All the chemicals used for the experiments were of analytical grade, and they were directly used as received without any further purification. CaCl₂, NaOH, KH₂PO₄, Na₂HPO₄, NaCl, KCl, alendronate sodium (C₄H₁₂NaNO₇P₂·3H₂O, Adn), cyclophosphamide (C₇H₁₅C_l2N₂O₂P, Cpp), zoledronic acid monohydrate (C₅H₁₀N₂O₇P₂·H₂O, Zda), and foscarnet sodium (CNa₃O₅P, Fss) were purchased from Aladdin Industrial Co., Ltd.; 2× phosphate-buffered saline (2× PBS) was prepared by successively dissolving NaCl, KCl, Na2HPO4, and KH2PO₄ in deionized water with the concentrations of 272 mM, 5.2 mM, 16 mM, and 4 mM, respectively. The final solution has a pH value of about 7.4 at room temperature.

Theoretical calculations were performed employing the ORCA program (Neese et al., 2009) using a density functional theory (DFT)–based methodology. The geometries of all the structures were optimized with the hybrid B3LYP functional and the 6-31G (d) basis set to rationalize the interaction. Accordingly, the Gibbs free energies were further obtained with the hybrid B3LYP functional and the 6-311G (d, p) basis set.

In a typical synthesis procedure, 20 ml 2 × PBS solution was added to 20 ml of a solution containing CaCl₂ and phosphorus-containing molecules with a concentration of 33.4 mM, NaOH (0.2 M) was introduced to adjust the pH of the mixed solution to 7.4, and then the mixture was placed in a water bath at 37°C and subjected to magnetic stirring for 1 h. The products were collected via centrifugation and washed with deionized water and ethanol three times, and then freeze-dried for further characterization. Reactions were also conducted without the addition of phosphorus-containing molecules, and the products were denoted as Control CaPs. The CaP samples prepared under the regulation of Adn, Cpp, Zda, and Fss were marked as CaP/Adn, CaP/Cpp, CaP/Zda, and CaP/Fss, respectively. Moreover, the co-precipitation products formed at 10 min were also collected and denoted as CaP/And-0, CaP/Cpp-0, CaP/Zda-0, and CaP/Fss-0, respectively. Thereafter, Adn or Zda with a concentration of 3.34 mM in the mixed solution was also conducted to obtain CaP products with different drug contents, and the samples were marked as CaP/Adn-1 and CaP/Zda-1.

The dried powders (5 mg) of the as-prepared CaP/Cpp, CaP/Zda, and CaP/Fss were dispersed in 5 ml of PBS solution (pH 7.4) to reveal the release kinetics of the incorporated molecules from the corresponding CaP products. The suspensions were placed in a shaker with a constant shaking frequency of 140 rpm at 37°C. The supernatants and sediments were collected at given time points of 1, 6, 12, 24, and 48 h, respectively. The UV-Vis absorption curves of Cpp, Zda, and Fss with different concentrations were measured, and the related curves are shown in Supplementary Figures S5A,B, S6A. The UV-Vis absorption curves of the released medium at different time points were also recorded. Moreover, to detect the incorporated molecules in CaP samples, 0.25 mg of CaP/Cpp, CaP/Zda, CaP/Zda-1, and 1 mg of CaP/Fss were dissolved in 1 ml of HCl (1M), and the UV-Vis absorption curves were also recorded and shown in Supplementary Figures S5C,D, S6D.

The X-ray powder diffraction (XRD) patterns of the CaP products were carried out using an X-ray powder diffractometer (Bruker advance D8, Germany) supplemented with Rigaku D/max 40 kV and Cu Kα radiation. Fourier transform infrared (FTIR) spectra were collected via an FTIR spectrometer (Nicolet iS5, Thermo Scientific, USA). Transmission electron microscopy (TEM) micrographs of the as-prepared CaP products were taken with a field-emission electron microscope (JEOL JEM-2100F, Japan) associated with energy-dispersive spectroscopy (EDS, JED2300).

Mouse bone marrow–derived mesenchymal stem cells (BMSCs) were purchased from Cyagen Biosciences Incorporation (China) and cultured in a low-glucose Dulbecco’s minimum essential medium (DMEM (LG), Sigma Life Science), supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (PS). Sprague–Dawley rat–derived osteoblast-like UMR106 cells were obtained from the Cell Resources Center, Chinese Academy of Sciences (Shanghai), and cultured in a high-glucose DMEM (Sigma Life Science) containing 10% FBS and 1% PS. The cell incubator was set at 37°C and 5% CO₂ following the manufacturer’s instructions. BMSCs from passage 3 to 5 were used for further experiments.

BMSCs and UMR-106s were seeded in the 96-well plate with a density of 2,000 cells per well. After being sterilized under ultraviolet light for 30 min, CaP products were co-cultured with BMSCs and UMR-106s with different concentrations. CCK8 assay was applied for testing the biocompatibility and cancer-killing effects of the CaP nanocomposites. After incubation for 2 days separately, the cell viability was assessed by CCK8 assay kits (DOJINDO, Japan). The detailed values were recorded via a microplate reader (BioTek Instruments, United States) at a wavelength of 450 nm. The corresponding results documented here were presented as an average value of at least four parallel measurements. Data were analyzed by GraphPad Prism 8.0 software, and comparisons were evaluated by Student’s t ̵test and ANOVA. Only when the p value < 0.05, the results were regarded as statistical significance.

To theoretically reveal the role of phosphorus-containing molecules in the formation of CaPs, theoretical calculations were conducted employing the ORCA program (Neese et al., 2009) with the density functional theory (DFT) methodology at the B3LYP/6-31G (d) level. The optimized atomic structure and electrostatic density of phosphorus-containing molecules before and after the combination with calcium ions are shown in Figures 1A–H; the ionization state of each molecule at the pH of 7.4 was determined by their pKa values.The electrostatic density of certain oxygen atoms of the phosphate group becomes notably less positive after the combination with calcium ions, which indicates that the phosphate group and carboxyl group interact with calcium ions via the linkage of the Ca-O bond.

Gibbs free energies (G) were further obtained with the hybrid B3LYP functional at the 6-311G (d, p) basis sets to quantificationally illustrate the interaction, and the detailed data under each chemical structural formula are also shown in Figure 2. The value of delta G represents the binding energy of phosphorus-containing molecules and calcium ions. It can be seen that the Cpp has the weakest interaction with Ca²⁺, Fss³⁺ combines closest with Ca²⁺, and the binding energy of Adn⁻ and Ca²⁺ is close to that of Zda⁻. According to the simulation results, the binding energy is probably related to the ionization state, charged group, and spatial configuration of the phosphorus-containing molecules.

It is reported that the phosphorus-containing molecules compete with the phosphate ions to react with the calcium ions (Fleisch, 1998), the simulation result also provided quantitative data about how strong the interaction is between each phosphorus-containing molecules and calcium ions. For Cpp, it has the weakest interaction with calcium, and the obtained CaP/Cpp has been rarely influenced by Cpp; Adn and Zda have a stronger binding affinity with calcium, while the binding energy is still lower than inorganic HPO₄ ^2- and PO₄ ^3- (Supplementary Figure S1); thereby, the crystal growth process of CaP is disturbed or some of the inorganic phosphates are replaced by Adn or Zda during the co-precipitation process, which leads to the formation of amorphous calcium phosphate or other new composites consisting of calcium and phosphorus-containing molecules; Fss and calcium have the strongest binding energy, which is higher than that of calcium and HPO₄ ^2- but close to that of calcium and PO₄ ^3- (Figures 2E,F). Considering the pH of the reaction system, HPO₄ ^2- extensively exists in PBS (PH 7.4), Fss molecules may have an advantage over HPO₄ ^2- to bind with calcium ions, and a new complex consisting of calcium and Fss could form.

To verify the results of the theoretical simulation, Adn⁻, Cpp, Zda⁻, and Fss^3- were selected as additives to regulate the formation of CaPs. Figure 3 provides the TEM micrographs of CaP Control prepared without adding phosphorus-containing molecules (Figure 3A), suggesting a nanosheet structure which is the typical morphology of CaP in bones. Both the CaP/Adn (Figure 3B) and CaP/Zda (Figure 3D) have a nanosized spheroidal structure with a diameter of 20–50 nm, whereas the grain size of CaP/Adn is obviously larger. Supplementary Figure S3 shows the TEM micrographs of CaP/Adn-0 and CaP/Zda-0, which also indicate the morphology of nanoparticles. It is found that Cpp has limited influence on the morphology of CaP/Cpp (Figure 3C). In contrast, CaP/Fss-0 (Supplementary Figure S3E) possesses a defective nanoplate structure, which is quite different from the other CaP products, but can be transformed into a perfect nanoplate with prolonged co-precipitation time (Figure 3E).

As the XRD patterns of CaP Control and CaP/Cpp given in Figure 3F show, their related characteristic diffraction peaks are located at 2θ of ∼32° but with weak intensities and large peak width, indicating a low-crystallinity apatite phase. CaP/Adn exhibits a typical amorphous structure, while CaP/Zda and CaP/Fss exhibited totally different features within the XRD patterns, which cannot be detected in the database of Joint Committee on Powder Diffraction Standards (JCPDS). As shown in Supplementary Figure S2, the XRD patterns of CaP/Zda matches those of calcium–Zda complexes extracted from cif data provided by Freire et al. (2010), indicating that CaP/Zda is a complex of calcium and Zda. According to the simulation results, Adn and Zda have close binding energy with calcium, while Adn decreased the crystallinity of CaP, which is also extensively reported (Huang et al., 2022), and Zda and calcium formed a complex with a higher crystallinity. CaP/Ada and CaP/Zda may have different atomic arrangements due to the different spatial configuration of Adn and Zda. The DFT simulation only calculated the binding energy of calcium and phosphorus-containing molecules, while the 3D arrangement of CaP products has not been taken into consideration, and the crystal structure may not be predicted via binding energy. Thereafter, FTIR spectra were collected to further investigate the chemical structure.

As shown in Figure 4, the intense absorption peaks of CaP Control at about 1,122, 1,024, 601, and 560 cm⁻¹ are attributed to the presence of the PO₃ ^4- group. The typical features of Adn at 1,549 cm⁻¹, Zda at 1,386 cm⁻¹, and Fss at 973 cm⁻¹ also appear on the spectra of CaP/Adn, CaP/Zda, and CaP/Fss, respectively, which reveals that the CaP products in Figure 4A are assigned to organic-inorganic complexes except for CaP Control. With comparisons, the typical features of Cpp are not obvious on the spectrum of CaP/Cpp, indicating the low content of Cpp and the weak interaction between Cpp and CaP. These experimental results are consistent with those calculated by a theoretical simulation.

To further investigate the physicochemical property of the CaP products, loading capacity and drug-releasing profile of the molecules from the as-prepared CaP have been evaluated and shown in Figure 5 and Supplementary Figures S5, S6. In terms of the theoretical and experimental data, CaP/Adn and CaP/Zda are organic–inorganic complexes; meanwhile, Adn and Zda have similar binding energy with calcium, and the drug-release kinetics of CaP/Zda was examined due to the convenient detection. The R square of the absorbance-concentration curves of both Zda and Fss are above 0.999, indicating that an excellent linear fitting correlation has been obtained. Thereafter, the incorporated Zda and Fss in CaP/Zda, CaP/Zda-1, and CaP/Fss were tested and calculated; the corresponding drug-loading capacity are 175, 101, and 411 mg/g, respectively. Figures 5C,D show the release curves: the release of Zda and Fss reaches a plateau, or the release rate decreases after 24 h. CaP/Zda released about 40% of Zda after 24 h, while CaP/Fss only released 16% of Fss, suggesting a different transformation occurred in the system over 72 h. As shown in Figure 5H, after being shaken in PBS at 37°C for 72 h, the nanosized spherical CaP/Zda transformed into a similar nanosheet structure as CaP Control, indicating that Zda competes with the inorganic phosphate groups to react with calcium and the spherical morphology needs to be maintained by Zda. Meanwhile, CaP/Fss developed into a structure with a bunch of nanowires/nanorods (Figure 5I), and Fss which has a stronger binding affinity with calcium and played an essential role in the formation and transformation of CaP/Fss in the physiological environment. Moreover, Supplementary Figure S6 shows the UV-Vis absorption curves of the released medium of CaP and CaP/Cpp at different time points. Obviously, the UV-Vis absorption peak of Cpp is influenced by the dissolved CaP, and the peak intensity between dissolved CaP/Cpp and CaP are similar, revealing that the content of incorporated Cpp is quite low in CaP/Cpp. These results also explain the phenomenon that CaP/Cpp has a similar morphology to Cpp.

To evaluate the biocompatibility of CaPs with different structures, BMSCs and UMR-106 cells were co-cultured with the CaP products at a series of concentrations. As shown in Figure 6, CaP Control and CaP/Cpp exhibit excellent biocompatibility, and thus could promote the proliferation of BMSCs and UMR-106. CaP/Fss is non-toxic at a low concentration, while the biocompatibility is reduced when a concentration is set above 20 μg/ml. CaP/Adn and CaP/Zda show obvious cytotoxicity, and the cell viability is decreased with the concentration of nanoparticles increasing. The cytotoxicity of CaP/Adn and CaP/Zda is derived from the high content of Adn and Zda in the organic–inorganic complex. The nanocomplex can release phosphorus-containing drugs with degradation, as CaP is pH sensitive (Gou et al., 2016; Mi et al., 2016). Hence, Adn and Zda can realize controlled release under acid conditions(Huang et al., 2022). Since the bisphosphonates are reported to inactivate human epidermal growth factor receptors (human EGFR or HER) to perform antitumor actions (Yuen et al., 2014), CaP/Adn and CaP/Zda can be considered to be promising drug delivery systems for osteosarcoma therapy. CaP/Adn-1 (Supplementary Figure S4) and CaP/Zda-1 (Supplementary Figure S4B) loaded with fewer drugs showed better biocompatibility and are more suitable for biomedical uses.