Discovery of (meth)acrylate polymers that resist colonization by fungi associated with pathogenesis and biodeterioration

Anti-attachment materials that are sprayable and 3D-printable passively prevent colonization by harmful fungi.


INTRODUCTION
The capacity for fungi to cause disease, spoilage, and biodeterioration is a major scourge for society. Fungal infections of humans are asso ciated with high mortality rates (~50% in hospitalized patients), killing more than 1.5 million people annually (1,2). Fungi also destroy crops and postharvest foods sufficient to feed 600 million people annually (3,4). This has spawned the development of major antifungal and fungicide industries with a combined worth of ~$30 billion globally, even without accounting for fungicides used to tackle fungal biodeterioration of valuable products and materials. Antifungal drugs and fungicides provide our first line of defense against fungi. However, efficacy of the current arsenal of approved agents is being eroded by drug resistance (2). The issues of resist ance, tightening antifungal/fungicide regulations, and mounting concerns for human and environmental health issues resulting from excessive chemical use have combined to underscore the need for alternative, sustainable strategies for fungal control.
Fungi need to attach to surfaces, both biological (e.g., epithelia and leaf surfaces) and inert (e.g., medical devices and household surfaces), to initiate many of the problems that they cause. Further more, in some scenarios (e.g., human infection) attached fungi can form biofilms, communities bounded by a biomaterial matrix. This property plays a crucial role in fungal virulence (5,6). Therefore, limiting attachment of fungal cells or spores to surfaces is a key but relatively unexploited target for combatting fungal colonization. Most strategies for tackling fungi rely on antifungals and fungicides, often incorporated into or applied to surfaces. In the case of surface colonization by human pathogens, "lock" therapy is used to eradi cate biofilm formation on catheters before their contact with patients. This involves pretreating the devices with high concentrations of antifungal drug (7,8). Medical devices can also be coated or impreg nated with inhibitory agents (9). To control fungal phytopatho gens in agriculture, fungicides are commonly sprayed onto crops, but resistance is a major concern here. Another chemicalbased crop protection strategy is the use of actives that perturb attachment, celltocell communication, or dispersion without necessarily killing the fungi. The plantderived bioactive zosteric acid, which alters oxidative balance of cells by targeting the NADH:quinone reductase (10,11), has been shown at sublethal concentrations to reduce adhe sion of the phytopathogens Magnaporthe grisea and Colletotrichum lindemuthianum (12) and food spoilage fungi Aspergillus niger and Penicillium citrinum (13). These strategies all rely on the use of bioactive agents that, as outlined earlier, hold diminishing appeal for longterm fungal control. However, since fungal attach ment is essentially a passive process, it is reasoned that passive approaches could provide effective control of fungi at the crucial surface attachment step. A passive intervention like an attachment resistant material should exert less selective pressure for resistance than bioactive drugs, for example. This is because nonresistant fungi would not be killed by an anti attachment surface, and development of resistance would typically require a gain of new function (i.e., ability to attach). Despite these advantages, such materials are difficult to design rationally because of our limited mechanistic knowledge of how fungi interact with different surfaces.
Several of the above issues also apply to the control of bacteria, which are evolutionarily distant from the fungi. An activefree ap proach has been developed recently, which shows promise for control of pathogenic bacteria (e.g., Pseudomonas aeruginosa). Specifically, using polymer microarrays (14), a copolymer of ethylene glycol dicyclopentenyl ether acrylate (EGDPEA) and di(ethylene glycol) methyl ether methacrylate (DEGMA), i.e., poly(EGDPEAcoDEGMA), was developed following a highthroughput screen for bacterial anti attachment properties among a large panel of monomers (15). The copolymer decreased bacterial biofilm formation when coated to silicone urinary catheters, compared with existing commercial catheters coated with silicone rubber and silvercontaining hydro gel (15)(16)(17). Clinical trials in Europe are currently underway using catheters coated with this antiattachment polymer. This resistance of the developed material to bacterial attachment has been attributed to their combined weakly amphiphilic nature and the molecular rigidity of the polymers' pendant groups (18,19). The use of the highthroughput screening approach allowed this discovery of the novel polymer without requiring indepth understanding of the underlying biologicalmaterial interactions. More recently, data mining of the structural characteristics exhibited by these resistant polymers has led to the identification of a predictive tool called the alpha parameter that has allowed the first "designed" resistant polymer to be synthesized from first principles (19,20).
A similar antiattachment approach has not previously been re ported for fungi despite the fact that key human pathogens such as Candida albicans avidly form biofilms, including on biomedical de vices. Fungal biofilms necessitate replacement of expensive indwell ing devices like voice prostheses every few months (21). Moreover, a passive antiattachment technology for fungi could have far wider applications such as food security and longevity of commercial ma terials, which are substantially affected by fungal attack. It is not obvious that antiattachment materials developed against bacteria, such as poly(EGDPEAcoDEGMA), should also be effective against fungi. Bacteria and fungi have very different cell surface character istics, e.g., their cell walls comprise distinct major components, in peptidoglycan and chitin, respectively.
In this study, we apply an acrylate/methacrylatebased polymer microarray screening approach (15) to fungi, to identify fungal anti attachment materials. We characterize a group of polymeric materials that substantial reduce the attachment and biofilm formation of key fungi onto diverse surfaces [e.g., plant surfaces and threedimensional (3D)-printed parts of medical devices], subject of a U.K. patent application (GB2002011.1). We demonstrate the potential of these materials to combat pasand sively fungal colonization and the pos sibility to 3D print medical devices from antiattachment materials.

Identification of candidate fungal anti-attachment polymers by microarray screening
To identify materials that may resist fungal attachment, we screened 281 acrylate and methacrylate homopolymers printed in a micro array format (Fig. 1A). These encompassed bacterial antiattachment candidates described previously (15) and other commercially avail able monomers that exhibited a wide chemical diversity within the side chains attached to the polymer backbone (table S1). Polymer spots were evenly distributed on arrays (Fig. 1B). Atomic force mi croscopy (AFM) images for selected materials are presented in Fig. 1C. Analysis by AFM of all array materials (Fig. 1C, right) showed that 90% had a root mean square (RMS) roughness less than 4.0 (including the six presented examples) and a mean modu lus of 2.8 ± 0.7 GPa. Fungal attachment was determined after incu bating suspensions of cells (C. albicans, yCherrytagged) or spores (Botrytis cinerea, Congo red-stained) with the polymer microarrays for 2 or 6 hours, respectively (Fig. 1A). These incubation times were sufficiently short to allow attachment to spotted materials while precluding subsequent overgrowth of hyphae and mycelium onto neighboring polymer spots (Fig. 1D). Fluorescence signals from the labeled fungi were used to quantify relative attachment to the 281 homopolymers (table S1). The distribution of attachment levels across the array was broad: 3.9 and 1.1% of the polymers gave very high attachment (>1000% of the median attachment value), while 2.5 and 3.9% were strongly resistant to attachment (<10% of the median) for C. albicans and B. cinerea, respectively (Fig. 1E). Levels of attachment of the two fungi across all polymers were only weakly positively correlated, albeit significantly (Pearson correlation, R 2 = 0.097, P < 0.0001) (Fig. 1F). Differences in the responses of the two fungi were expected given that the representative forms initiating their attachment (cells versus spores) have quite different surface properties (22) and the organisms themselves are phylogenetically distant among the ascomycete fungi (23). Nonetheless, the correla tion between these fungi was closer than that we have analyzed in a similar way between the present dataset for cells of C. albicans and previous attachment data (15) for vegetative cells of the pathogenic bacteria Escherichia coli (R 2 = 0.002), P. aeruginosa (R 2 = 0.040), and Staphylococcus aureus (R 2 = 0.002). Therefore, the bacterial results were very different to those obtained for fungi in this study.
Machine learning (ML) methods were used to generate predictive models for C. albicans and B. cinerea attachment, to assess the rela tionship between surface chemistry and the attachment of each fungus. Models using molecular signature descriptors (24) and time offlight secondary ion mass spectrometry (ToFSIMS) data collect ed from the polymers were generated for the fungal attachment to the polymers investigated. Before modeling, sparse feature selection was used to eliminate less informative descriptors. An independent test set was used to determine the predictive power of the fungal attachment models. For both fungal species, signature descriptor (computed molecular fragments) produced the best models. Non linear ML models produced a small performance improvement over linear partial least square regression (figs. S1 to S3). The predictive performance of the extreme gradient boosting (XGBoost) models for C. albicans and B. cinerea attachment using signature molecular descriptors is presented in fig. S4. There was only a moderate rela tionship between signature molecular descriptors and observed log attachment values for B. cinerea, with R 2 = 0.43 and RMS error (RMSE) 0.29 log fluorescence ( fig. S4). A fragment descriptor rela tively strongly associated with low B. cinerea attachment was a keto ether associated with the polymer backbone ( fig. S4, C and D), al though the weak predictive power of the model means that this ob servation should be treated with caution. For C. albicans, the model was stronger, with R 2 = 0.70 and RMSE 0.29 log fluorescence. The molecular features that were most strongly associated with low C. albicans attachment were methylene nitrile C(CN) and car bonyl C(O).
Because of the noise associated with fluorescence data (figs. S3 and S4) and the relatively weak predictive power of the compu tational models, we adopted a reasonably liberal approach in se lecting candidate materials from the screen for further interrogation, reasoning that autofluorescence, differences in polymerspot geom etry, and/or instability evident with certain polymers during the as says introduced elements of error. Therefore, for further study via scaleup, we selected 80 polymers that gave the lowest attachment for each fungus; twentyseven of these were common to both fungi (Fig. 1F).

Assay scale-up indicates acrylate and methacrylate polymers that are most resistant to fungal colonization
To investigate the scalability of the polymers' physiochemical prop erties and biological performance, the 80 polymers supporting the least attachment of each fungus from the microarray screen were deposited as a coating covering the 6.4mmdiameter wells of 96well microplates. Several of the polymers proved to exhibit surface crack ing after polymerization and under the vacuum step intended to remove unreacted monomer and so were excluded from the analy sis because of the presence of these additional topological features. Incubations of fungus with polymers for 24 hours were longer than in the microarray screen to allow some outgrowth and biofilm for mation for a more sensitive measure of preceding attachment events; nonadherent cells or spores were removed by washing at the end of an initial attachment phase ( Fig. 2A). Biofilm was detected with a metabolic XTT (tetrazolium salt, 2,3bis[2methyloxy4 nitro5 sulfophenyl]2Htetrazolium5carboxanilide) reduction assay, which eliminated the issue with autofluorescence of certain polymers. Materials of interest were designated as those supporting <25% biofilm formation compared to the control (noncoated well): <25% is equivalent to a biofilm that would result from a >95% re duction in attachment by the test fungi ( fig. S5). For C. albicans, nine of the scaledup polymers supported <25% biofilm formation ( Fig. 2B and table S2), whereas 19 of the test polymers had such efficacy against B. cinerea ( Fig. 2Band table S3). Of the 27 materials that were common to the two organisms in the screen and did not exhibit surface cracking (see above), only four yielded <25% biofilm formation for both organisms in the scaleup assay. In the scaleup assay, there was a similarly weak positive correlation between re sults for C. albicans cells and B. cinerea spores as in the preceding screen, but this was not significant in the case of the scaleup where only 23 polymers were assayed (Pearson correlation, R 2 = 0.077, P = 0.20) (Fig. 2B). We extended the scaleup assay to test the 19 polymers giving <25% B. cinerea biofilm also against another major plant pathogen, Zymoseptoria tritici, and an environmental filamentous fungus that colonizes diverse materials, Aspergillus brasiliensis. There were stronger correlations between results for spores of these three fungi than with C. albicans, with the strongest correlation between B. cinerea and A. brasiliensis (R 2 = 0.558, P = 0.0002) (Fig. 2C). Fifteen of the 19 polymers tested were re sistant to the attachment of at least two of the three filamentous fungi (table S3). On the basis of available information on the polymers' costs, chemistries, and toxicities, we selected nine leads deemed suitable by these criteria for further investigation: six from the C. albicans assay and five from B. cinerea (two were common to both fungi).  (Fig. 3). As the focus was on materials that passively resisted fungal attachment, we tested for potential toxicity effects to exclude polymers that might be actively inhibit ing the fungi. Here, the main technical difference with the above "antiattachment" assays was the omission of the washing steps; all cells including any dead ones were therefore retained in the wells (Fig. 3, A and C). As there were no washing steps, the potato dextrose broth (PDB) medium could not be replaced with phosphate buffered saline (PBS) to perform the XTT assay for toxicity in B. cinerea; therefore, B. cinerea was cultivated for 15 days on the mate rials in the presence of PDB and growth effects assessed visually (Fig. 3C). C. albicans growth was not inhibited by any of the poly mers; there were no significant differences between the polymer and control (uncoated well) incubations (Fig. 3A). Furthermore, conditioned supernatant (CS) from wells coated with these polymers did not show any significantly elevated cytotoxicity to mouse 3T3 cells compared with CS from uncoated wells, either at a standard 10 or 100% CS (Fig. 3B), supporting broader biosafety. However, pEGPhEA strongly inhibited the growth of B. cinerea (Fig. 3C). This may have been caused by residual or leached toxic material in the medium rather than any action by the polymer. To test this hypothesis, we added PDB medium to a well containing ultraviolet (UV)polymerized pEGPhEA and, after 24 hours, transferred this medium to wells containing B. cinerea spores. After 24 hours of subsequent growth, the OD 600 (optical density at 600 nm) was 2.5fold lower than in a control incubation (i.e., spores incubated with polymer free medium) (Fig. 3D). This result corroborated that materials present in the medium (e.g., monomers or short oligomers leached from pEGPhEA) were toxic to B. cinerea growth. Given this com plication, pEGPhEA was excluded from further study. In summary, the attachment data obtained for several polymers indicated that scaling up materials could alter their biological performance. However, several lead materials maintained a strong fungal antiattachment effect, which was not attributable to toxicity.

3D-printed polymer forms resist colonization by C. albicans
To evaluate the potential of the polymers of interest (above) as biofilm resistant coatings (e.g., for medical devices), we attempted to dip coat silicone coupons using polymer solutions prepared with the candidate materials. The resultant surfaces proved either to be cracked (AODMBA and tBCHMA) or to produce coatings that were both poorly adhered and prone to exhibit high levels of creep (DEGMA and TEGMA). Consequently, instead, we attempted to 3D print the target geometry directly. All the candidate monomers showed stable droplet formation during initial assessment of print ability. However, during the actual printing process for 3D struc tures, only an AODMBAbased formulation solidified and formed stable geometries. The other candidates either remained as a tacky solid phase or collapsed during the printing, which was attributed to a low level of polymerization or low glass transition temperature (T g ); analysis by differential scanning calorimetry (DSC) indicated T g values of 86°C for AODMBA, −36°C for DEGMA, −53°C for TEGMA, and 159°C for tBCHMA. Coupons (diameter, 3 mm) manufactured with AODMBA were used initially to test the anti attachment properties of the printed polymer (Fig. 4A). Polyethylene glycol diacrylate (PEG 575 DA) gave good C. albicans attachment in the previous tests (comparable to the attachment on a noncoated well) (Fig. 2B), so coupons 3Dprinted with PEG 575 DA were used as attachment positive control. A ~100% reduction in C. albicans attachment was apparent with the AODMBAprinted coupons com pared to those that were PEG 575 DAprinted (Fig. 4B). As a more commercially relevant example, next, we used AODMBA for print ing valve flap forms for voice prostheses. This example was chosen because commercial silicone-manufactured valve flaps are highly susceptible in vivo to develop C. albicans biofilms (21). Our results showed that AODMBAprinted valve flaps were more resistant to fungal attachment than a standard silicone-manufactured product, with a mean 84% reduction in biofilm biomass and up to 100% re duction in some replicates (n = 8) (Fig. 4C). In conclusion, AODMBA was demonstrated to (i) be 3Dprintable and (ii) exhibit strong anti attachment properties that were retained in AODMBAprinted forms. As drugresistant C. albicans poses particular problems for therapy, we also tested the lead material against strains resistant to azole drugs (Fig. 4D). These tests showed that antiattachment by AODMBA is just as effective against the drugresistant isolates as against a standard C. albicans strain, further supporting the poten tial for clinical application.

Lead polymers can protect plant leaves from fungal infection
We then hypothesized that the lead polymers could find novel applications for protecting plant (crop) surfaces from fungal infec tion. To explore this possibility, first, we tested for potential plant toxicity with polymers that had given good antiattachment against B. cinerea in vitro (Fig. 2B). Polymer solutions (≥85% monomer conversion; table S4) were sprayed on to lettuce leaf discs. No leaf lesions were observed up to 3 days after spraying with polymer, sug gesting an absence of toxicity (leaf samples deteriorated after 3 days regardless of polymer application) (Fig. 5A, left). To test for effects on fungal infection, leaf discs treated with either polymer or solvent alone (control) were inoculated with B. cinerea spores. Whereas fungal lesions were evident after 2 days in all of the controls, leaves treated with either DEGEEA, DEGMA, or TEGMA were signifi cantly resistant to B. cinerea infection [ Fig. 5, A (middle) and B]. Fewer than 15% of TEGMAtreated leaf samples showed any sign of infection up to 3 days. In contrast, mMAOES did not confer any apparent protection as lesions appeared after 2 days: The outcome for mMAOES was similar to the untreated leaves or leaves treated with ethylene glycol methyl ether methacrylate (EGMMA), which had been selected as an attachment positive control. We confirmed that B. cinerea could grow in the presence of these synthesized poly mer batches in vitro, so the effects on infection could not be as cribed to toxicity to the fungus. Examination of leaf surfaces coated with the bestperforming polymer (TEGMA) by scanning electron microscopy (Fig. 5C) and ToFSIMS ( fig. S6) suggested good, al though not complete, surface coverage. Last, we tested the resilience of TEGMA to rinsing with water. TEGMA was sprayed onto the leaf discs and airdried as above before the leaves were rinsed three times with water and subsequently infected. After 3 days, no lesions were observed (Fig. 5D). This indicated that the antiattachment property of the polymer conferred to the leaf surface was resilient to rinsing with water, such as may occur in the natural environment during rainfall. The presence of TEGMA after washing was confirmed by ToFSIMS; no significant change was observed between the washed or unwashed leaf sections (fig. S6). The data were consistent with a potential for application of these materials in agriculture.

DISCUSSION
This study identified polymer materials with the potential to stop fungal colonization by blocking attachment and demonstrates po tential applications for tackling at least some of the diverse socio economic problems that fungi cause. Important findings that have emerged from this study include our identification of materials that block attachment of problematic microbial spores (e.g., B. cinerea spores), whereas previous work was restricted to vegetative bacterial cells; the lead antiattachment materials identified are different from those described previously for bacterial pathogens, indicating sub stantial differences in the mechanisms of pro/antiattachment; we identify both novel applications and routes of manufacture, including formulating the material in a form that offers protection to the sur faces of edible produce from fungal infection, and 3Dprinted per sonalized devices that are notoriously prone to fungal deterioration.
Currently, antifungal and fungicidal agents are widely used to combat fungal pathogens, fungi causing biodeterioration, and spoil age fungi by killing them. However, with an increased incidence of fungal isolates resistant to current treatments and tightening anti fungal and fungicide regulations, novel methods for fungal man agement are needed. Controlling fungal attachment to surfaces in a passive manner (i.e., without active killing of organisms) presents an alternative, attractive intervention at the initial step of fungal colo nization. We also demonstrated absence of active killing by the poly mers in cytotoxicity assays with mammalian cells, supporting the biosafety of this approach. Attachment via adhesion is a prerequisite for most adverse effects of fungi, including formation of biofilms that is an important virulence factor in microbial pathogenesis. There fore, inhibition of attachment should be an effective target for con trolling most fungi. The passive control described here could reduce the potential development of resistant organisms, as selection pressure for resistance (to antiattachment polymers) should be considerably lower because nonresistance is not fatal and may have negligible disadvantage in many scenarios. Furthermore, resistance in this case could require organisms to gain a new function, to achieve attach ment, potentially raising greater evolutionary hurdles (25).
For the prevention of C. albicans attachment on acrylic resins, chemical, or physical surface alterations such as modification of surface charge (26,27), increasing surface wettability or decreasing surface energy (28)(29)(30)(31) previously gave lower C. albicans adhesion. The present study used highthroughput screening to identify polymers that are able to reduce fungal attachment from a library of more than 250 materials, the largest study of materialfungi interac tions to date. Highthroughput synthesis, assessment, surface charac terization, and chemometrics have accelerated discovery of polymers resistant to bacterial adhesion and helped characterize chemical moieties that reduced bacterial attachment to coated medical devices in vivo (15). This class of materials could not have been pre dicted from the current understanding of bacteriamaterial interac tions. Furthermore, variation in molecular weight between 3000 to 50,000 affected material properties but not biological anti attachment performance (16). Adopting this highthroughput informaticsbased approach for fungi, we identified both acrylate and methacrylate polymers that resisted fungal attachment to either biological or inert surfaces; in the latter case, the polymer described here outperforms the current marketleading silicone-manufactured material. Hydrophilic polymers have also been used in surface mod ification (29,30), as hydrophobic fungi preferentially adhere to hy drophobic surfaces (32). The lead polymers identified in this study that resisted attachment by several fungi, including TEGMA or DEGMA, were consistent with this, being broadly classified as hy drophilic with a water contact angle (WCA) of 20° to 50° (table S3). There were, however, exceptions where polymers with a WCA between 62° and 72° could still prevent fungal biofilm. The lead polymers resisting attachment only of C. albicans were more hydro phobic, with WCAs of 62° to 96°. Thus, the ability of these polymers to prevent fungal biofilm formation indicates that hydrophilicity alone is not sufficient to predict fungal attachment to a particular material.
In this study, we were able to use AODMBA for inkjetbased 3D printing to demonstrate the capability to manufacture bespoke fun gal attachment-resistant devices. The antiattachment properties of AODMBA were retained after printing, including in a printed med ical device component (valve flap for a voice prosthesis). One of the advantages of manufacturing these parts with polymer rather than coating the polymer onto target devices is that the object is homog enous, and thus, it is less likely that there will be exposure of regions of (potentially attachmentprone) native surface. The AODMBA Printed valve flaps showed >80% reduction of biofilm formation compared with a standard silicone-manufactured product. However, AODMBA forms a hard, glassy polymer that would therefore be, in its homopolymer form, too inflexible for valve flap applications. This is analogous to the antibacterial polymer EGDPEA development, in which the homopolymer was also not suitable to produce a viable catheter coating (16); rather, an optimized copolymer had to be de veloped. Similarly, to develop a commercially viable coating, the mechanical properties of that material were improved by copolymer ization with a comonomer (DEGMA) that has a lower T g . In practice, the T g values that were exhibited by poly(EGDPEAcoDEGMA) polymers, synthesized in various different comonomer ratios, were used as a highthroughput screening guide to predict the copolymer compositions that should match the flexibility of the commercial catheter material. Pin printing assays using those copolymers with acceptable T g confirmed that they still retained the bacterial attach ment resistance (16). Similar optimization could be an aim of future work to improve the mechanical properties of AODMBAbased forms. This could include copolymerization with lower T g comono mers, and the screening data presented here suggest a number of potentially suitable candidate comonomers that themselves exhibit some antiattachment properties. By observing the latter point, any loss of antiattachment in copolymer materials should be minimized. Design guidance for materials may, of course, need to be tailored for different fungal attachment scenarios. The present work is a key step toward deriving a set of molecular design rules to define even more relevant molecular structures to be incorporated in materials to improve performance. Our analyses showed that surface chemistry is not a very strong differentiator for fungal attachment, suggesting that material properties will play a more significant part in perform ance compared with the bacterial work.
In agriculture, polymer materials have found applications for improving physical properties of soil and as adjuvants in polymeric biocide and herbicide formulations. These latter are controlled release formulations designed to reduce the possible side effects accompanying the overuse of biologically active agents. However, the passive application proposed in the current study is a novel po tential replacement for active agents in formulations. These materials were effective against B. cinerea and Z. tritici, two major crop patho gens. Furthermore, three of four selected polymers conferred plant protection against B. cinerea infection. TEGMA, the bestperforming polymer, showed resistance to the attachment of all four fungi used in this study, suggesting a broad spectrum of action of this meth acrylate material. Broadspectrum agents are particularly valued in common antimicrobial applications, including crop protection.
In conclusion, this work identified a panel of polymers that are resistant to fungal attachment and therefore reduce fungal biofilm formation and infection. Besides the therapeutic and crop protec tion potential, such acrylate and methacrylate polymers have wider applications, as exemplified by their effect on the attachment of A. brasiliensis, known to colonize synthetic products and materials. This study is an important first step toward the targeted design of novel materials tailored for different antifungal applications.

Polymer array synthesis
Polymer microarrays were prepared using a modified version of the previously described procedure (33). Polymer microarrays were printed using an XYZ3200 dispensing station (BioDot) using quilled steel pins (Arrayit, 946MP6B). Printing was carried out under an argon atmosphere maintaining O 2 < 2000 parts per million (ppm), 25°C, and 30 to 35% relative humidity. Diluted polymerization solutions were composed of monomer [50% (v/v) for oils and 50% (w/v) for solids] in N,N′-dimethylformamide, 1:1 N,N′-dimethylformamide: water or 1:1 N,N′-dimethylformamide:toluene depending on solu bility. The photoinitiator 2,2dimethoxy2phenyl acetophenone [1% (w/v)] was added to all solutions. A total of three replicates were printed on each slide, each replicate comprising 281 different polymers (table S1). Monomers were purchased from SigmaAldrich, Scientific Polymers, Acros Organics, or Polysciences and were used as received. Spacing between the printed spots in each row was 1500 m in the x axis, with an alternating +750m/−750m offset in the x axis between each row and a 750m spacing between each row in the y axis. After printing was completed, arrays were dried in a Heraeus vacuum oven (35°C, 0.3 mbar) for 7 days.

High-throughput surface characterization
ToFSIMS measurements were conducted using a ToFSIMS IV (IONTOF GmbH) instrument operated using a 25kV Bi 3 + primary ion source exhibiting a pulsed target current of ∼1 pA. Samples were scanned at a pixel density of 100 pixels/mm, with eight shots per pixel over a given area. The analysis area was 20,000 m by 20,000 m. An ion dose of 2.45 × 10 11 ions/cm 2 was applied to each sample area, ensuring that static conditions were maintained throughout. Both positive and negative secondary ion spectra were collected (mass resolution of >7000 at mass/charge ratio of 29). Owing to the non conductive nature of the samples, a lowenergy (20 eV) electron flood gun was applied to provide charge compensation. WCA was measured as described previously (34).

Computational modeling
The replicate fungal fluorescence values for each of the polymers screened (three replicates for B. cinerea and six for C. albicans) were averaged, and the SDs were calculated. As the fluorescence values spanned a wide range, the log of the fluorescence values was used as the dependent variable in the computational models, as is common practice for quantitative structureactivity relationship modeling. Polymers with low signaltonoise ratio (<1.5) were excluded from the B. cinerea (173 polymers) and C. albicans (197 polymers) attach ment datasets. For modeling, least absolute shrinkage and selection operator (LASSO) was used to select sparse subsets of features from larger pools of possibilities in a contextdependent manner.
Partial least square regression was conducted using MATLAB R2018a 9.4.0.813654. ToFSIMS positive and negative data were concatenated into a single data matrix to be used as the X variables for the model. X variables were meancentered and variancescaled before analysis. Data were randomly split into training and test sets (3:1) to validate the model produced. The number of latent variables used in the model was selected on the basis of a minimum in the RMSE of cross validation. Three latent variables were selected for models for each fungal species.
XGBoost regression ML, a robust nonlinear ML method (35), was used to generate models relating chemical features to fluorescence (and, therefore, attachment). The chemical features used to train the models were of two types: signature molecular descriptors (36,37) generated by computationally fragmenting molecules and ToFSIMS ion peaks derived from actual molecular fragmentation by probe ions. Although independent test sets are the best way of assessing the predacity of ML models, because of the high variability and noise present in the datasets, leaveoneout (LOO) crossvalidation was used for this purpose.
The XGBoost algorithm (version 0.22) (35) with default param eters was used to generate the models, and LOO cross validation was implemented using the package LeaveOneOut from sklearn.model_ selection (both codes were implemented in Python 3.7). LASSO feature selection was implemented in MATLAB R2017a using the lassoglm function, selecting the features that provide the minimum value for the squared error for the lambda parameter. Their rank in importance is given by the XGBoost descriptor importance param eter, which provides a score indicating how useful each descriptor was in constructing the boosted decision trees within the model, using Gini as performance measure. This importance was calculated for each descriptor and averaged across the multiple trees, allowing attributes to be ranked and compared to each other.

Free-radical polymerization scale-up for performance validation, inkjet 3D printing, and leaf coating Polymerization method for biological performance validation
The synthesis of selected compounds was upscaled to allow the val idation of the biological performance observed in the pin printing assays. This was achieved by coating the 6.4mmdiameter wells of 96well plates. Plates were prepared by adding 50 l of monomer solution into each well. Polymerization was initiated by addition of 2,2dimethoxy2phenylacetophenone (SigmaAldrich) to a final concentration of 1% (w/v). Samples were irradiated with UV (BlakRay XX15L UV Bench Lamp; 230 V, ~50 Hz, 15 W, 365 nm) for 1 hour with O 2 < 2000 ppm. The samples were dried at <50 mtorr for 7 days. Wells were then washed briefly with isopropanol and left for 2 days at 37°C in distilled water. Plates were then washed briefly with isopropanol and distilled water and airdried before irradia tion with UV for 20 min to sterilize the samples.

Polymerization method for validation of inkjet 3D printing performance
Exploring the potential printability of a monomer for inkjetbased 3D printing requires consideration of several key factors including viscosity, surface tension, printing conditions, etc. Following exist ing methods for the efficient formulation development of inkjet based 3D printing inks (38)(39)(40)(41), candidate monomers that were suitable for the inkjet 3D printing process were identified, and then, associated ink formulations were prepared by dissolving 1% (w/v) 2,2dimethoxy2phenylacetophenone (SigmaAldrich) into 5 ml of the candidate monomer. The mixture was stirred at 800 rpm at room temperature until the initiator was fully dissolved. The ink was then purged with nitrogen gas for 15 min and filtered through a 5m nylon syringe filter. The final ink formulation was left at 4°C over night to degas. A Dimatix DMP2830 material printer was used for printing, equipped with a 10pl cartridge containing 16 nozzles, each with a square cross section with a side length of 21 m. The jetting voltage and waveform were adjusted until stable droplet formation was achieved. A 365nm UV lightemitting diode (LED) unit (800 mW/cm 2 ) was used for inline swathbyswath ink curing after deposition. The whole printing process was carried out in a nitrogen environment, where the oxygen level was 0.2 ± 0.05%.

Polymerization method for leaf coating
For investigating the fungal infection of polymercoated plant leaves, polymerization of the materials identified as candidates for resist ance to fungal infection was performed by freeradical polymeriza tion using a thiol chain transfer agent to limit the molecular weight of the final material and ensure that it was processable. Candidate monomers were dissolved in cyclohexanone (Acros Organics; 1:3, v/v), and the chain transfer agent [1dodecanethiol (Acros Organics), 5% (mol/mol) with respect to the monomer] and initiator [2′azobis (2methylpropionitrile; SigmaAldrich; 0.5% (w/w)] were added. Argon was purged into the mixture for 30 min to remove oxygen, before holding at 75°C for 24 hours. Isolation of the polymer was achieved by precipitation into an excess of either (i) heptane (Fisher Scientific; DEGEEA, DEGMA, EGMMA, and TEGMA) or (ii) chloroform (Fisher Scientific; mMAOES). The nonsolvent to reaction mixture ratio used for the precipitations was 5:1 (v/v). Precipitated materials were collected in vials and incubated in a vacuum oven for at least 24 hours before use. Nuclear magnetic resonance (NMR) spectro scopic analysis was performed with the crude polymerization solution to determine polymer conversion and on the final precipitate to assess purity. To evaluate the molecular weight of the materials, purified samples were dissolved in highperformance liquid chromatography (HPLC)-grade tetrahydrofuran (THF) for analysis by gel permeation chromatography (GPC).

Differential scanning calorimetry
Polymer thermal properties were investigated by DSC (Q2000, TA Instruments, Leatherhead, UK), at a heating rate of 10°C min −1 . Data analysis was done with TRIOS software (version 4.4.0.40883). Pans with holed lids (TA Instruments, Brussels, Belgium) were used for sample analysis, with empty pans as the reference. Glass transi tions were determined by performing two heating/cooling cycles between −90° and 200°C. 1 H NMR spectroscopy 1 H NMR spectra were recorded at 25°C using a Bruker DPX300 spectrometer (400 MHz). Chemical shifts were recorded in H (in ppm). Samples were dissolved either in deuterated chloroform (DEGEEA, DEGMA, EGMMA, and TEGMA) or in deuterated acetone (mMAOES) to which chemical shifts were referenced (residual chloroform at 7.26 ppm and residual acetone at 2.05 ppm).
Gel permeation chromatography GPC analysis was performed using an Agilent 1260 Infinity instru ment equipped with a double detector in the light scattering con figuration. Two mixed columns at 25°C were used, using THF as the mobile phase at a flow rate of 1 ml min −1 . GPC samples were pre pared in HPLCgrade THF and filtered before injection to the GPC system. Analysis was carried out using Astra software. The molecular weight (number average, M n ) and polydispersity (Ð) were calculated, with reference to a calibration curve created using commercially purchased poly(methyl methacrylate) standards.

Microscopy
Optical microscopy of polymer microarray slides was with a GX microscope (GXML3201 LED). AFM measurements of polymer microarrays were acquired as previously reported (42) using a Bruker FastScan Icon AFM with Bruker TAP150A spring (5 N/m) constant tips. Polymer spots were analyzed in batches of 100, with polystyrene controls taken at 15measurement intervals to ensure that tips were not damaged during a run. Images were analyzed using Gwyddion 2.55 software. Scanning electron microscopy imaging of polymercoated lettuce leaves was conducted with a JEOL JSM6490LV. Samples were goldcoated before imaging using a Polaron SC7640 sputter coater.

Polymer microarray screening for fungal attachment
Before testing against fungi, the microarray slides were washed by immersion in distilled water for 10 min, airdried, and UVsterilized. For screening with C. albicans (yCherrytagged), single colonies were used to inoculate YPD broth cultures in Erlenmeyer flasks and incubated at 37°C with orbital shaking at 150 revolutions min −1 . Overnight cultures were washed twice in RPMI 1640 (SigmaAldrich) and adjusted to OD 600 ~ 10. Microarray slides were incubated statically at 37°C for 2 hours with 15 ml of the cell suspension. For tests with B. cinerea, spores were harvested from 7dayold PDA plates, washed twice in PDB medium, and resuspended in PDB at a concentration of 2 × 10 7 spores ml −1 . As with C. albicans, microarray slides were incubated statically with 15 ml of the cell suspension but at room temperature for 6 hours and stained for 10 min with 0.5% Congo red. As controls, slides were also incubated with noninoculated medium. After the period of attachment, the slides were removed and washed three times with 15 ml of PBS at room temperature. After rinsing with distilled water to remove salts then air drying, fluorescence images from the slides were captured using either a GenePix Autoloader 4200AL (C. albicans; Molecular Devices, USA) or 4000B (B. cinerea; Molecular Devices, USA) scanner, with a 635nm red laser and red emission filter. The total fluorescence signal from each polymer spot was determined using GenePix Pro 6 software (Molecular Devices, USA). The fluorescence signal attributable to fungal attachment to each polymer was determined by subtracting the fluorescence signal in the mediumonly control incubation from that in the incubation with fungus. For polymers where the fluorescence was below the limit of detection, fluorescence was recorded as zero, as discussed in (15). Fungal attachment to each polymer is expressed as a percentage relative to the median value (equal to 100%) across all polymers for each fungus.

Fungal biofilm assessment
Biofilm metabolic activity was measured by the XTT (SigmaAldrich) reduction assay. For C. albicans, single colonies were used to inoculate YPD broth cultures in Erlenmeyer flasks and incubated overnight at 37°C with orbital shaking at 150 revolutions min −1 . Cultures were washed twice in RPMI 1640 and diluted to 125,000 cells ml −1 . Aliquots (100 l) of the cell suspension were transferred to 96well microtiter plates (Greiner BioOne, Stonehouse, UK), either coated with the polymers of interest or containing coupons 3Dprinted with polymer as described above and then incubated statically for 2 hours. Similarly, 100 l of fungal spores (2.5 × 10 6 spores ml −1 in PDB) from 7dayold PDA plates were transferred to coated 96well plates for 6 hours at room temperature. In all cases, coupons were subsequently transferred to fresh 96well plates. Nonadherent cells or spores were removed by three gentle washes with PBS; then, 100 l of fresh medium was added to each well, and plates were incubated at 37°C up to 24 hours after inoculation. Coupons were again trans ferred to fresh plates. The wells were washed three times with PBS, and the XTT reaction was initiated by adding XTT and menadione to RPMI (for C. albicans) to final concentrations of 210 g ml −1 and 4.0 M, respectively, or to PBS (for B. cinerea) to final concentra tions of 400 g ml −1 and 25 M (final volume per well, 200 l; PBS was used instead of PDB, as the XTT reaction does not work in PDB medium). After 2 and 6 hours, respectively, 100 l of the reaction solutions was transferred to fresh 96well plates, and the absorbance at 490 nm was measured using a BioTek EL800 microplate spectro photometer. To assess the impact of the polymers on fungal growth, washing steps were omitted as presented in Fig. 3. Contrary to RPMI and as mentioned above, the XTT reaction cannot be performed in PDB medium, and fungi are not able to grow in PBS. Therefore, B. cinerea was cultivated for 15 days on the polymers in the presence of PDB, and growth effects were assessed visually.
Biofilm formation was also assessed on prosthesis valve flaps, either printed (above) or commercial manufactures from silicone (provided by Atos Medical; raw material is Silastic Q74735, Dow Corning). The latter was used as the control material. The materials were immersed in the presence of 1 × 10 6 cells in RPMI 1640 (final volume, 1 ml) in 12well plates (Greiner BioOne). After 2 hours of static incubation at 37°C, valve flaps were transferred to new plates and washed three times with PBS to remove nonadherent cells. Fresh RPMI 1640 was added. After 46 hours at 37°C with orbital shaking at 100 revolutions min −1 , RPMI 1640 was removed and biofilm stained with 0.5% (w/v) crystal violet for 1 min. The valve flaps were washed three times with PBS to remove nonadherent biofilm and excess stain before image capture. For quantification, the crystal violet was dissolved with 1 ml of ethanol and 100 l of the reaction was transferred to 96well plates. Absorbances at 600 nm were measured using a BioTek EL800 microplate spectrophotometer.

Cytotoxicity assessment
Immortalized National Institutes of Health 3T3 mouse embryonic fibroblast cells (passage 25) were cultured in Dulbecco's modified Eagle's medium (DMEM; SigmaAldrich) supplemented with 10% fetal bovine serum, 2 mM lglutamine (SigmaAldrich), and penicillin (100 U ml −1 ) + streptomycin (100 g ml −1 ) (Gibco) in 75cm 2 cellculture flasks (Greiner BioOne) at 5% CO 2 and 37°C. Cells were detached with trypsin/EDTA (SigmaAldrich) and washed with DMEM before dis pensing 100 l of cell suspension at 5000 cells per well in a 96well plate. After 24 hours, either 10 or 100 l of the medium was replaced with CS (below) and incubated for a further 24 hours. For CS, 200 l of supplemented DMEM (above) was added to polymercoated or uncoated wells of plates (prepared as described in the "Polymerization method for biological performance validation" section) and collected at intervals, with recovered CS replaced by fresh medium for a sub sequent interval. Toxicity of CS to the 3T3 cells was according to release of lactate dehydrogenase (LDH) into the cell culture medium, determined with the CyQUANT LDH Cytotoxicity Assay Kit (Thermo Fisher Scientific) according to the manufacturer's protocol. Cytotoxicity was calculated relative to controls for spontaneous LDH activity (cells grown in fresh medium) and maximum activity (cells treated with CyQUANT lysis buffer).

Fungal infection of plant leaves
Polymer solutions [20% (w/v), prepared using 20% (v/v) isopropanol as solvent] were sprayed onto 1.5cmdiameter leaf discs prepared from fresh lettuce. Discs were placed onto water agar [sterile dis tilled water, 2% (w/v) agar] in square Petri plates (Greiner) and then incubated at room temperature for up to 3 days. To measure resil ience of coated polymer to rinses with water, some lettuce leaf discs were washed by submersion in water. Spores of B. cinerea were har vested from 7dayold PDA plates, washed twice with PDB, and adjusted in PDB to a concentration of 5 × 10 5 spores ml −1 . Once dried, leaf discs were infected with B. cinerea by aliquoting 5 l of spore suspension to the middle of the discs (2500 spores per leaf disc). Images were captured every day up to 3 days after infection to assess lesions. To assess potential toxicity of polymers to the plant material, leaf discs were sprayed with the polymers but not infected with B. cinerea.

Statistical analysis
Statistical analyses were carried out by Student's t test, Pearson cor relation, or twoway analysis of variance (ANOVA), using Prism software, from a minimum of three independent replicate values. Regression models used for the computational modeling are detailed above and in the Supplementary Materials.

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ content/full/6/23/eaba6574/DC1 View/request a protocol for this paper from Bio-protocol.