All geometries considered in this study have been optimized at the long- range corrected CAM-B3LYP (Yanai et al., 2004) and the standard hybrid B3LYP (Becke, 1993) functional level of theory in combination with Ahlrichs SVP (Schäfer et al., 1992) and Pople 6-31G* (Hehre et al., 1972) basis set qualities as implemented in the Gaussian16 program package (Frisch et al., 2016). The B3LYP functional is well-established and tested in the field of transition metal computational chemistry, especially in the context of cobalt-containing systems of different multiplicity (Vargas et al., 2006). All minimum structures have been verified by analysis of the Hessian. Based on the crystal structure (PDB code 1DZI (Emsley et al., 2000) the model systems have been constructed using Hyperchem 8.0 (Hypercube (2007). Hyperch, 2007) to mimic the electronic environment of the cobalt and magnesium complexes respectively. All amino acid residues except the ones indicated in Figure 1A have been cleaved and the ends have been capped with methyl groups. The amino acids (Asp A151, Glu A152, Ser A153, Asn A154, Ser A155, Thr A221, Asp A254, Gly A255, Glu A256, Ser A257, His A258, Arg B12 und Glu C11) and two water molecules were used for the model. All residues are completely free to move during the calculations. Solvation effects have been addressed by means of the PCM (Miertuš et al., 1981; Miertus̃ and Tomasi, 1982) solvation model. In line with literature findings our benchmark calculations have indicated satisfying overall performance of the chosen hybrid and long-range corrected functional quality regarding geometric and electronic features of the metal complexes under consideration.

The protein-complex 1DZI (Emsley et al., 2000) MD-simulation was run with YASARA 20.10.14 (Krieger and Vriend, 2014). The setup included an optimization of the hydrogen bonding network (Krieger et al., 2012) to increase the solute stability, and a pKa prediction to fine-tune the protonation states of protein residues at the chosen pH of 7.4 (Krieger et al., 2006). Na⁺ and Cl⁻ ions were added with a physiological concentration of 0.9%, with an excess of either Na⁺ or Cl⁻ to neutralize the cell. After steepest descent and simulated annealing minimizations to remove clashes, the simulation was run for 100 nanoseconds using the AMBER14 force field (Maier et al., 2015) for the solute, GAFF2 (Wang et al., 2004) and AM1BCC (Jakalian et al., 2002) for ligands, spezial amino acids (hydroxyprolin) and ions (Mg²⁺, Co²⁺, and Co³⁺) and TIP3P for water. The cutoff was set to 8 Å for Van der Waals forces (the default used by AMBER (Hornak et al., 2006), no cutoff was applied to electrostatic forces (using the Particle Mesh Ewald algorithm (Essmann et al., 1995). The equations of motion were integrated with a multiple timestep of 1.25 fs for bonded interactions and 2.5 fs for non-bonded interactions at a temperature of 310 K, 0,993,249 g/ml density and a pressure of 1 atm (NPT ensemble) using algorithms described in detail previously (Krieger and Vriend, 2015). After inspection of the solute RMSD as a function of simulation time, the first 10 ns were considered equilibration time and excluded from further analysis.

Graphical representation of the molecular geometries has been produced with the aid of CylView 1.0b (Legault, 2009) and Chemcraft (Zhurko, 2005).

The Methods of the Cell assay part are described in the Materials and Methods section of reference (Petridis et al., 2011).

Seven days old mice were sacrificed by decapitation under hypothermic anesthesia. The brains were removed and transferred into serum-free DMEM/F-12 medium supplemented with 1% penicillin/streptomycin (GIBCO-BRL; Carlsbad, CA), and 400 μm thick sagital slices were cut with a vibratome. Circular pieces (about 1 mm²) of the subventricular zone region l were extracted and put in DMEM/F-12 medium supplemented 1% penicillin/streptomycin. The cultures were treated with 6 U/ml of a polysialic acid (PSA)-specific endoneuraminidase (endo N). This amount of endo N is sufficient to completely remove PSA from the tissue. Armenian hamster monoclonal anti-β1 (10 μg/ml; BD Biosciences) was added together with endo N to the cultures to assess the role of β1-integrin in endo N-induced neuritogenesis. Untreated cultures were used as controls. All animal care and experimentation were carried out according to MSKCC (Memorial Sloan Kettering Cancer Center) institutional guidelines.

The cell line used in this study was the human neuroblastoma SH-SY5Y. The cell line was cultured in Dulbecco`s modified medium DMEM/F12 supplemented with 15% heat inactivated fetal calf serum (FCS) and 1% penicillin/streptomycin and Zeocin.

Cells were fixed in 4% paraformaldehyde at RT for 10 min, washed in PBS, then incubated with primary antibodies at 37°C for 2 h. After washing with PBS and incubation for 1 h at 37°C with the secondary antibodies, the cells were mounted in mowiol (Calbiochem; San Diego, CA). Primary antibodies used: mouse monoclonal anti-PSA (5A5, 1:1,000), armenian hamster monoclonal anti-β1 Integrin (1:200 BD Biosciences), mouse monoclonal anti-NF-M (145k) antibody for neurofilament staining (Chemicon), mouse monoclonal anti-NCAM (123C3; 1:150; Santa Cruz, Santa Cruz, CA). Secondary antibodies were FITC and Cy3 conjugated (Jackson Laboratories; West Groove, PA). Immunostaining with secondary antibodies alone was used as negative control. Cells were observed under a Zeiss 510LSM confocal microscope.

The PSA-positive neuroblastoma cell line SH-SY5Y (ATCC; Rockville, MD) was plated on glass coverslips and cultivated in Dulbecco’s modified medium DMEM/F12 supplemented with 15% heat inactivated fetal calf serum (FCS) and 1% penicillin/streptomycin. After cells reached a confluency of 60%, all-trans-retinoic acid (RA) (10 μM) was added. Retinoic acid treated and completely untreated cells were used as controls. Four days after addition of retinoic acid, 20 U/ml endo N was added together with serum and antibiotic free DMEM: F12 medium with or without blocking antibodies for 24 h. Antibodies used for blocking were, armenian hamster monoclonal anti-β1 (10 μg/ml; BD Biosciences) and rabbit polyclonal anti-NCAM Fab’-antibody 10 μg/ml (R004).

All experiments were repeated at least four times in triplicate. Statistical significance was assessed using analysis of variance (ANOVA).

The Methods of the AFM are described in the Materials and Methods section of reference (Stötzel et al., 2012).

As far as the immunohistochemistry experiments with the antibodies against ADAM12 and GFAP in the oligoastrocytoma tissues are concerned, these are described in every detail in the study published by Kanakis et al. (2013). In particular, regarding the quantification of the two antibodies (i.e ADAM12, GFAP) in the two different cell populations (i.e. neoplastic astrocytes, neoplastic oligodendrocytes) this was performed by estimating the number of immunostained cells for each antibody separately, which was then presented as a percentage value in a scale from 0 (no immunostained neoplastic cells) to ++++ (>75 positive neoplastic cells).

As expected, the Co³⁺ and Mg²⁺ hexaaqua complexes are characterized by essentially symmetric octahedral ligand fields according to their electronic configurations while the Co²⁺ exhibits pronounced Jahn-Teller distortion due to its d 7 configuration giving rise to a singly occupied d z 2 orbital. This in turn is reflected by the significant bond elongation of axial ligands to approximately 2.1 Å as depicted in Figure 1B. In contrast, almost identical bond lengths of 1.924 and 2.100 Å are encountered for both axial and equatorial ligands in the Co³⁺ and Mg²⁺ complexes respectively. Table 1 summarizes the bond lengths with respect to the employed level of computational treatment highlighting an essentially method-independent behavior of the geometric parameters. As can be deduced from the numbers, the long-range corrected CAM-B3LYP functional slightly contracts all the bonds by roughly 0.01–0.02 Å while utilization of the Ahlrichs SVP basis set shows mixed trends in comparison to the 6-31G* Pople basis set. Further employment of polarization functions at the hydrogen atoms via the 6-31G** basis set results in slight and systematic overall bond contraction throughout our study which is attributed to the increased flexibility in the description of the electronic structure but does not change the observed trends.

As opposed to the cobalt based systems, the magnesium hexaaqua complex is largely stabilized by ion-dipole interactions indicated by the straight geometric orientation of the surrounding water molecules, cf. Figure 1B as opposed to the essentially covalent interactions that dominate the [Co(OH₂)₆]²⁺ and [Co(OH₂)₆]³⁺ geometries which exhibit bent orientations of the hydrogen atoms in the ligand water molecules with respect to the metal center.

Regarding the relevant electronic configurations for the [Co(OH₂)₆]²⁺ and [Co(OH₂)₆]³⁺ complexes respectively, the typical scheme is obtained in terms of the five d-orbitals of the central cobalt ion that are intermixed with the ligand orbitals as outlined in panels A and B of Figure 2. The resulting orbital occupations patterns can be rationalized as d x y 2 d x z 2 d y z 2 d z 2 1 d x 2 − y 2 0 for the Co²⁺ and d x y 2 d x z 2 d y z 2 d z 2 0 d x 2 − y 2 0 for the Co³⁺ systems respectively, perfectly in line with the expected geometries for distorted and undistorted d 7 and d 6 ligand fields.

The depiction of the net spin density (cf. panel A of Figure 3) emphasizes the notion of the singly occupied d z 2 -orbital strongly localized at the cobalt site, a result that is anticipated given the allowed mixing pattern in an octahedral coordination sphere for the set of e g orbitals at the cobalt center and the donor orbitals (that stem from the oxygen in the axial water ligands) that constitute σ -type interactions. Regarding the overall positive sign of the spin density, the predominance of spin delocalization to the adjacent ligands is apparent at the chosen cutoff in Figure 3.

Potential energy surface scans reveal distinct energetic differences in bond dissociation curves for axial and equatorial ligand cleavage from the [Co(OH₂)₆]²⁺ complex (Figure 4). Due to the elongated bond distances at the axial positions the net dissociation energy is expected to be considerably lowered by approximately 12 kcal/mol indicative of possible biological relevance in the Co²⁺-mediated processes in the full protein environment.

Our integrin-collagen model system consists of an inner layer of electronic interaction patterns inspired by the original chemical site at the crystal structure from the PDB code 1DZI by Emsley et al. (2000). As can be deduced only two coordinating water molecules are directly surrounding the central ion while the side chain residues of the protein chain occupy the four remaining positions in the hexacoordinate ligand sphere. The incorporation of the metal cation in the ligand sphere of the protein background affects the geometric arrangement of the complex and slightly alters the electronic interactions. Most prominently, the complex symmetry is lowered to certain degrees as the various bond lengths are determined by the nature and electronic properties of the respective ligand that is either a water substituent or an amino acid type residue. Table 2 displays the relevant bond lengths for the three investigated collagen-integrin model complexes. While the d 7 configuration in case of the cobaltous system still dominates the Jahn-Teller type distorted structural motif, the geometries of the Co³⁺ and Mg²⁺ complexes effectively adapt to the heterogeneous coordination spheres including the isolated water molecules. The cobaltic system for instance consequently shrinks the R C o − O distances facing the oxygen donor atom of the water substituents with respect to the oxygen atoms originating from the amino acids in the geometric proximity. This matter, however, correlates well with the notion of flexible movement of disjoint water ligands in contrast to the rigid backbone of the protein chain. In case of the magnesium system, the overall R M g − O distances are effectively larger thus facilitating the structural adjustments of the complex given the steric and geometric demands of the respective ligand framework. The inclusion of solvation effects by means of the PCM model (for a water environment) does not alter the overall picture conducting only minimal changes in the bond lengths. In general, the bond contraction originating from the long-range corrected functional seems to outnumber the influence of the continuum model. Therefore, the already compact structure of the model complex does not seem to significantly benefit from additional solvation effects during geometry optimization relying on the PCM approach.

Analysis of the three model cases does, however, demonstrates the general tendency to asymmetrize the complex from its either perfect octahedral or Jahn-Teller distorted shape into somewhat tilted hybrid structures that qualitatively feature two subsets of bond lengths. This finding raises the doubt for a distinct cobaltous character of the central ion in the embedding of the experimental system. In Figure 5 the frontier orbital occupation (with significant Co contribution) are depicted for the cobaltous system. As can be deduced, the orbital shape and orientation is clearly affected by the asymmetric electronic environment due to the amino acid residues in direct proximity of the metal center. Thus, the tilting and bending of orbital character is apparent to optimize the interactions upon protein embedding of the cation unlike in the hexaaqua system where orbital deformation is less distinct (cf. panel A of Figure 2).

The spin density distribution in the Co²⁺ model complex (cf. panel B of Figure 3) is likewise altered in comparison to the hexaaqua ligand sphere: while the spin delocalization seems to dominate the situation in a coordination sphere of six water molecules as stated above, the asymmetric surroundings as well as the bond elongation with respect to the paramagnetic center induce a more complex spin density distribution. For instance, the additional dark green lobes are indicative of substantial spin polarization phenomena in the vicinity of the Co center. Given the more involved interaction patterns the Mulliken spin density of the cobalt center decreases ever so slightly from approximately 0.966 in the hexaaqua system to 0.961 in the model system based on our computations at the B3LYP/6-31G* level of theory. This trend, however, depends crucially on the quality of the long-range interaction description as provided by the CAM-B3LYP functional. In the latter case the numbers change from 0.962 to 0.971 thus showing a reversed picture with a more isolated spin density accumulated at the metal center. The effect of PCM solvation further intensifies the spin density aggregation at the Co center. The numbers are inferred from Table 3. As the computation of the spin density is, in general, less sensitive with respect to the employed level of theory and basis set quality (Cano et al., 1998; Ruiz et al., 2005; Giner and Angeli, 2016; Philips et al., 2010) the overall method dependency is, however, comparably negligible.

Further inspection reveals the marked deviations from the predicted structural arrangement: as Figure 6 suggests, the experimental cobalt complex extracted from the PDB file provided by Emsley et al. (2000) lacks the intrinsic symmetric properties that are characteristic for the [Co(OH₂)₆]²⁺ system in several ways. Firstly, the pseudo-equatorial ligand atom positions lead to a substantial cobalt out-of-plane relocation (featuring a dihedral angle of approx. 6° with respect the plane) in contrast to the hexaaqua complex. Secondly, the angle of the axial ligands (denoted as β O − C o − O in Figure 6) is significantly readjusted from essentially linear to a tilted orientation of approximately 165°. The angle contraction hence correlates with the attenuated distinction of axial and equatorial bond lengths as already illustrated before.

Comparison to the biological magnesium complex is reasonably reflected in the cobalt complex since it similarly allows for geometric rearrangements that are present in the more flexible embedding of the metal ion in an augmented hexacoordinate ligand sphere with elongated bonds as compared to the hexaaqua situation.

The MD simulation of 1DZI (Emsley et al., 2000) has been carried out to assess the structural stability of the protein complex complementary to our DFT analysis of the transition metal center. Our findings are in line with similar MD studies by (Siebert et al., 2010; Zheng et al., 2018) The explicitly mapped coordination sphere of the cobalt ion in our fully quantum mechanical structure turns out to be well-behaved during our optimization and no expansion or compression of our model system is observed indicative of unbalanced estimation of interaction patterns induced by the chemical environment in the native complex. To analyze the MD simulation, the RMSD was determined as a function of 100 ns runtime (Figure 7A). Figure 7A shows an average RMSD of about 2 Å. Higher RMSD values are only present within the range of 27 and 79 ns. In order to determine whether higher RMSD values affect the environment of the Co²⁺ ion, the Co-ligand bond distances have been included for these times in Table 4, displaying overall good agreement with the other evaluated bond distances. To analyze where the structural fluctuations come from, the RMSF values for the individual residuals were calculated (Figure 7B). In particular, the ends of the proteins are shown to be very flexible, with the exception of the N-terminal end of the integrin, which is well stabilized by two neighboring beta sheet structures. In particular, the ends of the proteins are shown to be very flexible, with the exception of the N-terminal end of the integrin (Mol A magenta), which is well stabilized by two neighboring β-sheet structures. The amino acids Ser A153, Ser A155, Thr A221, and Glu C11 exhibit a significantly reduced flexibility as integral parts of the transition metals coordination sphere. During the simulation over 100 ns, the bond distances of the above-mentioned amino acids and two water molecules fluctuated by approx. 0.1 Å and the bond angles by approx. 10 degrees around their equilibrium position (Figures 8A,B). The average bond distances extracted from our MD simulations (up to 100 ns) the complex based on the 1DZI code from the PDB displays satisfactory agreement with our findings based on the density functional treatment of the model system for all levels of theory employed. Incidentally, the asymmetrical coordination sphere of the metal center featuring six gradually different interatomic distances rather than a perfect (Jahn-Teller-based) octahedral geometry is reproduced in both computational approaches. Our integrin-collagen model system consists of an inner layer of electronic interaction patterns inspired by the original chemical site at the crystal structure from the PDB code 1DZI by Emsley et al. (2000).

When SH-SY5Y neuroblastoma cells were treated with endo N, which specifically cleaves polysialic acid (PSA) from NCAM, neuritogenesis of these cells occurred (Figure 9A). Induction of neuronal processes, which were neurofilament M (NF-M)-positive (Figures 9B,C), was significantly inhibited by the use of an NCAM-blocking antibody (Figure 9A). Surprisingly, β1-integrin blocking antibody significantly inhibited the endo N induced effect on neuritogenesis (Figure 9A), but not at the same scale as anti-NCAM. Inhibition of endo N-induced neuritogenesis by β1-integrin blockage were not only applied in neuroblastoma cells but also in subventricular zone (SVZ) progenitor cells. Figure 9D shows that neurite outgrowth in SVZ explant cultures is induced by endo N and that addition of β1-integrin blocking antibody reduces the effect of endo N.

Our first approach for evaluating the relation between NCAM and β1-integrin in neuroblastoma cells was to immunostain cells for both molecules. We found that in the absence and presence of endo N, some NCAM colocalized with β1-integrin (Figure 10A). This colocalization was present at points where there was cell–cell contact and was independent of the presence of endo N (Figure 10B). β1-integrin was not expressed in isolated cells, whereas NCAM was expressed and was present on cell surface. In the case of cell–cell contact, NCAM was localized on the cell surface and β1-integrin was expressed in contact areas between the cells (Figures 9A, 10B).

For such cell systems it makes sense to carry out experiments with Co²⁺, Mg²⁺ and Mn²⁺ as well as Co³⁺ ions in order to identify the bio-medical role of heavy metal cations on a cell-tissue level.

The cell biological/histological data presented in this study avenues for further experimental settings in which the theoretical molecular modeling results are tested on tumor tissues obtained from neuroblastomas, oligodendrogliomas (WHO grade II-III), glioblastomas (WHO grade IV) and various metastatic cells.

It is known that cell differentiation and migration as well as tissue repair processes strongly depend on the occurrence of polysialic acids and the HNK-1 epitope which are contact-structures of the corresponding cell surfaces. The concentration of the HNK-1 epitope and the length of the polysialic acid chain on a tumor cell surface are suitable prognostic markers to determine the malignancy of such cells. Integrins which are glycoproteins and major players in cell differentiation as well as cell migration processes interact in various ways with the HNK-1 epitope and polysialic acids, thereby, influencing the fate of the cell. A number of questions are still unanswered in respect to this glycobiological processes. Especially, concerning the interaction processes on a sub-molecular level it is essential to evaluate potential tumor cell experiments in relation to the new insights provided by our DFT results. As shown already with SPR experiments it is possible to replace the Mg²⁺ cations of integrin by Co²⁺ cations. In the same way this can be performed for the integrins in tumor cell tissues. According to our molecular modeling studies in which parts of the metalloprotein integrin are analyzed even on sub-atomic size scale it has now tested by staining techniques whether there is an impact on the cells and tissues. This is of particular interest because potential therapeutic options with cobalt salts in cancer research could be a promising option and provide at least new insight in the structure–function relationship of tumor spread and migration. The answer why nature has chosen Mg²⁺ instead of Co²⁺ could then even be answered.

In histology, the glial fibrillary acidic protein (GFAP stain) is carried out to determine whether cells contain the glial fibrillary acidic protein, a protein found in glial cells (Kanakis et al., 2013). The GFAP staining is useful for determining whether a tumor of glial origin. ADAM (A Disintegrin And metalloprotease) 12 has also been implicated in the development of pathogen various forms of cancer, hypertension, liver fibrogenesis, and asthma (Estrella et al., 2009). Since collagen–integrin (Siebert et al., 2010) and collagen–ADAM interactions (Eckert et al., 2021) play a dominant role in cell experiments (Petridis et al., 2011; Martinez-Olivera et al., 2018) where Mg²⁺ is replaced by Co²⁺ we have already discussed their feasibility here.

Replacing the Mg²⁺ by a Co²⁺ cation in an integrin it is less complicated than the replacement of such cation in a complete tumor cell tissue probe which is necessary to study a potential impact on cell differentiation and migration processes. Special staining methods which are introduced in Figures 11A,B as well as in Figures 9, 10 can also be used to document the effects of these cation replacements. Nevertheless, one has to keep in mind that in a tumor cell tissue also other proteins with cations are triggering cellular processes (Kanakis et al., 2013). Figure 11A show cells of a previously diagnosed oligoastrocytoma WHO grade II. A vast majority of cells mainly from the oligodendroglioma component expresses the metalloprotease ADAM 12 which interacts with integrins in a specific way (Estrella et al., 2009; Kanakis et al., 2013). The results of our molecular model calculations clearly show that, especially, with integrins, a clear effect can be expected through the replacement of the Mg²⁺ ions by the Co²⁺ ions.