I) have similar crystal structures. The centrosymmetric monoclinic crystals consist of the same number of two chiral isomers of the molecule. The unit cell contains two right-handed isomers and two left-handed isomers as shown in Supplementary Figures 10a and 11a. An independent molecule in the asymmetric unit has a disordered structure in which two chiral isomers overlap with the ratios of 0.8 and 0.2 as shown in Supplementary Figures 10b and 11b. In all, 95 C–C, 15 C–F, and 4 C–Gd (La) bond lengths of the major part of the disordered Gd@C60(CF3)5 (I) and La@C60(CF3)5 (I) are listed in Supplementary Tables 2 and 3, respectively. A total of 30 short 6:6 bonds fusing two hexagons and 60 long 6:5 bonds fusing a hexagon and a pentagon on the C60 cage are separately shown. The CIF deposition numbers at the Cambridge Crystallographic Data Centre (CCDC) are 1587428 for Gd@C60(CF3)5 (I) and 1587430 for La@C60(CF3)5 (I).

Gd@C60(CF3)5 (II): Single crystals of Gd@C60(CF3)5 (II) were obtained from CS2 solution as co-crystals with Ni(OEP) by vapor diffusion. Results of the XRD measurement are summarized in Supplementary Table 4. The crystal structure was determined by using SIR and SHELX with a good reliable factor. The centrosymmetric triclinic crystals consist of the same number of two chiral isomers of Gd@C60(CF3)5 (II). The unit cell contains a right-handed isomer, a left-handed isomer, three Ni(OEP), and half toluene molecules as shown in Supplementary Figure 12a. An independent molecule in the asymmetric unit has a disordered structure in which two chiral isomers overlap with the ratios of 0.8 and 0.2 as shown in Supplementary Figure 12b. In all, 95 C–C, 15 C–F, and 4 C–Gd bond lengths of the major part of the disordered Gd@C60(CF3)5 (II) are listed in Supplementary Table 5. A total of 30 short 6:6 bonds fusing two hexagons and 60 long 6:5 bonds fusing a hexagon and a pentagon on the C60 cage are separately shown. The CIF deposition number at CCDC is 1587429.

Gd@C60(CF3)3: Single crystals of Gd@C60(CF3)3 were obtained from CS2 solution by vapor diffusion. Results of the XRD measurement were summarized in Supplementary Table 6. The tetragonal unit cell can contain 16 Gd@C60(CF3)3 molecules. An expected molecular arrangement was obtained by charge flipping using Superflip and maximum entropy method using ENIGMA as shown in Supplementary Figure 13. The figure shows an electron charge density surface obtained by maximum entropy method. A structure model with 16 double uniform shells of Gd@C60 located at (1/4 ± 1/8, 1/4 ± 1/8, 0), (−1/4 ± 1/8, −1/4 ± 1/8, 0), (1/4 ± 1/8, −1/4 ± 1/8, 1/2), and (−1/4 ± 1/8, 1/4 ± 1/8, 1/2) was used in the analysis. Detailed structure of Gd@C60(CF3)3 could not be determined due to a severe orientation disorder and lack of resolution. We also attempted to obtain co-crystals of Gd@C60(CF3)3 with Ni(OEP); however, that have not been obtained.

19F-NMR measurements

NMR spectra were recorded on a JEOL ECS-400 spectrometer; chemical shifts for 19F-NMR (376 MHz, CDCl3/CS2) are expressed in parts per million (ppm) relative to C6F6 (δ = –164.9 ppm). Data are reported as follows: chemical shift; multiplicity (s = singlet, q = quartet, m = multiplet); coupling constant (Hz); and integration. La@C60(CF3)3: δ –75.5 (s, 9F). La@C60(CF3)5 (I): δ –79.4 (s, 3F), –76.0 (q, J = 10 Hz, 3F), –74.9 (q, J= 12 Hz, 3F), –67.4 (m, 3F), and –66.9 (m, 3F). La@C60(CF3)5 (II): δ –78.9 (s, 3F), –78.0 (q, J= 12 Hz, 3F), –73.6 (q, J = 10 Hz, 3F), –68.3 (m, 3F), and –67.3 (m, 3F).

SQUID magnetic measurement

Sample dissolved in CS2 was transferred into quartz tube and the solution was vaporized under argon flow. Then the sample in quartz tube was evacuated by using turbo molecular pump at 260 °C for 12 h and vacuum sealed. By this heat treatment, weight of Gd@C60(CF3)3 sample was decreased by ~65% (finally 0.71 mg of sample was loaded in the quartz tube), while no such weight change was observed for La@C60(CF3)5 one. Former finding suggests that the CS2 molecules were incorporated in a Gd@C60(CF3)3 solid, forming Gd@C60(CF3)3(CS2)6.4 just after vaporizing CS2 in Ar flow, and such incorporated CS2 molecules have been completely removed by heating at 260 °C in vacuo. On the other hand, from the latter fact, La@C60(CF3)5 solid did not incorporate any CS2 molecule. That is, all SQUID measurements were carried out by using solution free samples. As described in the Discussion, magnetic behavior can be explained as an intramolecular feature and the interaction between the molecules is very small. Therefore, even if the sample incorporates some CS2 molecules, these molecules do not affect the present results.

ESR measurements

The X- and W-band ESR measurements were performed using a Bruker E500 and a E680 spectrometer, respectively. The temperature was controlled by helium flow cryostat (Oxford Instruments model ITC500). The X-band ESR spectra were measured at 4.0 K, and W-band ESR spectra were measured at 20 K. The spectral simulation were performed by using MATLAB software package with EasySpin toolbox. The observed ESR parameters are shown in Supplementary Table 7.

Data availability

Structural data are available in Supplementary Information. All additional data such as calculated xyz structures are available from the authors on request. The X-ray crystallographic coordinates for structures reported in this study have been deposited at the CCDC, under deposition numbers 1587428–1587430. These data can be obtained free of charge from The CCDC via www.ccdc.cam.ac.uk/data_request/cif.