**3. Results**

## *3.1. Producing Gold-Functionalized Mesoporous Surfaces*

Using electrochemical etching techniques described in Section 2, we produced porous silicon surfaces. Tuning the parameters of the electrochemical etching, we obtained two di fferent pore morphologies: (i) mesoporous silicon samples with a pore size that oscillates around the central value PS = 11 nm (MeP1 silicon with a pore size in the lower nanometer range) and (ii) mesoporous silicon samples with an average pore size PS = 21 nm (MeP2 silicon with a pore size in the higher nanometer range). Scanning electron micrographs (SEM) of MeP1 and MeP2 samples taken at di fferent magnifications reveal the morphology of the porous surface at di fferent scales (Figure 2a–f). Pores on the surface of MeP1 silicon are less uniformly distributed, are less dense, and result in a porosity of the sample of about P ≈ 12% (Figure 2a–c). The porosity or void fraction is a measure of the void spaces in a material, it is a fraction of the volume of voids over the total volume, expressed here as a percentage between 0% and 100%: it was calculated following the image analysis algorithms reported in the Supporting Information 1 of Supplementary Materials. Di fferently, pores on the surface of MeP2 silicon are more uniform and are more densely packed compared to the pores found in the MeP1 morphology: for this category, the porosity of the sample soars to *P* ≈ 40% (Figure 2d–f).

**Figure 2.** (**<sup>a</sup>**–**<sup>c</sup>**) SEM micrographs of MeP1 substrates at different levels of magnification, the pores are less uniformly distributed over the sample surface with an average pore size of PS ≈ 11 nm. (**d**–**f**) SEM micrographs of MeP2 substrates at different levels of magnification, the pores are uniformly distributed over the sample surface with an average pore size of PS ≈ 21 nm. (**g**–**i**) SEM micrographs of clusters of gold nanoparticles deposited over the porous sample surfaces, the particles follow the profile of the samples without occluding the pores, the average particle size is ≈8 nm.

Thus, the room potentially available to accommodate drugs or other molecules is significantly larger for MeP2 silicon than for MeP1 samples. After porosification, MeP1 and MeP2 samples were functionalized with gold nanoparticles using electroless deposition techniques. Electroless deposition is a technique that enables the reduction of metal ions on a solid surface as bulk metal without the application of external electric fields or forces [33]. Samples were treated with a solution of gold chloride and hydrofluoric acid for 3 min at 50 ◦C (Section 2). The process resulted in the homogeneous deposition of clusters of gold nanoparticles on the surface of the porous samples. SEM images of the sample surface (Figure 2d–f) and a convenient analysis of data (Supporting Information 2 of Supplementary Materials) indicate that the average diameter of the gold nanoparticles is snp = 8 nm, with a small deviation around the mean σ(snp) ≈ 1 nm. In no case do the gold nanoparticles occlude the pores, therefore, preventing the correct release of drugs of molecules from the porous matrix. Porous surfaces functionalized with gold nanoparticle were verified using atomic force microscopy (AFM). AFM imaging enabled to resolve the structure of the samples at the nanoparticle level for both MeP1 (Figure 3a) and MeP2 (Figure 3c) silicon. Fast Fourier transform of AFM data enabled to derive the power spectrum associated to MeP1 (Figure 3b) and MeP2 (Figure 3d) silicon functionalized with gold. The power spectrum reports the change of information content as a change of size in a bi-logarithmic

scale thus indicating how much of the originating complexity is maintained by changing the degree of detail of a surface [46]. Upon analysis of AFM data we found the values of roughness (Ra) and fractal dimension (Df) of the samples as RaMeP1 = 7 ± 2 nm and DfMeP1 = 2.48 ± 0.4 for MeP1, and RaMeP2 = 13 ± 3 nm and DfMeP2 = 2.15 ± 0.2 for MeP2 silicon. Thus MeP1 samples exhibit values of roughness and fractal dimension larger than the corresponding values found for MeP2 silicon. For comparison, nominally flat silicon surfaces, used as a control, have significantly smaller values of roughness (RaSi = 1 ± 0.1 nm) and fractal dimension (DfSi = 2.1 ± 0.2) (Figure 3h). The luminescence properties of porous silicon samples were verified under UV light (Figure 3e). The intense luminescence emission from MeP1 samples, compared to the low emission of MeP2 and to the no-emission from simple silicon, indicates that MeP1 samples may have—in the long tail of their pore size distribution—pores with a size smaller than 2 nm [31]. The wettability of mesoporous silicon samples was verified using contact angle measurements. Before oxidation, porous silicon samples as made exhibit a marked hydrophobicity with values of contact angles (CA) approaching 120 (Figure 3f). After treatment (Section 2), contact angle values measured on the sample surface shift to smaller values (CA ≈ 35◦) typical of a hydrophilic surface.

**Figure 3.** (**a**) Atomic force microscopy (AFM) profile of MeP1 substrates functionalized with gold nanoparticles imaged over a sampling area of 1 × 1 μm, the height values of the profile fall within the 0–25 nm range. (**b**) Power spectrum density function associated to the topography of the MeP1 substrate, the slope of the function in the linear regime is indicative of the fractal dimension of the samples. (**c**) AFM profile of MeP2 substrates functionalized with gold nanoparticles imaged over a sampling area of 1 × 1 μm, the height values of the profile fall within the 0–50 nm range. (**d**) Power spectrum density function associated to the topography of the MeP2 substrate. (**e**) Luminescence of MeP1 and MeP2 samples under UV light, compared to the light emission of silicon. Contact angle of a drop of D.I. water measured on the porous surfaces before (**f**) and after (**g**) sample oxidation. (**h**) Values of porosity, pore size, roughness, fractal dimension, and characteristic size of the gold nanoparticles of the porous surfaces determined through analysis of SEM and AFM images of samples.

## *3.2. Controlling Cell Organization on Au-Mesoporous Silicon Surfaces*

The substrates that we produced exhibit di fferent values of pore size (PSMeP1 ≈ 11 nm, PSMeP2 ≈ 21 nm), gold nanoparticles size (snp ≈ 8 nm), roughness (RaMeP1 ~ 7 nm, RaMeP2 ≈ 13 nm), and fractal dimension (DfMeP1 ≈ 2.48, DfMeP1 ≈ 2.15). To examine whether the nano-topographical characteristics of the surfaces have the ability to direct cell behavior on the substrate in a controlled way, we put in culture on both Au-MeP1 and Au-MeP2 silicon MCF-7 breast cancer cells. We then examined the topological characteristics of the networks that cells formed 24 h from seeding and we correlated them to the topography of the surface. We used nominally flat silicon substrates as a control. Figure 4 shows the spatial layout of cell-nuclei on flat silicon, Au-MeP1 and Au-MeP2 silicon imaged 24 h after culture. The initial number of cells deposited in each well for incubation was the same for all the substrates (Section 2). Fluorescence images in Figure 4 show that cells are homogeneously distributed on flat silicon surfaces, showing no preferential points of accumulation. Di fferently, cells on mesoporous surfaces form complex structures of those cells with a correlation length, cluster size, and topological characteristics that seem to vary from Au-MeP1 to Au-MeP2 silicon.

**Figure 4.** Fluorescence images of MCF-7 cancer cells over MeP1, MeP2, and silicon surfaces after 24 h from seeding.

We used image analysis algorithms and the methods of networks analysis, described in [30,40,47,48] and Section 2 of this article, to measure the characteristics of cell-networks quantitatively. Starting from the fluorescence image of a cell configuration, we segmented that image to find the cell-centers. Then, we routed cell-centers using the Waxman algorithm (Figure 5a). The algorithm selected the cell pairs to be connected basing on their distance: cells that were closer than a threshold were connected as described in Section 2. We analyzed more than 30 images per substrate. For each image, we extracted from the resulting network the number of cells in a region of interest ( *N*), the clustering coe fficient (*Cc*), the characteristic path length (*Cpl*), the small-world-ness (SW). *N* measures the adhesion strength of cells to a substrate [17]. The clustering coe fficient, characteristic path length, and small world coe fficient are a quantitative measure of the characteristics of the networks that cells form on a surface. The clustering coefficient is the ratio of active links to the total combinations of connections that cells, around a reference node, can possibly establish, averaged all over the cells of the network [37]. The characteristic path length is the mean shortest distance between nodes of network [37]. The small-world coefficient, found as a combination of the *Cc* and the *Cpl* [42], is a metric that tests whether the distance between nodes grows with the logarithm of the number of nodes in a graph: *Cpl*∝log(*N*). Typically, small-world-networks are characterized by a few clusters with a high number of elements for a cluster, the identification of small-world networks is of interest because networks with small-world-characteristics communicate more efficiently than equivalent random or ordered graphs of the same size [48,49].

**Figure 5.** (**a**) Image analysis of fluorescence images: the original images were segmented with a watershed algorithm to identify individual cells, and cell nodes were then connected using the Waxman algorithm to obtain the equivalent graph for each sample. (**b**) Values of adhering cells, (**c**) clustering coefficient, (**d**) characteristic path length, and (**e**) small-world-ness determined for the cultures of MCF-7 cell on MeP1, MeP2, and silicon surfaces 24 *h* from seeding.

After network analysis of cell images, we found that the number of adhering cells in a region of interest of 1174 × 882 μm is *N* ≈ 663 on flat silicon, *N* ≈ 884 for the Au-MeP1 substrate, *N* ≈ 544 for the Au-MeP2 substrate (Figure 5b). The maximum number of adhering cells is found for the Au-MeP1 substrate with intermediate values of roughness and higher values of fractal dimension, in line with previous reports [17,29,30,48,49]. Notably, the difference between the number of cells found on Au-MeP1 and Au-MeP2 silicon is statistically significant (*p* < 0.05). For the same sets of images, we found that the values of clustering coefficient reach a maximum for the Au-MeP1 substrate, with *Cc* ≈ 0.74, while the clustering coefficient is lower for the Au-MeP2 substrate (*Cc* ≈ 0.65), and significantly lower for simple silicon (*Cc* ≈ 0.58) (Figure 5c). At the same time, the values of characteristic path length are nearly identical for the Au-MeP1 (*Cpl* ≈ 2.43) and Au-MeP2 (*Cpl* ≈ 2.15) substrates, and they are statistically different from the values found on flat silicon with *Cpl* ≈ 8 (Figure 5d). The resulting small-world-coefficient of cell networks on mesoporous substrates is SW ≈ 1.29 and SW ≈ 1.35 for MeP1 and MeP2, respectively, while SW ≈ 0.35 for flat silicon. Thus, cell networks on nanostructured, mesoporous surfaces passed the small-world test, differently from cells on flat silicon that settle on a surface without any appreciable large- or small-scale structure.

#### *3.3. SERS Analysis of Cell Adhesion on Au-Mesoporous Silicon Surfaces*

The substrates that we produced induce cell clustering. The increased susceptibility of cells to condensate into compact structures is in turn ascribed to the intermediate values of roughness and large values of fractal dimension of mesoporous silicon compared to flat silicon substrates [17,30]. Since cell clustering and condensation is a side e ffect of the of the increased adhesive properties of a substrate [48,49], we performed a chemometric analysis of cells cultured on MeP1 and MeP2 substrates to examine whether cell adhesion molecules are preferentially expressed from cells on nanostructured surfaces. We mapped the Raman intensity of MCF-7 cells cultivated on MeP1 and MeP2 silicon functionalized with gold, compared to the same cells on silicon sputtered with a continuous layer of gold, used as a control (Figure 6a). Raman spectra of cells were acquired following the procedure reported in Section 2 36 h after seeding, that is a su fficiently long time to assure complete adhesion of cells on the substrate. In each point, the Raman maps reported in Figure 6a are proportional to the intensity of the spectra measured at 1569 1/cm, that is typical of integrins as explained in the following of this section. Raman spectra were then subjected to a principal components analysis (PCA), and the principal components resulting from the analysis were in turn clustered into groups using classical k-means algorithms. This allowed to identify in the cell under analysis of five di fferent regions, where points in a region have similar chemometric characteristics (Figure 6b). Moreover, since the principal components are sorted in order of decreasing information content and variance [50], regions in Figure 6b define the portions of the cell that exhibit the more intense and the more vibrantly varying Raman signal.

Integrins are one of four principal cell adhesion molecules families, they play a major role in the process of adhesion of cells to the extracellular matrix (ECM), and especially in tumor cells where they are overexpressed during the process of adhesion [51]. α5β1 and α3β1 are integrins specifically expressed by tumour and epithelial cells. In particular, α3β1 is overexpressed in tumours spreading in ECM with a high content of collagen and laminin, so that an elevated concentration of α3β1 is a hallmark of cell proliferation and migration [52]. While each of those adhesion molecules have their own distinctive features, nonetheless they exhibit a certain number of peaks that do not vary from spectrum to spectrum representing a fingerprint for those molecules. Those peaks are found at (i) 1126 1/cm related to C-N bond, (ii) 1175 1/cm associated to Tyrosine or Phenylalanine, (iii) 1306 1/cm attributable to amide III, (iv) 1506 1/cm related to Phenylalanine or Hystidine, (v) 1569 1/cm originating from tryptophan, (vi) 1645 1/cm due to the amide I signal [53]. The Raman analysis that we performed on cells on di fferent surfaces was enhanced by the interaction of the electromagnetic (EM) field with gold nanostructures, which amplify the Raman signal by several orders of magnitude in a SERS (surface enhanced Raman spectroscopy) e ffect [54]. SERS analysis of cells enabled the identification of adhesion markers that are otherwise inaccessible to classical spectroscopy techniques. The Raman intensity profile in Figure 6a for MeP1 and MeP2 follows a characteristic and distinguishable spatial distribution. Points in the map with higher values of Raman intensity may be indicative of the expression of integrin cell-adhesion-molecules suggesting that in those spots adhesion is established. The map relative to simple silicon with gold shows less preferential points of adhesion, indicating that smoother unstructured surfaces impair cell adhesion and proliferation. Moreover, SERS maps of cells measured on MeP1 and MeP2 substrates show a very high correspondence to the principal components distribution in Figure 6b, and especially to the first two components PC1 and PC2. Regions in the maps with greater overlap are at the borders of the cells (Figure 6a,b), with their membrane actively involved in the process of adhesion. The diagram in Figure 6c reports the loading associated to the first principal component measured for the MeP1, MeP2, and Si substrates. The loading is a statistical measure of how much di fferent frequencies contribute to a certain principal component. The curves in Figure 6c indicate that the frequency that above all is responsible for the signal is 1569 1/cm for the MeP1 substrate, while the signal content associated to 1569 1/cm is gradually weaker for the MeP2 and the simple silicon substrate. Recalling that 1569 1/cm is the distinctive frequency for the integrins, the form of the diagrams of Figure 6c suggests that cell adhesion molecules are preferentially expressed for the foremost on MeP1 substrates, followed by MeP2 substrates and by flat silicon. For each substrate we calculated the ratio *r* between the value of loading intensity measured at 1569 1/cm and the loading averaged over the entire spectral range (Figure 6d). *r* is a quantitative measure of the

relative abundance of integrins at the interface of a cell with a surface. Values of *r* much larger than one for MeP1 (*r* ≈ 4.53) and MeP2 (*r* ≈ 4.34), compared to the lower value of *r* measured on silicon (*r* ≈ 1), show that the multiscale architecture of mesoporous substrates, with meso-pores functionalized with metal nanoparticles, facilitate cell adhesion compared to flat geometries. The values of *r* determined for the MeP1 and MeP2 substrates are significantly different from that determined for silicon, with *p* < 0.05.

**Figure 6.** (**a**) Raman maps of MCF-7 cells acquired over a region of interest of 10 × 10 μm for MeP1, MeP2, and silicon surfaces, the maps show the Raman intensity measured at 1569 1/cm. (**b**) We show, for each of the considered surfaces (MeP1, MeP2, and Si), the first five principal components extracted from the Raman maps, and the spatial distribution of the principal components correlate with the Raman intensity maps previously reported. (**c**) The loading associated to the first principal component measured over MeP1, MeP2, and Si surfaces. (**d**) Ratio between the maximum and the mean intensity of the PC1 loading correspondent to MCF-7 cells cultivated over MeP1, MeP2, and Si samples.

#### *3.4. Kinetics of Drug Release from the Mesoporous Silicon Matrices*

The devices that we produced incorporate networks of nano-pores penetrating deep within their structures. We verified the capability of the device to accumulate and consequently release drug molecules over time using UV spectroscopy techniques as described in Section 2. We incubated MeP1

and MeP2 silicon substrates with the antitumor drug PtCl(O,O-acac)(DMSO) for 60 h. We used two different concentrations of the originating drug during the loading process: *c*1 = 30 μM and *c*2 = 50 μM. We then measured the release of the drug in D.I. water up to 10 days from the activation of the process. The diagram in Figure 7a shows the cumulative dose–response curves for different substrates and different initial values of the loading concentration. The dynamics of release from the MeP2 silicon system is faster compared to MeP1 silicon, consistently with the fact the pores in MeP2 silicon are larger (≈21 nm) than those contained in MeP1 silicon (≈11 nm). Wanting to approximate the curves of release with a function of time of the form *c*(*t*) = *co* + *cs*1 − *<sup>e</sup>*<sup>−</sup>*t*/τ, we found after nonlinear fitting of data the following solutions for cs and τ: (i) *c*s ≈ 4.94 μM and τ ≈ 46 h for MeP1 silicon loaded with an initial concentration equal to *c*1, (ii) *c*s ≈ 2.81 μM and τ ≈ 50 h for MeP1 silicon with an initial *c*2 concentration, (iii) *c*s ≈ 6.7 μM and τ ≈ 38 h for MeP2 silicon with an initial *c*1 concentration, *c*s ≈ 7.2 μM and τ ≈ 20 h for MeP2 silicon with an initial *c*2 concentration. *c*s is the steady state value of the concentration increment with respect to a zero reference value. τ is the time constant of the drug delivery system, i.e., the time necessary to the system to reach 66% of its final value of concentration. The values that we found for cs and τ for the different combinations of substrate morphology and initial loading concentration that we used in our study, indicate that the rate of drug release (1/τ) increases moving from MeP1 to MeP2 and, for the same substrate, it is higher for an initial higher concentration of the loaded drug. This behavior can be easily described by the first law of Fick, J = <sup>−</sup>D∂*c*/∂*<sup>x</sup>*, where the intensity of the flux (J) is proportional to the gradient of concentration from the substrate to the external environment, and the absolute number of molecules transported through the system per unit time depends on the area of the surface actively releasing the drug. In the equation D is the molecular diffusion coefficient [50]. Moreover, the values of *c*s and τ and Figure 7a indicate that the total amount of drug released in a system (*c*s) is higher for MeP2 silicon, that has a higher porosity compared to MeP1, and is higher for an initial higher concentration of the loaded drug. In the neighbor of *t* = 0, the drug release profile can be expanded in a Taylor series yielding, neglecting higher order terms of the expansion, *c*(*t*) ≈ *c*s⁄τ: this approximate formula enables calculation of the velocity of release (*v*) at the early stage of the delivery process. Using data from the model fit, we obtained *v* ≈ 0.06 μM/h (MeP1, *c*1), *v* ≈ 0.11 μM/h (MeP1, *c*2), *v* ~ 0.17 μM/h (MeP2, *c*1), v ≈ 0.37 μM/h (MeP2, *c*2). Thus, the kinetics of initial release can be varied in the 0.06-0.37 μM/h interval by changing the parameters of the process. Figure 7b displays the drug released from the systems over time, normalized to the initial concentration of the loaded drug. Values in the figure are a measure of the efficiency of the drug delivery system. Data show that the maximum efficiency of release varies between ≈0.12 for MeP1 silicon with an initial loading concentration *c*2, and ≈0.23 for MeP2 silicon with an initial loading concentration *c*1. While the efficiency of delivery is still larger for MeP2 silicon with higher values of pore size and porosity, it decreases for increasing values of initial loading concentration, possibly because for larger amounts of initial payload the losses associated to the process are also larger. Thus, one of the points of strength of this bio-chip, is that it can artificially increase the half-life of a drug. The half-life (*t* 12 ) is the time required to change the amount of a drug in the body by one-half during elimination. Drug clearance from the body is the result of elimination by renal excretion and by non-renal pathways, the latter most often represent clearance by the liver. Remarkably, the characteristic half-life—the duration of action—of anticancer drugs is, on average, small. Clinical studies and reports [55] indicate that the mean half-life of more than 140 small-molecule drugs approved for oncology indications is ≈15 h, with an even smaller value of median of about ≈5 h. The drug delivery system set-up in this study enables the active release of drugs for more than ≈50 h, depending on the configuration. Thus, the chip that we produced can possibly increase the half-life of most anticancer treatments by 300% on average.

**Figure 7.** (**a**) Absolute and (**b**) normalized release profile of the anti-tumor drug *PtCl*(*<sup>O</sup>*, *O* − *acac*)(*DMSO*) measured for MeP1 and MeP2 substrates up to 15 days from the beginning of the delivery.

#### *3.5. Timely Delivery of Drugs Impairs Tumor Cells Adhesion*

The theranostics device presented in the work has the ability to deliver over time the antitumor drug PtCl (O,O-acac) (DMSO) for several hours from the initial release. We examined the effects of a controlled release of drugs on the adhesion and growth of tumor MCF-7 cells on the mesoporous substrates. MCF-7 cells were cultured on MeP1 silicon and MeP2 silicon functionalized with gold nanoparticles and on flat silicon surfaces used as a control. Half of the mesoporous substrates used in this study were loaded with PtCl (O,O-acac) (DMSO) in a concentration of 50 μM. We then imaged cell-nuclei on different substrates and at different times from incubation. The different number of cells adhering on the substrates is the effect of a combination of factors: (i) the drug released from the system and (ii) the different nano-topographical characteristics of the substrates. Visual examination of samples reveals that already 6 h from culture the number of adhering cells on flat silicon is lower than that observed on MeP1 with gold, that is in turn different from the number of cells on MeP1 silicon loaded with drugs (Figure 8). This difference is exacerbated by time. In Figure 9a we report the

bar-chart of the number of cells (*N*) measured on different substrates 36, 48, and 72 h from incubation. *N* was estimated from more than 20 images per substrate and five technical repeats per sample.

**Figure 8.** Fluorescence images of MCF-7 cancer cells growing on Au-MeP1 substrates with and without the delivery of the antitumor agen<sup>t</sup> PtCl(O,O-acac)(DMSO), at different time frames. In the experiments, unfunctionalized silicon was used as a control.

A total of 36 h after culture, the number of cells on MeP1 and MeP2 silicon without drug is not statistically different from that measured on flat silicon, with NMeP1 ≈ 1871, *N*MeP2 ≈ 2100, and *N*Si ≈ 1951. Diversely, *N* is significantly lower for the mesoporous substrates loaded with drug, being *<sup>N</sup>*MeP1drug ≈ 675 and *<sup>N</sup>*MeP2drug ≈ 424. A total of 36 h from incubation the effect of topography is negligible compared to the release of drug.

A total of 48 h from incubation, the number of cells on MeP1 silicon without drug increases to *N*MeP1 ≈ 2928, significantly larger than the number of cells measured at the same time on simple MeP2 and flat silicon, with *N*MeP2 ≈ 1983 and *N*Si ≈ 2090. At this time step, the number of cells measured on the mesoporous substrates loaded with drug falls to *<sup>N</sup>*MeP1drug ≈ 451 and *<sup>N</sup>*MeP2drug ≈ 255, that are significantly different from the values measured for the unloaded substrates. A total of 48 h after cell culture, the improved adhesive characteristics of MeP1 silicon over simple MeP2 and simple silicon become apparent, while the delivery of drugs from MeP2 silicon achieves maximum effects.

**Figure 9.** (**a**) Number of adhering cells on Au-MeP1 and Au-MeP2 substrates under and without the influence of the antitumor agen<sup>t</sup> PtCl(O,O-acac)(DMSO), at 36, 48, and 72 h from cell seeding, compared to plain silicon used as a control. All *p* values less than 0.05 are summarized with two asterisks. <sup>E</sup>fficacy of the antitumor drug on the cancer cells at different times from the initial release, for the (**b**) Au-MeP1 (**c**) and Au-MeP2 substrates.

A total of 72 h from incubation, the number of adhering cells on MeP1 silicon with drug hits a minimum, *<sup>N</sup>*MeP1drug ≈ 362, while *N* surges to *N*MeP1 ≈ 3962 for simple MeP1 silicon without drug. These values are significantly lower and significantly larger than the control: *N*Si ≈ 2697. At this time of the analysis, the MeP2 morphology does not enhance significantly adhesion with respect to the control, with *N*MeP2 ≈ 2462, while the release of drug from MeP2 silicon still bears appreciable effects, with NMeP2drug ≈ 1355 significantly lower than the number of cells measured on flat silicon.

Thus, the efficacy of the drug delivery system depends on both of the substrate morphology and the time of the process. For each time, we calculated the efficiency of delivery as the number of adhering cells on the substrate loaded with drug, to the number of cells measured on the same substrate without drug (Figure 9b,c). We observe that the efficiency of the system gradually increases for MeP1 silicon, varying from ≈0.64 at 36 h, to ≈0.84 at 48 h, to ≈0.90 at 72 h from incubation. For this substrate, more than 90% of cells are killed compared to the same substrate in absence of drug delivery. The curve of efficiency for the MeP2 silicon is different. For this system, the efficiency of delivery is sufficiently large already 36 h from the initial release (≈0.80), it attains a maximum value at 48 h (≈0.87), to subsequently drop to ≈0.45 at 72 h from incubation. The different efficiency measured for MeP1 and MeP2 silicon reflects the different kinetics of release measured for those devices and reported in previous sections of this work.
