Experimental study on the cytocompatibility of milling surface of poly-ether-ether-ketone (PEEK)

Poly-ether-ether-ketone (PEEK) is widely used in biomedical science because of its biocompatibility. However, the material is biologically inert. Many works have showed the potentiality of surface modification technique to enhance its biological performance, such as surface coating and impregnating bioactive materials into its substrate. Even so, these techniques are low efficiency or inapplicable in processing parts with complex structure. This thesis presents an experimental work of milling for the surface modification of PEEK, and its impact on the biological response is addressed. First, the influence of milling parameters on machined surface topography features is discussed based on the Taguchi method and end milling experiments. And then, the effect of the machined surface topography on the viability of fibroblasts is investigated by using cells from a mouse fibroblast L929 and the agar overlay test, MTT test. The experimental result indicated that there is a certain surface topography (characterized by Ssk = 0.011), which ensures the highest possible cells viability, but a significant decrease of L929 viability started with the surface roughness of Sa = 1.18 μm. The axial and radial depths of cut are the two most critical factors to affect surface topography parameters. It is concluded that the milling process significantly affects surface topography of PEEK, which in turn influences the L929 viability.


Introduction
Poly-ether-ether-ketone (PEEK) is considered an important engineering material for orthopedic applications due to its excellent combination of properties such as low density, high strength and durability, good biocompatibility, and better environmental resistance in addition to high service temperature [1]. Despite these excellent properties, PEEK is still categorized as bio-inert due to its very low reaction with the surrounding tissue, which limits its potential applications [2,3].
Adopting appropriate manufacturing practices offers the possibility to activate the surface of PEEK implants, including by surface coating, surface treatment, and impregnating bioactive materials into its substrate [4,5].
Furthermore, the existing literature reporting on increasing PEEK implants' bioactivity indicates that implant surface topographical features such as grooves, ridges, and pits can affect osteoblast proliferation rate to a large extent. Riveiro et al. [6] used laser surface texturing (LST) technique to control the surface properties of biomedical polymers and enhance their biological performance. Scott [7] observed pre-osteoblast cell adhesion to a textured surface to identify cell preference for surface geometry of PEEK. The surface textures with different shape, spacing, size, depth, and surface convexity/concavity were embossed into PEEK using micro-etched aluminum molds. The results were that 25 µm and 10 µm features discouraged cell adhesion while 325 µm and 120 µm features encouraged cell adhesion with pillars performing better than holes. Prochor et al. [8] indicated that surface roughness could significantly influence the viability of human osteoblasts, and there is a certain surface topography which ensures the highest possible viability of human osteoblasts. Sidambe et al. [9] studied the effect of SLM-induced surface topographies on the in vitro biological responses of mouse fibroblasts (L929).
The surface roughness amplitude parameters were found to discriminate cell viability and cytotoxicity better. Even so, all these techniques mentioned above are low efficiency or inapplicable in processing parts with complex structure.
Because the milling process has high efficiency and great flexibility and diversity of structure processing, it has been more widely used. Due to the different milling parameters, the traces left on the machined surface are different in depth, density, shape, and texture [10]. Two-dimensional (2D) surface parameters such as profile average height R a and the average maximum height of the profile R z are always used to characterize the machined surface topography of PEEK and its composite material [11,12]. Li et al. [13] studied the milling process of biocompatible PEEK materials, reporting that the significance order of the effects on surface roughness from large to small is the axial depth of cut, feed rate, and radial depth of cut, respectively. Cao et al. [14] found that cutting speed is the key factor affecting the machined surface topography in milling carbon fiberreinforced PEEK. Izamshah et al. [15] claimed for milling PEEK that the geometric characteristics of end milling cutter have an important influence on the machined surface roughness and the formation of burr. Increasing the rake angle and helix angle will improve the machined surface roughness. However, Rahman et al. [16] claimed the opposite to Izamshah's conclusion that the smooth surface finish of the PEEK workpiece is caused by some form of polymer softening action rather than determined by the tool geometry of the cutter. The surface of the processed material has microscopic geometric features composed of small spacing, tiny peaks, and valleys, etc.
2D surface roughness parameters provide a simple and valuable method for quantifying contour height distribution, but they are inadequate for functional surface evaluation. Milling-induced surface topographical features are orientation dependent that surfaces with small 2D surface roughness values do not guarantee high performance at the same time [17,18]. As a consequence, it makes sense to evaluate functional surfaces with threedimensional (3D) surface parameters. Recent advanced detection equipment and techniques such as white light interferometer and laser-scanning confocal microscopy have enabled us to measure and characterize surface topography in 3D [19,20]. In addition, a series of 3D surface topography parameters, including amplitude parameters, spacing parameters, and hybrid parameters, have been proposed [21]. Notwithstanding, no systematic study has been reported to characterize the machined surface defects and the three-dimensional (3D) surface topography parameters of PEEK in milling.
This investigation aimed to characterize the surface topographical features and to assess the biocompatibility of PEEK processed using end milling technology. The influences of process variables like cutting speed, feed rate, and axial and radial depth of cut are investigated simultaneously to study their effects on 3D surface parameters using the Taguchi optimization method and multiple linear regression mathematical model. Furthermore, agar overlay test coupled with MTT assay was used to evaluate the biocompatibility of PEEK with different milling surface features. The study offers a new insight on the surface modification of PEEK material and gives technical references to enhance its biological performance.
The research outline of the present study is shown in Fig. 1.

Milling experimental design
Multiple objective optimizations of 3D surface topography parameters are considered within the framework of the Taguchi method and linear regression mathematical analysis. The Taguchi method is an effective experimental design method. The number of special experiments can be significantly reduced by using the orthogonal design. Taguchi parameter design is widely used in cutting test design and test result in analysis. It provides a simple and systematic approach for optimizing machined surface topography [22,23].
In the present study, four milling parameters, including spindle speed n, feed rate f, axial depth of cut a p , and radial depth of cut a e , were studied. Three levels for each factor were applied. Taguchi L9 (3 4 ) orthogonal test table for scheduling parameters and the orthogonal array are shown in Table 1. According to the Taguchi method, nine experiments can provide the optimized parameters. Each run was repeated three times to exclude the influence of random factors (or noises) on the results.
The regression analysis method, a predictive modeling technology, is always mentioned in statistics. The multiple regression method can determine the correlation between a continuous dependent variable and two or more continuous or discrete independent variables [24].
In this investigation, the multiple linear regression mathematical models were developed to study the influence of four process parameters, namely, spindle speed (n), feed rate (f), axial depth of cut (a p ), and radial depth of cut (a e ) on 3D surface topography parameters (S j ). The non-linear response surface equations of S j for milling of PEEK are given by Eq. (1): where k 0 , k 1 , k 2 , k 3 , and k 4 are the regression coefficients to be determined. The absolute value of the regression coefficients in the established models reflected the sensitivity of the surface roughness to the cutting parameters.

End milling process of PEEK
The PEEK used for studies is a natural Ketron 1000 PEEK plate from Quadrant Engineering Plastic Products (EPP). The main mechanical and thermal properties of this workpiece material supplied by the manufacturer are summarized in Table 2.
E n d m i l l i n g ex p e r i m e n t s we re c a r r i e d o u t in rectangular block workpieces with a size of 50 mm × 50 mm × 5.0 mm, as shown in Fig. 2a. A vertical type machining center DAEWOOACE-V500 was used, and the experiments were performed with dry cutting to avoid the pollution of the machined surface caused by the use of cutting fluid. The workpiece was fixed on the machine table with medical glue (adhesive) to make the mechanical clamping force small. The workpiece was processed by an end milling cutter, which is made of cemented carbide. The milling mode is down milling. The diameter of the end milling cutter is 10 mm, and as shown in Fig. 2b, there are four flutes with variable helix angles (β 1 = β 3 = 38° and β 2 = β 4 = 41°).

Machined surface characteristics
After fabrication, the optical microscope KEYENCE VR-3000 was used to measure the machined surface morphology. The evaluation area of 3D surface roughness parameters was set as 1.43 mm × 1.07 mm. The evaluation area was picked at three different locations on the machined surface, and the average values of surface roughness parameters were obtained.
Milling processes with end mills of defined geometry generate severe anisotropic patterns. Considering the existence of anisotropy on the machined surface and their topographic properties that are statistically dependent on the measuring direction over the surface, 3D surface parameters are measured. Not only will it be closely related to the properties of a functional surface, including reflexivity, friction and wear, lubrication, and fatigue corrosion [25], the amplitude parameter of surface roughness can also better distinguish cell viability and cytotoxicity [9]. Therefore, the present study used 3D amplitude surface topography parameters S a and S z together with S sk and S ku to characterize the machined surface topography of PEEK.
S a is the average height difference of each point on the machined surface and represents the average surface relative to the reference datum. S z is the average maximum height of the sampling area. Skewness parameter S sk describes the asymmetry of the surface that is deviating from the mean plane. S ku is  an index reflecting the steepness or flatness of the topography height distribution, which is always presented in conjunction with the skewness S sk . The mathematical expressions of S a , S z , S sk , and S ku are as follows: where Z(x,y) is the machined surface topography data, Z si and Z vi (i = 1, 2,…,5) are the five highest surface summits and deepest pits, respectively, and A is the sampling area. S q is the root mean square deviation parameter, which is defined as follows:

In vitro cytotoxicity tests
In vitro cytotoxicity test is a preliminary screening test to evaluate the biocompatibility of biomaterials. The main techniques involved are the agar overlay method and MTT assay. The agar overlay method was developed by Guess et al. as a semi-quantitative detection method for the evaluation of the cytotoxicity of biomaterials [26], and the MTT assay is a quick and quantitative method for biocompatibility test in vitro, which Mosmann proposed in 1983 [27].
In the present study, the agar overlay test is first applied to determine the leachability from PEEK affecting cell activity.
After the agar overlay test is finished, the MTT test, a direct contact test, is performed to investigate the effect of the machined surface topography on the viability of fibroblasts. Before cytotoxicity tests, the workpiece was first cleaned by ultrasonication of 240W for 30 min. Then, samples were exposed to ultraviolet (UV) light for 30 min at room temperature for sterilization. After cleaning with ultrasonication, some adhered material particles can be removed.
In agar overlay tests, mouse fibroblast L929 cells were cultured in DMEM medium (10% fetal bovine serum, penicillin 100 U/mL, and streptomycin sulfate 100 μg/mL) at 37 °C in wet air with 5% CO 2 and then digested with 0.25% trypsin containing EDTA (ethylenediaminetetraacetic acid) to obtain a 2.5 × 10 5 cells/mL suspension. The suspended cells were dispensed 10 mL in a 100-mm plate and cultured at 37 °C, 5% CO 2 for 24 h. The sterilized agar solution was cooled to 48 °C and mixed with preheated DMEM at 48 °C. The original culture medium was discarded, and 10 mL of mixture was added and solidified at room temperature. Then, the samples and the filter paper with extract of negative control and positive control were put on the agar. This was done in triplicate in each group and cultured at 37 °C, 5% CO 2 for  Fig. 2 Workpiece to be machined and the end milling cutter another 24 h. After 24 h of incubation, the sample outline was marked at the bottom of the Petri dish and then the sample was taken out. The growth states of the cells were observed microscopically. According to ISO 10993-5:2009 standard [28], the cytotoxicity was indicated with grade "0" to "4," which means that the degree of reaction is from none to severe.
In MTT tests, L929 cells were cultured in DMEM (10% fetal bovine serum, penicillin 100 U/mL, and streptomycin sulfate 100 μg/mL) at 37 °C in a humidified atmosphere of 5% CO 2 and then digested by 0.25% trypsin to obtain a single cell suspension. Also, 1 × 10 5 cells/mL suspension was obtained by centrifuging (1000 rpm, 5 min) and finally re-suspended in DMEM. The suspended cells were dispensed at 100 μL per well in a 96-well plate and cultured in a cell incubator (37 °C, 5% CO 2 ). The cell morphology was evaluated to verify that the monolayer was satisfactory. After 24 h of culture, the cell morphology was observed first, and then the culture medium was discarded. Also, 50-μL aliquots of MTT (1.0 mg/mL) were added to each well and then incubated at 37 °C in a humidified atmosphere of 5% CO 2 for 2 h. The liquid was poured out from each well and 100% isopropyl alcohol was added to each well per microliter.
By contrast, blank control, negative control, and positive control tests were performed to compare the difference of cell proliferation in each group. The control samples' information is shown in Table 3.
The optical density (OD) value was evaluated using a dual-wavelength spectrophotometer with a measurement wavelength of 570 nm and a reference wavelength of 650 nm. Mean-standard deviation ( x ± SD) and the cell RGR (relative growth rate), which is defined by Eq. (7), are recorded.
Here, OD 570 and OD 650 are the OD values measured with the wavelength at 570 nm and reference wavelength at 650 nm, respectively.

Correlation analysis
To describe the linear correlation between the viability of fibroblasts and the machined surface roughness, correlation coefficient r(S j ,OD) was adopted, which is defined as Eq. (8): where S j is a surface roughness symbol, j = a, z, sk, and ku. Cov(S j ,OD) is the covariance of S j and OD. Var[S j ] and are the variances of S j and OD, respectively. The value of r(S j ,OD) lies in the range [0,1], with "0" indicating no correlation and "1" indicating perfect correlation.

Machined surface defects
Machining processes generated severe anisotropic patterns. Figure 3 shows the milling-induced surface topographies of partial PEEK samples. Delicate grooves and ridges can be observed on the machined surface, and they are indicators of the dynamic track of cutting edge. The distance between the two adjacent ridges is determined by the feed rate of the milling cutter, and the larger the feed rate, the larger the interval of the adjacent ridges. The surface topographies recorded for the minimum and the largest feed rate are shown in Fig. 3b and c, respectively. Optical micrographs of machined surfaces obtained under different cutting parameters are given in Fig. 4. Feed marks are visible defects in all typical images. In addition to the feed marks left on the machined surface, scratch marks, surface abrasion, plucking, budlike protuberances and adhered surface particles, etc. are easily observed.

Surface topography parameters
S a , S z , S sk , and S ku are in the ranges of 1.18 to 2.49 μm, 22.54 to 47.61 μm, 0.011 to 0.675, and 2.93 to 4.398, respectively. Also, their responses to cutting parameters are plotted, as shown in Fig. 5. The plots show the variation of each surface roughness parameter with the four cutting parameters, including cutting speed (spindle speed n), feed rate, axial depths of cut, and radial depth of cut, respectively. In the plots, the x-axis represents the value of each process parameter at three levels, and the y-axis represents the response value of surface roughness. The main effect diagram is used to determine the optimal design milling parameters to obtain a lower surface roughness value. As shown in Fig. 5a, values of S a fluctuate around 2.0 μm and almost no change with the increased cutting speed and feed rate, while an increase of a p or a e causes a rapid increase of S a . S z initially decreases with the cutting speed (up to 3000 rev./min) and then increases with the further increased cutting speed. S z is a negative proportion to feed rate. S z has the minimum value when the axial depth of cut a p takes the intermediate value 1.0 mm. S z increased with the increase of radial depth of cut a e , and the increase rate accelerates when a e exceeds 4.0 mm, as shown in Fig. 5b. S sk is the parameter to judge the concave-convex tendency of roughness shape. The record values of S sk under different cutting conditions are all positive in the present study, which means that there are more peak data higher than the mean plane. The higher the value of S sk , the greater the degree of data deviation from the mean plane. As seen from Fig. 5c, S sk initially increases with the cutting speed and then decreases with the further increased cutting speed. S sk almost linearly increases with the increased feed rate because of the enlarged scallop height by milling with a high feed rate. For the effects of a p and a e , S sk is slowly decreasing after a rapid increase. The maximum value of S sk exists in the condition of a p = 1.0 mm and a e = 4.0 mm. The parameter S ku represents the extension of the height distribution, and the kurtosis value of the Gaussian surface is 3. In the present study, the threshold value (that is 3) lies in the variable range of S ku . A kurtosis value that is larger than 3 indicates a centrally distributed surface, while the one On the basis of the recorded values of S a , S z , S sk , and S ku , the mathematical models of cutting parameters for the optimized surface roughness were established using multiple linear regression analysis, as described in Eqs. (9)- (12). Here, S j,expt is the measured surface roughness of response corresponding to jth test, and S i,pre is the predicted surface roughness of response corresponding to jth test. m is the number of tests. By calculation, the prediction errors of S a , S z , S sk , and S ku models are 5.2%, 7.3%, 14.2%, and 4.8%, respectively.

In vitro cytotoxicity evaluation
Agar overlay tests determined reactivity grades of cytotoxic. After 24 h of incubation, decolorization zone and cytolysis were easily observed around positive control materials. Meanwhile, no obvious cytolysis in the negative control group was observed. For the test group, some malformed or degenerated cells were observed, which means that a mild/slight reaction has occurred.
The activity of L929 cells on different test samples was detected with the MTT method. Before inoculation, fibroblasts were homogeneous round cells, and the newly inoculated fibroblasts were spherical, suspended in the culture medium. After 24 h of incubation, the cells grew well, and the number of cells increased significantly in the negative control group, as shown in Fig. 6a. Yet, some of the cells in the positive control group gradually subsided and became spherical. It can be seen from Fig. 6b that the number of cells disintegrated and living cells decreased significantly. In an aspect of the experimental group, the cell growth in the extracts was poor, and some of the cells were deformed and shrunk, as shown in Fig. 6c and d, compared with the cells on the machined surface (obtained under the cutting condition of test T02 and T07, respectively).
For the experimental group, a difference between the cell's viability of the test samples is observable. Figure 6 also presents the mean-standard deviation ( x ± SD) of the optical density value (OD) and the cell RGR . It should be noted that the recorded OD and the RGR are the average (9) S a = 0.412n 0.045 f 0.026 a 0.238 Figure 7 shows the calculated correlation coefficients of the viability of fibroblasts to the machined surface roughness S j (j = a, z, sk, and ku). The results of correlation analysis indicated that the correlations order of the viability of fibroblasts with the machined surface roughness parameters were S sk , S a , S ku , and S z in large to small.
The viability of L929 cells to the surface roughness of the tested samples is plotted in Fig. 8. It can be noticed in Fig. 8a that the viability of fibroblasts falls with the increase of roughness S a . A significant decrease of L929 viability started with the surface roughness of S a = 1.18 μm, and the changing trend will become not pronounced when S a increases to a particular value. Surface roughness characterized by S sk = 0.011 (namely, S sk closed to zero) also achieved the highest possible cell viability. When analyzing the effects of S ku and S z , it can be noticed that L929 viability decreased after an initial increase of S ku and S z . Further analysis can also indicate that a proper centrally distributed surface is conducive to enhancing L929 viability.

Discussion
Currently, there is continuous research to activate the surface of PEEK implants by adopting appropriate manufacturing practices. This study aimed to investigate the in vitro cytotoxicity of PEEK that was produced using end milling technology. As presented in Figs. 3 and 4, typical defects of the machined PEEK are feed marks, scratch marks, budlike protuberances, adhered surface particles, etc. Feed marks are inherent in the cutting process, which are produced by the combination of spindle rotation and cutter movement, namely, the plane helical motion of the end milling cutter. Scratch marks are also common in all machined samples, especially under the cutting condition of high feed rate, as shown in Fig. 4c, which can be attributed to the high-frequency friction between the flank of the cutting tool and the machined surface of the workpiece. Severe plastic deformation of machined surfaces results in the formation of budlike protuberances. Due to the low yield strength of PEEK, and the role of the cutting force, it is easy to produce plastic deformation in processing.
As the cutting process goes on, a built-up edge (BUE) causes the formation of plucking and scratching. Surface abrasions are typically formed by plucking in the process of cutting. Hard particles slide between the tool and the machined surface, resulting in scratches and wear on the machined surface. According to Ref. [17], these particles are a small part of the cutting tool (made of carbide) and/or the part of BUE (deformed chip), which are harder than the bulk material. There are presumed to be two kinds of adhered material particles caused by the debris of microchips or detached builtup edge. Due to the high cutting temperature, debris and the detached built-up edge are welded to the machined surface, which will lead to an increment of the values of the amplitude parameters of the surface roughness. Cutting temperature is in direct proportion to the speed of cutting. As a consequence, more material particles are deposited on the machined surface at high milling speed conditions. From the relationships between cutting parameters and surface topography parameters, it is clear that different surface topography parameters give different responses to cutting parameters. Thus, all these topography parameters' simultaneous reduction or increment requires a trade-off. The mathematical models, presented in Eqs. (9)- (12), are used to assess the sensitivity of the surface roughness to the cutting parameters. The parametric analysis reveals that all the studied surface parameters are most sensitive to the variation of radial depth of cut a e . For S a and S z , the remaining sensitive factors are axial depth of cut a p , feed rate f, and spindle speed n. However, for S sk and S ku , the second sensitive factor is the axial depth of cut a p and spindle speed n, respectively. Feed rate f is the most insensitive factor for S sk and S ku .
The study of agar overlay tests produced no evidence of cell damage caused by PEEK, and a conclusion can be drawn that PEEK has no cytotoxicity to L929 cell. L929 cells viability and cytotoxicity test results, presented in Fig. 6, show that milling-induced surface topography can affect the attachment and growth of cells on the PEEK surface for L929 fibroblasts could respond to surface topography features by adjusting adhesion, migration, morphology, and cytoskeleton [29]. Further analysis (as shown in Fig. 7) can also indicate that the asymmetry of the surface topography discriminated cell viability and cytotoxicity better, verifying that L929 fibroblast cell growth on machined surfaces was more influenced by the asymmetry rather than the average maximum height of the machined surface.
Many biocompatibility assessment reports showed that more proteins might be adsorbed on rough surfaces and increased proliferation and osseointegration on micron-scale features compared to smooth controls [29]. The reason may be that L929 cells are forced to locate the features of small cells guided by existing features when they are highly similar to cells at the micron level. In an aspect of the height difference parameter S a , the larger height differences led to continuous cell proliferation and attachment until the cells could elongate and further spread on the surface [30]. It is thus clear that there is a specific surface roughness threshold to define L929 viability, and as presented in Fig. 8a, the threshold value of S a is 1.18 μm.
It is clear from Fig. 8b that surface roughness characterized by S sk = 0.011 (namely, S sk close to zero) also allowed to achieve the highest possible cell viability. This means that high cell viability could be obtained when the machined surface height distribution exists symmetrically with respect to the mean plane. This phenomenon can be attributed to the local adhesion mechanism of cells. It has been reported that the cell adhesion mechanism affects the cytoskeleton according to the size of the groove [31].
The results suggest that there may exist a specific condition for milling PEEK to obtain a surface with excellent cytocompatibility. In consideration of the relationship between the fibroblast's viability and the machined surface roughness and the influence of milling parameters on machined surface roughness, the milling condition of n = 5000 rev./min, f = 0.05 mm/rev., a p = 0.5 mm, and a e = 3.0 mm was proposed in the present study.

Conclusions
This study is devoted to characterizing the machined surface topography features and assessing the cytotoxicity of PEEK processed using end milling technology. A detailed analysis, including milling-induced surface topographical features, the change of surface topography parameters over milling parameters, and the viability of fibroblasts to the machined surface roughness, has been carried out. The conclusions from this research are drawn as follows.
Machined surface topographical features in milling of PEEK were revealed, such as feed marks, scratch marks, surface abrasion, plucking, budlike protuberances, and adhered surface particles. All these features determine the roughness of the machined surface.
The significance sequences of the key milling parameters on surface roughness were identified. S a , S z , S sk , and S ku are the most sensitive to the change of a e . Different surface parameters have different sensitivity sequences. For S a and S z , the remaining sensitive factors are axial depth of cut a p , feed rate f, and spindle speed n, respectively. However, for S sk and S ku , the second sensitive factor is the axial depth of cut a p and spindle speed n. Feed rate f is the most insensitive factor for S sk and S ku .
The agar overlay test results showed that the end milling process could be used successfully to prepare safe and biocompatible PEEK. A difference between the cell's viability of the test samples was observable by MTT tests. There was a certain surface topography (characterized by S sk = 0.011), which ensures the highest possible cell viability, but a significant decrease of L929 viability started with the surface roughness of S a = 1.18 μm.
Author contribution All of the authors contributed to the study conception and design. Methodology was proposed by QL, and the cytotoxicity evaluation and data analysis were performed by XL. The first draft of the manuscript was written by XL and revised by QL. All authors read and approved the final manuscript.