Combining Inducible Lectin Expression and Magnetic Glyconanopar-ticles for the Selective Isolation of Bacteria from Mixed Populations.

: The selective isolation of bacteria from mixed populations has been investigated in varied applications ranging from differential pathogen identification in medical diagnostics and food safety to the monitoring of microbial stress dynamics in industrial bioreactors. Selective isolation techniques are generally limited to the confinement of small populations in defined locations, may be unable to target specific bacteria, or rely on upon immunomagnetic separation which is not universally applicable. In this proof-of-concept work, we describe a novel strategy combining inducible bacterial lectin expression with magnetic glyconanoparticles (MGNPs) as a platform technology to enable selective bacterial isolation from co-cultures. An inducible mutant of the Type 1 fimbriae, displaying the mannose-specific lectin FimH, was constructed in E. coli allowing for ’on demand’ gly-can-binding protein presentation following external chemical stimulation. Binding to glycopolymers was only observed upon fimbrial induction, and was specific for mannosylated materials. A library of MGNPs was produced via the grafting of well-defined catechol-terminal glycopolymers prepared by RAFT polymerization to magnetic nanoparticles. Thermal analysis revealed high functionalization ( ≥ 85% polymer by weight). Delivery of MGNPs to co-cultures of fluorescently labelled bacteria followed by magnetic extraction resulted in efficient depletion of Type 1 fimbriated target cells from wildtype or afimbriate E. coli . Extraction efficiency was found to be dependent on the molecular weight of the glycopolymers utilized to engineer the nanoparticles, with MGNPs decorated with shorter Dopa-(ManAA) 50 mannosylated glycopolymers found to perform better than those assembled from a longer Dopa-(ManAA) 200 analogue. Extraction efficiency of fimbriated E. coli was also improved when the counter-part strain did not harbor the genetic apparatus for expression of the Type 1 fimbriae. Overall, this work suggests that modulation of the genetic apparatus encoding bacterial surface associated lectins coupled with capture through MGNPs could be a versatile tool for the extraction of bacteria from mixed populations


INTRODUCTION
Industrial biotechnology has transformed we manufacture fine chemicals and pharmaceuticals. 1 The engineering of microbial consortia to express complex pathways in distinct modules is now enabling the production of biochemicals, such as cis,cis-muconic acid 2 , hydroxybenzoic acid 3 , biobutanol 4 and the flavonoid sakuranetin 5 , at significantly higher yields than other biosynthetic production methods. Moreover, engineering metabolic modules across synthetic microbial consortia as opposed to a single chassis can reduce metabolic load upon the host organism(s) as well as limiting the crosstalk between native and non-native pathways or signalling mechanisms. [6][7] However, the growing complexity of microbial cocultures in bio-industrial applications will likely require specialised tools for organism differentiation, capture and processing. On-demand selective removal of bacteria from a biosynthetic fermentation culture could improve the product yield and efficiency of processing by the removal of modules which may generate non-desirable reactions. 8 This is particularly true for multi-step biosynthetic pathways split across multiple hosts, whereby depletion of a single module could push the enzymatic equilibria towards the next intermediate or the desired product. 7,9 Furthermore, selective removal of organisms containing the desired product could aid in the mediating the shift from batch-based to continuous bioreactors further increasing synthetic yields. 10 Several physical entrapment techniques involving microfluidics 11 and hydrogels 12 , often coupled with soft lithography methods 13 , have been utilised to capture bacteria in specific locations. However, these approaches are often limited to the confinement of small populations of bacterial cells in defined locations and low volumes; for microfluidic channels this is typically less than 10 µL. 14 Furthermore, physical and temporal separation lacks the ability to differentiate between phenotypically divergent cells within a population, thus limiting utility in microbial consortia.
Antibody coated magnetic nanoparticles (MNPs) have been widely used for selective bacterial isolation from mixed populations 15 , with applications ranging from detection and depletion of pathogenic organisms from foodstuffs 16 to rapid, diagnostic bioprobes. 17 Immunomagnetic capture systems have also been coupled with innovative technologies such as continuous flow microfluidics 18 or transiently magnetizable pipette tips 19 for high-throughput bacterial isolation from mixed populations at the benchtop level. However, the high cost, manufacturing issues and cold-chain requirements of immunomagnetic separation may limit the scalability of this technology. Furthermore, immunomagnetic separation techniques typically target natively expressed surface antigens on bacteria, limiting the ability to couple immunomagnetic capture to a signal.
In contrast, by utilizing the sugar-encoded biological language, the glycocode 24 , glycomimetics may confer specificity based on the glycan preference dependent on the bacterial species. 25 Sugar-recognizing proteins, lectins, typically exhibit weak binding affinities for target monosaccharides 26 . However, lectins demonstrate high avidities and specificity for multivalent ligands, such as glycodendrimers and glycopolymers, due to the multivalent effect leading to nanomolar KD values for some synthetic glycomaterials. 27 Kiessling and co-workers utilized glycopolymers to modulate chemotactic responses in bacteria through clustering of lectin chemoreceptors, 28 while von der Ehe et al developed glycopolymer-cryogels as potential materials for affinity chromatography of bacterial cells. 29 We showed that modulation of glycopolymer sugar patterns can lead to preferential binding to either S. mutans or E. coli, 23 and utilized thermoresponsive polymers with pendant glucose groups to reversibly assemble/disassemble target bacteria through temperature cycling. 30 Recently, Young et al demonstrated that glycopolymer funtionalised gold surfaces could be used to recruit Shewanella oneidensis colonization and biofilm formation. 31 Furthermore, fluorescent mannose-containing glycopolymers 32 and monovalent mannose decorated magnetic nanoparticles 33 have been used as an analytic tool for the detection of fimbriated E. coli. To couple a stimulus to bacterial capture, Paul et al developed thermoresponsive mannosefunctionalized p(NIPAM) microgels which could display or mask bacterial lectin ligands in a temperature dependent manner, facilitating reversible capture of E. coli from solution. 34 With these developments in mind, we postulated that glycopolymer coated MNPs (MGNPs) could provide a scalable, yet selective alternative to existing materials for specific sequestration of bacterial sub-populations. Our strategy entails tightly controlled, 'on-demand' expression of lectins by bacterial sub-populations, enabling their subsequent selective sequestration and removal using appropriately sugar-encoded MGNPs. In this proof-of-principle work we first constructed an inducible mutant of the Type 1 fimbriae in Escherichia coli, in which the native regulatory mechanism of the fim operon was replaced with a regulatory system which we could control. Accordingly, chemical stimulation using anhydrotetracycline (ATc)-induced Type 1 fimbriae network expression, and our engineered strain demonstrated specific affinity for mannosylated glycomaterials due to surface expression of Type 1 fimbriae containing FimH lectin. Next, MGNPs were engineered by grafting glycopolymers to superparamagnetic iron (II,III) oxide nanoparticles (SPI-ONs). We showed that mannose-containing MGNPs selectively depleted Type 1 fimbriated target cells from their afimbriate counterparts in bacterial co-cultures following magnetic separation, thus highlighting the potential of this approach for selective isolation of specific bacteria form mixed populations.

RESULTS AND DISCUSSION
Genetic engineering of E. coli produces a controlled, mannosespecific binding phenotype Initially, we genetically engineered an E. coli strain in which we could induce the expression of a lectin as a model system for bacterial separation. One of the best-characterized fimbrial systems identified in prokaryotes is the Type 1 fimbriae which includes the mannose-specific adhesin FimH 35 , and was chosen as a model lectin in this study. FimH is displayed at the distal tip of the organelle, and includes a lectin-like domain which binds to monovalent mannose residues with KD values calculated at around 2.3 µM but has significantly higher affinities for mannose polymers and heptyl or aryl mannosides. 36 In E. coli expression of the fimAICDFGH operon is under the control of a phase-variable regulatory circuit. Phase variation in the expression of Type 1 fimbriae is associated with the inversion of a 314-bp element, which has been dubbed the fim switch (fimS). 37 Inversion, and hence orientation, of the promoter determines the transcription levels of this operon. Furthermore, the switching of this invertible element is not dependent on RecA; it instead requires recombination by FimB or FimE. The genes encoding these regulatory recombinases are located proximally upstream of the switch. FimB and FimE specifically recognize two 9-bp inverted repeats either side of the fim switch. 37 FimB catalyzes the inversion of this switch in both directions at equivalent frequencies, whereas FimE catalyzes inversion at a higher frequency than FimB and only in favor of the on to off recombination. 38 In order to eliminate the stochastic switching of the fim operon seen in wild type E. coli strains (Fig. 1A) 37 elimination of the fim switch and replacement with an inducible regulatory mechanism was necessary ( Fig. 1B).We designed a synthetic construct containing the inducible tetR-PtetA system, a well-characterized inducible promoter with minimal leakiness and easily tunable expression levels which is activated in the presence of tetracycline or its non-bactericidal analogue anhydrotetracyline (ATc). 39 A synthetic bidirectional transcription terminator was included upstream of this promoter to minimize polar effects upstream of the chromosomal replacement. This synthetic terminator/promoter construct was flanked by regions of homology (~500 bp) to the regions of MG1655 chromosome flanking the fimB promoters and the start codon of fimA facilitating allelic exchange with the native regulon without alteration of fimA sequence. A suicide plasmid derived from the R6K replicon, pDM4Cm 40 , was selected as the delivery vector. The mutation cassette was cloned into this vector by standard subcloning procedures as described, yielding the plasmid pJEP1. Homology-mediated allelic exchange with pJEP1, resulted in the replacement of the native regulon of the fim operon with the inducible tetR-ptetA system ( Fig. 1A-C). and associated regulatory elements. An invertible promoter element, fimS, regulates the downstream fimAICDFGH operon. Inversion is mediated by the actions of the two recombinases, FimB and FimE, which are encoded by genes upstream of the operon. B) Schematic representation of the fim operon in the regulatory mutant (mt), in which the native fimS regulatory system has been replaced by the inducible tetR-PtetA system. A synthetic bidirectional transcription terminator (red hexagon) is included upstream of tetR. Half arrows represent the primer pair FimF2/FimR2. C) PCR of mutant (mt) and wildtype (wild) E. coli (MG1655 background) with the mutation screening primer pair FimF2/FimR2, confirming chromosomal rearrangement. Representative confocal microscopy images of E. coli incubated with mannose-(D) and galactose-containing (E) Oregon Greentagged glycopolymers binding. These glycopolymers were prepared and characterized in a previous work 23 and used here as fluorescent probes to confirm selective bacterial lectin binding. E. coli cells are stained with DAPI (blue nucleic acid dye) and FM-464fx (red membrane dye) then incubated with 10 µM poly(glycan)@Oregon Green for 1 hour to facilitate lectin mediated binding. Scale bars represent 5 μm. i) Uninduced mutant (mt), ii) Mutant (mt) induced with ATc (100ng/ml), iii) wildtype (wild), iv) ∆fimH lectin deficient knockout (JW4283-3). The relative fluorescence intensity (RI) of total polymer bound was plotted as the sum green intensity (polymer signal) divided by sum red intensity signal (membrane signal), error bars represent standard error of the mean (SEM), (n≥3 in all experiments). Analysis performed by one-way ANOVA with Tukey's Multiple Comparison Test. A significant difference in bound mannose polymer was observed between the induced mutant and all other strains (p=0.002-0.0022) but no significant difference was observed between the uninduced mutant and the other strains. No significant capture of galactose polymer was identified under any conditions.
To determine whether the replacement of the native fim switch in E. coli MG1655 with the tetR-PtetA regulatory system would result in decreased fitness of the non-native regulatory mutant, a growth assay was performed. Growth of JEP01 (the regulatory mutant), in the presence and absence of the inducer compound anhydrotetracycline (ATc) was examined in parallel with the wild type strain (MG1655) and a lectin deficient (ΔfimH) strain (JW4283-3). No difference in growth rate was observed under the tested conditions (Supplementary Fig. S8).
To confirm affinity of mannosylated polymers for the inducible Type 1 fimbriae mutant in the presence of the inducer, a binding assay was performed utilizing fluorescently-labelled mannosylated glycopolymers described previously by our group. 23 Glycopolymer-bacteria association was only observed in the case of the inducible Type 1 fimbriae mutant, when the medium was supplemented with ATc, a non-bactericidal inducer of the inserted tetR-PtetA regulatory apparatus (Fig. 1D). A statistically significant difference in polymer signal (sum green fluorescence intensity) divided by cell membrane signal (sum red fluorescence intensity) was observed in the case of induced mutant compared to the uninduced mutant, the wild type (MG1655) and the ∆fimH strains (p=0.002, 0.002 and 0.022 respectively, n≥5). These iii) iv) data suggest that only the mutant, when induced by ATc, is able to bind significant quantities of mannosylated polymer and that this binding is mediated by expression of the Type 1 Fimbriae. In the absence of the inducer and, hence, absence of Type 1 fimbriae expression polymer-bacteria binding was very low (Fig. 1D). In contrast, when the same experiment was repeated with an analogous fluorescent polymer which possesses galactose pendant groups, as opposed to mannose, green fluorescence was markedly diminished (Fig 1E). In addition, the sum green fluorescence intensity over sum red fluorescence intensity for each strain, and hence the total of galactose-functional polymer bound to the bacteria as a function of cell density, was not significantly different between each strain and condition (Fig. 1E). Taken together, these data suggest that the constructed inducible fim mutant was able to bind mannosylated but not galactosylated glycopolymers, which is consistent with previous reports on the binding preferences of FimH. 36,41 This binding phenotype was exclusively displayed upon induction, suggesting tight, controllable expression of Type 1 fimbriae expression by replacement of the native regulon by the tetR-PtetA system.

Synthesis of MGNPs by grafting dopamine terminal glycopolymers to superparamagnetic iron (II,III) oxide nanoparticles (SPIONs)
To decorate MNPs with the required glycopolymers, we followed a 'grafting to' strategy where dopamine-terminal polymers were attached to superparamagnetic iron (II,III) oxide nanoparticles (SPIONs) via covalent linkages between the catechol group of dopamine and the surface of SPIONs. Recent examples of similar approaches include the use of glycopolymers made by ATRP, 42 and monodisperse PEG. 43 Here, the required catechol-terminal glycopolymers were synthesized by RAFT polymerization. Accordingly, firstly a catechol-containing RAFT chain transfer agent (Dopa-PABTC, (3)) was synthesized, by reaction of butanethiol with carbon disulfide then with bromopropionic acid in acetone, under basic conditions. The carboxy group of the resulting intermediate was treated with N-hydroxysuccinimide and DCC, and the resulting activated ester was reacted with dopamine, to give the desired catechol-containing Dopa-PABTC chain transfer agent ( Fig. 2A). Then, a library of four glycopolymers were synthesized by ultrafast aqueous RAFT polymerization, which allows almost quantitative monomer conversions, while retaining good chain-end fidelity and narrow molecular weight dispersity 44 . For the synthesis of the required glycopolymers both the nature of the repeating units and the length of the polymer chains were varied. The monomers utilized were mannose (ManAA, (5)) and galactose (GalAA,(7)) acrylamides, and non-sugar control hydroxyethyl acrylamide (HEAA). Targeted degrees of polymerization (DP) were 50 and 200. 1 H NMR analysis confirmed a good agreement between theoretical and experimental DPs of the final glycopolymers, while SEC analysis revealed narrow and monomodal molecular weight distributions and narrow dispersities (Ð) ( Table 1, Fig.  2A). For catechol-terminal glycopolymers with DPth = 50, number-average molecular masses estimated by 1 H NMR (Mn,NMR) and SEC (Mn,SEC) were in good agreement with the theoretical ones (Mn,th). for the short chain glycopolymers (Table 1). However, for control polymer Dopa-(HEAA) 50 and longer glycopolymer Dopa-(ManAA) 200 a certain deviation between Mn,SEC and Mn,th was observed, likely due to the difference in hydrodynamic volume between these polymers and the narrow standards used to calibrate the SEC.
The ability of these glycopolymers to selectively recognize mannose-binding lectins was then tested by surface plasmon resonance (SPR). To the best of our knowledge no methods have been reported in the literature for the immobilization of heterologously expressed bacterial FimH onto SPR sensor chips, nor is purified FimH commercially available. Thus, here Dendritic Cell-Specific Intercellular adhesion molecule-3-Grabbing Non-integrin (DC-SIGN) and mannosebinding lectin (MBL) were tested as model mannose-binding lectins. SPR results revealed that the mannose containing glycopolymers displayed a relatively strong binding avidity, with KD in the sub-nanomolar range for both lectins ( Fig. 2B and 2C, and Tables S4 and S5 in the supporting information). The longer mannosylated polymer (Dopa-(ManAA)200) demonstrated stronger binding than the shorter analogue Dopa-(ManAA)50, potentially due to its higher glycovalency. As expected, the non-sugar polymer (Dopa-(HEAA)50 ) and the non-mannose glycopolymer (Dopa-(GalAA)50) used as negative controls did not show any specific binding to DC-SIGN and MBL. Dopa-(ManAA)200 demonstrated the lowest KD value, thus, the highest binding avidity, for DC-SIGN (0.051 nM). Taken together, these initial experiments confirmed that Dopa-(ManAA)50 and Dopa-(ManAA)200 glycopolymers were able to recognize mannose-binding lectins. In principle, it is possible, or indeed likely, that other polymeric materials bearing glycans with structural similarity to mannose, for example glucose, which only differs from mannose in the stereochemistry of carbon C2 of the pyranose ring, could also bind to these model lectins and our fimbriated bacteria. However, investigation of the glycan preference and promiscuity of lectins was beyond the scope of the present work.
Next, SPIONs were synthesized by a thermal decomposition method, incorporating oleic acid for nanoparticle stabilization as described. Thermal decomposition produces small, monodisperse SPIONs of uniform shape. 43,45 However, oleic acid-stabilized SPIONs are dispersible in relatively non-polar solvents, and only a few examples of grafting of very hydrophilic materials such as glycopolymers have been reported. Recently, Galli and co-workers displaced oleic acid on SPIONs with catechol-functionalised polyamidoamines in an aqueous system using the phase transfer catalyst tetramethylammonium hydroxide (TMAOH). 46 This method was adapted in our work here for nanoparticle decoration with catechol-terminal glycopolymers. Following exchange of oleic acid for the glycopolymers, FTIR analysis revealed a decrease of C-H stretching bands in the 2800-2900 cm -1 consistent with the loss oleic acid (Fig. 3B). The concurrent emergence of bands at 3300 cm -1 corresponding to the sugar O-H stretch, and at 1660 and 1550 cm -1 for C=O stretch and N-H bend of the amido groups suggested successful decoration of the hydroxy and amido functionalities to the SPIONS consistent with catechol-terminal glycopolymer decoration. Thermogravimetric analysis (TGA) revealed a twostep TGA profile (below 400°C) for the oleic acid capped SPI-ONs, similar to that described in previous studies. 43,47 A twostep profile was also observed for Dopa-(HEAA) 50 coated nanoparticles. The final temperature at which no further weight loss is observed, around 595°C, was also similar for this material and the oleic acid capped SPIONs (Fig. 3C). This may suggest that incomplete ligand exchange was achieved between Dopa-(HEAA)50 and oleic acid, leading to some retention of oleic acid on the particle surface. Conversely, complete vaporization of the coating material was seen by 400°C for glycopolymer-functionalized SPIONs, suggesting high replacement of oleic acid. Furthermore, TGA plots do not reveal multistep patterns for these materials, suggesting a monophasic coating of the SPION surface. Polymer weight percentage by total organic content was determined to be 89, 85, 83, and 92%, for SPION@(HEAA)50, SPION@(GalAA)50, SPION@(ManAA)50 and SPION@(ManAA)200, respectively, demonstrating high functionalization for all coatings (Fig. 3C).
Particle imaging and sizing by transmission electron microscopy (TEM) did not reveal any significant change in the diameter of the SPION core of MGNPs (~6 nm) (Fig.  3D), with the glycopolymer corona which was not clearly visible under the conditions tested. TEM also revealed that all coated nanoparticles were monodisperse and presented a consistent, spherical morphology, with no significant aggregation (Fig. 3D)    Oleic acid coating

Fluorescent labelling of target bacteria
In order to distinguish bacterial strains within a mixed population we first aimed to tag strains of interest with constitutively expressing genes coding for red and green fluorescent proteins.
The Tn7-mediated chromosomal tagging system developed by Crépin and co-workers 49 was thus modified to include expression cassettes for mRuby2 50 and eGFP under the control of the synthetic strong, insulated, constitutive promoters, P11 and P24. 51 By inserting these constitutive expression systems as a single copy in the chromosome, we aimed to minimize variation and noise in fluorescence data which can commonly be observed in plasmid-based expression systems due to stochastic variance in copy number between cells in a clonal population. 52 PCR screening of the resultant transposon mutants revealed the integration of all fluorophore expression cassettes at the correct genomic locus (Supplementary Fig. S9). Growth and fluorescence analysis revealed no discernable difference in growth rates for any of the transposon mutants compared to their parental backgrounds regardless of promoter/fluorophore fusion inserted (Supplementary Figs S10 and S11). However, fluorescence cassettes under the control of the stronger promoter (P24) uniformly demonstrated higher fluorescence outputs (Supplementary Fig. S11). Hence only P24-controlled fluorescent protein tagged strains were taken forward for use in mixed population extraction experiments.

Specific cellular isolation of bacteria from mixed populations mediated by MGNPs
To test the ability of the produced MGNPs to selectively isolate bacteria from a mixed population, a pulldown experiment was performed, as illustrated in Fig. 4A. An initial 1/1 co-culture of the eGFP green tagged inducible mutant and a mRuby red tagged non-target strain (wildtype or ∆fimH) was prepared and treated with ATc inducer to selectively promote expression of the mannose-binding Type 1 fimbriae in the eGFP green tagged mutant. The mixed culture was then incubated with MGNPs with different glycopolymer coatings, followed by magnetic extraction of SPI-ONs-bacteria aggregates. Spatial separation of green mannose-binding fimbriated cells from their red afimbriate neighbors was quantified by analysis of the supernatant by flow cytometry, and expressed as red:green cell ratio ( Results showed successful depletion of eGFP green tagged fimbriated mutant following magnetic extraction with mannosylated MGNPs, as demonstrated by a 80.5 and 49.3% decrease for Dopa-(ManAA)50 and Dopa-(ManAA)200 coated nanoparticles, respectively in the mixed population containing wildtype E. coli cells (Fig. 4B). SPION@Dopa-(HEAA)50 control nanoparticles were unable to deplete the green labelled inducible mutant under any conditions, confirming the interaction is glycan dependent (Fig. 4B and 4C). Some depletion of the target strain from mixed populations was also observed in the case of galactosylated materials, however this effect was both small (5% on average) and highly variable ( Fig. 4B and 4C). Consequently, capture of fimbriated bacteria by galactosylated nanoparticles was statistically insignificant compared to the non-glycan decorated control, SPION@Dopa-(HEAA)50.
Increasing mannose polymer chain length decreased isolation of the Type 1 fimbriated E. coli. MGNPs coated with Dopa-(ManAA)50 were able to deplete 80.5% of the green labelled strain from the starting mixture, whereas MGNPs coated with the DP200 mannosylated polymers only depleted 49.3% of the green labelled (Fig. 4C). Potentially, the observed difference in fimbriated E. coli depletion by the two mannosylated materials could be attributable to the magnetic separation step, as in Dopa-(ManAA)50 coated MGNPs have higher proportion by weight of the magnetic iron oxide core (18%) compared to Dopa-(ManAA)200 coated MGNPs (8%) (Fig. 3C), thus increasing the attractive magnetic force exerted on these nanoparticles. While the presence of coatings on SPIONs does not by itself appear to diminish superparamagnetic properties, some studies have shown a decrease in saturation magnetization with the presence of coating materials or increasing polymer graft densities. [53][54][55] Kim and co-workers suggest that coatings decrease the uniformity of SPIONs due to quenching of surface moments, thus reducing the magnetic moment of these particles 56 . It is possible that reduced magnetic saturation of Dopa-(ManAA)200 coated SPIONs may have an effect on the efficiency of magnetic bacterial separation.
It is also possible that variance in the hydrodynamic radii of SPION@Dopa-(ManAA)50 and SPION@Dopa-(ManAA)200 may have some impact on the lectin interaction as this parameter could alter the spatial availability of glycan epitopes presented by nanoparticles in solution. For linear, brush and star glycopolymers those of greater size which display multiple sugar domains tend to have a higher affinity to target lectins than smaller, less valent materials, presumably due to an increase in binding epitope density 57-60 . This effect is remarkably pronounced when the size of polymers in solution matches, or exceeds, the minimum distance between binding sites on target multivalent lectins, which are typically in the order of 2-10 nm [61][62] . However, to the best of our knowledge the relationship between the hydrodynamic radius/volume of MGNPs and their affinity to target lectins has not been explored in existing literature. A decrease in grafting density was observed between SPIONs coated with glycopolymers with increasing chain length -0.65 chains nm -2 for SPION@Dopa-(ManAA)200 compared to 1.00 chains nm -2 for SPION@Dopa-(ManAA)50 (Fig.3). Although the former MGNPs would display a larger number of mannose molecules, in principle a different arrangement of sugar epitopes on SPION@Dopa-(ManAA)200 and SPION@Dopa-(ManAA)50, could potentially lead to a different affinity for bacterially associated Type 1 fimbriae and contribute to the decreased extraction efficiency of Dopa-(ManAA)200 coated MGNPs.
Intriguingly, when expression of the Type 1 fimbriae in the eGFP labelled mutant was not induced, a small proportion of the red labelled wildtype E. coli (MG1655::P24mRuby) was still depleted by the MGNPs as demonstrated by a negative green depletion of 11.7 and 9.5% by Dopa-(ManAA)50 and Dopa-(ManAA)200 coated nanoparticles, respectively ( Fig. 4B and 4C). Although depletion of the wildtype strain could in principle be attributed to the background stochastic expression of mannose binding Type 1 fimbriae 37 , leading to a small, fimbriated subpopulation which can be bound by mannosylated MGNPs, this effect is not statistically distinct to depletion of red labelled cells by the control materials, Dopa-(HEAA)50 and Dopa-(GalAA)50 functionalized MGNPs. Poly(hydroxyethyl acrylamide) (HEAA) coated MGNPs were included as a non-sugar negative control. Data shown as mean of three independent experiments with standard deviation. **** p<0.0001; *** p=0.001; * p<0.05 compared to non-sugar material control in one-way ANOVA with Tukey's multiple comparisons test.

Specific cellular isolation of bacteria from mixed populations by MGNPs is improved when the non-target strain is lectin deficient
In an attempt to further improve the specificity of magnetic isolation mediated by MGNPs magnetic isolation experiments were performed using an initial starting mixture of the eGFP tagged ATc-inducible Type 1 fimbriae mutant, and JW4283::P24mRuby -the mRuby2 tagged ∆fimH mutant described in the first part of this manuscript.
Similar to the effect seen for the mutant/wildtype E.coli, extraction experiments isolation of the ATc-inducible fimbriate strain was successfully achieved with mannosylated MGNPs as demonstrated by a decrease in green fluorescent cells (Fig. 4D and 4E). SPION@Dopa-(ManAA)50 was able to remove 94 % of the target strain from the starting mixed bacterial population, whereas MGNPs coated with Dopa-(Ma-nAA)200 a lower 34.8 % depletion was observed (Fig. 4E). These improved capture dynamics suggests a greater specificity of the mannosylated MGNPs to fimbriated cells when the counterpart strain does not harbor the genetic material for the expression of a lectin with competing specificity to the binding-sugar coated nanoparticles. However, a 17.9% depletion of green labelled cells was observed with SPION@Dopa-(GalAA)50. This finding may suggest that galactosylated materials could also be capable of fimbriated cell depletion by non-specific interactions. Nevertheless, this effect is not statistically significant compared with the separation obtained with the other control nanoparticles, SPION@Dopa-(HEAA)50, or with all materials in the absence of the inducer. This suggests that if non-specific binding is involved, its contribution to the overall binding is minimal. The higher depletion efficiency observed for the nanoparticles decorated with Dopa-(ManAA)50 may again be attributable to a more efficient magnetic separation step, due to the higher content in iron oxide in Dopa-(ManAA)50 compared to Dopa-(ManAA)200, or differences in spatial availability of sugar binding units between these two MGNPs.
Additionally, in the absence of the ATc inducer, no depletion of the afimbriate strain from the starting mixed population is observed by mannosylated glyconanoparticles (Fig. 4D and 4E). Thus, lack of depletion in the absence of the inducer further supports that specificity of the magnetic isolation of fimbriated bacteria from a mixed population is improved when the counterpart strain is afimbriate.
Taken together, these results show that coupling inducible lectin expression with MGNPs mediated capture could provide a viable, highly efficient tool for bacteria isolation from a mixed population with a maximum 94% target depletion achieved under the conditions tested. Depletion of fimbriated E. coli was achieved with higher efficiency by MNPs coated with shorter glycopolymers and may be influenced by the sugar binding capacity of the non-target strain.

CONCLUSIONS
In this report we have shown that bacterial subpopulations can be successfully isolated from mixed co-cultures through a strategy involving first the induction of the Type 1 fimbriae by coupling expression to an external chemical stimulus (ATc), followed by magnetic capture using SPIONs decorated with glycopolymers with specific chain length and pendant sugar ligands.
To this aim, we first constructed an inducible fim mutant in E. coli which specifically adhered to mannosylated, but not galactosylated, glycopolymers upon induction. Next, glycosylated magnetic nanoparticle were successfully engineered from SPIONs and a family of catechol-terminal glycopolymers synthesized by RAFT polymerization, through a 'grafting to' approach. Finally, the general strategy proposed in this work was demonstrated with a mixed bacterial culture containing two differentially fluorescent protein labelled E. coli strains which was selectively depleted of fimbriate (induced mutant) bacteria by magnetic capture using the produced MGNPs.
Isolation of fimbriated E. coli (fim + ) from afimbriate (fim -) E. coli was near complete under the best conditions tested and isolation was most efficient when non-target strain did not harbour the genetic apparatus for expression of Type 1 fimbriae. This may suggest that genetic manipulation of target microbes may improve selectivity by MGNPs. Interestingly, in terms of initial structure-function relationship for these materials, SPION@Dopa-(ManAA)50 proved more effective, with almost complete depletion of fimbriated E. coli from solution, than nanoparticles decorated with longer polymer chains, SPION@Dopa-(ManAA)200. This was partially ascribed to a higher iron oxide content in the former, which would facilitate the magnetic extraction of target bacteria, and suggest that the efficiency of the isolation process may depend not only on the avidity by which glycopolymer ligands bind bacterial lectins, but also on the iron oxide:glycopolymer mass ratio. However, this relationship is complex and is likely also to be impacted by both the packing and orientation of glycopolymers onto the nanoparticle surface, as demonstrated by the observed difference in grafting density for MGNPs bearing different length glycopolymers, and the hydrodynamic volume of the coated MGNPs in solution.
This system complements the range of available bacterial extraction systems by utilizing materials which are synthetically scalable. Additionally, coupling varied bacterial lectin expression systems to stimulus responsive promoters could allow for 'on demand' activation of MGNP binding, leading to depletion of a target strain at a specific time or in sequential rounds as desired. It may also be possible to heterologously express non-native lectins 63 or other ligand capture proteins such as streptavidin 64 on bacterial surfaces, thus mediating their selective isolation by decorated MNPs and broaden the applications of this system further.
Finally, in principle, bacterial isolation by MGNPs may not be limited to biotechnology and could be utilized for the specific capture of bacteria from mixed populations in 11 a range of applications. These include replacement of immunomagnetic nanoparticles in diagnostics 65 ; treatment of contaminated drinking water 66 ; pathogen detection agent in the food industry 16,67 ; bioremediation 68 ; extraction of biosensing strains from microbial consortia 69 and in other fundamental microbial research.

Instruments and Analysis
1 H and 13 C NMR spectra were recorded on a Bruker 400 Ultrashield spectrometer using deuterated solvents. High-resolution 1 H NMR for calculation of polymer DP was performed on a Bruker Av(III)500 spectrometer equipped with a dual 1 H/ 13 C helium-cooled cyroprobe. ESI-TOF mass spectrometry was performed on a Bruker microTOF II mass spectrometer in either HPLC grade MeCN, MeOH or milliQ H2O with a sample concentration of ≤0.01 mg/mL. Fourier Transform Infrared Spectrometry was performed on an Agilent Cary630 FTIR with an ATR module between the wavelengths of 4000-650 cm -1 with a resolution of 2 cm -1 with 32 background scans and 64 sample scans per sample. Thermogravimetric analysis was performed on a TA Instruments Q500 TGA in the temperature range 25-650°C with a ramp of 10 K/min under inert gas. Platinum pans were filled with ~ 5 mg of sample and the total organic content was evaluated as the mass loss fraction at 500°C by the horizontal setting. Imaging of coated and uncoated SPIONs was performed on a Tecnai G2 transmission electron microscope (Tecnai) with an acceleration voltage of 100 kV. 13 μL of aqueous sample was deposited on a Cu200 grid with a carbon support film and negatively stained with uranyl acetate (2%) where applicable. The suspension was allowed to settle onto the grid for ~10 minutes, excess liquid removed, and the grid allowed to air dry prior to imaging. Particle sizing analysis was performed in ImageJ for >800 individual particles. Nanoparticle grafting densities were calculated using the TGA grafting density equation reported by Benoit et al 48 . Dynamic light scattering analysis was performed using a Malvern Zetasizer Nano-ZS at 25°C with a 4 mW He-Ne 633 nm laser at a scattering angle of 173° (back scattering). Measurements were taken assuming the refractive index (1.440) and absorption (3.0) of iron oxide. Nanoparticles were measured at 0.5 mg mL -1 in deionised water and measured after filtration through a 0.22 µm cellulose filter in low volume (70 µL) polystyrene cuvettes. Samples were incubated for 60 seconds at 25°C prior to measurement. Measurements were repeated three times with automatic attenuation selection and measurement position. Results were analysed using Malvern DTS 6.20 software.

Bacterial Strains and Growth Conditions.
E. coli strains produced and used in this work are described in Supplementary Table S1. Bacteria were grown in Lysogeny Broth (LB) liquid medium and agar plates (1.5% w/v), at 37°C. For counter-selection after biparental mating, bacteria were grown in M9 minimal media supplemented with fructose (M9F) or M9F agar plates (1.5% w/v). When needed for plasmid selection, antibiotics were added to the media at the following concentrations: Ampicillin (Amp); 25 µg/mL Chloramphenicol (Cm); 25 µg/mL Kanamycin (Km); 15 µg/mL Trimethoprim (Trim) and 20 µg/mL Tetracycline (Tet). Replicative plasmid propagation was performed in strain DH5α. For cloning and propagation of suicide vectors, containing the conditional R6K origin of replication, strains S17-1(λpir) or ST18 were used.

Plasmids, DNA Constructs, and Oligonucleotides.
The plasmids constructed and used in this study are summarized in Supplementary Table S2. Oligonucleotides were synthesized by Sigma-Aldrich and are listed in Supplementary  Table S2. PCRs were performed with GoTaq Green Master Mix (Promega) for standard amplifications in screenings and with Q5 High-Fidelity DNA Polymerase (New England Biolabs) for cloning purposes. All plasmid identities were confirmed by restriction digestion and Sanger sequencing (Source Bioscience) prior to use.

Construction of an inducible fim mutant in the E. coli background MG1655
The mutation cassette for construction of an anhydrotetracyline inducible mutant of the Fim operon was synthesised by Thermo Fischer using the GeneArt Custom Gene Synthesis service. Design of this cassette is shown in Scheme S.1. The inducible mutation cassette was amplified from the pMKFimAE by PCR using primers Fim1_Fw and Fim1_Rv, (Table. S.3) then subcloned into the suicide plasmid pDM4 by restriction cloning with XhoI and XbaI. Biparental mating was performed using 3 x 10 7 CFU/mL of donor strain E. coli S17-1 carrying pJEP1 and 1 x 10 7 CFU/mL of the recipient strain (MG1655) at 30°C on M9F agar plates for 18 h. After incubation, the mating lawn was resuspended in sterile PBS and serially diluted to 10 -3 , spread onto M9FCm plates 30°C until individual colony growth was observed (typically 2 to 3 days) to select against the donor and screen for single crossovers. To enrich the cultures for clones having accomplished double crossover events, in which the suicide plasmid was excised, the toxic effects of the sacB gene on cells carrying the suicide vector was exploited. Single crossover isolates were grown overnight in no salt LB medium (NSLB) at 30°C. This culture was diluted 100-fold in fresh NSLB medium supplemented with sucrose 15% and grown overnight at 30°C twice. The cultures were then serially diluted to 10 -6 and 100 µL of each serial dilution was plated onto a NSLB sucrose 15% agar plate and grown overnight at 30°C. Sucrose-resistant colonies were picked from these plates and patched onto NSLB agar, and replica-plated on NSLB sucrose 15% agar and NSLB agar supplemented with chloramphenicol 25 µg/µL. Following the isolation of several Cm S , Suc R colonies which had undergone double crossover homologous recombination, PCR screening was able to confirm the presence of the desired mutation at the correct chromosomal location, with an electrophoretic profile consistent with the expected size of the insertion-deletion mutation. Furthermore, Sanger sequencing of the PCR 12 product isolated from mutant clones aligned with the expected sequence with full identity, confirming the chromosomal presence of the designed construct (data not shown).

Confirmation of mutant phenotype by fluorescent glycopolymer binding assay
A 5-mL culture of the strains of interest including the fim inducible mutant; wild type E coli, MG1655; and the ∆fimH mutant, (JW4283-3) were grown overnight in LB media at 37 °C with agitation at 200 rpm for aeration. The following morning, 5 mL of fresh LB media was inoculated with 50 μL of the overnight culture and this procedure was repeated for all strains. The fresh cultures were grown at 37°C with agitation at 200 rpm until the mid-logarithmic growth phase (typically ~0.4-0.6 at OD600). ATc (to 100 ng/mL) was then added to the inducible mutant culture and grown for a further 2-4 hours to induce expression of the Type 1 Fimbriae. 1 mL of each culture was then dispensed into a sterile 1.5 mL microfuge tube, pelleted by centrifugation at 5000 xg for 5 min and

Chromosomal tagging of bacterial strains with constitutive fluorescent protein expression cassettes by tn7 transposition
The constitutive promoters P11 and P24 were obtained cloned in a pBBR1 replicon vector fused to eYFP 51 . eYFP was excised from these plasmids by digestion with EcoRI and BamHI. The fluorescent proteins mRuby and eGFP were amplified by PCR from pSW002-Pc-mRuby2 and pJH257-2 respectively. These primer sets had a 5' extension on the forward primer to restore the RBS (Ribosome Binding Site) lost from the pBBR1 vector series by EcoRI/BamHI digestion. The PCR products were digested with EcoRI and BamHI and then ligated into the cut pBBR1-P11 and pBBR1-P24 vectors yielding the pBBR1 P11/P24 mRuby2/eGFP vector series. Analysis of plasmid borne cassette fluorescence revealed that strong fluorescence was only seen in the case of the two strongest promoters P11 and P24. Next the insulated constitutive fluorescent protein expression cassettes were amplified from the pBBR1 intermediate vectors with the PxxF/R primer set, digested with XhoI/NsiI and ligated into pGP-Tn7-Cm digested with XhoI/NsiI yielding the vector series: pGP-Tn7-Cm-P11mRuby; pGP-Tn7-Cm-P11eGFP; pGP-Tn7-Cm-P24mRuby and pGP-Tn7-Cm-P24eGFP. Since the promoter insulation motif upstream of the promoters and fluorescent protein Ctermini were identical for all the constructs, a single primer set was sufficient for all mRuby/eGFP constructs A 5' extension was included on the primer set for this amplification to restore the RBS and start/valine codon which was omitted by digestion with this enzyme set. The sequence of all the resulting plasmids was verified by Sanger sequencing prior to Tn7-mediated transposition of the constitutive fluorescent protein expression cassettes onto the E. coli chromosome.

Delivery of mini-Tn7 constitutive fluorescent protein expression cassettes into desired E. coli recipient strains
Biparental mating was performed using 3 x 10 7 CFU/mL of donor strain E. coli ST18 carrying the pGP-Tn7-Cm derivative vectors harboring the mRuby2/eGFP expression cassettes and 1 x 10 7 CFU/mL of the recipient strains carrying the transposase-encoding vector pSTNSK at 30°C on LB agar plates for 18 h. After incubation, the mating lawn was resuspended in sterile PBS and serially diluted to 10 -3 , spread onto M9FCm plates, incubated at 42°C for 5h and then at 37 °C for 18 h. Colonies were then screened for resistance to Cm and sensitivity to Amp and Km. Since the Amp R marker is located outside of the transposable mini-Tn7 cassette on the suicide vectors, sensitivity to Amp indicates the proper integration of the mini-Tn7 cassette at the intended chromosomal attTn7 site instead of any unwanted chromosomal integration of the vector. Additionally, since the transposases are encoded on a temperature-sensitive helper plasmid, incubation at 42°C was undertaken to promote the loss of these plasmids from the recipient strains, which is verified by loss of Km resistance. Furthermore, the use of M9F Cm plates selected for the growth of the recipient strains as the S17λpir donors are thi-, pro-and hence are auxotrophic for these amino acids. Chromosomal integration of the mini-Tn7 transposons at the attTn7 site downstream of glmS within various fluorescent clones was confirmed by PCR with the glmS/pstS primer pair (Table. S.3).

General method for aqueous RAFT mediated polymerization of acrylamide glycomonomers
The synthesis of the Dopa-PABTC CTA, 2′-acrylamidoethylα-D-mannopyranoside and 2′-acrylamidoethyl-β-D-galactopyranoside glycomonomers and corresponding precursors is described in the Supporting Information. In a typical polymerization a glass vial was charged with mannose acrylamide (1.00 g, 3.61 mmol, 50 eq.), Dopa-PABTC (27 mg, 0.072 mmol, 1 eq) (from a 100 mg mL -1 stock solution in 1,4-dioxane) and VA-044 (1.16 mg, 3.61 x 10 -3 mmol, 0.05 eq.) (from a stock solution of 20 mg mL -1 in water). A stirrer bar and a water:1,4-dioxane (4:1) mixture was added, to a total volume of 3.6 mL. The reaction vessel was sealed with a rubber septum and degassed with N2 for 15 min before being immersed in a preheated oil bath at 70 °C for 2 hours. Polymers with different DP were synthesized by varying the Monomer conversion was determined by 1 H NMR analysis in DMSO-d6 of the final reaction mixture, by comparing one of the monomer vinyl peak at 6.07 ppm to the CTA CH3 Zgroup chain end group peak (δ =0.86 ppm), in all cases ≥97% final monomer conversion was obtained. All polymers were analyzed by SEC with aqueous 0.10 M NaNO3 as the mobile phase, at 35 °C. Exemplar 1 H spectra are shown in Fig. S20-22 in the Supporting Information. DP was calculated by comparing the integral of the catechol protons (δ 6.79, 6.56, and 6.66 ppm), each set to 1, to those of the -CH and -CH2 signals for the polymeric backbone (δ 2.17 and 1.70 ppm, respectively).

Synthesis of superparamagnetic iron(II,III) oxide nanoparticles (SPION)
Oleic acid stabilized magnetite (Fe3O4) nanoparticles were obtained as per the thermal decomposition method reported previously, 43 with some modification. A mixture of dioctyl ether (50 mL) and oleic acid (9.36 mL, 8.38 g, 29.7 mmol, 4 eq.,) were heated to 100°C in a two neck RB flask with constant N2 bubbling. After 15 min Iron(0) pentacarbonyl (7.61 mmol, 1.49 g, 1 eq., 1 mL) was then injected rapidly into the oleic acid/dioctyl ether solution and the resulting mixture was heated to 290°C with a temperature ramp of 3 °C /min. After 1 h the nanoparticle dispersion was allowed to cool to room temperature. The nanoparticles were precipitated 3x with acetone from toluene to remove the excess of oleic acid. The average size and morphology of the isolated particles were analysed by TEM.

Decoration of SPIONs with RAFT polymerization catechol terminal glycopolymers
Modification of the hexane soluble SPIONs stabilized with oleic acid was performed via a ligand exchange reaction with the catechol-terminal polymers by a two-step ligand exchange method adapted from Galli et al. 46 with some modification. In a typical reaction, a suspension of 15 mg of oleic acid-stabilized SPIONs in approximately 3 ml of hexane was dried under reduced pressure, via rotary evaporation. Subsequently aqueous solution of TMAOH (0.125 M, 20 mL) was added and the vial was sonicated for a few minutes to redisperse the SPIONs, yielding a turbid mixture. To the water-dispersed SPIONs, DOPA-poly(HEAA)50 (39 mg, 0.1 eq to Fe3O4) was added and left stirring for 48 hours. The particles were then purified from free polymer/oleic acid by dialysis against water for 3 days in 100 kDa MWCO Spectra-Por Float-A-Lysers (Merck). The particles were then lyophilized to yield a fine black powder which responded to an external magnetic field and was easily redispersible in water. A sample of the lyophilized powder was analyzed by FTIR, TGA and TEM to confirm presence of the polymer coating, estimate the amount of polymer in polymer-SPIONs conjugates, and for size distribution analysis.

Surface Plasmon Resonance (SPR) and kinetics
The extent of interaction between the glycopolymers and lectins were analyzed on a BIAcore 3000 system (GE Healthcare). DC-SIGN and MBL (0.025 mg/ml) were immobilized via a standard amino coupling protocol onto a CM5 sensor chip that was activated by flowing a 1:1 mixture of 0.1 M N-hydroxysuccinimide and 0.1 M N-ethyl-N'-(dimethylaminopropyl)carbodiimide over the chip for 6 min at 25 °C at a flow rate of 20 µL/min after system equilibration with filtered HEPES buffer (10 mM HEPES, 150 mM NaCl, 5 mM CaCl2) supplemented with 0.005% TWEEN® 20 at pH= 7.4. Subsequently, channels 1 (blank), 2, 3 and 4 were blocked by following a solution of ethanolamine (1 M pH 8.5) for 10 min at 5 µL/min to remove remaining reactive groups on the channels. Sample solutions were prepared at varying concentrations (10 nM-0.625 nM) in the same HEPES buffer to calculate the binding kinetics. Sensorgrams for each glycopolymer concentration were recorded with a 300 seconds injection of polymer solution (on period) followed by 150 seconds of buffer alone (off period). Regeneration of the sensor chip surfaces was performed using 10 mM HEPES pH 7.4, 150 mM NaCl, 10 mM EDTA, 0.01% P20 surfactant solution. Kinetic data were evaluated using a single set of sites (1:1 Langmuir Binding) model in the BIAevalulation 3.1 software

Magnetic Isolation of target bacteria from mixed populations by magnetic glyconanoparticles (MGNPs)
A total of 5 ml cultures of each JEP01::p24eGFP and non-target controls (JW4283::p24mRuby/MG1655::p24mRuby) were grown in LB media overnight supplemented with chloramphenicol (25 μg/ml). The following morning the OD600 of these cultures was measured and used to inoculate 5 ml of fresh LB for individual cultures and the two mixed cultures, with OD600 normalized to 0.05. These day cultures were then incubated at 37°C for 2 hours, prior to the addition of ATc (50ng/ml), where appropriate, and further incubation for another 3 hours to induce expression of the fimbriae in one of the mixed cultures. After 5 hours of growth, 100 μL of each culture (JEP01::p24eGFP alone, JW::p24mRuby alone, mixed culture with ATc, mixed culture without ATc) were transferred into 1.5 mL Eppendorf tube. The cell pellet was collected by centrifugation and washed twice with sterile PBS. The cell pellet was then resuspended in 200 μL of nanoparticle solution for each coated nanoparticle sample (SPION@Dopa-(HEAA)50, SPION@Dopa-(GalAA)50, SPION@Dopa-(ManAA)50, SPION@Dopa-(Man)200; all at 5 mg/ml in sterile PBS) and left on a magnetic extraction rack 14 overnight at 4 °C. The following morning, the supernatant and pellet were recovered for each sample. Cells from were then collected by centrifugation, washed twice with sterile PBS and resuspended in 500 μL sterile PBS. The relative proportion of each strain was then estimated by flow cytometry using an Astrios EQ cell sorter (Beckman Coulter).

ASSOCIATED CONTENT Supporting Information.
Further information on plasmids, bacterial strain construction, validation and additional material construction data. This material is available free of charge via the Internet at http://pubs.acs.org.