Principal Component Analysis to Determine the Surface Properties that Influence the 1 Self-Cleaning Action of Hydrophobic Plant Leaves

23 It is well established that many leaf surfaces display self-cleaning properties. However, an 24 understanding of how the surface properties interact is still confounding. Consequently, twelve 25 different leaf types were selected for analysis due to their water repellency and self-cleaning 26 properties. The most hydrophobic surfaces demonstrated splitting of the v s CH 2 and v CH 2 27 bands, ordered platelet-like structures, crystalline waxes, high surface roughness values, high 28 total surface free energy and apolar components of surface energy, and low polar and Lewis 29 base components of surface energy. The surfaces that exhibited the least roughness and high 30 polar and Lewis base components of surface energy had intracuticular waxes, yet still 31 demonstrated self-cleaning action. Principal component analysis demonstrated that the most 32 hydrophobic species shared common surface chemistry traits with low intra-class variability, 33 whilst the less hydrophobic leaves had highly-variable surface chemistry characteristics. 34 Despite this, we have shown through partial least squares regression that leaf water contact 35 angle (i.e. hydrophobicity) can be predicted using attenuated total reflectance Fourier transform 36 infrared spectroscopy surface chemistry data with excellent ability. This is the first time that 37 such a statistical analysis has been performed on a complex biological system. This model 38 could be utilised to investigate and predict the water contact angles of a range of biological 39 surfaces. An understanding of the interplay of properties is extremely important when 40 producing optimised biomimetic surfaces. 41


INTRODUCTION 45
There has been significant interest directed towards producing biomimetic surfaces 46 with controlled surface wetting properties. 1 Much of this work has concentrated on altering 47 surface topography and chemistry to produce superhydrophobic surfaces. It is generally 48 considered that the topography of plant surfaces is the main factor influencing water contact 49 angle, and hence water repellency. 2,3 Specifically, hierarchical structures at the macro and 50 micro levels (the Lotus effect) are associated with superhydrophobicity of leaf surfaces. 4 The 51 leaves are also self-cleaning, meaning that rolling droplets can remove microorganisms and 52 other contaminants from their surfaces. Numerous biomimetic surfaces have been developed 53 which emulate the topography of superhydrophobic leaves to achieve self-cleaning, water 54 repellency, and anticontamination properties. [5][6][7][8] However, many self-cleaning surfaces 55 produced with biomimetic topographies still require chemical modification to exhibit 56 superhydrophobicity. Many plant surfaces are hydrophobic (WCA >110°) or 57 superhydrophobic (WCA > 150°). 9 However, in nature, there are also several leaf surfaces that 58 display self-cleaning and water-repellent behaviours, and yet they are not superhydrophobic 59 and may not have predominant topographical features. 60 It is well established that the wax layer on leaf surfaces, in particular epicuticular wax 61 crystals, makes an essential contribution to surface hydrophobicity. 10 The chemical 62 compositions of such waxes from numerous leaf surfaces have been determined. 11-17 However, 63 the exact relationship between the surface chemistry and topography, in addition to their 64 influence on surface physiochemistry is not fully understood. Consequently, producing 65 biomimetic surfaces that maintain their anti-wetting features still presents a significant 66 challenge. Thus, an understanding of the key surface properties that result in the water 67 repellency of natural surfaces is essential to further the development of biomimetic surfaces. The aim of this work was to determine the relationship between the surface topography, 69 chemistry, and physiochemistry of a selection of plant leaves that demonstrated self-cleaning 70 properties. This was implemented through a combination of complimentary experimental 71 techniques and modelling methods to identify the key parameters that resulted in the self-72 cleaning properties of these natural surfaces. This information is vitally important to many 73 aspects of industry where producing low-cost and consistent biomimetic surfaces is a priority.  The total surface free energy (γs) and the apolar (γs LW ), polar (γs AB ), Lewis acid (γs + ), 94 and Lewis base (γs -) free energy components of the adaxial surfaces of the leaves were 95 determined using contact angle goniometry. The surface free energy components of these three liquids were taken from Bos et al. 21 107 (Supporting Information: Table S1).

108
The contact angles of each test liquid were obtained from five different areas on the 109 leaf, therefore average values were used to obtain the physicochemical parameters. The 110 statistical error in the calculated surface energy components was estimated from the contact 111 angles of each test liquid by using propagation of error principles. The interfacial free energy 112 (ΔGiwi) was used as a measure of the hydrophobicity of a leaf surface where greater (negative) 113 ΔGiwi values related to more hydrophobic surfaces.

115
Optical profilometry and scanning electron microscopy (SEM) 116 The surface topographies of the leaves were investigated using a previously described 117 method with a MicroXAM (phase shift) surface mapping microscope (ADE corporation, XYZ 118 6 model 4400 ml system, USA). 22 The optical profilometer used an AD phase shift controller 119 (Omniscan, UK). A MAPVIEW AE 2.17 (Omniscan, UK) image analysis system was utilised 120 to obtain the average surface roughness (Sa), root mean square roughness (Sq), and average 121 peak-to-valley roughness (Spv) (n=10).

122
SEM images were obtained using a Supra 40VP SEM (Carl Zeiss Ltd., UK) with an 123 adapted protocol. 23 The leaf samples were soaked for 24 h at 4 °C in 4 % v/v glutaraldehyde 124 (Agar Scientific, UK). The leaf samples were removed and the excess glutaraldehyde was 125 washed from the leaf surface using sterile water. The leaf samples were then dried overnight.   Error bars were representative of the standard deviation or ± 5% error. One-way 146 analysis of variance (ANOVA) followed by Newman-Keuls tests were performed using R.        showed a very strong peak in the C-H rocking (CH2) region at ~720 cm -1 , an area that also 322 influences PC2 and further accounts for its marked positioning in PC2 away from other leaves.

323
While differences in various methylene frequencies were largely demonstrated by PC2, 324 the broad hydrogen bonded OH stretching band centred at 3300 cm -1 was a main contributor to 325 PC1 (Figure 6, top). As such, the positioning of the species along the horizontal axes of all the 326 score plots in Figure 5, representing PC1, was demonstrative of the strength of this peak. As the influence of various bands in the ATR-FTIR spectra that were most influential to this 347 model, in relation to the WCA of the leaf surface. As seen following PCA, PC1 for this model 348 was strongly influenced by the prominent OH stretching band centred at 3300 cm -1 , whilst PC2 349 was largely influenced by this peak but also the asymmetric and symmetric methylene CH2 C-

350
H stretching peaks at ~2916 cm -1 and ~2846 cm -1 , respectively. These bands within the loading 351 plots were the main contributors to these principal components which attests to their given surface (see Figure S2 for the plot depicting the performance of the PLSR model).  showed excellent performance in validation (using a leave-one-out technique) and based on 459 these results, it could be suggested that leaf WCA (i.e. hydrophobicity) can be predicted using 460 ATR-FTIR surface chemistry data. As such, this finding has the potential to change the way in