SEARCH

SEARCH BY CITATION

Keywords:

  • biofilm;
  • cell proliferation;
  • gene expression;
  • titanium alloy;
  • surface analysis

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

Titanium-based implants are widely used in modern clinical practice; however, complications associated with implants due to bacterial-induced infections arise frequently, caused mainly by staphylococci, streptococci, Pseudomonas spp. and coliform bacteria. Although increased hydrophilicity of the biomaterial surface is known to be beneficial in minimizing the biofilm, quantitative analyses between the actual implant parameters and bacterial development are scarce. Here, the results of in vitro studies of Staphylococcus aureus and Staphylococcus epidermidis proliferation on uncoated and coated titanium materials with different roughness, porosity, topology, and hydrophilicity are shown. The same materials have been tested in parallel with respect to human osteogenic and endothelial cell adhesion, proliferation, and differentiation. The experimental data processed by meta-analysis are indicating the possibility of decreasing the biofilm formation by 80–90% for flat substrates versus untreated plasma-sprayed porous titanium and by 65–95% for other porous titanium coatings. It is also shown that optimized surfaces would lead to 10–50% enhanced cell proliferation and differentiation versus reference porous titanium coatings. This presents an opportunity to manufacture implants with intrinsic reduced infection risk, yet without the additional use of antibacterial substances.


1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

Infections associated with implantations remain a potentially serious complication, even if their incidence is relatively low.1–3 For total hip arthroplasty (THA) operations being on the rise (from over 260 000 made in 2004 to ∼423 000 in 2008 in the US alone) up to 1.3% were treated for reasons of infection. For total knee arthroplasty (TKA), “primary” infections were reported up to 2% of ∼490 000 cases in 2004 and more than 600 000 in 2008.1, 3, 4 About 10% of the arthroplasties were revised later due to implant failures;5 8–15% of which were reported to be a direct result of an infection.1, 5 These “primary” and “secondary” infections together indicate a failure about 2.4% of revised THA and 2.3% of revised TKA (various numbers could be found in different sources). Implant-related infections were linked with a mortality rate of 7–63% for THA and of 2.5% for TKA. Furthermore for 80% of the infected cases for TKA, the implants were no more functional.3 Most of the published data suggest the percentage of the direct infection-caused complications and the following revisions to be ∼1–2%, which has been stable for many years. However, the rising total number of orthopedic operations worldwide clearly means more patients would potentially suffer from these infections, unless that percentage will be significantly reduced.

Most of the microorganisms causing implant infections are present in the host flora, of which the most frequent are staphylococci (Staphylococcus aureus and Staphylococcus epidermidis) and enteric bacteria.2, 3 Staphylococci infections are particularly associated with metallic and polymer implants and it is a common practice to apply an antibiotic treatment to prevent them. However, with emerging amount of multiple antibiotic-resistant strains (MRSA and others), the efficacy of antibiotic treatment might be low and being even further suppressed due to biofilm formation.1, 3, 6 The biofilm is formed after microbial attachment and further colonization of the surface. During formation of microcolonies, microbes are encapsulated in an extracellular polymeric substance layer existing of polysaccharides, proteins, and other products, possibly making bacteria up to thousand times more resistant to antibiotics compared to their planktonic counterparts and inaccessible for the host immune system.3, 4, 6–9 The biofilm is extremely stable and virulent with little possibilities to be removed from the implant surface. It is generally understood that the key factor in preventing biofilm formation is the bacterial adhesion to the surface. Therefore, if the bacterial adhesion to an implant surface is being properly understood and controlled, there is a potential to significantly reduce risks of infections since lower adhesion potential practically means less driving force for biofilm formation and therefore higher efficacy of applied antibiotic treatment.1, 2, 9

The attachment of bacteria to the implant surface is a very complex process. Different factors such as intermolecular forces (surface potential, surface energy, charge, and activity), local environment variables (pH, ionic strength, and competitive species such as proteins), surface topography, porosity, hydrophobic properties, and microfluidics (shear rates, vorticity, and velocity) are all affecting the adhesion. Because of the huge variety of these factors, most of the studies directed at bacterial attachment to the implant surface were limited to specific surface conditioning only (e.g., micro- or nanotopology, patterned surface treatment, hydrophilic treatment, etc.) since it is difficult to vary many parameters at the same time.1, 3, 4, 7–10 Studies have also demonstrated that surface roughness might be modified in a way to decrease bacterial adhesion,1, 7, 8, 11 but the approach of optimized “biofilm-inert” implants still remains largely unexploited.3, 9 Many recent studies also focused on nanotopology on micro-flat and macro-flat surfaces, and good results were obtained at specific nanopatterned surfaces.1, 10 However, other studies showed more biofilm formation on nanocrystalline surfaces.12 Exact nanopatterning in mass production might be too complex and difficult to implement at the implant manufacturers facilities, adding costs and uncertainty in the implant's “intrinsic” ability to really minimize bacterial adhesion. In orthopedic implants, however, proper porosity and pore topology are also important to enhance osteoblast activity, improve the bone-to-implant contact, provide proper biomechanical properties, and finally improve the osteointegration.9, 13

This is particularly challenging when data were gathered on bacterial adhesion in vitro for one to three varied parameters, which might not be fully relevant for in vivo situations. Nevertheless, several parameters, such as low roughness and wetting of the surface (hydrophilicity), were identified to be favorable for biofilm prevention.1, 4, 10 Unfortunately, some of the studies have involved either too exotic surfaces, or treated data in a semiquantitative way, which makes it difficult to compare them to and extrapolate them on real implants, of which specific surface parameters might be not so easy to check and monitor from case to case.

Analysis of bacterial adhesion and biofilm formation inside porous coatings is very challenging and might not be feasible with some standard techniques. Besides the “optimal” implant anti-bacterial function (here, drug-carrying surfaces are not considered), the requirements for osteoblast fixation, mechanical, and biomechanical constraints set additional challenges for implant developers, because in many cases almost incompatible demands have to be met. For osteointegration, whether it is described by biomechanics, contact osteogenesis or surface signaling theories,10 it is important to promote osteoblast fixation and to avoid bacterial adhesion at the same time. As pointed out by Wu et al.,4 cells and bacteria respond differently to implant surface parameters and this gives an opportunity to look for an “optimal” solution: would it be possible to create a robust, proper implant surface capable of repelling bacteria and attracting osteogenic cells, at least immediately after implantation (within few minutes or hours)?

At the present state-of-the-art, it is known (at least qualitatively) that roughness, wettability, porosity, pore topology, and other surface conditions are the key factors to adjust. However, it is not easy to produce numbers of specific surfaces for which these parameters are independently varied and their specific effect on both bacteria and cells is properly evaluated and extracted.

Statistical analysis of data from biomaterial studies is usually limited to a very simple one for single hypothesis. This is difficult to apply to scattered data, depending on several cross-linked variables (particularly when the error distribution is unknown or clearly being different from the norm). Thus, taking into account the variety of the factors affecting the implant behavior, it is not straightforward to elaborate the “best” or “optimized” specimen. The term “optimization” is often being understood in the narrowest way, i.e., just as a selection of the proper solution from the list of few results. Mathematically defined optimization is the task of finding one or more solutions, which minimize (or maximize) one or more specified objectives, satisfying all existing constraints (if any). A multiobjective optimization task considers several conflicting objectives simultaneously: in such a case, there is usually a set of alternative solutions with different trade-offs (“Pareto optimal” solutions).14 The advanced meta-analytical methods applied to experimental data have an improved ability to determine the relative influence of different parameters on a given phenomenon, to find existing relationships among the considered variables, and to resolve the unavoidable noise present within a vast set of numerical data. The challenge of applying meta-analysis to the testing of biomaterials is to overcome limitations due to a considerable amount of time needed to perform an optimization analysis,14–16 when the number of free parameters is noticeably high, when these parameters are linked with each other, and when output functions (e.g., bacterial colony features) are unknown or not quantified.

The present study is aimed towards the vitro analysis of S. aureus and S. epidermidis bacterial biofilm formation kinetics on selected commercial and experimental titanium implants surfaces, combined with parallel analysis of cell adhesion, proliferation, and gene expression (cell differentiation). By meta-analysis application to these test data, the effect of the parameters of the surfaces is revealed and the “optimal” implant surface is suggested.

2. Materials and Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

2.1. Titanium Specimens: Preparation and Treatment

Commercial purity titanium (grade 2, UNS R50400, marked as Ti2) and Ti-6Al-4V (grade 5, UNS R56400, marked as Ti5) alloys certified for orthopedic applications have been supplied by three independent implants producers (Lima Lto Ltd., Italy; Helipro Ltd., Slovenia; Alhenia Ltd., Switzerland) in a form of disks (15.5 mm diameter, 2 mm thick).

The surface of the titanium disks has been as-received (machined (M), bead-blasted (B) and polished (P) according to the manufacturers own protocols), Table1. The part of the disks was hydrothermally treated (index -HT) by a procedure developed in Jozef Stefan Institute, Slovenia17 to create a thin layer of anatase TiO2 (confirmed by phase analysis) with thickness up to 350 nm (characterized by X-ray photoelectron spectroscopy, XPS). Specimen types and their surface properties are listed in Table 1. The addition of a thin anatase nano-layer did not significantly affect the roughness neither porosity topology parameters, but substantially changed hydrophilic properties.

Table 1. Key parameters of the titanium specimens (numbers “± value” where marked, are standard deviations; IPC shows three values for minimal, mean and maximal interconnective pore channel diameter respectively).
MaterialRa [μm]Rz [μm]PorosityWetting angle [deg.]Max pore size [μm] IPC [μm] 
      minmeanmax
  • a)

    a)Ti5-HT specimens have TiO2 layer of 100–350 nm thickness, Ti2-A specimens ∼60 nm, and VPS-HT ∼100 nm.

Ti5 machined (M) 1.0 ± 0.04.5 ± 0.36054.70000
Ti5 beadblasted (B) 4.9 ± 0.645.1 ± 1.40 44.7 ± 8.80000
Ti5 polished (P)0.05 ± 0.00.2 ± 0.020 41.1 ± 8.30000
Ti5 treated (HT)a)1.1 ± 0.04.8 ± 0.36041.0 ± 5.00000
Ti2 machined (M)0.32 ± 0.053.24 ± 0.38075.5 ± 6.20000
Ti2 anodized (A)a)0.32 ± 0.043.49 ± 0.61045.1 ± 10.00000
Plasma sprayed Ti (VPS)27.9 ± 1.5 221.7 ± 15.00.208143.0 ± 5.210011030
Ti VPS treated (VPS-HT)a)27.9 ± 1.5221.7 ± 15.00.20815.0 ± 5.010011030
EPD, grade P4.4 ± 0.340.8 ± 2.40.510100.5 ± 3.13027.58
EPD, grade VM1.8 ± 0.129.3 ± 7.10.37889.1 ± 5.610134
EPD, grade U3.9 ± 0.337.7 ± 1.30.47181.2 ± 2.72026.57
EPD, mixed (EPDTi)7.1 ± 0.982.0 ± 7.00.597102.7 ± 4.95018.330

For porous coated specimens, state-of-the-art vacuum-plasma sprayed (VPS) titanium-coated disks (Alhenia Ltd.) were used as the reference. In addition, a new technology of manufacturing porous titanium coatings with characteristics similar to VPS-Ti coatings from titanium hydride by electrophoretic deposition (EPD) of TiH2 powder suspensions followed by dehydrogenation (500–550 °C) and sintering in vacuum (850 °C) was applied.18, 19 Distinct TiH2 powder grades of different particle sizes, all supplied by Chemetall GmbH (Germany), have been used: grade type P (average particle size 8.0 ± 2.0 μm, maximal 60 μm), type U (5.0 ± 1.0 μm, maximal 45 μm) and type VM (1.8 ± 0.2 μm, maximal 45 μm). The same technique was applied on TiH2 powder stabilized emulsions in order to obtain additional spherical pores in the pure Ti coating.20 For the emulsion technique, TiH2 type VM emulsions and type U suspensions were co-deposited by EPD. Specimens with mixture of these powders were made to get different porosity and porosity topology distribution (Table 1). All specimens were gamma-sterilized (25 kGy) by Helipro Ltd. prior to testing.

2.2. Surface Characterization

The surface roughness (Ra and Rz) was measured by white light interferometry (WLI) with Wyko NT 3300 Optical Profiler (Veeco Metrology Inc., USA). A total of four equidistant locations were measured per every disc. The surface wettability, i.e., the contact angle with demineralized water, was measured from the sessile drop tests using optical imaging (CaM200, KSV, USA). The mean contact angle was determined by averaging ten measurements obtained from three different disc samples, taking into account three different areas on each surface. The wetting properties of the specimens are shown in Table 1.

Characterization of the porosity and porosity topology was done by image analysis (PPM2OOF software, NIST, USA) in combination with the mercury intrusion porosimetry (MIP) using AutoPore IV 9500 (Micromeritics, Germany). For all porous specimens, the total open porosity and maximal pore size was determined by image analysis of backscattered electron images obtained by scanning electron microscopy (SEM, XL30-FEG, FEI, The Netherlands) of representative metallographic cross-sections while MIP was used to assess the minimal, mean and maximal interconnecting pore channel sizes.

2.3. Bacterial Strains and Growth Conditions

Bacterial strains S. aureus ATCC25923 and S. epidermidis 1457 were grown in Bacto tryptic soy broth (TSB, Becton Dickinson) or on BBL tryptic soy agar (TSA, Becton Dickinson) plates. Prior to testing biofilm formation on Ti-based discs, an overnight culture grown in TSB (5 mL) at 37 °C and 250 rpm, was 25-fold diluted in TSB and re-incubated for 1 h at the same conditions. The early exponential-growth-phase culture was next adjusted to OD600 of 0.3 and further 10000-fold diluted in TSB, giving a bacterial suspension of (1–3)·104 cells mL−1 and this suspension (2.5 mL) was added to a disc placed in a 12-well CELLSTAR cell culture suspension plate (Greiner Bio-One, Germany). To allow bacterial adhesion to the test discs, plates were first incubated for 2 h at 37 °C under static conditions. Thereafter, plates were further incubated at 30 °C on a 3D-rotating platform (Belly Dancer, Stovall Live Sciences).

2.4. Evaluation of Biofilm Formation

Upon 24, 48 and 72 h (1, 2, and 3 days), three discs of each type were removed from wells and carefully rinsed by dipping the disc in saline (40 mL). Upon repeating the latter step for 2 times, the discs were finally placed in saline (10 mL). Detachment of the biofilm-embedded bacterial cells occurred by sonification twice (2 min at 40 kHz, Bransonic 2510E-MT, Branson, USA) followed by mixing (15 sec; longer sonification might cause bacterial cell killing). Biofilm formation was quantitatively evaluated first by viable colony forming unit (cfu) counts upon incubation (20 h) at 37 °C via plating 10-fold serial dilutions of the obtained bacterial suspension on TSA. A more qualitative evaluation of the biofilms after 2 days was additionally done by CLSM (confocal laser scanning microscopy) and SEM visualizations. Biofilms were analyzed in function of time taking the VPS discs as reference and setting the number of viable bacteria recovered from the VPS discs to a unity. The “LIVE/DEAD BacLight bacterial viability and counting kit” (Molecular Probes) was used to quantify biofilm formation and to simultaneously get an idea of the ratio of live/dead cells within a biofilm. Biofilm-derived cells are stained with Syto9 (green; stains all cells, live and dead) and propidium iodide (PI, red; only enters dead cells with a damaged membrane) and mixed with 106 microspheres (∼6 μm). The concentration of cells was determined from the ratio of bacterial events to microsphere events during laser flow cytometry (LFC) using a FACSCalibur apparatus (Becton Dickinson). To test the suitability of this method in comparison to viable cfu counts, biofilms formed on 3 VPS-Ti discs were evaluated by both methods with measurements done in triplicate for each disk. The LFC was found to produce highly adequate result compared to cfu count method with differences less than standard deviation.

2.5. Cell Culture and Viability Analysis

Cell cultures used were human osteogenic cells (HOC)21 arising from bone marrow (6th passage) and human endothelial cells (HEC) (9th passage).22 Samples were completely covered up with Iscove's Modified Dulbecco's Medium (IMDM, Gibco) supplemented for HOC with fetal calf serum (FSC, 10% vol.; Gibco) and for HEC - with 20% FCS and 0.4% of endothelial cell growth supplement/hepatin (ECGS/H, Promocell), incubated at 37 °C. Pre-coating with serum proteins mimics an in vivo situation following implantation (adsorbed protein layer). HOC and HEC were seeded (100 μL) onto the disc specimens at 104 and 2.0·104 cells cm−2, respectively. Thereafter, the complete culture medium was carefully added to the samples, and plates were incubated at 37 °C in a humidified atmosphere containing 5% vol. CO2.

Quantitative tests were made at the end of the incubation period (1, 3, 9, 15, and 27 days). The culture medium was removed and discarded and cells were detached using trypsin (0.2%wt.) in Hank's balanced solution (Ca2+ and Mg2+ free), incubated for 2 min with trypan blue (0.2%wt., Sigma–Aldrich) in NaCl (0.15 M). Thereafter, dead cells (blue) and living cells (uncolored) were counted using a hemocytometer.

Following the above precoating process, the study of HOC and HEC adhesion to the coating or substrate system was evaluated using the “LIVE/DEAD cell viability/cytotoxicity kit” (Invitrogen; Molecular Probes), where live cells are distinguished by the presence of ubiquitous intracellular esterase activity, determined by the enzymatic conversion of the nonfluorescent cell-permeant calcein AM to the intensely fluorescent calcein. Nuclei of dead cells are stained by passive transport and subsequent binding of ethidium homodimer-1 (EthD-1). These qualitative tests (LIVE–DEAD iconography) were made after the same incubation periods (1, 3, 9, 15, and 27 days). Samples were incubated for 20 min at 37 °C in the dark, within “LIVE–DEAD” solution (calcein AM and EthD-1; 4 μM of each in PBS). After incubation, the samples were analyzed by fluorescence microscopy. Plastic of the culture wells was used as the negative control in these tests. The experiments were performed in triplicate for each time point.

The analysis of cytoskeleton and focal adhesion formation was also performed for HOC and HEC, using 24 h incubation times. The detection of α-tubulin, β-actin, integrin αvβ3, and vinculin was performed using specific monoclonal antibodies conjugated to fluorescein isothiocyanate, isomer I (FITC; Sigma–Aldrich) for α-tubulin, β-actin, integrin αvβ3, or crystalline tetramethylrhodamine isothiocyanate (TRITC; Sigma–Aldrich) for vinculin. Monoclonal concentrations were used: for integrin αvβ3 1/300, α-tubulin 1/200, β-actin 1/800, and vinculin 1/200. A nuclear labeling was performed at the same time using Hoechst staining (50 ng mL−1). The data obtained for the cytoskeleton were used for additional qualitative and semiquantitative analysis and for control of the HOC and HEC reactions, but they were not numerically analyzed.

2.6. Cells Differentiation Analysis

HOC and HEC differentiation analysis was performed with the same cells cultures following the precoating process described above. A concentrate of cells (105 cells cm−2) in 100 μL was seeded on the samples. Thereafter complete culture medium was carefully added and the plates were incubated at 37 °C in a humidified atmosphere with 5% vol. CO2. Total RNA was isolated from HOC and HEC after 1 and 3 days of contact using the RNeasy Micro kit (Qiagen). cDNA was synthesized from total RNA (1 ng), using the cDNA Synthesis kit (Thermo Scientific Verso).

Polymerase chain reaction (PCR) was performed (Thermocycler- Techne TC 3000G) with of cDNA (1 μL) to analyze the genes related to HOC and HEC.23 Analyzes of genes related to HOC (collagen type I (here marked as ColI), alkaline phosphatase (here marked as APh), and for HEC: CD31 and von Willebrand factor (vWF)) were carried out. Specific primers and annealing temperatures are used for each target gene, Table2. Products of amplification were analyzed and quantified (Gene Tools; Syngene) using a standard of calibration (low molecular weight DNA ladder; Biolabs). Results were normalized to housekeeping gene β-actine.

Table 2. Human primers pairs and PCR conditions used in the present study.
Target genesPrimers sequencesAnnealing temperature [°C]Product length [bp]
β-actine5′-AGT CCT GTG GCA TCC ACG AAA-3′57186
5′-GGA GCA ATG ATC TTG ATC TTC-3′  
Alkaline phosphatase (APh)5′-AGC CCT TCA CTG CCA TCC TGT-3′6568
5′-ATT CTC TCG TTC ACC GCC CAC-3′  
Collagen type I (Col-I)5′-TGG ATG AGG AGA CTG GCA ACC-3′65223
5′-TCA GCA CCA CCG ATG TCC AAA-3′  
CD315′-CCT GCT GAC CCT TCT GCT CTG-3′6574
5′-TAC AGT CGT GGT GGA GAG TGC-3′  
von Willebrand factor (vWf)5′-CCC CTG AAG CCC CTC CTC CTA-3′65124
5′-ACG AAC GCC ACA TCC AGA ACC-3′  

2.7. Data Analysis

For the biofilm formation factor (BFF) and cytological reactions (CyR) analysis, the following factors have been considered, Table3. All the data collected from bacterial studies were normalized to VPS titanium coated specimens; all the results are thus relative to VPS Ti taken as unity. The data from cytological reaction studies were also normalized to VPS Ti, except for cell differentiation data, which were first normalized to ß-actine and then to VPS Ti. The selection of specific incubation days for cytoreactions was carried out on the basis of the most full and consistent datasets, yet in random order. The qualitative and semiquantitative data (such as for cytoskeleton) were not used due to the larger uncertainty and possible higher scatter in the interpretation of these results.

Table 3. Parameters of the experiments included in the data meta-analysis.
Surface parametersNormalized biofilm formationNormalized cytological reactions
Roughness (Ra, Rz, their deviations) [μm]Formation and colonization of S. aureus at 1, 2, 3 daysHEC proliferation after 9 days
TiO2 thickness [nm]Formation and colonization of S. epidermidis at 1, 2, 3 daysHOC proliferation after 9 and 27 days
PorositySum of all Staphylococci at all days (= biofilm formation factor, BFF)CD31 expression after 3 days
Maximal pore size [μm] vWf expression after 1 day
Interconnective pore channel sizes [μm] APh expression after 1 day
Wetting angle, its deviation [deg.] Collagen-I expression after 1 and 9 days
  Sum of all the above (= cytoreactions sum, CyRsum)

All specimen data were entered into a data file as a single record, which contains all information, as shown in Table 3. If the data have been averaged, the normal average and the standard deviation limits were also determined (Table 1). During the analysis and visualization, no assumptions about the data internal structure have been made (meta-analysis does not require hypotheses about normality of the distributions).14, 15

The analysis has been performed in the following order: 1) surface parameters only, 2) surface parameters with BFF, 3) surface parameters, BFF and CyR altogether. This has given an opportunity to analyze and separate the effect of different factors and their cross-correlations. Data visualization, multivariate analysis (self-organized maps) and neural network training was carried out with a reactive search algorithms using Grapheur software 1.0.293 (Reactive Search Ltd., Italy). Deeper data analysis and meta-modeling was made with modeFrontier 4.3.0 (ESTECO Ltd., Italy).

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

3.1. Specimens Surface Characterization

Specimens' surface and porosity parameters (Table 1) were analyzed on their possible cross-correlations before moving to the analysis of microbiological experiments. The surface parameters and the wetting-angle cross-correlations are shown in Figure1. As expected, roughness Ra and Rz are highly correlated, independently on the surface type (coated, treated, or not), porosity, or porosity internal topology. This means it is not possible to vary these two roughness parameters independently from each other. Similar high correlation is obtained between the interconnective pore channel (IPC) mean size and maximal pore size. Thus, the number of independent parameters for surface characterization of the specimens could be reduced to one roughness value (either Ra or Rz), the total porosity, and either IPC mean size or maximal pore size. Other parameters of Table 1 are dependent on these or have practically no relevance within the range studied. The correlations in Figure 1 show that the wetting angle is affected by all major parameters. Figure2 shows the dependence of the cosine of the wetting angle on porosity, roughness, and titania layer thickness. Increase of roughness and porosity leading to higher wetting angles (more hydrophobic surfaces) might be overruled by creating a titania layer using any feasible method.

thumbnail image

Figure 1. Spearman's correlation coefficients (color scale) between the surface parameters (Ra, Rz, porosity, pore size and anatase thickness) and wetting angle.

Download figure to PowerPoint

thumbnail image

Figure 2. Wetting angle dependence on total porosity, roughness Ra [μm] (as color) and TiO2 layer thickness [nm] (as bubble size).

Download figure to PowerPoint

Therefore, surface characterization factors such as Ra versus Rz, maximal pore size versus Rz, or IPC mean size are strongly correlated, and some of them have to be excluded from the analysis. Titania layer thickness and total porosity, on the other hand, can be considered as independent variables.

3.2. Staphylococci Colonization and Biofilm Formation

The data of the staphylococci biofilm formation show a clear increasing dependence versus incubation time. There is also a nearly linear dependence between the S. aureus and S. epidermidis population for a respective incubation day (Figure3 demonstrates example for day 1 (24 h of incubation) for all specimens; normalized to VPS titanium), which is invariant to the type of the specimen or its surface parameters. Besides, very clear correlation between different staphylococci incubation times (Figure4), higher roughness, IPC mean size, and porosity are all observed to affecting the proliferation positively. Since all studied bacteria do behave in a very similar way, a combined indicator - biofilm formation factor (BFF) - was employed. BFF was calculated as the sum of all staphylococci for all days for every specimen, and this sum has been normalized to its value for VPS titanium coating. This is just a measure to reduce the number of output parameters and to combine the indirect influence of all incubation times and different stains. The example of the BFF value (relative to VPS Ti), changing as a function of wetting angle, porosity (as color), and roughness (Ra as bubble size), is shown in Figure5. Note that smoother, nonporous specimens have generally the lowest BFF, whereas for porous specimens the roughness effect might overrule wetting and porosity. If only results as Figure 5 would have been considered, one might conclude that wetting angles 30–90° are the best option for expecting the lowest BFF value. However, this holds only for nonporous specimens, which also have the lowest surface roughness, and thus being not suitable for orthopedic applications, where a porous surface layer is required to ensure bone in-growth.

thumbnail image

Figure 3. Example of relative (normalized to VPS Ti) formation and colonization of S. aureus and S. epidermidis (after the first day each). The dashed line serves as a guide to the eye. Symbols reflect average values for different specimen categories.

Download figure to PowerPoint

thumbnail image

Figure 4. Spearman's cross-correlation between values of S. epidermidis and S. aureus after 1, 2, and 3 days of colonization (numbers in brackets) and with surface parameters. Note the high positive correlation with porosity, roughness, and IPC mean size.

Download figure to PowerPoint

thumbnail image

Figure 5. The combined BFF value (relative to VPS Ti) as function of wetting angle, porosity (color), and roughness (Ra [μm], as bubble size)

Download figure to PowerPoint

The overall Student's factor analysis shows that the total porosity contribution is most significant for the first day of incubation only, whereas roughness (Ra) and IPC size are important for all incubation days. This might be explained by the larger surface-area fraction available for bacteria attachment at the beginning – after 24 h of colonization and biofilm formation no more “fresh” porous surface area is available because the first-coming bacteria have already occupied some pores. The effect of TiO2 is mostly relevant to provide a higher hydrophilic state, which is known to decrease the level of staphylococci attachment.

Inter-relations between many different input and output variables are difficult to understand using just simple graphs. Therefore a multivariate analysis (MVA) of the data has been carried out. One of the most useful representations of multi-dimensional data could be achieved via unsupervised learning with self-organized maps. A self-organizing map (SOM) is an unsupervised neural network for ordering of high-dimensional data in such a way that similar data are grouped spatially close to one another. Here the training data set contains only input variables, so what the SOM learns is the structure of the data. The training arranges the neural network so that units representing neighbors in the input space are situated close together on the map. This is achieved by folding the map into the multi-dimensional input space, so as to adapt, as far as possible, to the input space structure. Although any representation of a multi-dimensional space in two dimensions results in the loss of detail, SOMs are very useful in allowing the visualization of data, which might otherwise be impossible to understand.14, 15

The SOM involving these factors have been trained, resulting in maps as seen in Figure6. Every variable is allocated with its position onto a 2D projection map (color scale: minimum–blue, maximum–red), where the cell position is correlated with other variables position and color-map locations. Maps of variables located closest to one another and with a similar color scale (e.g., Ra and Rz at the bottom left) are highly correlated and thus depending on each other. Maps with opposite color allocations (e.g., cosine of the wetting angle has color distribution opposite to porosity) are negatively correlated.

thumbnail image

Figure 6. The self-organized map of a) all relevant variables (all specimens) and b) with enlarged BFF SOM showing the best matching units (squares with specimens ID), right. The highest BFF value is in red.

Download figure to PowerPoint

For the biofilm formation, the prediction of BFF values is of the highest interest. The enlarged SOM for BFF combines simulated (SOM trained and predicted) results and real specimens, which experimental BFF data are projected on the same map as best matching units (BMU). The presence of BMU (squares) in any SOM cell shows that a realistic specimen has been matched calculated results with specific input parameters combinations.

For specimen optimization and BFF prediction one is targeting such surface combinations, which result in a low BFF but with biomechanically relevant porosity and roughness values at the same time. Figure 6b clearly shows that non-optimized control VPS Ti has the highest BFF observed in the experiments (cells at the lower left of the map). The lowest BFF (region at lower right of the map Figure 6b) is undoubtedly associated with nonporous, smooth specimens. The best specimens that satisfy the objective (minimal BFF) and the constraint (the presence of at least 10–15% porosity) could be selected at the right top of the map (four marked hexagons with four specimens ID No. 8 to 11). These BMU match the specimens of EPD-processed titanium coatings using TiH2 powder grades P, U, and VM, but not with the EPDTi specimens made of these mixtures of TiH2 P+U+VM powders (Table 1). The calculated SOM values (ID No. 49, 59, 79 and 97) represent surface designs located on the Pareto frontier,14, 15Figure7. This means it is not possible to change any more of the input parameters (roughness, porosity, anatase layer, IPC mean size, etc.) without worsening the performance of the specimen (i.e., without increasing the BFF value) and without violation of the objectives and constrains set.

thumbnail image

Figure 7. Real (experimental) and virtual (SOM-predicted) values of BFF (all staphylococci, all three days) normalized to VPS Ti as a function of total porosity, roughness [μm] (as bubble diameter) and wetting angle [deg.] (as color scale). Marked bubbles with numbers represent the best matching units (BMU) with real experiments (Figure 6). Dashed line represents Pareto-dominated limits (BFF minimum), crossed dotted lines indicate biofilm formation porosity onset (∼0.47), after which the BFF values are getting significantly higher.

Download figure to PowerPoint

It is evident that for an average porosity level of 0.20–0.50 the lowest BFF would be achieved, with a wetting angle of the 75–100° and a roughness (Ra) within the 0.5–5.0 μm limits. Even lower BFF values are possible, but this would require less porosity, higher hydrophilicity, and less roughness. It is also interesting that for porous specimens the best apparent wetting angle would be close to amphoteric (90 ± 10°), and in this case neither too hydrophilic nor too hydrophobic surfaces are desirable, Figure 5 and Figure 7. The onset threshold of porosity is about 0.47 for these experiments, which means lower porosity specimens are recommended to be more hydrophilic, whereas highly porous specimens should be more hydrophobic (but in all cases roughness parameter Ra should be minimal). This might be adjusted by application of a titania nanolayer, e.g., by hydrothermal treatment.

To predict the variation of biofilm formation beyond the experimental limits, metamodeling (response surface modeling, RSM) has been carried out for the whole data set. Figure8 shows an example of RSM output for untreated specimens using a radial basis functions method.14, 15 One can see that the worst cases, where most of biofilm is expected (“red area”), are specimens with a porosity level of 0.20–0.35 and a roughness Ra > 20 μm.

thumbnail image

Figure 8. Response surface model (RSM metamodel) for BFF vs. roughness (Ra [μm]) and porosity fraction, for IPC mean size = 8.8 μm and without anatase. Dots represent both experimental and SOM (virtual data) values.

Download figure to PowerPoint

3.3. Cell Viability, Proliferation, and Differentiation

Although the biofilm formation might be greatly reduced with the optimal surface topology as shown above, this alone does not yet lead to the “optimal” implant surface. The interaction of cells, i.a., osteogenic stem cells arising from human bone marrow (HOC) and human endothelial cells (HEC), with the surface of biomaterials, is of great relevance for osteointegration. It is not enough to have the lowest BFF only, the surface of the biomaterials also has to be tailored to enhance cell adhesion, growth, differentiation, and elaboration of a mineralized extracellular matrix (ECM). Adhesion of cells to the material plays a fundamental role in regulating ECM synthesis. The physicochemical characteristics of a material surface regulate serum protein adsorption, and therefore cell adhesion.

As for bacterial attachment, HOC and HEC attachment and proliferation is a complex process. Figure9 shows examples of normalized HOC proliferation (after 9 days), alkaline phosphatase, and collagen I (1 day). The HOC proliferation data seems to favor low-porous, smooth specimens (although with a larger data scatter). However, as shown above, porosity or its topology do not act independently and there is a “hidden” contribution from other factors such as wetting. Similar pictures are observed for other incubation times. Alkaline phosphatase (Figure 9b) and collagen I gene expression (Figure 9c), after the first day, show opposite trends in respect to porosity variations, although differences for alkaline phosphatase are not as high as observed for collagen I (the latter is similar to HOC proliferation, Figure 9a).

thumbnail image

Figure 9. Example of a) HOC proliferation after 9 days of incubation (HOC9), b) alkaline phosphatase after the 1st day (APh1) and c) collagen I expression after 1st day (ColI_1) versus specimen porosity, IPC mean size [μm] (as color scale) and roughness (bubble size – Ra [μm]). All data are normalized to VPS Ti.

Download figure to PowerPoint

The overall Student's factor analysis of analyzed cell proliferation and differentiation variables with surface parameters is shown in Figure10. There is an additional composite variable (CyRsum) – the sum of all cytoreactions data (normalized proliferation + normalized gene expression). The goal of this variable is to summarize all data for HOC and HEC incubation days to be maximized. This is expected to give a rough indication of the best relative osteointegration ability versus control VPS Ti specimens. Student's analysis gives a hint towards higher importance of porosity for HOC proliferation, von Willebrand factor, alkaline phosphatase, and CD31, Figure 10. The IPC mean channel size, on its turn, seems to be the most important for HOC, HEC proliferation, and less for collagen I, which is dominated more by roughness values.

thumbnail image

Figure 10. Student's overall factor analysis on HOC and HEC proliferation and differentiation for all specimen groups.

Download figure to PowerPoint

3.4. Combined Effect of all Factors

Cell proliferation and differentiation data were finally analyzed together with biofilm formation data. This allows searching for the best optima related to the titanium implant surface parameters. The target objectives set in this study are the minimization of biofilm formation (minimum BFF, implying least S. aureus and S. epidermidis formation at any and all incubation days) and maximization of osteointegration potential (CyRsum). The latter variable may not be straightforward, and there might be several opinions of its proper meaning. As defined above, it was simply chosen to be the sum of all “positive” results, i.e., higher HOC and HEC proliferation values in addition to a higher gene expression and differentiation. No specific weights were assigned to these separate input and output variables (Table 3), although this could be done if a reasonable justification would be provided, e.g., whether higher alkaline phosphatase should be more important than collagen I expression. The sample plot of all these cytoreactions versus BFF with impacts of porosity and roughness is shown in Figure11. Here, one can see a slight trend of cytoreactions to proceed better when the BFF values are lower. Although there is some scatter, especially at low porosity levels (uncoated substrates) and for smooth titanium specimens, there is a clear trade-off area (outlined) with average porosity values, where CyRsum > 1 and BFF ≪ 1. In this work, we are particularly concerned about the reduction of biofilm formation first, with lower importance on the improvement of the cell osteogenic potential. The same data can be shown in different coordinates but the optimal specimens' area will be the same.

thumbnail image

Figure 11. Variation of total cytoreactions output versus BFF output (normalized to VPS Ti) for different porosity (color scale) and roughness Ra [μm] (as bubble size). The optimal specimen area is outlined.

Download figure to PowerPoint

3.5. Discussion

By application of MVA to the combined dataset, including all cytoreactions and bacterial reactions, it is possible to find the optimal area, which should “guarantee” the lowest biofilm formation probability (i.e., the lowest infection risk) while having the same or better osteointegration potential (higher cell proliferation, gene expression, differentiation values, etc.). Taking into account all these objectives and setting up the additional constraint, i.e., porosity must be present for bone in-growth, the optimal implant surface could be identified, Table4.

Table 4. Results of optimization of the titanium implant surfaces with respect to different objectives (used constraint: porosity level should be at least 0.15).
Variable (independent)For minimizing biofilm formation (BFF [RIGHTWARDS ARROW] min)For maximizing osteointegration potential (CyRsum [RIGHTWARDS ARROW] max)For both objectives simultaneously (“best implant”)
Roughness (Ra) [μm]0.5–5.00.5–5.01.5–3.0
TiO2 thickness [nm]10–20010–805–65
Porosity (total open)0.20–0.500.30–0.500.30–0.45
IPC mean size [μm]1.5–7.52.5–6.52.5–5.0

Thus, the optimized titanium implant surface should have a total porosity in the range of 0.35 ± 0.10, an interconnecting pore channel mean size of ∼3.5 μm, a surface roughness Ra between 1.5 and 3.0 μm, and a surface treatment to form an anatase layer (∼40 ± 20 nm thick). This combination of parameters for coatings and surfaces similar to used in this study would automatically ensure proper values of Rz, the wetting angle, and the maximal pore size, if no other extra factors (e.g., a specific nanoroughness, which was not analyzed in this study) are introduced additionally.

According to the experimental data, this combination would lead to a decrease in combined biofilm formation (normalized sum for S. aureus and S. epidermidis) by ∼80% compared to the state-of-the-art VPS Ti substrate at all incubation days. At the same time, this will also promote a higher human osteogenic cell proliferation (+10–30%) and a higher cell differentiation (e.g., CD31 expression +15–40% and collagen I expression +50–80%) compared to VPS Ti. Thus it is possible to reduce the risk of biofilm formation by almost five times keeping the high osteointegration potential, just by producing the optimized surface topology of titanium implants.

Regarding the validity of these findings, it is necessary to note that the conclusions obtained in this meta-analytical procedure were based solely on our own in vitro experiments, which nevertheless had a limited number of surface variations and specimens, and might not cover all relevant implant surface characteristics. Furthermore, for some implants (such as dental implants) there are different requirements (i.e., porosity level) than for orthopedic ones, and thus the application of these results to dental implants would result in a different optimal set of parameters.

The optimal topology of the coating for a good bone in-growth remains one of the most debated topic. Black24 mentions soft tissues found in pore sizes as small as 1 μm, whereas osteonal bone requires pore sizes of ∼250 μm, and mineralized tissue occupies a porosity range between these extremes. However, in many previous publications a more detailed characterization of topology of the pores (such as IPC and connectivity) is missing. An additional combination of micromotion, shear stresses, microfludics, etc., in vivo might significantly affect the optimal parameters values (Table 4). Therefore the benefits of the method used could be better exploited if a larger dataset would be obtained, which also has to be complemented by in vivo experiments and longer clinical trials. The analysis has revealed some nonlinear dependence between surface parameters, bacteria, and cell reactions, and it would be useful to expand such treatment to existing independent data, providing the specimen history and characteristics are documented. At the moment, similar algorithms are used mostly to biomechanical properties analysis, optimization and prediction, and genomics or pharmacological data analysis,25–27 but the potential of multi-objective optimization and data visualization application to biomaterials analysis, in particular involving bacterial and cell interactions, is very high and it provides an excellent opportunity to look at the experimental data from a new perspective.

4. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

Several porous and non-porous commercial titanium substrates have been studied from the point of view of a combined effect of variations in porosity, pore topology, roughness and hydrophobicity on bacterial attachment and cell reactions (proliferation and gene expression). Due to mutual correlations between the surface parameters, a meta-analysis of the data has been applied to assess the effect and contribution of these factors on biofilm formation and cell reactions. In addition to expected dependencies, such as hydrophilic state and probability of bacterial attachment, complex links between these input variables with output data have been elucidated and quantitatively evaluated.

The experimental in vitro data are clearly indicating the possibility of decreasing biofilm formation by 80–90% for flat substrates versus untreated and non-optimized plasma-sprayed porous titanium, but also by 65–95% for other porous titanium coatings. Moreover, these porous surfaces are shown to lead to a 10–50% enhanced cell proliferation and gene expression versus state-of-the-art vacuum plasma sprayed porous titanium coatings. This presents an opportunity to manufacture implant surfaces with intrinsic reduced infection risk without the additional use of antibacterial substances.

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

Supporting Information (including the experimental data matrix containing the measured parameters) are available from the Wiley Online Library or from the author.

Acknowledgements

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

Financial support of the European Commission under FP6 integrated project No. 026501-2 “MEDDELCOAT” is gratefully acknowledged. Disclosure: The authors confirm that there are no known conflicts of interest associated with this publication and there has been no significant financial support for this work that could have influenced its outcome.

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Materials and Methods
  5. 3. Results
  6. 4. Conclusions
  7. Supporting Information
  8. Acknowledgements
  9. Supporting Information

Detailed facts of importance to specialist readers are published as ”Supporting Information”. Such documents are peer-reviewed, but not copy-edited or typeset. They are made available as submitted by the authors.

FilenameFormatSizeDescription
adhm_201100006_sm_suppl.pdf186Ksuppl

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.