RND Pump-Mediated Efflux of Amotosalen, a Compound Used in Pathogen Inactivation Technology to Enhance Safety of Blood Transfusion Products, May Compromise Its Gram-Negative Anti-Bacterial Activity

ABSTRACT Pathogen inactivation is a strategy to improve the safety of transfusion products. The only pathogen reduction technology for blood products currently approved in the US utilizes a psoralen compound, called amotosalen, in combination with UVA light to inactivate bacteria, viruses, and protozoa. Psoralens have structural similarity to bacterial multidrug efflux pump substrates. As these efflux pumps are often overexpressed in multidrug-resistant pathogens, we tested whether contemporary drug-resistant pathogens might show resistance to amotosalen and other psoralens based on multidrug efflux mechanisms through genetic, biophysical, and molecular modeling analysis. The main efflux systems in Enterobacterales, Acinetobacter baumannii, and Pseudomonas aeruginosa are tripartite resistance-nodulation-cell division (RND) systems, which span the inner and outer membranes of Gram-negative pathogens, and expel antibiotics from the bacterial cytoplasm into the extracellular space. We provide evidence that amotosalen is an efflux substrate for the E. coli AcrAB, Acinetobacter baumannii AdeABC, and P. aeruginosa MexXY RND efflux pumps. Furthermore, we show that the MICs for contemporary Gram-negative bacterial isolates for these species and others in vitro approached and exceeded the concentration of amotosalen used in the approved platelet and plasma inactivation procedures. These findings suggest that otherwise safe and effective inactivation methods should be further studied to identify possible gaps in their ability to inactivate contemporary, multidrug-resistant bacterial pathogens. IMPORTANCE Pathogen inactivation is a strategy to enhance the safety of transfused blood products. We identify the compound, amotosalen, widely used for pathogen inactivation, as a bacterial multidrug efflux substrate. Specifically, experiments suggest that amotosalen is pumped out of bacteria by major efflux pumps in E. coli, Acinetobacter baumannii, and Pseudomonas aeruginosa. Such efflux pumps are often overexpressed in multidrug-resistant pathogens. Importantly, the MICs for contemporary multidrug-resistant Enterobacterales, Acinetobacter baumannii, Pseudomonas aeruginosa, Burkholderia spp., and Stenotrophomonas maltophilia isolates approached or exceeded the amotosalen concentration used in approved platelet and plasma inactivation procedures, potentially as a result of efflux pump activity. Although there are important differences in methodology between our experiments and blood product pathogen inactivation, these findings suggest that otherwise safe and effective inactivation methods should be further studied to identify possible gaps in their ability to inactivate contemporary, multidrug-resistant bacterial pathogens.


RESULTS
Assessment of amotosalen activity against contemporary multidrug-resistant E. coli, K. pneumoniae, A. baumannii, Pseudomonas aeruginosa, Strenotrophomonas maltophilia, and Burkholderia isolates was performed using MIC testing (Table 1 and Table S1). Notably, MIC values were near, and for some isolates, exceeded the 150 mM concentration of amotosalen used in the Intercept pathogen inactivation system. The modal MICs of multidrugresistant (MDR) E. coli and K. pneumoniae (128 mM) isolates exceeded the MICs of broadly-susceptible ATCC strains of the same species (8 mM); however, broadly-susceptible A. baumannii 17978 (34) had an MIC of 128 mM, identical to the modal MIC of MDR A. baumannii isolates. Pseudomonas aeruginosa strains generally had either high MICs or were variably killed directly by UV light (Table 1 and data not shown).
We then examined whether well-characterized, major efflux pumps from E. coli, A. baumannii and P. aeruginosa were capable of influencing susceptibility to psoralen compounds. These pumps are classified as resistance-nodulation-cell division (RND) efflux pumps, and consist of 3 components residing, respectively, in the inner membrane, periplasm, and outer membrane.
In E. coli, multiple RND efflux pumps depend on a shared outer membrane channel protein, TolC (35). We, therefore, compared psoralen MICs in an E. coli K-12 parent strain and a TolC-knockout (DtolC) strain that were otherwise genetically identical (i.e., isogenic strains). For amotosalen, 8-MOP and AMT (Table 2), the MIC in the DtolC background was reduced by up to 64-fold, consistent with TolC-dependent efflux of psoralens. The magnitude of MIC reduction was similar to that observed for positive controls, clindamycin, fusidic acid, ethidium bromide, and minocycline, known efflux substrates (33,36,37). As expected, based on prior observations (36), non-effluxed, gentamicin and apramycin aminoglycosides, and meropenem; and poorly effluxed, rifampin negative controls showed minimal or no TolC-dependent change in MIC (38)(39)(40). Furthermore, phenylalanine-arginine b-naphthylamide (PAbN), a competitive inhibitor of RND efflux pumps (41), reduced the MICs of psoralens including amotosalen in the wildtype strain by approximately 4-fold, but without significant additional effect on the tolC mutant, also supporting a major contribution of RND efflux pumps to observed psoralen resistance in the E. coli K-12 strain background.
In A. baumannii and other Acinetobacter species, multidrug resistance is commonly associated with upregulation of the AdeABC RND efflux system (42)(43)(44). To examine whether AdeABC could efflux amotosalen, we cloned adeAB and adeC on separate plasmids under the control of inducible promoters. These plasmids were then transformed alone or in combination into the E. coli AG100AX strain, which is deleted for the main RND efflux pumps (DacrAB, DacrEF) that partner with TolC, thereby removing confounding effects of native efflux pumps in the host strain (45).
For amotosalen, the modal MIC of the AdeC expressing control strain, i.e., an incomplete, inoperative pump, was found to be 8 mM, which was comparable with the modal MIC of the AG100AX parent E. coli strain (Table 3). In contrast, a strain expressing an induced, intact AdeABC pump, exhibited a 32-fold higher modal MIC of 256 mM. Therefore, psoralen MIC was substantially increased in the presence of the complete heterologous AdeABC RND pump, providing evidence for efflux of amotosalen by this alternative system. Similarly, expression of adeABC increased the MIC of the known efflux substrate, minocycline, approximately 10-fold, while having relatively little effect on rifampin, a substrate of a different pump in A. baumannii, adeIJK (46), and apramycin, a poor substrate and therefore a negative control. Lastly, we noted that expression of E. coli acrAB and P. aeruginosa mexXY in AG100AX led to a substantial increase in amotosalen MIC, while expression of the A. baumannii AdeIJK and P. aeruginosa MexAB pumps or a vector expressing an unrelated fluorescent protein used as a negative control had negligible, if any, effect (Table 4). In these experiments, minocycline MIC was modestly increased in the AdeIJK and MexXY strains as expected (32,47). Surprisingly, an increase in gentamicin MIC was minimal, or not observed, with expression of MexXY, a pump that is responsible for efflux of aminoglycosides in P. aeruginosa. This was the case even when the outer membrane protein, OprM, believed to partner with MexXY in P. aeruginosa, was expressed from a separate plasmid in a DtolC deletion strain. In contrast, the MIC of amotosalen was increased 64-fold, when MexXY and OprM were co-expressed in this strain background (data not shown).
Traditionally, MIC testing is performed in the absence of serum proteins; however, such testing may overestimate activity if lower amounts of free drug are available for antimicrobial activity due to binding to circulating proteins, such as albumin. Therefore, as psoralens are relatively hydrophobic, we considered the potential effect of human serum binding on its activity. We performed experiments in the DtolC E. coli K-12 strain background to allow us to evaluate the extent of MIC increase within the testable range of amotosalen concentrations. However, E. coli K-12 was inhibited by 50% serum. It has been observed previously that many strains do not grow in the presence of serum (48), potentially necessitating the use of alternatives, such as human serum albumin (HSA) as a surrogate or adaptation of strains to serum containing medium. We found, however, that addition of 1 mg/mL of sodium polyanethole sulfonate (SPS), a standard additive in blood culture medium, alleviated observed serum inhibition, which is consistent with its known ability to block complement-and non-complement-mediated serum bactericidal activity (49). Notably, MIC of amotosalen was not consistently increased in the presence of serum, but increased 16-fold with 2% HSA (Table 5). Therefore, there were discordant effects of serum and HSA on amotosalen activity. Fusidic acid, protein binding of 97 to 99% (https://go.drugbank.com/drugs/DB02703), was used a positive control, and showed a similar 8-to 16-fold increase in MIC in the presence of either serum or HSA. In contrast, chloramphenicol with a plasma protein binding of roughly 50% (https://go.drugbank.com/ drugs/DB00446) showed a much smaller increase in MIC, while the activity of apramycin, observed to have , 10% serum protein binding in a recent study (50), was modestly enhanced in the presence of serum or HSA. It should be noted that the observations for serum and HSA may differ from potential effects of plasma present in blood products on amotosalen during the pathogen inactivation procedure.
Transposon mutants in RND pumps in widely researched strains of A. baumannii AB5075 (51, 52) and P. aeruginosa PAO1 (53) from the University of Washington transpo-  son bank were also examined for effects on amotosalen MIC (Table 6). Independent mutants in adeB showed a 4-fold lower MIC, while mutants in other A. baumannii and P. aeruginosa pumps examined did not result in an observable change, although it should be noted that the baseline level of resistance in the P. aeruginosa PAO1 strain background was beyond the measurement range of our assay.
In the A. baumannii AdeABC efflux pump, the AdeB protein is the drug-binding channel and pump, energized by a proton motive force, which in coordination with other pump components, extrudes compounds from the bacterial cell (54). We, therefore, assessed the ability of amotosalen to bind purified AdeB using a fluorescence polarization assay, taking advantage of the inherent fluorescence of amotosalen (Fig. 1). The dissociation constant (K D ) for binding to AdeB was 27.9 6 1.8 mM, in the same range as the K D of rhodamine 6G (3.1 mM) and ethidium bromide (2.5 mM) for AdeB (55), and the K D of ethidium bromide (8.7 mM), proflavin (14.5 mM), and ciprofloxacin (74.1 mM) efflux substrates for the homologous, AcrB (30).
The previously described cryo-EM structure of AdeB identified a pathway for efflux pump substrate extrusion with entrance through a cleft in the periplasmic domain, and sequential binding to proximal and distal multidrug-binding sites (54,55). Computer modeling of the molecular docking of amotosalen using Extra Precision Glide (56) demonstrated binding to both the proximal and distal multidrug binding sites within the AdeB periplasm domain (Fig. 2). AMT and 8-MOP were also found to dock with the proximal and distal binding sites, and cleft and distal binding sites, respectively (data not shown). These data are consistent with binding of known efflux substrates to AdeB and homologous RND pumps (30).

DISCUSSION
Microbial contamination of blood products remains a critical transfusion safety issue. Here, we provide evidence that psoralens including amotosalen are multidrug efflux substrates. Noted, we did not directly demonstrate efflux using physicochemical methods; however, heterologous expression and gene mutation data, and the inhibitory effects of the well-established efflux pump inhibitor, PAbN, are highly supportive of an efflux mechanism. It is of interest that the derivatization of psoralen scaffold to create amotosalen and AMT included the instillation of a primary amine into an existing planar structure with low  globularity and few rotatable bonds (Fig. 3). These features, taken together, are now known to be associated with enhanced penetration of antibiotics across the Gram-negative membrane barrier (57). Therefore, the increased activity (lower MIC values) of amotosalen and AMT compared with 8-MOP is fully consistent with our current understanding of antimicrobial penetrance into Gram-negative pathogens. Nevertheless, access to an intracellular target (in this case, DNA) appears especially vulnerable to efflux mechanisms. The major RND efflux pumps that efflux amotosalen, AcrAB-TolC, AdeABC, and MexXY, have overlapping substrate specificity, and 40 to 50% amino acid identity (58,59). Therefore, finding that these separate systems all efflux psoralens is not surprising. Interestingly, amotosalen appeared to be selectively effluxed by MexXY rather than MexAB. Previously, these 2 pumps were found to have some substrates in common and some unique substrates (59). Amotosalen now joins aminoglycosides among compounds selectively effluxed by MexXY. Genetic evidence suggested the MexXY-dependent efflux of amotosalen occurs in E. coli with either native TolC or heterologously expressed, P. aeruginosa OprM serving as outer membrane partner.
Binding of amotosalen to AdeB showed a micromolar dissociation constant, consistent with known affinities of antibiotics to RND pumps (30) and MATE family transporters (60). This affinity is likely located along a dissociation continuum that allows for the  ability to bind to several sites within the pump itself with eventual handoff to the outer membrane pump protein, and release into the extracellular space (54). Molecular docking to AdeB was consistent with what has been either determined or predicted for other canonical efflux substrates (55). AcrAB-TolC AdeABC, and MexXY are upregulated in resistant pathogens. For example, acrAB is upregulated in Enterobacter species resistant to colistin (61); E. coli resistant to multiple antibiotics (62)(63)(64); and in drug-resistant Salmonella (65, 66) and Klebsiella (67,68). The expression of AdeABC is often upregulated in resistant Acinetobacter isolates (42)(43)(44). MexXY is normally relatively suppressed, but when upregulated is responsible for inducible pan-aminoglycoside resistance in P. aeruginosa (69), and such strains may be even more resistant to amotosalen than those in our analysis. The lack of a substantial effect of mexXY mutants in P. aeruginosa PAO1, i.e., bringing the MIC into a measurable range, may reflect contributions of additional pumps (there are 12 RND pumps in PAO1) not examined in our study (70). In addition, Stenotrophomonas maltophilia, and Burkholderia spp. express or overexpress multiple efflux pumps, resulting in characteristic intrinsic resistance to many antibiotics (71)(72)(73).  (55) using Induced Fit Docking and XP scores for ranking. The sites are within the hydrophobic cleft between PC1 and PC2 subdomains in AdeB that are thought to form the entry site and pathway for efflux from the periplasm (54). Amotosalen is predicted to bind proximal and distal drug binding sites, as previously defined through autodocking and experimental analysis of antibiotic efflux substrates for AdeB, and homologous RND transporters (54,92,93). Specifically, critical contacts are made with hydrophilic residue D664 in the PAIDELGT sequence defined "F-Loop," which forms both part of the cleft entrance and the proximal drug binding site. Amotosalen also binds to the distal multidrug binding site inclusive of hydrophobic interactions with F178, I607, and W610. This hydrophobic patch in the homologous AcrB is critical for stable binding of all substrates, and is further highlighted in corresponding residue interactions in the AcrB-minocycline crystal structure and corresponding MtrD-erythromycin cryo-EM structure (54,(92)(93)(94). The predicted binding affinities (XP scores) to the 2 sites are 210.63 and 211.66, respectively.

amotosalen AMT 8-MOP
FIG 3 Chemical structures of amotosalen, AMT, and 8-MOP. These related psoralens share a core planar structure with few rotatable bounds. However, amotosalen and 4'-aminomethyltrioxsalen (AMT) also contain primary amino groups, which when combined with molecular planarity and small numbers of rotatable bonds, are associated with enhanced penetrance into Gram-negative pathogens (57). Therefore, structural properties appear to explain the observed enhanced activity of amotosalen and AMT, compared with 8-MOP.
Therefore, it is perhaps not surprising that we found relative resistance to amotosalen among MDR isolates where a significant fraction had MICs exceeding the concentration of amotosalen used in the Intercept pathogen inactivation procedure. Although we cannot, without extensive genetic and expression analysis, definitively conclude that higher MIC values across this strain set were due to efflux activity alone, it is reasonable to hypothesize that efflux was a major contributor.
It was of additional interest that we identified the addition of SPS as a method to potentially address growth inhibition during MIC testing with serum, an issue in the field (48). Clearly, these findings should be replicated with additional pathogens to determine its broad applicability; however, we do note the use of SPS as standard additive in blood culture medium that appears to alleviate growth inhibition by human blood (49). Oddly, human serum albumin, but not serum, significantly increased the amotosalen MIC. This potentially may indicate that the binding site for amotosalen in the serum tested was not accessible.
Interestingly, a minority of Burkholderia spp. isolates and Pseudomonas aeruginosa were either always or variably killed by UVA light in the absence of amotosalen exposure (Table 1 and data not shown). Based on prior literature, we speculate this may result from free radical generation during UVA excitation of bacterial fluorescent pigments and/or endogenous photosensitizers (74,75).
Though generally effective against most pathogens, psoralens are ineffective against non-enveloped viruses (e.g., HAV, HEV, parvovirus B19, and poliovirus) and relatively imporous bacterial spores (76)(77)(78). Taken together, our data now raise the possibility that contemporary multidrug-resistant bacterial isolates have reduced susceptibility to psoralen inactivation based on efflux. Collectively, our data also suggest the need for further study of psoralen-efflux pump interaction, and that future chemical optimization of pathogen inactivating compounds should specifically explore Gramnegative penetrance in the presence of efflux pumps.
Our study has several limitations. Notably, we did not employ the INTERCEPT Illuminator for UVA exposure. We tested inactivation of pathogens in standard antimicrobial susceptibility testing medium, and in lid-less microwell plates with a very short path length for UVA exposure, not in blood products contained in bags with potentially greater UVA opacity, and therefore, to compensate, used a slightly lower 2J/cm 2 dose of UVA compared to the Intercept 3J/cm 2 dose. As such, our results may differ from inactivation achieved in the INTERCEPT system. Also, our results should be taken in context. MDR Gram-negative pathogens, thus far, are relatively rare causes of transfusion associated sepsis. Furthermore, even partial killing might be sufficient to prevent a sepsis event, even if not demonstrating full susceptibility in a presumably more stringent MIC test. Finally, pathogen inactivation strategies have provided significant benefit in reducing the overall frequency of transfusion-associated bloodstream infection (79).
Nevertheless, emerging antimicrobial resistance, including resistance associated with efflux mechanisms, is becoming increasingly common. The failure of pathogen reduction to completely eliminate contaminating pathogens depending on bacterial load has been recently reviewed (80), and we suggest that these observations may in part be related to the ability of some pathogens to efflux inactivation agents. Our findings serve as an alert to a potential vulnerability in pathogen inactivation methods that may explain some instances of pathogen inactivation breakthrough and should be an area of further research.
Bacterial strains. Clinical bacterial isolates are listed in Table S1, and were obtained from the American Type Culture Collection (ATCC), the CDC-FDA Antimicrobial Resistance Isolate Bank (ARIB), Walter Reed Army Institute of Research (WRAIR), and BEI Resources. The Keio strain, BW25113, and isogenic, JW5503-KanS DtolC E. coli were obtained from the Coli Genetics Stock Center (Yale University, New Haven, CT) (81). E. coli AG100AX DacrAB DacrEF (82) was from Ed Yu (Case Western University, Cleveland, OH). We gratefully acknowledge receipt of A. baumannii AB-7075 and P. aeruginosa PAO1 transposon mutants from the Manoil Laboratory (University of Washington, Seattle, WA).
Creation of isogenic AdeABC efflux strains. Vectors for regulated expression of adeAB and adeC were created as follows: To create the isopropyl b-D-1-thiogalactopyranoside (IPTG)-inducible, pAdeC vector, pBMTL-2 (83) was first converted to pBMTL-2NTC by replacing the kanamycin resistance gene with a nourseothricin acetyltransferase resistance gene. Specifically, pBMTL-2 was amplified by PCR using "F pLAC (NAT)" and "R pLAC (Nat)" primers (Table S2), and the nourseothricin resistance gene was amplified from plasmid pMOD3-mNeptune2-nat (84) (Addgene #120335) using primers "F Nat" and "R Nat" with inclusion of 59 tails encoding overlap between vector and nourseothricin amplicons, respectively. pBMTL-2 was a gift from Ryan Gill (Addgene #22812). All amplification reactions were performed using Q5 high-fidelity DNA polymerase (New England Biolabs), followed by DpnI digestion, column-purification (Qiaquick PCR Purification kit, Qiagen) assembly using the HiFi reaction kit (New England Biolabs). Transformants were selected on 50 mg/mL nourseothricin. The adeC gene from A. baumannii strain AYE (ATCC BAA-1710) was then similarly amplified from genomic DNA using primers "F AdeC" and "R AdeC," and cloned downstream from the vector pLac site and Shine-Delgarno sequence in pBMTL-2NTC using vector primers, "F pLAC (AdeC)" and "R pLAC (AdeC)," with the new vector, again, assembled using HiFi as described above.
MIC determination. For MIC determination, bacterial stocks frozen at 280°C were streaked onto tryptic soy agar plates containing 5% sheep blood (Remel), and grown overnight at 35°C in ambient air. Colonies were then suspended in 0.9% saline to 0.5 McFarland, measured using a Densicheck (bioMérieux); diluted 1:300 in cation-adjusted Mueller-Hinton broth (BD); and dispensed into 384-well polystyrene plates (Greiner Bio-One) at 50 mL per well with the following exceptions: For MIC determinations with AG100AX strains containing pAdeC and pAdeAB plasmids, 280°C frozen stocks were inoculated directly into non-cation-adjusted Mueller-Hinton broth containing 100 mg/mL ampicillin, 50 mg/mL nourseothricin, 5 mM calcium chloride, and 5 mM magnesium chloride, with or without adeABC induction using 1% L-arabinose and 1.0 mM Isopropyl b-D-1-thiogalactopyranoside (IPTG), and grown overnight at 37°C in 15 mL conical tubes with continual rotation. Bacterial cultures were then adjusted to 0.5 McFarland and diluted 1:300 in the same medium, and dispensed as above. Induction of other cloned efflux pumps was with 1% L-arabinose (or 0.5% L-arabinose for cloned P. aeruginosa pumps) in the absence of supplemental divalent cations. Lastly, P. aeruginosa PAO1 strains were dispensed into 96-well plates to avoid skipped-well effects observed in 384-well plate format.
Filled microplates were centrifuged at 1250 rcf for 4 min to ensure the entire inoculum was in continuity with the bottom of the well. Then, 2-fold doubling dilutions of stock solutions of amotosalen, AMT, or 8-MOP supplemented with 0.3% Tween 20 (Sigma-Aldrich) were dispensed into microwells using the HP D300 digital dispensing system (HP, Inc.), as previously described (87)(88)(89)(90), and the microplates were mixed for 5 min on a microplate shaker to ensure complete mixing of psoralens with the inoculum. Microplates were then exposed to UV light in a UV Stratalinker 1800 (Stratagene), retrofitted with UV BL F8T5 CFL 12-inch UVA 365 nm Blacklight Bulbs (Coolspider), and calibrated according to manufacturer's instructions with an UVA365 UV Light Meter (Amtast) for a total exposure of 2 Joules/cm 2 of UVA, to approximate exposure during platelet pathogen inactivation (10). Testing of control antibiotics was performed in an identical fashion, however, without UVA exposure.
After incubation at 35°C for 20 h, the A 600 was measured using a TECAN M1000 microplate reader. MICs were determined based on a growth inhibition A 600 cutoff of 0.06, which is consistent with our prior determinations of Clinical and Laboratory Standards Institute reference standard equivalent, MIC determinations in 384-well plate format (88,90,91). Notably, the AdeABC pump was inactive in E. coli AG100X in Mueller-Hinton broth without the addition of divalent cations, consistent with prior use of magnesium supplementation when expressing this A. baumannii efflux pump in E. coli (32). The concentrations of MgCl 2 and CaCl 2 used were empirically determined previously to be optimal for efflux of minocycline and ethidium bromide (data not shown).
AdeB binding affinity and molecular docking. His-tagged AdeB protein was purified as described previously (30). Fluorescence polarization assays were performed in a ligand binding solution consisting of 20 mM HEPES-NaOH pH 7.5 and 0.05% n-dodecyl-b-d-maltoside (DDM) and 3 mM amotosalen. The experiments were done by titrating the AdeB protein in solution containing 20 mM HEPES-NaOH pH 7.5 and 0.05% DDM into the ligand binding solution while keeping DDM concentration constant.
Fluorescent polarization was measured at 25°C using a PerkinElmer LS55 spectrofluorometer, coupled with a Hamamatsu R928 photomultiplier. The excitation wavelength for amotosalen was 350 nm and the fluorescent polarization signal (DP) was measured at 470 nm. Titration data points represent 15 measurements, and 3 biological replicates were performed to determine the K D as previously described (30). ORIGIN Version 7.5 (OriginLab Corp.) was used for curve fitting.
Molecular docking. A structure of the "binding" protomer of AdeB-Et-I (PDB ID: 7KGH) was used as the template, in which the bound Et ligands were removed from the protomer (55). Induced-fit-docking was performed using Extra Precision Glide (56) with default parameters in the Protein Preparation Wizard Module of Maestro (Release 2019-3) (Schrödinger) to predict binding modes of amotosalen, AMT, and 8-MOP to AdeB. For each calculation, residues within 5 Å of the bound ligand were selected for side chain optimization using Prime refinement. The docking results with the lowest XP scores were selected as predicted poses.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. The content is solely the responsibility of the authors and does not necessarily represent the view of the National Institutes of Health, the Department of Health and Human Services, or the U.S. Federal Government.
The HP D300 digital dispenser and TECAN M1000 used in the experiments were provided by TECAN (Morrisville, NC). TECAN had no role in the study design, data collection/interpretation, manuscript preparation, or decision to publish.