Design of a Functionalized Metal–Organic Framework System for Enhanced Targeted Delivery to Mitochondria

Mitochondria play a key role in oncogenesis and constitute one of the most important targets for cancer treatments. Although the most effective way to deliver drugs to mitochondria is by covalently linking them to a lipophilic cation, the in vivo delivery of free drugs still constitutes a critical bottleneck. Herein, we report the design of a mitochondria-targeted metal–organic framework (MOF) that greatly increases the efficacy of a model cancer drug, reducing the required dose to less than 1% compared to the free drug and ca. 10% compared to the nontargeted MOF. The performance of the system is evaluated using a holistic approach ranging from microscopy to transcriptomics. Super-resolution microscopy of MCF-7 cells treated with the targeted MOF system reveals important mitochondrial morphology changes that are clearly associated with cell death as soon as 30 min after incubation. Whole transcriptome analysis of cells indicates widespread changes in gene expression when treated with the MOF system, specifically in biological processes that have a profound effect on cell physiology and that are related to cell death. We show how targeting MOFs toward mitochondria represents a valuable strategy for the development of new drug delivery systems.

S3 microscope. The microscope was equipped with 405 diode, argon and HeNe lasers. Leica LAS AF software was used to analyze the images.

Super-Resolution Microscopy:
Measurements were carried out using a custom-built 3-color Structure Illumination Microscopy (SIM) setup that has previously been described 1 . The structured illumination patterns were generated by a spatial light modulator (SLM: SXGA-3DM, Forth Dimension Displays). A 60x/1.2NA water immersion lens (UPLSAPO 60XW, Olympus) focused the structure illumination pattern onto the sample. This lens also captured the samples' fluorescence emission light, which was imaged onto a sCMOS camera (C11440 Hamamatsu). Laser excitation wavelengths used were 488 nm (iBEAM-SMART-488, Toptica), 561 nm (OBIS 561, Coherent), and 640 nm (MLD 640, Cobolt), to excite the fluorescence emission of MOF, mitochondria, and DNA, respectively. Images were acquired using custom SIM software previously published 1 . Nine raw images were collected at each plane and recombined using a custom implementation of fairSIM 2 .

S2. Materials and Synthesis
All reagents unless otherwise stated were obtained from commercial sources and were used without further purification. The synthesis of UiO-66 -[Zr6O4(OH)4(C8H4O4)x]nwas adapted from a literature procedure 3 .

UiO-66 Synthesis
UiO-66 was synthesised by adaptation of a literature procedure to include different modulators as follows. For all samples, after cooling the reaction mixture, particles were collected by centrifugation (4500 rpm, 15 minutes), and washed (sonication centrifugation cycles) with fresh DMF (x1) and MeOH (x3). The NMOFs were dried for at least 24 hours under vacuum before analysis.

Calcein Loading
10 mg of either DCA5-UiO-66 or TPP@(DCA5-UiO-66) were dispersed in 5 mL of a methanolic solution of calcein (1 mg/mL) and left to stir at room temperature for 1 day. For TPP@(DCA5-UiO-66), the loading solution also contained 1 mg/mL of TPP in order to make sure that the calcein did not completely replace the TPP at the external surface. The MOFs were harvested and dried as described previously.  c.

b.
S7 allowed to dry in the oven at 60 ºC for 5 minutes. Generally, increasing the concentration of DCA gives smaller particle sizes. For particles synthesized without TPP, increasing the concentration of DCA from 2.5 to 10 eq decreases particle size from 139 to 81 nm. Modulators of MOF crystal growth act by competing with the organic linker for coordination to the zirconium nodes. They have two antagonistic and co-occurring effects, promoting crystal growth by preventing nucleation through coordination to zirconium in solution and limiting crystal growth by capping zirconium nodes on the surface of already formed particles. Within the range of concentrations used, it seems that the capping effect of DCA is more dominant than the nucleation-preventing one. All the particles are within the correct size range for cellular uptake.

S8
To determine the incorporation of DCA and TPP in the structure, we performed an 1 HNMR.  TGA was also performed to estimate the amounts of DCA and TPP in the structure. Figure S4 shows the TGA profiles of pristine UiO-66, DCA5-UiO-66, TPP@(DCA5-UiO-66) and free TPP. The  Table 1, Main Manuscript), by measuring the phosphorus and chlorine contents respectively. The DCA content of DCA5-UiO-66 as determined by TGA is ca. 12 w/w %; slightly higher than values obtained by ICP-OES (ca. 10 w/w %), possibly due to solvent mass loss increasing DCA content upon determination by TGA, but still within the same range. FT-IR was then performed to confirm the hypothesis that TPP is attached to the metal nodes.

S4. Control experiments. Cytotoxicity of MOF, drug and targeting agent
In order to investigate whether the MOF rather than the drug is toxic to cells when directed to mitochondria, UiO-66 with benzoic acid as a modulator was synthesised, TPP was attached to its surface (TPP@UiO-66), and the toxicity of the material was assessed using the MTS assay. The results are shown in Figure S7. For all DDS concentrations, the viability of MCF-7 cells decreased to the same level when incubated with both UiO-66 and TPP@UiO-66, indicating that the addition of TPP does not increase the toxicity of the MOF when it is not loaded with DCA. This supports the hypothesis that the increase in efficacy of the DDS seen in Figure 1 is due to the DCAor a synergistic effect of the MOF and DCAand not the MOF alone.

Bioinformatics
GCCN-STT correction was applied to the raw data using the Affymetrix powertools in order to correct probes for GC content bias and adjust the dynamic range of the array to give better spread of intensity in order to identify differentially expressed genes. After scanning the files generated by the scanner (CEL files), GCCN-STT corrected files were loaded in R using the oligo package from bioconductor. 7 No background correction or normalisation was applied at this stage. In order to assess the quality of the data, plots of the control probes were generated along with boxplot, MAplot and intensity distribution plot. Variation within biological replicates was also investigated using clustering methods. The raw data was then processed using the Robust Multichip Analysis (RMA) method. 8 The data was background corrected, normalised using quantile, and summarized.
Once the data was processed, the comparisons were performed using the limma package 9 and the results corrected for multiple testing using False Discovery Rate (FDR) 10 .

Gene Ontology (GO) analysis:
GO analysis was performed by submitting gene lists to online tool David Bioinformatics Resources 6.8 (https://david.ncifcrf.gov/).

Functional protein association networks:
Free online tool STRING was used to identify and predict functional protein association networks. 11 For network analysis in this study we used 2 shells of interactors (proteins) to identify a functional network of proteins. 1 st shell of interactors was from the protein lists significantly identified by our microarray analysis. 2 nd shell of interactors, connecting to the 1 st shell of proteins, was automatically identified by STRING database based upon experimentally established or predicted associations between a pair or group of proteins.