*2.4. AFM Sample Characterization*

Sample nanotopography was verified using atomic force microscopy (ICON Atomic Force Microscope, Bruker, Coventry, UK). We measured the surface profile over a sampling area of 1 × 1 μm2, in a dynamic tapping mode in air. All measurements were performed at room temperature. During image acquisition, the scan rate was fixed as 0.5 Hz, while images were discretized in 1024 × 1024 points. We used Ultra-sharp Si probes (ACLA-SS, AppNano, Mountain View, CA, USA) with a nominal tip radius less than 5 nm to assure high resolution. Multiple measurements were done

in di fferent scan directions to avoid artefacts. At least four images were recorded per sample to reduce uncertainty. After acquisition, images were analyzed using the methods developed in [17] to determine the average surface roughness (Ra) and fractal dimension (Df) for each sample.

## *2.5. Contact Angle Characterization of Samples*

The wettability of the samples was verified using an automatic contact angle meter (KSV CAM 101, KSV Instruments Ltd., Helsinki, Finland). A drop of 5 μL of D.I. water was gently positioned on the sample surface at room temperature. After 5 s from deposition, the contact angle of the drop at the interface with the substrate was measured.

## *2.6. MCF-7 Cell Culture and Staining*

Breast carcinoma MCF-7 cells were grown on the porous silicon surfaces. Cultures were carried out at 37 ◦C in a humidified 5% CO2/air atmosphere in a Dulbecco's modified eagle's medium (DMEM, Euroclone) supplied with 10% heat-inactivated fetal bovine serum (Euroclone, Pero (Mi), Italy), streptomycin (0.2 mg/mL) and penicillin (200 IU/mL). When cells on the petri dishes reached 90% confluence, they were dissociated: medium was removed and MCF-7 were treated with a solution of 0.25% Trypsin-0.53mM EDTA (Euroclone) for about 5 min at 37 ◦C. Trypsin was deactivated by adding medium and completely removed after centrifugation of the cell suspension (1300 rpm, 5 min, 18 ◦C). Then, trypsin/growth medium solution was removed. Single sterilized porous Si wafer specimens with and without loaded drugs, having a size of around 15 × 15 mm, were individually placed into each well of a 6-well plate (Corning Incorporated) and washed with phosphate-bu ffered saline solution (PBS, Invitrogen). After that, cells were seeded in complete cell medium and cultured up to 15 days in a humidified incubator at 37 ◦C with 5% of CO2. After the incubation period, cell culture medium was removed and the MCF-7 cells were washed twice in PBS, fixed with 4% PFA (paraformaldehyde), and left for 30 min at room temperature (RT). Subsequently, cells were washed twice in PBS and permeabilized with 0.05% triton (Invitrogen, Milano, Italy) for 5 min at RT. Fixed and permeabilized cells were stained with 100 μL DAPI (40, 6-Diamidino-2-phenylindole, Sigma Aldrich, Milano, Italy) solution for 10 min at 4 ◦C in dark environment. In the end, the DAPI solute ion was removed and each sample was washed with PBS. The total number of cells ntot initially seeded in each well for incubation was approximately ntot ≈ 105. Cells were sub-confluent for the duration of the experiment. MCF-7 cells were chosen as a cell-model because they are characterized by a moderate expression of the integrins. As previously reported [35], the change in the intensity and type of expression of integrin is the basis of the cancer disease progression. Notably, MCF-7 are a secondary cell line. The choice of another cell line, perhaps a primary cell line, while could possibly enhance clustering, may not have—at the same time—the same e ffect on the expression of integrins in the system.

## *2.7. Imaging Cells on the Substrates*

An inverted Leica TCS-SP2 ® laser scanning confocal microscopy (Wetzlar, Germany) system was used to image cells adhering on the substrates. All measurements were performed using ArUv laser (Leica, Wetzlar, Germany). The pinhole was set to ≈ 80 μm (1.5 Airy units) and the laser power to 80% of the maximum, these values of the parameters were maintained constant throughout each acquisition. Confocal images of blue (DAPI) fluorescence were acquired using a 405 nm excitation line and a 10× dry objective, so that several cells could be simultaneously imaged in the region of interest, that was of 1174 × 882 μm2, resulting in a pixel size of ≈ 1.72 μm. For each substrate, a large number of images was taken for statistical analysis. Each image was averaged over four lines and 10 frames to reduce noise. Images were acquired with a resolution of 1024 × 768 pixels, and were exported to a computer for processing and analysis.

#### *2.8. Image Analysis and Topological Characteristics of Cell Networks*

Confocal images of cell nuclei stained with DAPI were analyzed with Matlab ® to extract the cell positions. Images were preprocessed to enhance contrast and low-pass filtered to remove constant power additive noise. Each image was partitioned into k = k di fferent segments (going gradually from bright, k = 1, to dark, k = k) by k-means segmentation algorithms [30]. The information content of the image was thus associated with a gray level k = t and all the segments brighter than a certain threshold t were considered as background and shifted to 0. The values of the remaining segments, representing the cells, were shifted to 1. After that, the resulting image (*g*) was downsampled: if *f* was the average operator, *f* was shifted over *g* by steps of size *r*, where *r*2 was the expected area of a cell nucleus in pixels. The pixel intensity (ranging from 0 to 1) of the resulting image indicated the probability that a pixel is a cell. If this probability is greater than a threshold, that pixel is considered being a node of the graph in a bi-dimensional grid. At this point, the links between the nodes can be established using the Waxman model [36], according to which the probability *P*(*<sup>u</sup>*,*<sup>v</sup>*) of being a connection between two nodes *u* and *v* exponentially decays with their Euclidean distance *d*. If *L* is the largest Euclidean distance:

$$P(\mathfrak{u}, \mathfrak{v}) = \alpha \mathfrak{e}^{-d(\mathfrak{u}, \mathfrak{v})/\mathfrak{\beta}\mathfrak{L}} \tag{1}$$

where *d* is the Euclidean distance between nodes *u* and *v*, and *L* is the largest possible Euclidean distance between two nodes of the grid. In the equation, α and β are the Waxman model parameters. According to α and β, which have to be chosen between 0 and 1, the density of links in a graph changes. In particular, low values of these parameters result in a low number of connections. For our study, α = 1 and β = 0.025. The probability *P* ranges between 0 (if the distance between the pair of nodes is ideally infinite) and 1 (if the distance between the pair of nodes is zero). The connectivity information of a graph is described in the adjacency matrix A. A is a square matrix in which each element *aij* indicates whether two nodes *i* and *j* are connected (*aij* = 1) or not (*aij* = <sup>0</sup>). In the analysis, diagonal elements were all zero, since links from a node to itself were not allowed. Moreover, graphs were considered being undirected, so that information could bidirectionally flow from *i* to *j*. As a consequence A was symmetric and *aij* = *aji*. As the Euclidean distances *dij* in the networks were extracted, we could decide if a pair of nodes is connected by the subsequent formula

$$
\alpha e^{-d\_{i,j}/\beta L} - R \ge 0 \tag{2}
$$

in which *R* is a constant that we chose as 0.1 so that the probability of being a connection is *P* = 0.9. Once established the connections between the nodes, the network parameters including *c*lustering coe fficient, characteristic path length, and small-world-ness can be extracted. The definition and significance of these terms may be found in influential textbooks [37] and papers [38–41]. Once obtained the *Cc* and *Cpl* values, we found a precise measure of 'small-world-ness', the 'small-world-ness' coe fficient (SW), based on the trade-o ff between high local clustering and short path length as described in [42]:

$$\text{SW} = \frac{\text{C}c\_{\text{graph}}}{\text{C}c\_{\text{rand}}} / \frac{\text{C}pl\_{\text{graph}}}{\text{C}pl\_{\text{rand}}} \tag{3}$$

where *Cc*grap<sup>h</sup> and *Cpl*grap<sup>h</sup> are the clustering coe fficient and the characteristic path length of the graph *G* under study, and *Cc*rand and *Cpl*rand are the equivalent values for a random Erdös-Rény<sup>i</sup> graph with the same number of nodes and edges of *G*.

## *2.9. Raman Analysis of Samples*

MCF-7 cells fixed on the Au-mesoporous sample surface were analyzed by a WITec Raman microscope Alpha300 AR equipped with a 50×/0.7 N.A. (Numerical Aperture) objective. The signal was excited by a 633 nm laser, set to a power of 1 mW. For each sample, SERS (Surface Enhanced Raman Spectroscopy) maps of a portion of a cell were acquired in the *<sup>x</sup>*-*y* plane with a 0.5 μm stepsize, with the aim to analyze the SERS spectra coming in particular from the cell membrane, searching the di fferent biochemical composition of each point to evidence the possible presence of adhesion proteins [43]. The Raman spectra were collected in the spectral range from 700 to 3250 1/cm, with an integration time of 1 s.

## *2.10. Principal Components Analysis of Raman Spectra*

Spectra were pre-processed to minimize the e ffect of the fluorescence of samples by normalization on the total spectrum area. A principal component analysis (PCA) and a clustering analysis were performed on the spectral collection to highlight the chemical di fferences between the membrane's points [44]. The first five principal components (PCs) accounted for nearly 90% of the total spectral variation and they were then used to implement the clustering analysis by the Kmean method, imposing a number of five classes. All the pre-processing steps, the PCA, and the clustering analysis were carried out using the free software package Raman Tool Set (available on http://ramantoolset.sourceforge.net).

## *2.11. UV Characterization of Drug Release*

To assess the drug-delivery capability of the device we verified the release over time of the anti-tumor drug PtCl(O,Oˆ-acac)(*DMSO*). PtCl(O,O-acac)(*DMSO*) is a platinum(II) complex containing acetylacetonate (*acac*), characterized by a high toxicity both in immortalized cell lines, as human cervical carcinoma (HeLa) cells or human breast cancer (MCF-7) cells, and in primary cultured human breast epithelial cells [45]. In this work, we incubated the mesoporous silicon samples in 30 μM/50 μM solutions of PtCl(O,O-acac)(DMSO) in D.I. water for 60 h to load the drug. The release kinetic was tested over time up to 15 days, by immersion of the loaded sample in DI water and monitoring the drug concentration at di fferent time, through the analysis of drug in solution by a spectrophotometer UV/Vis (LAMBDA 25 UV/Vis PerkinElmer), after standard calibration procedures of the samples.
