Understanding, inhibiting, and engineering membrane transporters with high-throughput mutational screens

Abstract

Promiscuous membrane transporters play vital roles across domains of life, mediating the uptake and efflux of structurally and chemically diverse substrates. Although many transporter structures have been solved, the fundamental rules of polyspecific transport remain inscrutable. In recent years, high-throughput genetic screens have solidified as powerful tools for comprehensive, unbiased measurements of variant function and hypothesis generation, but have had infrequent application and limited impact in the transporter field. In this primer, we describe the principles of high-throughput screening methods available for studying polyspecific transporters and comment on the necessity and potential of high-throughput methods for deciphering these transporters in particular. We present several screening approaches which could provide a fundamental understanding of the molecular basis of function and promiscuity in transporters. We further posit how this knowledge can be leveraged to design inhibitors that combat multidrug resistance and engineer transporters as needed tools for synthetic biology and biotechnology applications.

Introduction

The cell membrane controls the flow of energy, information, and material between the cytoplasm and the external environment. Membrane transporters mediate the movement of solutes across the lipid bilayer, either by facilitating diffusion of substrates down their concentration gradients (uniport) or by catalyzing concentrative transport by coupling solute movement to an energy source, such as electrochemical gradients or ATP. Many transporters are polyspecific, meaning that a single protein can transport a broad range of substrates with varying chemical structures and physical properties. Crucially, these are not indiscriminate non-specific transporters, but selectively promiscuous proteins which act on a defined set of substrates, even though they may greatly differ in structure. Polyspecific transporters may be uniporters, ion-coupled transporters, or ATP-driven transporters and are found in mammalian and microbial systems alike (Fig 1A). Substrates of polyspecific transporters include a variety of drug-like hydrophobic organic molecules, which bind flexibly within large hydrophobic binding sites. Binding often occurs with very dissimilar binding poses (Fig 1B) or even separate binding sites for different substrates^1-3. This degeneracy has clouded a deep understanding of the specific interactions that govern specificity for one molecule or another, and has made polyspecific transport challenging to therapeutically target.

Figure 1. Diversity and flexibility of polyspecific transporters.

A) Classes of transporters and examples of polyspecific transport involved in uptake and efflux of diverse metabolites and xenobiotics in microbial and mammalian systems. B) Five substrates of the human hepatic uptake transporter OCT1 superimposed in the binding site to illustrate the flexibility of polyspecific transporter ligand recognition. Substrates are shown as sticks and proximal residues are shown as lines. Ligand poses are from cryo-EM structures of OCT1^136,137: 8SC2 (diltiazem), 8SC3 (fenoterol), 8SC4 (metformin), 8SC6 (thiamine), and 8JU0 (spironolactone).

Due to their ubiquity in cellular life and the bioactive nature of their substrates, it is not surprising that promiscuous transport plays central roles across human health and biotechnology. In multidrug resistance, drug efflux pumps may export antibiotics and cancer therapeutics^4,5; in pharmacology, uniporters can mediate the uptake and disposition of drugs^6-8; and in industrial biosynthesis, transporters can control the flow of reactants and products or confer resistance to toxic impurities^9-11. Through extensive structural, genetic, computational, and biochemical efforts, some general models for transporter function have been established^12-14, and some are now being translated into therapeutics^15,16. Despite this, key questions in transporter biology persist: How is specificity encoded in sequence and structure? What is the mechanistic basis of polyspecificity? How can we create specific inhibitors of transporters? Can we design entirely new transporters with novel functions? These questions call for large-scale, unbiased methods to broadly probe sequence-function relationships. High-throughput screens have proven powerful tools to provide just this type of insight, and have already transformed our understanding of virology^17,18, signal transduction^19,20, drug development^21,22, enzymology^23,24, and more. Despite this, they have not been widely applied to study transporters.

In this primer, we provide an overview of available high-throughput methods for transporters and argue that these are necessary and particularly effective for addressing the open questions in transporter biology, especially polyspecificity and drug resistance. To encourage these approaches to be more broadly adopted, we describe several possible approaches and contrast them with prior work in the membrane field, including a discussion of prior screening efforts in transporters. We then discuss how our understanding of polyspecificity will benefit from large screens with an emphasis on health-related systems, and how we may translate this insight into specific inhibitors. We conclude by discussing broader impacts for biotechnology, drug design, and synthetic biology, and highlight the major gaps that must be addressed to unlock the full potential of high-throughput screens for the transporter field.

Applying high-throughput methods to transporters: advantages and current limitations

Biochemistry in the membrane.

Traditional biochemical approaches for membrane proteins have shaped our current understanding of transporter function. Electrophysiology, radioactive uptake assays, and equilibrium binding experiments can provide quantitative biophysical insight into transporter energetics, kinetics, and specificity. However, these are also technically demanding and low-throughput methods that often require purification and reconstitution of the target transporter. Thus, only a limited number of mutants can be feasibly assayed with these approaches, requiring focus on defined regions of the protein. Previous works have rightfully concentrated on important regions like substrate binding pocket^25,26 and have proven extremely useful in understanding how promiscuous transporters interact with substrate directly. Other foundational works have applied scanning mutagenesis in vivo, as with the cysteine scans of LacY²⁷, SERT²⁸, P-gp²⁹, EmrE³⁰, and others. In the absence of solved structures, these provided crucial structural insight for mechanistic modeling^28,31,32. However, early methods of variant construction and testing were also limited by throughput, requiring focus on individual mutations (usually to cysteine or alanine). More in-depth mutagenesis was limited to a few dozen positions at most, often an individual transmembrane helix or sequence motif. This technical requirement of focusing on isolated regions risks overlooking the multidomain complexity of transporter function, including the well-documented impact of distal mutations on recognition, gating, and regulation of transport^33-38.

Modern high-throughput screens have alleviated these restrictions, revolutionizing protein science by enabling comprehensive, unbiased, and high-content datasets³⁹. Pooled screens can evaluate thousands to millions of variants simultaneously by linking function to abundance under selection pressure, measured using deep sequencing (Fig 2A). Since each variant is an independent hypothesis of protein function, this approach allows us to build a comprehensive mutation-function landscape of a target – essentially a molecular “blueprint” of the protein. These advantages of scale, flexibility, and lack of bias are especially important when studying functions that emerge from complex, spatially distributed mechanisms not constrained to an individual protein domain. This is often true of transporters, especially promiscuous ones, which combine substrate recognition, energetic coupling, local gating rearrangements, and global conformational switching to achieve biologically relevant transport^40-43. Despite these advantages of modern screening methods, only a handful of applications to transporters have been reported^44-47.

Figure 2. High throughput screening tools for studying polyspecific transport.

A) General principle of high-throughput screens described herein. An initial distribution of transporter variants is subjected to selective pressure based on growth (dis)advantage or fluorescence, and resulting abundance changes are measured using next generation sequencing. B) Types of transporter variant libraries that can be screened. C) Screening methods that can be used to select for various transporter functions. Further development is particularly needed for screening conformational dynamics, coupling efficiency, and transport of nontoxic/nonessential substrates.

Designing high-throughput screens for transporters.

High-throughput pooled screens map genotype to phenotype at scale by linking changes in function to changes in abundance. After building a large library of genetic variants – an increasingly facile task with advancements in bulk DNA synthesis – the library is transformed into a pooled culture of cells such that each cell expresses a different protein variant. Next, variants are enriched or depleted according to their function. This can be achieved by leveraging proliferation differences during competitive growth, or by selectively sorting cells with flow cytometry. The differences in abundance before and after applying this selection can be measured using deep sequencing to count the relative frequencies of each variant. The design of the variant library (which genetic variants are present) and the selection scheme (how the function of interest is linked to abundance) are the foundations of high-throughput screens, and each requires careful attention.

Variant library design.

Library design determines how the experiment samples sequence space. Many types of libraries exist, each addressing different questions of transporter function (Fig. 2B). The most common are deep mutational scanning (DMS) libraries, which contain all single amino acid changes to a reference wildtype sequence⁴⁸. These are comprehensive and unbiased libraries that provide a high-resolution view of sequence-function landscapes, identifying trends in the location or character of mutations that impact function (e.g., transport). DMS libraries are straightforward to design, construct, and interpret, and are appropriate for a wide variety of questions. Because they only contain single-mutants, however, they cannot measure non-additive epistatic interactions between multiple mutations. Combinatorial libraries, which contain multi-mutant sequences, are better suited for this^49,50. Because they grow exponentially, comprehensive combinatorial libraries are usually too large for current technologies, but rational design of smaller combinatorial screens can still be highly informative. For example, complete mutational trajectories between sequences differing by ~20 mutations can be studied with current techniques, offering systematic insight into adaptive change across alleles or homologs⁵¹. Combinatorial libraries can also be targeted to specific regions for deeper insight into a known functional hotspot.

In contrast to mutational screens, which examine variation around a single point in sequence space, other libraries include more disparate sequences (Fig. 2C). One example is ancestral reconstruction libraries, which contain computationally reconstructed ancestors of a protein at various evolutionary distances⁵². Ancestral libraries address questions about how properties emerge and change over evolutionary time. This would be especially informative in polyspecific transporters, which often show opportunistic activity on non-natural molecules which did not exist for most of evolutionary history⁵³. Metagenomic libraries take an even broader approach, exploring functional differences across species or gene families. These could pinpoint interacting transporters within microbial ecosystems or identify natural pumps with useful properties for biotechnology⁵⁴. Lastly, libraries of engineered proteins can be screened to extract successful design candidates. This is an essential step of the protein design process, which can produce practically useful tools for synthetic biology and evaluate our understanding of rules for transporter design. While massive library sizes are feasible, the potential sequence space to be explored by these approaches can still exceed DNA synthesis and sequencing capacity. Continued development of focused library designs within these limits – guided by prior knowledge and specific hypotheses – will accelerate and deepen the investigations of transporter function discussed herein.

Linking function to abundance.

The second key element of high-throughput pooled screens is a selection, which links a phenotype of interest to abundance of the causal genotype (Fig. 2D). Competitive growth is a simple selection method which may directly reflect the relevant physiology: fitness change due to transport. By passaging libraries with a toxin or nutrient whose transport is mediated by the protein of interest, these screens assess variants’ ability to confer resistance or improve nutrient uptake. Additional complexity in competitive growth selections can address finer details of polyspecific transport. For example, selections can be performed in multiple physiologically relevant environments or host backgrounds to clarify the interaction of sequence variation with external factors like pH⁵⁵, lipid composition^56-58, and presence of other host transporters⁵⁹. Selections can also use multiple substrates concurrently (combined or sequentially) to study the phenomenon of sacrificing high activity on individual substrates in favor of lower activity on a wider range of molecules. Understanding the molecular and evolutionary basis of such a tradeoff could explain the prevalence of polyspecificity in transporters, even though most proteins evolve towards specialized functions⁶⁰. Because many open questions concern substrate selectivity, which is broad by definition, introducing small molecule diversity is an essential aspect of designing transporter screens. Arrayed chemical libraries offer an uncomplicated way of parallelizing existing selection protocols. This approach would establish a thorough specificity profile for all library members and connect sequence variation to specific chemical and structural features of substrates. Selection for phenotypes that are not easily linked to cell viability remains a problem, such as transport of substrates that are neither toxic nor essential. Fluorescence-activated cell sorting (FACS) can be a useful selection tool in these cases, if the substrate is fluorescent or if a biosensor is available for the substrate of interest⁶¹, but there is ample room for more innovative selection schemes in this area.

Extracting mechanistic detail from pooled screens.

Additional screens can complement growth-based selections to better understand the mechanistic details behind sequence-function relationships. Ideally, changes in abundance should be due to variation in the specific function of interest, and nothing else – but this is rarely the case. While the phenotype examined by these selections is usually the most relevant (the total effective fitness change from transporter expression), this could be due to changes in abundance, aggregation, pleiotropy⁶², or a completely orthogonal function⁶³. Furthermore, changes in overall transport are a sum of changes to multiple processes necessary for effective efflux – ligand binding, energy coupling, conformational changes, etc. Nonetheless, well-designed multiparametric screens can deconvolute a single readout (e.g., differential growth) into mechanistic elements by measuring multiple phenotypes⁶⁴. Competitive growth in the absence of substrate, apart from being an essential control, can measure fitness effects from variant expression alone. This could be due to energy loss from unproductive transport cycles, a phenomenon common in promiscuous transporters^14,65 and an area of significant interest for transporter biophysics⁴⁰. These energetic phenotypes may be amplified by chemically disrupting the proton motive force⁶⁶ or ATP production⁶⁷. Using FACS to screen libraries with split-GFP or FRET-based tags can measure abundance, folding, membrane insertion, and trafficking – key elements of transporter fitness⁶⁸. In concert with the growth-based screens above, these measurements can begin to describe changes of overall transport in terms of basic properties.

Analyzing the large datasets from pooled screens to interpret the underlying biology is a last major challenge. While current pooled screens generally involve one protein and one substrate, studies of polyspecific transporters should include many substrates, increasing the dimensionality of the data. To this end, statistical approaches such as unsupervised hierarchical clustering and dimensionality reduction can be used to identify groups of mutations with consistent effects on substrate-specific functions. More recently, machine learning (ML) algorithms have been shown to effectively learn sequence-function relationships from high-throughput screening data⁶⁹. In combination with generic zero-shot protein language models that predict stability and structural properties from protein sequences^70,71, task-specific ML models trained on high-throughput screening data can guide design of specific transporter functions – a desired substrate range, for example. The models themselves can also be studied using feature analysis methods to determine which sequence and structural features are most relevant for the function in question, extracting basic molecular rules from the algorithm’s training⁷². Further research is needed to identify the most efficient methods for generating high-content data with ML training in mind. Another area of interest is understanding whether ML models can learn substrate structural features from sequence-function data and extrapolate to unseen ligands, and what the training data requirements are to do so. We expect that advances in library size and screening types will also drive improvements in deep learning-based analysis methods.

Understanding the molecular basis of polyspecificity

Protein-ligand interactions are generally understood in terms of carefully tuned electrostatic and steric interactions that confer precise specificity for a ligand or a class of structurally similar ligands. Soluble enzymes and ion channels generally fit this description, but many membrane transporters do not. Both efflux and uptake transporters commonly exhibit broad and complex specificity profiles, acting on several dissimilar molecules while maintaining selectivity for a defined range of substrates. The sequence and structural bases of this broad yet selective specificity are difficult to characterize, as it is often the result of decentralized, dynamic, or allosteric interactions.

Advancements in cryo-electron microscopy (cryo-EM) of membrane proteins have generated many high-quality structures of promiscuous transporters and shed some light on where the determinants of specificity might lie. Structures show large flexible binding pockets containing mostly hydrophobic and aromatic residues, which coordinate nonspecific hydrophobic interactions with diverse ligands^73,74. Occasional hydrophilic or charged residues make more specific contacts, likely playing a role in constraining substrate range^75-79. These findings have clarified many aspects of promiscuous recognition and structures continue to serve as an essential reference, but structural information alone is not adequate to explain the molecular basis of function, as evidenced by the importance of mutations outside the binding site for rationally engineering specificity⁸⁰. Static structure cannot describe the huge conformational changes of alternating-access transporters, and often fail to resolve unstable transition states or disordered regions⁸¹. Additionally, the environmental conditions of structure determination are far from physiological, often containing disruptive crystallization chaperones or cryo-EM fiducial markers and ignoring the electrical, lipid, and pH conditions of membranes in vivo, which may be essential for proper function^55,57. Crucially, the determinants of polyspecificity are not solely structural. Protein and environmental energetics, kinetics, and the rate and magnitude of conformational rearrangements all vary depending on substrate, and therefore may play a role in determining substrate range^40,41,82. While hydrophobic binding sites may explain the molecular determinants of ligand binding, the features enabling overall polyspecific transport are likely decentralized.

Small-scale mutational studies focus on specific residues or motifs, typically in the binding site, which limits their ability to understand substrate recognition comprehensively. Several specificity-driving residues were identified in a distal unstructured loop of the model E. coli efflux pump EmrE using a scanning mutagenesis technique that substituted all residues to alanine, valine, or glycine³⁴. Deep mutagenesis – substitution to all twenty amino acids – was performed on the transmembrane cavity of the ATP-driven efflux pump EfrCD of E. faecalis, revealing altered-specificity phenotypes such as enhanced ethidium efflux by mutants with additional negative charge at the binding site, supporting the role of charge-charge interaction in export of cationic substrates⁴⁷. At the time of writing, full-length deep mutagenesis has been published for only one promiscuous transporter, the human hepatic drug importer OCT1 (SLC22A1). The comprehensive mutagenesis performed here fully mapped the sequence determinants of function for a single substrate. A complementary abundance screen revealed a segmented organization of abundance- and functiondetermining residues across the sequence, demonstrating the utility of full-length mutational scanning and setting the stage for broader multi-substrate screens⁴⁶.

Complete scanning mutagenesis in the context of several dissimilar substrates could fully map non-intuitive effects on specificity and begin to describe the relationship between transporter sequence and the chemical structure of its substrates (Fig. 3A). Selections could also incorporate combinations of substrates to identify substrate-independent functional hotspots and deepen our understanding of how polyspecific transporters balance activity on multiple ligands. Screening libraries for activity on molecules not normally transported could identify rare gain-of-function mutations that expand specificity to include a new substrate. This would not only provide further insight into the functional landscape of polyspecificity, but also help to anticipate the acquisition or resistance to antibiotics and antineoplastics in drug-rich clinical settings. Selections that target specific aspects of transporter function – rather than a summative growth-based readout – will be especially informative. In addition to understanding the molecular bases of these functions in their own right, these studies can evaluate their roles in tuning substrate range. For example, split-GFP based membrane insertion screens⁶⁸, when run in parallel with functional assays, can distinguish altered expression and trafficking from genuine functional differences. Energetically driven phenotypes may be amplified by disrupting a transporter’s energy source, clarifying the hypothesized connection between energy coupling and specificity^65,83. In the case of proton-coupled transporters, this could be achieved using pH stress or decoupling ionophores valinomycin and nigericin. ATP-driven transporters could be targeted in a similar way using chemical inhibition or knockdown of ATP synthase. Alternating-access rates⁴¹, gating mechanisms⁴², and lipid environment^56-58 may also impact specificity, but these biophysical properties are challenging to independently measure in high-throughput with current technologies. In vitro translation of variant libraries directly into liposomes may facilitate bridging existing biophysical assays for these functions to high-throughput selection schemes⁸⁴.

Figure 3. Applications of high-throughput investigation for polyspecific transport.

A) High throughput functional screens can reveal the sequence and structural bases of polyspecificity and identify mutational trends that produce predictable changes in specificity profile. B) Peptide inhibitor screens can identify candidate peptide-based efflux pump inhibitors, which can be optimized for stability and bioavailability with computer assisted drug design methods. C) Engineered transporters, which are both informed by and evaluated with high-throughput methods, offer significant advantages in industrial bioprocessing.

Designing clinically useful efflux pump inhibitors

Many promiscuous transporters mediate efflux of drug-like molecules, resulting in multidrug resistance (MDR)^4,5. Efflux-mediated MDR in cancers and microbial pathogens pose significant challenges for human health, and there is great interest in developing efflux pump inhibitors (EPIs) to maintain the efficacy of chemotherapies and antibiotic treatments in the face of resistance. EPIs are theoretically very promising, having been shown to restore drug sensitivity to MDR organisms and cancers in lab settings^85,86. However, while numerous EPIs have been identified and pursued, a variety of factors have led all to fail clinically.

Small molecule inhibitors of promiscuous efflux pumps are surprisingly easy to come by through traditional screening-based methods of drug discovery. Over 40 years ago, early generation EPIs reserpine⁸⁷ and verapamil⁸⁸ were discovered by testing panels of membrane-interacting molecules. Since then, chemical library screens have identified numerous potent broad-spectrum EPIs such as PAβN⁸⁹ and NMP⁹⁰. Virtual screening has scaled this process significantly, generating several promising EPIs in recent years^91-93. While these serendipitous EPIs have been invaluable for studying efflux mechanisms in the laboratory, toxicity and strong pharmacological effects have limited their usefulness in the clinic^86,94-97. These are overwhelmingly competitive inhibitors which behave like substrates^98,99, associating with binding sites that have evolved for flexibility. Therefore, while broad-spectrum EPIs are readily available, they frequently inhibit essential human transporters in addition to the intended target¹⁰⁰. To address this shortcoming, drug development should focus on synthetic inhibitors designed to precisely target a single transporter or small group of pathogenic transporters.

Peptide-based drugs may pose a solution to the problem of EPI selectivity. Several efflux-inhibitory peptides have been rationally designed to disrupt transporter oligomerization^101-103, demonstrating the potential of peptide drugs to reverse MDR through selective mechanisms beyond competition for the substrate binding site. Because they can be genetically encoded, peptides can be screened in high-throughput and are evolvable to make precise interactions with high affinity. In one example of this, phage display was used to evolve a transport-inhibitory antibody fragment, and a peptide mimicking the CDR loop reversed efflux-mediated MDR⁷⁷.

Candidate EPIs can be screened in high throughput using a library of peptide variants tethered to the membrane. This approach has been used for antimicrobial peptide (AMP) discovery¹⁰⁴ but may be adapted to identify EPIs by screening in the context of a target transporter. Cells expressing EPI peptides will self-sensitize and drop out of the population after selection with a toxic substrate. While random peptide libraries have successfully identified AMPs, this approach could be refined for specific targets using de novo protein design methods, which have proven quite reliable in producing high-affinity binders for a target protein¹⁰⁵. Cell-free screens can also select peptide libraries using mRNA display and in vitro translation systems, and this approach has yielded peptide-based inhibitors for multiple drug efflux pumps^106,107.

Several approaches exist to mitigate peptides’ generally poor drug properties: cyclization or introduction of hydrocarbon “staples” can improve stability, or computational peptidomimetic methods can extract pharmacophores from peptides to create traditional small molecule drugs with similar bioactivity¹⁰⁸. Another approach is to intentionally select for peptides which are likely to be active in physiological conditions, as with one AMP screen which was performed directly in human serum¹⁰⁹. By incorporating these techniques into a peptide-based drug development pipeline, EPIs that are potent, selective, and bioavailable are well within reach (Fig. 3B).

Prospects for high-throughput methods in industry and beyond: from understanding to engineering transporters

The task of basic science is to understand the world; here, how nature has designed transporters to perform the wide variety of tasks they do. A deep understanding should allow us to create something new. In this way, engineering and applied methods are united. Biotechnology, molecular therapeutics, and synthetic biology will all benefit from the application of high-throughput methods with the ultimate goal of enabling transporter design. Here we discuss the near-term prospects for applying high-throughput genetic screens in these fields, concluding with long-term prospects for protein design and engineering.

In the transporter context, engineering has historically focused on improving bioprocessing yields through metabolic engineering by enhancing import of precursors, export of toxic intermediates, and secretion of final products (Fig. 3C)^110,111. Current experimental approaches to optimize transporters tend to find marginal improvements⁵⁴, often due to the particular sensitivity of membrane proteins to overexpression and expression in heterologous hosts¹¹². Directed evolution, a popular approach to metabolic engineering, tends to first modify gene expression rather than specific functions, which likely also contributes¹¹³.

Broader, unbiased, mechanistic screens are a more appropriate tool for this task and should be tightly integrated into iterative metabolic engineering design-build-test-learn cycles¹¹⁴. These should focus on two aspects: first, understanding the basis for polyspecific recognition and export of compounds by endogenous transporters; second, understanding the mechanisms limiting heterologous expression of transporters. This will require improvements to inference models to learn the essential features from each screen as well as improved library design and generation techniques to allow updating designs for maximal inference, which are active areas of study. The ability to robustly introduce entire biosynthetic pathways producing arbitrary chemical diversity would transform human health and industry. Although this goal remains distant, its potential impact cannot be overstated, and high-throughput tools are essential to reaching it.

The vast majority of therapeutic compounds reach their targets with the help of transporters, and efficacy and resistance are strongly influenced by transporter genetic variation. A large number of human disorders are linked to transporter defects as well, and this variation is often due to non-enzymatic effects, such as changes to targeting or expression¹¹⁵. In CFTR, the transporter where mutations cause cystic fibrosis, variants acting on both channel activity and misfolding are known, and specific therapeutics against these paths are poised to make major clinical impacts¹¹⁶. This suggests that mechanistic high-throughput screening will be a general way to identify the molecular basis for transporter defects and guide the direction of therapeutic design^46,64. Genetic diseases of membrane proteins often involve broader network effects, and also physiological adaptations, that resist simple fixes¹¹⁷. A mechanistic approach, possibly involving measuring perturbations across entire genetic networks, may be necessary to guide gene therapies.

Synthetic biology is particularly poised to benefit from high-throughput transporter methods. The field has long adopted engineering frameworks and large screens, but incorporating transporter screens is increasingly important. Major goals, including the design of new biosensors, circuits, and creating synthetic microbial consortia all require a deep understanding of transporter function only available by leveraging high-throughput methods.

Some cell-based biosensors have been produced using natural transporters through laborious screening and optimization¹¹⁸. With a generative engineering framework, this approach could be formalized into a powerful broad platform. For cell-free biosensors encapsulated in liposomes, the requirement for specific, controlled transport of analytes is even more essential. Cell-free liposomal systems have also been productively used as reductive models for cells and prebiotic life, but introducing transporters remains difficult¹¹⁹. Quantitative multi-phenotypic models could be of use for designing transporters with desired parameters, such as integration efficiency, kinetics, or specificity.

Beyond the reconstitution of existing biology, efforts to generate entirely new features, such as expanded genetic codes or new metabolic pathways, will require engineering transporter function. The novelty of the biology often implies that associated pathways for uptake are nonexistent. Modifications to existing metabolite transporters can address this: for example, relatively minor changes to an ABC transporter accessory protein were sufficient for 5-fold increase of unnatural amino acid uptake in E. coli¹²⁰.

Microbiome analysis and engineering is a major area of current research. A major theme is the importance of communication and distributed interactions. The role of human microbiomes in mediating the effects of diets and drugs, as well as neurological disorders, is another surprising discovery^121-123. Both of these have major roles for transporters, through the uptake and efflux of chemicals, the exchange of metabolites and quorum sensing molecules, and the determination of microbial composition^124,125. Given the infancy of the field, high-throughput screens are poised to make major advances. Understanding the chemical specificities of microbial transporters at scale will be necessary to model community interactions. Predicting microbiome/drug interactions will also require methods to predict transporter range and function for compounds and metabolites. Ultimately, controlling the composition of microbiomes or designing entirely new synthetic consortia will require knowledge of the broad biophysical and biochemical properties of the transporters enabling interactions.

For all of these fields, the ultimate goal is the rational design of a system with user-defined properties and function. Protein engineering and design have made major advances with an eye towards this, but still have much distance to cover. Much of the advances have been through moving away from purely physics-based models to knowledge-based statistical potentials¹²⁶ and coevolutionary information^127,128, which has culminated in the transformer-based large language models trained on the known sequence universe¹²⁹. These methods have successfully designed pores, beta barrels, and multipass transmembrane proteins^130-132 (although clever knowledge-guided computational work made this possible earlier^133,134). Although the limits of these recently developed methods are still under investigation, we believe their full potential will require close integration with large-scale experiments. One reason for this is the diminishing interpretability as models become more complex with more parameters. Understanding how biological parameters are embedded in this high-dimension space is an open problem, and one where high-throughput experimentation will likely be important, especially for transporters, where polyspecific function and genetic redundancy make phenotypes less precise. The utility of this for integrating computation and experimentation for soluble proteins has been suggested by pioneering work, but the best implementations, as well as how to apply the framework to transporters, still remain open^69,135. Ultimately, closing the loop between experimentation to learn principles and computation to predict and guide experiment will be an essential method with revolutionary potential. To accomplish this, we believe the field should prioritize enabling efforts such as flexible library designs suitable for iteration, robust experiments suitable for automation, and better statistical models for experimental inference.

Outlook and conclusions

Despite the challenges in development and application of high-throughput methods, these approaches promise a wealth of deep insight into basic and applied biology and stand to address major challenges in fundamental protein biophysics, multidrug resistance, pharmacokinetics, industrial biotechnology, and synthetic biology. Although their motivations and approaches differ, these subfields can collectively leverage modern methodologies like deep mutational scanning, high-throughput functional profiling, and machine learning to drive progress. Systematic mutational library screens can provide fundamental understanding of functional promiscuity, while also serving as a powerful hypothesis-generating tool to nucleate focused biochemical and biophysical studies. Machine learning algorithms can then tap these large datasets to find meaningful patterns and create predictive models that guide drug discovery and transporter engineering. High-throughput phenotypic profiling assays may also be adapted to screen large libraries of genetically encoded inhibitors, allowing EPI development to scale dramatically.

Future work should focus on developing novel adaptations that address the current barriers of fully exploiting these approaches. Namely, methods of building larger and more diverse libraries that include multimutant and larger structural variation, as well as additional innovative selection schemes that target precise biophysical properties of transport in high-throughput. Despite these challenges, many of the assays described in this primer can and should be approached using existing technologies. Ultimately, with appropriate and extensive application of high-throughput tools to the field of transporter biology, we can rapidly build our understanding of fundamental biophysical principles and pave the way for new solutions to pressing clinical and industrial challenges.

Promiscuous transporters play critical roles across human health and biotechnology. Here, Miller et. al discuss the need for high-throughput methods to address open questions in the field. They outline high-throughput screening for transporters, proposing potential screens and advocating broader adoption of these powerful techniques to understand, inhibit, and engineer transporters.