Internalization Characterization of Si Nanorod with Camouflaged Cell Membrane Proteins Reveals ATXN2 as a Negative Regulator

The fabrication of shape-controlled nanocarriers is critical for efficient delivery of biomolecules across the cell membrane. Surface coating of the nanocarrier can improve internalization efficiency. Here, we developed a facile method of silicon nanorod fabrication leading to a controlled size and shape. We then systematically evaluated five surface modifications with membrane proteins from different cancer cell lines including MCF7, MD231, Hela, Panc-PDX, and Panc-1. We demonstrated that silicon nanorods coated with either a homolytic or heterolytic membrane protein coating have significantly improved internalization efficiency as compared with uncoated Si nanorods. To elucidate the molecular mechanism of the improved efficiency associated with a modified coating, we analyzed the coating membrane proteins derived from five cell lines with proteomics and identified 601 proteins shared by different cell sources. These proteins may function as cell-substrate adhesion molecules that contribute to the enhanced internalization. We also tested the internalization efficiency of nanorods with different coatings in each of the five cell lines to determine the influencing factors from target cells. We found that the internalization efficiency varied among different target cells, and the ranking of the average efficiency was as follows: Hela > Panc-PDX > MD231 > MCF7 > Panc-1. The bioinformatics analysis suggested that the low internalization efficiency in Panc-1 cells might be associated with the upregulation of ATXN2, which is a negative regulator of endocytosis. We further demonstrated that ATXN2 knockdown with specific siRNA significantly improved nanorod internalization efficiency in Panc-1 cells suggesting that ATXN2 can be a reference for efficiency prediction of nanoparticle delivery to tumor cells. Thus, we studied the effect of different cancer cell membrane proteins on nanorod uptake efficiencies. These results can improve nanorod internalization to cancer cells, including a fundamental understanding of the internalization efficiency of cancer cells.


Introduction
Nanoparticles have been used therapeutically in various diseases [1,2]. The interaction between nanoparticles and target cells is critical to the understanding of the internalization efficiency in cells and tissues. The physicochemical properties of nanoparticles, such as shape, size, and surface modification, often determine the internalization dynamics [3,4]. For example, particle shape and size can modulate the internalization efficiency of nanoparticles [5]. Surface modification can also promote the nanocarrier

Fabrication of Nanorods
The silicon nanorod arrays were fabricated via a metal-assisted chemical etching method [39]. A P-type (100) oriented silicon wafer (electrical resistivity, 10-20 Ω·cm) was used in this research. The wafers were first ultrasonically cleaned in acetone and ethanol to remove surface contamination. The wafers were then immersed in boiling piranha solution (3:1, v/v, H 2 SO 4 /H 2 O 2 ) for 1 h to remove all organic contamination followed by a thorough wash with ultrapure water. Subsequently, monolayer polystyrene (PS) spheres (500 nm diameter) were coated on the silicon wafer via a self-assembly procedure. The PS spheres were then decreased to 400 nm dimeter via oxygen plasma etching (5 sccm oxygen, 100 W power) for 10 min. Thereafter, a 5 nm titanium layer was electron beam evaporated on the PS monolayer followed by 30 nm of silver thermal evaporated on the titanium layer. The PS spheres were removed by ultrasonic cleaning in ultrapure water for 1 min, and then the wafer was immersed in an etching solution (4.6 M HF and 0.44 M H 2 O 2 solution) for 10 min to get nanorods array. Silver was removed by HNO 3 for 1 h, and the organic residual was removed by chloroform for 1 h. Finally, the nanorods were scraped off the silicon wafer via a surgical blade and dispersed in ultrapure water for use. Some nanorods were further amine coated for membrane coating. Here, the wafer was first oxidized by boiling piranha solution (3:1, v/v, H 2 SO 4 /H 2 O 2 ) for 90 min. Then, nanorods were stepwise cleaned with DI water, methanol, methanol/dichloromethane (DCM) (3:1, v/v), and dichloromethane. Subsequently, nanorods were immersed in (3-aminopropyl) triethoxysilane (APTES, 20% in dichloromethane, v/v) and reacted for 12 h in nitrogen atmosphere. The nanorods were then sequentially cleaned with ethanol, isopropanol, and DI water and then nitrogen flow dried. The amino-modified silicon nanorods (SiNR-NH2) can be stored dry at 4 • C for at least one month.

Characterization of SiNRs by SEM and TEM
The Si nanorod arrays were observed using scanning electron microscopy (SEM) (JSM-7800F, JEOL, Tokyo, Japan). The coated proteins of SiNRs were examined with transmission electron microscopy (TEM) (JEM-1400plus, JEOL, Tokyo, Japan). Briefly, SNR aliquots (2.5 µL) were transferred onto 200 mesh Cu grids, and the excess moisture was removed with filter paper after 3 min of deposition. The grid was subsequently soaked in 1% uranyl acetate solution (Beijing Zhongjingkeyi Technology Co., Ltd., Beijing, China) for one minute, and the solution was then removed with filter paper. Finally, the grid was dried in a vacuum overnight for TEM characterization. The absorption spectra were characterized with UV-vis spectrometry (U3010, TECHCOMP LTD, Tokyo, Japan). The hydrodynamic size was measured on a nanosizer (NanoZS, Malvern, UK).

Cell Culture
The Panc-PDX cells were primary cells from a pancreatic cancer patient-derived xenograft mouse model (a gift from Dr. Wentao Dai in Shanghai Center for Bioinformation Technology, Shanghai Industrial Technology Institute). Panc-1, MCF-7, MDA-MB-231, and Hela were established cancer cell lines obtained from the cell bank of the Chinese Academy of Sciences (Shanghai, China). All five cell lines were maintained in Dulbecco's modified Eagle medium (DMEM) with 10% fetal bovine serum (FBS) without antibiotics.

Cell Membrane Protein Extraction
For the cell membrane protein extraction, all cell types were seeded in 15 cm plates. When they reached confluence, cells were harvested using a scraper in cold PBS and isolated by centrifugation at 700 g for 5 min. The supernatant was aspirated, and the cell pellets were frozen at −80 • C. The ProteoExtract ® Native Membrane Protein Extraction Kit (Millipore, USA) was employed following the manufacturer's instructions to obtain nondenatured functional membrane proteins. In brief, the cell pellet was washed twice with the washing buffer, and then incubated with ice-cold Extract Buffer I at 4 • C for 10 min under gentle agitation. The pellet was centrifuged at 16,000 g for 15 min (4 • C). The supernatant was discarded, and 1 mL ice-cold Extract Buffer II was added to the pellet. This membrane protein extraction step was allowed for 30 min at 4 • C under gentle agitation. Then the supernatant was collected after centrifugation at 16,000 g for 15 min at 4 • C. The membrane extracts were characterized with a BCA Protein Assay kit (Takara, Dalian, China) and then stored at −80 • C for SiNR coating.

LC-MS/MS
The proteins were precipitated with trichloroacetic acid solution (TCA, 6.1 N). The pellet was subsequently dissolved in 8 M urea and 100 mM Tris-HCl, pH 8.5. TCEP (final concentration of 5 mM, Thermo Scientific, Waltham, MA, USA) and iodoacetamide (final concentration of 10 mM, Sigma-Aldrich, St. Louis, MO, USA) were added to the solution and incubated at room temperature for 20 min and 15 min for reduction and alkylation, respectively. The protein mixture was diluted by a factor of four and digested with Trypsin at 1:50 (w/w) (Promega, Madison, WI, USA). The reaction was stopped by formic acid (FA), and the peptide mixture was desalted by monospin C18 column (SHIMADZU-GL, Tokyo, Japan).
The acquired MS/MS data were analyzed against UniProt Homo sapiens database (http://www. uniprot.org/) by Sequest algorithm integrated in the Protein Discoverer software (Thermo Scientific). A decoy database containing the reversed sequences of all the proteins was appended to the target database to accurately estimate peptide probabilities and false discovery rate (FDR); the FDR was set at 0.01.

GO Enrichment Analysis
The proteins identified from the MS data were further filtered as the parameter 'unipeptide' greater than or equal to 1. The Venn plot analysis were calculated by jVenn online tool [110]. The bioinformatics processes were carried out using R and Bioconductor tools. Briefly, the ID mappings were based on R package "org.Hs.eg.db" and the Gene ontology (GO) cluster and enrichment analysis were performed by R package "clusterProfiler" [111].

Membrane Protein Coating and Labeling
To fabricate membrane protein coated silicon nanorods, SiNR-NH 2 was firstly reacted with thiophosgene in (1% in acetonitrile (MeCN), v/v) for 2 h at room temperature. The phosgene-modified silicon nanorods, SiNR-NCS, can be stored dry at 4 • C for several months. After phosgene modification, the nanorods were immersed in a membrane protein solution (0.2 mg/mL in PBS) for 12 h at 4 • C on a shaker. Protein-modified nanorods were labeled with 5-carboxyfluorescein N-succinimidyl ester (5-FAM-SE, 1 mg/mL in 0.1 M NaHCO 3 solution, ex/em = 494/522 nm) for 2 h at room temperature. Finally, the membrane protein-coated SiNRs were washed with PBS at least 3 times. Fluorescence-labeled SiNRs without membrane protein was used as a control. All labeled nanorods were transferred to PBS and sonicated for 3 min to suspend the SiNRs into solution. The final working medium concentration of SiNR was about 2.0 × 10 7 nanorods per milliliter.

Quantitative Analysis on Nanorod Uptake
The cells were seeded in a 24-well plate at a 5 × 10 4 seeding density. The SiNRs (final working concentration~2.0 × 10 7 per milliliter) were added and incubated for 6 h. The live cells were stained with Alexa-594-conjugated WGA and Hoechst for 20 min and then detached by 3 mM EDTA in PBS followed by 4% PFA fixing for 15 min at 37 • C.
Imaging flow analysis was completed using an Amnis ImageStream X Mark II. Channel 0 bright field, SSC, Laser 405, 488, and 561. Cell samples were analyzed at a low speed for best sensitivity. The gating strategy is shown in Figure S4A. The data analysis was performed using Ideas 6.2 and Flowjo VX. To validate the cell-type specific internalization efficiency, we used the statistical module in Ideas software "spot count" to make an accurate algorithm of internalization. The internalization efficiency is the ratio of the fluorescence intensity inside the cell to that of the entire cell, i.e., internalization% = (cell count with indicated internalization level)/(total cells). This metric was binned to three levels: low (0 < spot count ≤ 3, blue), middle (3 < spot count ≤ 10, yellow), and high (spot count >10, red) (see Figure S4B).

Immunostaining, EGF Binding Assay, Colocalization Assay, and Confocal Microscopy
The cells were seeded on collagen-coated cover slides in a 24-well plate. The internalization protocol was as described above. The 4% PFA fixed cover slides were mounted with antifade reagent (Thermo, Waltham, MA, USA).
For the colocalization assay, the cells were seeded on collagen-coated cover slides in a 24-well plate. The internalization was examined as described. The cells were then incubated with Lyso-Orange (1:500 dilution, ab176827, Abcam, Cambridge, England) for 20 min combined staining with Alexa-633-conjugated WGA and Hoechst. The 4% PFA fixed coverslip was mounted onto a slide with antifade reagent (Thermo, U.S.).
Photographs were performed using a Leica SP8 laser confocal microscope at the National Center for Protein Science in Shanghai.

Total RNA Isolation and qRT-PCR Analysis
The total RNA was extracted using a Total RNA Isolater (Vazyme, Nanjing, China) according to the manufacturer's instructions. The RNA was collected, and 1 µg of RNA was reverse transcribed into cDNA with HiScript II Q RT SuperMix (Vazyme, Nanjing, China) according to the manufacturer's instructions. The real-time PCR analysis was performed using the LightCycler (Roche, USA) system with ChamQ SYBR qPCR Master Mix (Vazyme, Nanjing, China). The primer sequences for qRT-PCR procedures are listed in the Supporting Information Table S3.

Small Interfere RNA Transfection and Chemical Inhibitors Studies
The ATXN2-targeting siRNA and negative control siRNA were purchased from Genepharma (Shanghai, China). The detail sequence is shown in Table S4. The transfection of siRNA was performed using INTERFERin (Polyplus, USA) according to the manufacturer's protocol. Briefly, 2.5 × 10 4 Panc-1 cells per well were seeded in 12-well plates and transfected the next day with a 100 nM final concentration of siRNA using 2 µL INTERFERin. The cells were harvested 48 h after transfection for real-time PCR analysis and western blot assay.
For the imaging cytometry and chemical inhibitors assay, Panc-1 cells were seeded (8 × 10 4 cells) in 6 cm tissue culture plate and then allowed to adhere overnight. The cells were transfected the next day with a 100 nM final concentration of siRNA using 20 µL INTERFERin. The cells were cultured 48 h before the internalization assay. Inhibitors were used and added to the cells at the concentrations described in Figure 8E. The cells were incubated with Panc-PDX@SiNR and indicated inhibitors for 6 hrs. The internalization protocol was as described above.

Statistic Method
The data were expressed as mean ± SD based on at least three independent experiments. Statistical analysis used the Student's t-test. The differences were significant with a p value <0.05.

Fabrication of Size-Controllable SiNR Array
The SiNR fabrication techniques can be classified into the following two categories: bottom-up and top-down techniques [40]. The bottom-up method originated from the vapor-liquid-solid mechanism of Si nanorod growth [41], but its scale-up capacity is very limited. Top-down fabrication uses an etching process such as wet chemical etching or reactive ion etching. On the basis of previous methods [13,39], we introduced a facile chemical etching method for SiNR fabrication. Figure 1A shows that a single layer of polystyrene (PS) spheres is formed as a template via self-assembly on the silicon substrate. The following oxygen plasma etching process is used to shape the diameter of the PS spheres: a 5 nm titanium film and 30 nm silver film are sequentially e-beam processed and thermal evaporated onto the silicon substrate as the catalyst. A Ti/Ag film with a hexagonal array of holes is formed due to the PS monolayer mask. Subsequently, an etching step is conducted in a mixture of deionized water, HF, and H 2 O 2 . The Ag film catalyzes the etching of silicon beneath it. During the etching process, the "walls" of the honeycomb are gradually etched away, and the remnant silicon forms a nanorod array. Finally, the silver film and PS spheres are eliminated by HNO 3 and chloroform, respectively. Figure 1B,C show SEM images of silicon nanorod arrays from a tilted view (ca. 45 • ) and a plane view, respectively. The length of the fabricated nanorods is 3.495 ± 0.108 µm, and the diameter is 0.390 ± 0.009 µm. The results indicate that the nanorod distributes uniformly with a homogenous length and diameter.
Both the diameter and the length of the SiNR are key parameters that control size. The diameter of the SiNR is determined by the size of the PS sphere mask. The etching time is a precisely-controlled parameter ( Figure 1D). Therefore, we studied the change in the diameter of the PS spheres as a function of oxygen plasma etching time. In Figure 1D, the left panel shows selected SEM images of the PS spheres after various etching times. The right panel shows three parallel samples per measurement for a total of 10 time points (0-45 min); linear regression was performed to verify the correlation between the etching time and the PS sphere diameter. The modeling result is significant: The R-squared value reaches 0.9932, and the p-value is <0.0001. These metrics demonstrate that the diameter of the PS sphere can be precisely controlled via etching time.
The length of the SiNR arrays is also related to the duration of etching after silver film evaporation. Figure 1E shows cross-sectional SEM images of SiNRs fabricated with etching times of 5, 10, 15, and 20 min yielding nanorods with lengths of ca. 3.67, 7.05, 10.5, and 12.99 µm, respectively. The length of the nanorods varies linearly with the duration of the etching process and provides good control over the length of the nanorod arrays. This method is easy and practical for SiNR size control. etching process, the "walls" of the honeycomb are gradually etched away, and the remnant silicon forms a nanorod array. Finally, the silver film and PS spheres are eliminated by HNO3 and chloroform, respectively. Figure 1B and C show SEM images of silicon nanorod arrays from a tilted view (ca. 45°) and a plane view, respectively. The length of the fabricated nanorods is 3.495 ± 0.108 μm, and the diameter is 0.390 ± 0.009 μm. The results indicate that the nanorod distributes uniformly with a homogenous length and diameter.  Three sizes of SiNR were fabricated to investigate the SiNR internalization. Detailed morphology of SiNRs was observed by SEM (Figure S1F-H). As shown in Table 1, the large size (L-size) has the longest length, 3.495 ± 0.181 µm, and the middle size (M-size) is 2.229 ± 0.049 µm long. The L-size and M-size have the same diameter (0.409 ± 0.010 µm). The small size (S-size) is 0.105 ± 0.011 µm wide and 0.656 ± 0.126 µm long and its volume is about 1/54 of the M-size SiNR.

Nanofabrication of SiNR Coating with Membrane Protein
Different physiochemical factors influence the nanoparticle internalization such as surface charge, shape, size, and bioconjugate moiety [4,5,14,[42][43][44][45][46][47][48][49][50][51]. After preparing SiNR arrays of three different sizes, the SiNR were obtained by carefully scraping them off the array. We subsequently used surface modification via a cell membrane protein. Figure 2A shows the scheme to fabricate SiNR coatings with the same amount of cell membrane protein (see Materials and Methods Section 2.9). Five different cell membranes were used to coat the SiNRs using the same total protein starting concentration.

Proteomics Analysis of Coating Membrane Proteins Derived from Five Cancer Cell Lines
Using cell membrane proteins as a surface grafting moiety can be a practical internalization strategy, however, the exact composition and variation of the proteins in the membrane are often unknown [7]. To verify the membrane protein coating on SiNR, all five SiNR with different coating were boiled in loading buffer for PAGE followed by silver staining. The protein coating on the SiNR surface had the same molecular weight pattern as that in the membrane protein lysate ( Figure 3A and Figure S2). We also employed an antibody of the cell membrane marker ATP1A1 for immunostaining [52]. The confocal images ( Figure S3A-C) showing ATP1A1 staining on MP-coated SiNR indicate that the MP is coated on the SiNR surface. To confirm the biological function of MP after the coating process, we used the Cy3-labeled human EGF to incubate with MP-coated SiNR. The confocal pictures ( Figure S3D) show the fluorescence signal of human EGF was merged with the MPcoated SiNR but not the uncoated SiNR. This colocalization demonstrated that the EGFR proteins on SiNR remain functional in binding to its specific ligand. After SiNR nanofabrication and protein coating, the 5-FAMSE fluorescence was linked to a protein coating layer. We have already described the controlled fabrication and here focus on a unified size for testing internalization efficiency as a function of coating. Images of the SiNR, as shown in Figure 2B, had a uniform size (408 ± 10 nm in diameter and 2229 ± 49 nm in length). The confocal images indicate homogeneous fluorescence intensity for each SiNR in Figure 2C. The coating membrane proteins (CMP) on the surface of SiNR were investigated by TEM ( Figure 2D and Figure S1A-E), and the average thickness of the CMP layer was 19.09 ± 0.47 nm. Dynamic light scattering (DLS) was performed to analyze the particle size distribution of SiNR. Figure S1F shows SiNR has good homogeneity after CMP coating.

Proteomics Analysis of Coating Membrane Proteins Derived from Five Cancer Cell Lines
Using cell membrane proteins as a surface grafting moiety can be a practical internalization strategy, however, the exact composition and variation of the proteins in the membrane are often unknown [7]. To verify the membrane protein coating on SiNR, all five SiNR with different coating were boiled in loading buffer for PAGE followed by silver staining. The protein coating on the SiNR surface had the same molecular weight pattern as that in the membrane protein lysate ( Figure 3A and Figure  S2). We also employed an antibody of the cell membrane marker ATP1A1 for immunostaining [52]. The confocal images ( Figure S3A-C) showing ATP1A1 staining on MP-coated SiNR indicate that the MP is coated on the SiNR surface. To confirm the biological function of MP after the coating process, we used the Cy3-labeled human EGF to incubate with MP-coated SiNR. The confocal pictures ( Figure   Cells  We analyzed Panc-1 to validate that the protein extraction was indeed from the cell membrane. N-cadherin and E-cadherin are proteins that anchor the plasma membrane and act as cell junction markers [53] and they were highly expressed in the Panc-1 membrane extract ( Figure 3B and C). Vimentin is a cytoskeleton protein that acts as an intermediate filament. It was present on both the We analyzed Panc-1 to validate that the protein extraction was indeed from the cell membrane. N-cadherin and E-cadherin are proteins that anchor the plasma membrane and act as cell junction markers [53] and they were highly expressed in the Panc-1 membrane extract ( Figure 3B,C). Vimentin is a cytoskeleton protein that acts as an intermediate filament. It was present on both the membrane extract and the cytoplasmic lysate. Protein gel sets derived from the above setting were used for high-resolution LC-MS/MS analysis for protein identification and characterization. On the basis of this unique peptide match, the final proteomic data (Supporting Information Section 2) were used for matching in the uniprot (Homo sapiens) database.
We identified 1318 proteins from Panc-1, 1192 from Panc-PDX, 1450 from MCF-7, 1351 from MDA231, and 1506 from Hela cells. A Venn diagram was employed to compare the variation among the five cell membrane proteins, and a full list of uniprot proteins identified from each cell membrane was uploaded to Jvenn (http://jvenn.toulouse.inra.fr). The results indicate that 601 proteins were present in all five cell lines, and 861 proteins were present in only one of the five lines ( Figure 3D). Of the 601 shared proteins, two dominant clusters are proteins associated with the extracellular region (50%) and proteins belonging to cell junction function (16.7%). In Figure 3F, GO enrichment shows that the cell substrate junction term is the most enriched segment with a significant p-value (<1 × 10 −9 ). This implies that the cell-cell contact and cell-substrate adhesion proteins are key proteins in nanorod coating for internalization.
To further investigate the protein expression diversity among the five cell membranes, a few selected proteins were Western blotted for validation ( Figure 3G). Plectin is a biomarker for invasive pancreatic ductal adenocarcinoma [54,55]. It was highly expressed in the Panc-PDX cells which was the primary tumor sample. Pan-keratin is an epithelial marker of epithelial malignancies [56]. It was highly expressed in Panc-PDX and MCF-7 cells. The expression pattern of β-catenin was similar to pan-keratin, i.e., it was enriched in epithelial-like Panc-PDX and MCF-7 cells. Meanwhile, MCF-7 and Panc-1 had higher levels of Cdc42. These are a Rho family member of GTPases [57]. All of these results suggest that cell membrane proteomics are cell-type dependent.

SiNR Internalization Efficiency Is also Determined by Target Cells
The SiNR uptake is likely determined by coating and the intrinsic properties of target cells. To further characterize the interactions between coated SiNR and the target cells, we conducted quantitative and consecutive analysis of SiNR internalization efficiency in each of the five cancer cell lines using the AMNIS ImageStream system. (Figure S4A,B). Statistics on the internalization efficiency are shown in Figure 4, Figure S4H-I, and Table S1 for the SiNR coatings with different membrane proteins. We used the following index of the spot count to represent the SiNR amount internalized in a particular cell ( Figure S4B): low internalization 0-3, middle internalization 3-10, and high internalization >10. The L-size and M-size SiNR coatings with different cell membranes show enhanced internalization efficiency as compared with non-MP coated SiNR (FAM@SiNR) in all cell types ( Figure 4A,B). However, the MP coating does not improve the S-size SiNR internalization efficiency.
The internalization efficiency is different across cell types. Thus, we performed a quantitative analysis by ImageStream. The M-size SiNR average internalization of the five investigated cells are Hela (78.72%) > Panc-PDX (58.83%) > MDA231 (45.74%) > MCF7 (28.18%) > Panc-1 (14.44%). This ranking order corresponds well with ImageStream on internalization in Figure 5 and Figure S4D-H, which detail the internalization efficiency of all five cell types. The histogram plot of the M-size SiNR fluorescence intensity (MFI) in Figure S4C shows the cell-type specificity on internalization. The rank order was Hela>Panc-PDX > MCF7 > MDA231 > Panc-1. The MFI in Hela (9.44 × 10 4 ) is the highest and is twice as high as Panc-PDX (4.722 × 10 4 ).  This improved internalization efficiency expands from the homolytic coating proteins to the heterolytic proteins (Figure 4, Figure S4I and J, and Table S1). This result is confirmed by cytometry as well as confocal images. Figure 6 and Figure S5 show a detailed SiNR internalization process with different coatings across the five type cells. We used z-stack scanning of each type of single cell to identify the internalized SiNR. In Figure S5A, each picture crosshair points the SiNR presented near nucleus inside the cell body (cell membrane). Meanwhile, the lysosome fusion with MP-coated SiNR was observed in MCF7 cell ( Figure S5B). In Figure 6, the middle sections with observed nucleus were chosen to capture internalized SiNR near nucleus and inside the membrane staining. The SiNRs coating with membrane proteins are significantly aggregated inside the Hela and Panc-PDX cells and less aggregated inside MCF7 and MDA231 cells. For Panc-1, most SiNR spread outside the cells, and This improved internalization efficiency expands from the homolytic coating proteins to the heterolytic proteins (Figure 4, Figure S4I,J, and Table S1). This result is confirmed by cytometry as well as confocal images. Figure 6 and Figure S5 show a detailed SiNR internalization process with different coatings across the five type cells. We used z-stack scanning of each type of single cell to identify the internalized SiNR. In Figure S5A, each picture crosshair points the SiNR presented near nucleus inside the cell body (cell membrane). Meanwhile, the lysosome fusion with MP-coated SiNR was observed in MCF7 cell ( Figure S5B). In Figure 6, the middle sections with observed nucleus were chosen to capture internalized SiNR near nucleus and inside the membrane staining. The SiNRs coating with membrane proteins are significantly aggregated inside the Hela and Panc-PDX cells and less aggregated inside MCF7 and MDA231 cells. For Panc-1, most SiNR spread outside the cells, and a very small number of SiNRs were internalized. This is because Panc-1 cells have the lowest ranking and have an intrinsic mechanism at the molecular level. The internalization efficiency is cell-type dependent. a very small number of SiNRs were internalized. This is because Panc-1 cells have the lowest ranking and have an intrinsic mechanism at the molecular level. The internalization efficiency is cell-type dependent.

Gene Ontology on Cancer Cell Internalization
To understand the molecular mechanism underlying SiNR internalization in the system, we performed a bioinformatic analysis on the proteomic data of membrane proteins from five cell lines (Supplemental File 2). The GO (Supporting Information Section 2) mapping shows that all identified cell membrane proteins from five specific type can be divided into the following three categories: cellular component (CC), biological process (BP), and molecular function (MF). Figure 7A shows that these five cell types share 559 common GO terms. However, each type takes a different number of unique genes: 41 (Panc-1), 39 (Panc-PDX), 42 (MCF-7), 48 (MDA231), and 67 (Hela). In this research, Panc-1 shows much lower internalization efficiency than the others. The intrinsic cell-type specificity is attributed to this internalization variation. A GO term named "negative regulation of endocytosis" is found only in the gene enrichment result of Panc-1 membrane extracts.

Gene Ontology on Cancer Cell Internalization
To understand the molecular mechanism underlying SiNR internalization in the system, we performed a bioinformatic analysis on the proteomic data of membrane proteins from five cell lines (Supplemental File 2). The GO (Supporting Information Section 2) mapping shows that all identified cell membrane proteins from five specific type can be divided into the following three categories: cellular component (CC), biological process (BP), and molecular function (MF). Figure 7A shows that these five cell types share 559 common GO terms. However, each type takes a different number of unique genes: 41 (Panc-1), 39 (Panc-PDX), 42 (MCF-7), 48 (MDA231), and 67 (Hela). In this research, Panc-1 shows much lower internalization efficiency than the others. The intrinsic cell-type specificity is attributed to this internalization variation. A GO term named "negative regulation of endocytosis" is found only in the gene enrichment result of Panc-1 membrane extracts. By searching for the potential internalization driver gene (Tables 2 and 3, and Table S2) with alignment on the list of unique proteins from the cell membrane extraction (Figure 3D), we identified ATXN2 as a unique gene annotated via "negative regulation of endocytosis" in Panc-1 cells. ATXN2 is involved in endocytosis [58,59] and it functions as a repressor of the mTOR signal pathway [60]. By searching for the potential internalization driver gene (Tables 2 and 3, and Table S2) with alignment on the list of unique proteins from the cell membrane extraction (Figure 3D), we identified ATXN2 as a unique gene annotated via "negative regulation of endocytosis" in Panc-1 cells. ATXN2 is involved in endocytosis [58,59] and it functions as a repressor of the mTOR signal pathway [60]. One interesting question is whether its upregulation is associated with the lowest internalization efficiency of Panc-1. To validate ATXN2 expression in different cell lines, we analyzed ATXN2 mRNA levels using RT-RCR. The mRNA level of ATXN2 in Panc-1 cell is highest among the five cell lines ( Figure 7B). This trend is inversely correlated with the nanorod internalization efficiency, suggesting that ATXN2-associated down-regulation of endocytosis might be involved in the process. This finding supports the low internalization efficiency in Panc-1. At the same time, two other genes were upregulated in Panc-PDX which is the only primary cell of the five types. One was MELTF, a gene that encodes melanotransferrin [61,62] and plays a vital role in cell migration [63]. The other was C-Src, which is anchored at the inner face of the plasma membrane and regulates cell motility and cell-cell adhesion [64,65].

ATXN2 Is a Negative Regulator of Nanorod Internalization
To further investigate the role of ATXN2 in nanorod internalization, we first knocked down ATXN2 expression in Panc-1 cell with specific interfering RNA (siRNA). As validated with RT-PCR ( Figure 8A), siRNA1 and siRNA2 of ATXN2 inhibited ATXN2 expression by 77% and 45%, respectively. Meanwhile the Western blot ( Figure 8B) also showed the consistent result that the protein levels of ATXN2 were knocked down by the two siRNAs, respectively. Imaging flow cytometry was then performed to quantify the SiNR cellular uptake in control and ATXN2 knockdown Panc-1 cells ( Figure 8C). The ATXN2 knockdown from siRNA1 induced more efficient inhibition of ATXN2 expression and resulted in a greater improvement in internalization efficiency. The expression of siRNA2 was consistently less efficient in ATXN2 knockdown and led to modest or no improvement of internalization. The single cell images from imaging cytometer ( Figure 8D) also confirm the SiNR internalization pattern after the ATXN2 siRNA treatment. These results support a correlation between ATXN2 expression and SiNR internalization efficiency.
To understand the mechanisms that ATXN2 involved in cellular internalization of SiNR, we used chemical inhibitors and low temperature treatment to block the ATXN2 siRNA-induced SiNR internalization in Panc-1 cell. As Figure S6 shows, the ATXN2 siRNA1 induced SiNR uptaken was block at 4 • C. The results from low temperature treatments suggest that SiNR internalization affected by ATXN2 is energy dependent.
The chemical inhibitors are chlorampramazine, a specifically blocker of clathrin-mediated pathway [66]; filipin, a caveolae pathway inhibitor [66]; cytochalasin D which inhibits actin polymerization that is involved in various other endocytic pathways [66]; and nocodazole blocks microtubule polymerization and vesicular transport [45].
The chemical inhibitors were used to test their effects on SiNR internalization. As shown in Figure 8E, ATXN2 siRNA1 induced a 2.1-fold increase in cells with internalized SiNR, as compare with cells treated with NC scrambled siRNA. In contrast, chlorampramazine, cytochalasin D, and nocodazole, but not filipin, reversed the uptake improvement induced by ATXN2 siRNA. The results from the inhibitor assays suggest that ATXN2-modulated SiNR internalization may involve clathrin, micropinocytosis, and microtubules systems, but not the cavelae pathway in Panc-1 cells.

Discussion
Silica nanomaterials are safe and biocompatible in vitro and in vivo [12,14,67]. The cell membrane protein coating method is a powerful method to enhance internalization efficiency [6]. Here, we developed a facile method to fabricate silica nanorods with a controllable size, shape, and amount of coating protein from cell membranes for enhanced efficiency uptake. We then determined potential factors that may affect the internalization efficiency.
The diameter and length of SiNR can be precisely controlled by etch time, and three types of SiNRs were fabricated. Our method improved the internalization efficiency via SiNR size optimization. Our fabrication method employed the free -NH 2 of protein to form covalent bonds with the SiNR surface -OH via APTES and thiophosgene. We believe this is an unbiased method to link all types of cell membrane proteins to a SiNR surface.
We used silver staining, immunostaining, as well as TEM and EGF binding assays ( Figure 3A, Figures S1 and S3) to demonstrate that the cell membrane proteins were indeed linked to the SiNR surface. The free amine-mediated coating method used a proportional number of proteins coated to the nanorods. Due to the limited protein extracts available for coating the SiNR, we used direct cell membrane protein extraction to investigate the cell membrane proteomic fingerprints of the five cell lines.
The Amnis system is an establish method for internalization detection [68][69][70]. A practical protocol for eliminating the cells with background labeling and identifying specific internalization was published in 2012 [70]. We followed the protocol to identify the SiNR internalization, and the details are shown in Figure S4A. As confocal microscopy can provide more details on the SiNR uptake, the pictures from Z-stack scanning proved MP-coated SiNR was present inside the cell body in a three-dimensional view. Moreover, our lysosome colocalization assays indicate that the MP-coated SiNR internalization may be relative to the canonical endocytosis pathway [71].
Hela, Panc-1, Panc-PDX, MDA321, and MCF7 cells were used in this study to investigate the internalization efficiency. The FACS data showed that the high SiNR uptake in target cells led to a higher total internalized cell rate and a greater percent cell distribution (Figure 4). We found that the membrane coating significantly improved the L-size and M-Size SiNR internalization efficiency in all tested cells. This improved internalization efficiency holds for both homolytic and heterolytic proteins. However, this coating did not enhance the S-Size SiNR internalization which may be due to size-related different endocytosis mechanisms. For example, macropinocytosis can take about 2 µm particles but clathrin-mediated endocytosis prefers particles <300 nm [72]. This issue will be further investigated in future work.
To understand the mechanism underlying the altered internalization, we compared the proteomics data of the five cell membrane proteins and identified 601 proteins that are known as having cell substrate adhesion properties. These proteins include N-cadherin, E-cadherin, and integrin. These might contribute to the surface recognition and internalization. The homolytic or heterolytic SiNR coatings have the same improvement which may be due to these surface proteins.
In contrast to coating sources, the internalization efficiency of the target cells varied among the tested cell lines across all three sizes of SiNR. This motivated us to identify potential cellular factors that may affect nanorod internalization. Integration of the proteomics data and the internalization efficiency data suggested an association between the lowest internalization efficiency in Panc-1 cells and ATXN2 upregulation in these cells. The improved internalization efficiency in Panc-1 cells with ATXN2 knockdown provided specific evidence supporting ATXN2 as a negative regulator of nanorod internalization.
Previous studies of nanoparticle internalization have provided specific guidance on the mechanisms of endocytosis with inhibitors and temperature control [45,73,74]. To further understand the mechanisms of SiNR uptake in context with ATXN2 function, we used chemical inhibitors to explore the pathways the ATXN2 may participate in. The working concentrations of inhibitors were chosen based on a previous study [45]. As shown in Figure 8E, we showed that chlorampramazine, cytochalasin D, and nocodazole, but not flilipin, partially revered ATXN2 siRNA1 knockdown-induced MP uptake. The results from the inhibitor assays indicate that the ATXN2 may participate in Panc-1 cell uptake the MP-coated SiNR, including clathrin, micropinocytosis, and microtubules, however, not the cavelae pathway. Our results highlight the intrinsic properties of the target cells that modulate nanocarrier uptake. Our data provide clues to further identification of novel factors that may affect nanoparticle internalization to cancer cells.
In summary, we developed a facile method of silicon nanorod fabrication with controllable sizes. We found that coating the nanorods with cancer cell membrane proteins significantly enhanced the internalization of the nanocarriers. Using proteomics and bioinformatics approaches, we identified proteins that might facilitate the internalization. More importantly, our results highlighted the differences among different target cells in nanorod uptake and identified ATXN2 as a negative regulator of nanorod internalization. This was further supported by bioinformatic analysis and specific functional validation with siRNA experiments. These data collectively advance our understanding of factors that affect nanocarrier internalization and provide clues to further improve internalization efficiency.