Introduction

Rare earth elements (REEs) are critically important to the world economy on account of their unique luminescent, magnetic, and catalytic properties1,2,3. Comprising the fifteen elements of the lanthanide series, as well as scandium and yttrium, REEs have emerged as non-substitutable elements in a variety of consumer and defense applications, encompassing technologies as diverse as wind turbines, solar cells, high-capacity batteries, image-guided cancer therapies, telecommunications, and night-vision systems4,5,6,7,8,9.

Biomining presents a sustainable, potentially scalable path toward bolstering global lanthanide reserves via unconventional sources, though the absence of a system with appropriate dynamic range, especially in the environment, poses a significant barrier to practical biosensing and microbial engineering10,11,12,13. The lanthanide series elements are naturally luminescent and thus capable of ‘self-sensing’. However, on account of the Laporte selection rule, the radiative 4f–4f orbital transitions are spin-forbidden, and consequently, free lanthanide cations have a diminutive capacity for energy absorption14. To circumvent this photophysical constraint, lanthanide cations can be complexed with organic ligands to sensitize the lanthanide’s intrinsic luminescence via the well-studied antenna effect15,16,17. In other words, the organic ligand operates as an antenna, transducing energy to an otherwise sub-operative lanthanide beacon. Alternatively, lanthanides can themselves function as donors of energy transfer via overlapping and non-overlapping Förster resonance energy transfer (FRET) to closely positioned fluorescent protein acceptors18. Following the FRET model of energy transfer, molecular proximity gives rise to a highly efficient, distance-dependent energy exchange between donor and acceptor entities. Still, this system requires extensive tempering to bring the lanthanide chelate within the FRET signaling range, thereby limiting its use in vivo or at scale, but highlighting a promising avenue for biosensor development.

An attestation to the supreme adaptability of biology, the lanthanide-binding protein, lanmodulin (LanM), from the methylotrophic bacterium Methylobacterium extorquens has a sensitivity for lanthanides in the picomolar range19. Natively a periplasmic protein with a role in lanthanide utilization, LanM has been repurposed by biotechnologists for nascent applications in the sensing, separation, and sequestration of lanthanides20,21,22. For instance, the LaMP1 fluorescence-based sensor is a tri-fusion protein composed of a citrine and cyan fluorescent protein separated by the intrinsically disordered LanM domain20. Upon binding to the lanthanide series cations, LanM folds into a tight configuration, bringing the two adjacent fluorescent proteins within signaling range to initiate a measurable FRET. Alternatively, the naturally fluorescent amino acid tryptophan can be harnessed as a molecular antenna to sensitize the luminescence of trivalent terbium, yielding an apparent Kd of 10–100 pM23. Notably, this strategy employs sensitized lanthanide luminescence, instead of FRET; nonetheless, only terbium can be detected, and the energy transfer is inefficient due to tryptophan’s low fluorescence quantum yield24. Although LanM confers an ultra-high sensitivity to the fluorescence-based detection of lanthanides, its superlative nature could limit its utility as a biosensor in more concentrated environmental streams. Considering that acid mine drainage and other waste streams can contain an upwards of 20–100 µM of total REEs, a sensitivity range congruent with such conditions, whether achieved via LanM or some alternative approach, is needed for practical biosensing applications25,26,27. Moreover, the functionalization of microbes with the ability to concentrate lanthanides necessitates a platform capable of detecting multiple lanthanides in vivo at concentrations potentially spanning 5–5000 micrograms per kilogram of wet biomass or greater28. Recently, the capture of lanthanides on the surface of cells has yielded new insights, including, inadvertently, evidence that lanthanide cations can readily bind to the negatively charged side chains of aspartate and glutamate, found commonly in the extracellular matrix29. Based on the above, we hypothesized that protein charge engineering could be used to optimally position lanthanide cations for luminescence energy transfer to the internal chromophore of fluorescent proteins.

Here, we demonstrate a simple yet robust strategy for using fluorescent proteins as biosensors for lanthanides. The net charge of proteins can be augmented to a significant degree through extensively mutating the molecular surface30. This process, termed ‘supercharging’, generally preserves the overall structure and function of the protein while opening doors to customizable intermolecular behaviors that can meet specific engineering needs31,32,33,34. Hence, we modulate the surface of green fluorescent protein (GFP), yellow fluorescent protein (YFP), and cerulean fluorescent protein (CFP) to enable direct chelation between negatively charged residues and free lanthanide cations. Following an underutilized lanthanide-to-chromophore pathway of energy transfer35, ultraviolet (UV) excitation results in a detectable, dose-dependent signal transmission between trivalent terbium, thulium, dysprosium, and the chromophores of YFP and GFP, but not CFP. Notably, supercharged variants of YFP and GFP respond to lanthanides in the environmentally relevant 10 µM–5 mM range. The robust detection of lanthanides via a ready signal transmission to fluorescent proteins suggests that engineered protein surfaces may provide a green path forward in REE extraction, as well as broad opportunities in the development of biosensors and optically active materials.

Results

Supercharged fluorescent proteins demonstrate energy transfer with lanthanide antennae

Contrary to the established paradigm, in which lanthanides are the acceptors of resonance energy transfer, we hypothesized that lanthanides bound by electrostatic forces to a fluorescent protein surface could be functionalized into antennae for receiving an electromagnetic signal outside the normal excitation range of the internal chromophore (Fig. 1a). Subsequently, the signal could be transmitted via lanthanide-based resonance energy transfer (LRET) to the fluorescent protein for a quantifiable emission in the visible range. The efficiency of LRET, like FRET, is inversely dependent on distance35,36,37. The energy transfer is measurable by fluorescence microscopy only if the lanthanide and the protein are separated by a nanometer-scale distance (10–100 Å), thereby potentially enabling the detection of lanthanides upon deploying the fluorescent biosensors in various environmental and industrial settings38.

Fig. 1: Schema for energy transfer.
figure 1

a Schema of the posited lanthanide-based resonance energy transfer between the lanthanide antenna and the fluorescent protein (PDB: 2B3P). Antenna excitation gives rise to a detectable signal if the lanthanide and the biosensor are separated by 10–100 Å. Created in BioRender. Huang, K. (2023) BioRender.com/f23r502. b Spectral overlaps between terbium, thulium, dysprosium, and YFP, GFP, and CFP. The solid dashes represent select free-ion energy levels of trivalent terbium, thulium, and dysprosium. The vertical bars denote the excitation maxima +/− the spectral half-width of supercharged YFP, GFP and CFP. Source data are provided as a Source Data file.

Full size image

To interface with the lanthanide antennae, we aimed at establishing an expansive biosensor excitation range for a maximal chance of favorable spectral overlap with the emissions of Tb3+, Tm3+, and Dy3+ (Fig. 1b). To that end, we introduced the mutations T65G, V68L, S72A, and T203Y to red-shift and stabilize the series of supercharged GFP variants with net charges of −4, −10, −17, −31, +16, and +33 at pH 7.0, obtaining a series of YFP variants with identical net charges32,39. Likewise, we introduced the mutations Y66W, F146G, N147I, H149D to blue-shift and stabilize the same series of GFP variants, obtaining a series of CFP variants with net charges of −5, −11, −18, −32, +15, and +32 at pH 7.032,40,41,42. This set of eighteen fluorescent protein biosensors then served as our starting point for selecting the optimal candidates for detecting lanthanides (Fig. 2a, b).

Fig. 2: Charge engineering introduces lanthanide chelation sites.
figure 2

a Charge engineering schema for generating 18 protein biosensor variants of the base superfolder protein (PDB: 2B3P). b An illustration of potential lanthanide binding interactions and proximity to the protein chromophore (PDB: 2B3P). Solvent-exposed residues are mutated to increase the net charge while preserving the structure and function of the fluorescent protein. The charged surface enables lanthanide binding at various distances to the internal chromophore of the fluorescent protein. Created in BioRender. Huang, K. (2023) BioRender.com/f23r502.

Full size image

Initially, we measured the fluorescence output of the biosensors in response to excitation in the UV range (250–400 nm). When combined with any of the lanthanides (Tb3+, Tm3+, Dy3+), the negatively charged GFP and YFP variants exhibited a 2–3-fold increase in fluorescence intensity at select excitation wavelengths (Supplementary Fig. 1a and Supplementary Table 1). Similarly, the lanthanides Yb3+, Eu3+, and Sm3+ elicited a roughly 2–3-fold increase in fluorescence intensity when paired with the GFP−10 and YFP−31 variants (Supplementary Table 1). In contrast, the addition of Al3+ produced virtually unchanged excitation spectra at a concentration of 100 µM (Supplementary Fig. 1a), a value set to approximate or even exceed the levels of Al3+ commonly found in acid mine drainage43. Consistent with the Laporte rule, the lanthanides did not measurably luminesce on their own at any excitation wavelength14. The ratio of the fluorescence intensity upon adding the lanthanides to the fluorescence of the biosensors alone appeared to peak in the 340–400 nm excitation range, with 350 nm and 390 nm being the optimal wavelengths for achieving a maximal factor change in response upon exciting the lanthanide antennae bound to the GFP and YFP variants, respectively (Supplementary Fig. 1b). However, given that the wavelengths corresponding to these ratiometric maxima encroach on the excitation ranges of GFP and YFP, we opted to excite our lanthanide-bound system deeper in the UV range, at 340 nm and 360 nm for GFP and YFP, respectively. We rationalized that this approach would minimize crosstalk between FRET donor and acceptor channels while still staying within range for a sufficiently robust ratiometric response37. Interestingly, we observed a peak in the fluorescence output of YFP and GFP at an excitation wavelength of 280 nm, with or without the presence of lanthanides, though the addition of lanthanides enhanced the signal response across all variant conditions (Supplementary Fig. 1a). Hypothetically, the tryptophan residue at position 57, positioned approximately 12–13 Å apart from the chromophore within each of the fluorescent protein variants, could act as an antenna at this excitation wavelength for transferring Förster resonance energy directly to the internal chromophore, or in the presence of lanthanides, to the lanthanide, which in turn transmits the signal to the chromophore for an even greater response (Supplementary Figs. 2–4)44. A byproduct of our use of excitation wavelengths of 340 nm and 360 nm, we excluded the confounding effect of the tryptophan antenna in further analyses.

Importantly, time-resolved fluorescence spectroscopy revealed long-lived luminescence transmission from the lanthanide to the fluorescent protein biosensor across the excitation range of 250–400 nm (Fig. 3a, b), consistent with the intrinsically slow decay of the excited state of lanthanide series elements. In the presence of Tb3+, the GFP−10 and YFP−31 biosensors produced a markedly distinct fluorescence response relative to the protein alone. This difference in response was magnified by time delays of 100 µs and 300 µs between excitation and response detection. Of note, the biosensors alone produced negligible fluorescence when subjected to time-delayed detection, while the Tb3+:biosensor pairs exhibited a comparably modest attenuation in the signal response (Fig. 3, Supplementary Figs. 5 and  6). Thus, the long-lived signal responses of the GFP−10 and YFP−31 biosensors in the presence Tb3+ support the notion that signal detection is dependent on excitation and transmission via the lanthanide antennae.

Fig. 3: Time-resolved spectroscopy reveals long-lived lanthanide luminescence signaling.
figure 3

a Supernegative GFP−10 sensors fluoresce in the presence of 1 mM of Tb3+. Time-resolved luminescence responses measured at 0 µs, 100 µs, and 300 µs. GFP−10 concentration of 0.1 mg/mL (~3.7 µM) in 50 mM Tris-HCl at pH of 7. Green shading denotes the fraction of GFP−10 fluorescence attributable to Tb3+ placement. b Supernegative YFP−31 sensors fluoresce in the presence of 1 mM of Tb3+. Time-resolved luminescence responses measured at 0 µs, 100 µs, and 300 µs. YFP−31 concentration of 0.1 mg/mL (~3.7 µM) in 50 mM Tris-HCl at pH of 7. Yellow shading denotes the fraction of YFP−31 fluorescence attributable to Tb3+ placement. Source data are provided as a Source Data file.

Full size image

The excitation ratio, defined as the fluorescence of GFP at its emission maximum (510−520 nm) upon exciting at 340 nm versus 465 nm or the fluorescence of YFP (530−540 nm) upon exciting at 360 nm versus 500 nm, is a measure of LRET efficiency that permits comparisons between lanthanide:biosensor and biosensor-only conditions32. The factor change in excitation ratio, defined as the ratio of the excitation ratio calculated for the lanthanide:biosensor condition versus the biosensor-only condition, permits comparisons across biosensor variants. The addition of Tb3+ to YFP−31 produced the highest factor change in excitation ratio, 1.97 ± 0.25, whereas YFP variants with net charges of −17, −10, and −4 produced factor changes in the excitation ratio of 1.58 ± 0.25, 1.48 ± 0.15, and 1.30 ± 0.17, respectively (Fig. 4a and Supplementary Fig. 7). As expected, the non-interacting positively charged variants of YFP+16 and +33, produced factor changes in excitation ratio of 1.13 ± 0.04 and 1.05 ± 0.03, respectively, indicating decreased binding and detection of lanthanides compared to the wild-type YFP. The addition of Tm3+ and Dy3+ to the YFP series resulted in a similarly stepwise increase in signal as the magnitude of negative surface charge increased, with YFP−31 being the top-performing sensor for each of the lanthanides. The addition of Tb3+ to GFP−10 produced a factor change in excitation ratio of 2.58 ± 0.09, the highest signal among the GFP variants, whereas GFP−31, −17, and −4 produced factor changes in excitation ratio of 1.60 ± 0.06, 2.32 ± 0.25, and 1.82 ± 0.08, respectively, and the non-interacting GFP+16 and +33 produced responses of 1.06 ± 0.24 and 0.93 ± 0.11, respectively (Fig. 4a, Supplementary Fig. 7, Supplementary Fig. 8). Again, the addition of Tm3+ and Dy3+ to the GFP series manifested a similar trend, with the entire series of negatively supercharged GFP variants exhibiting an enhanced signal response relative to the positively supercharged and wild-type sensors (Fig. 4a and Supplementary Fig. 7). Importantly, our top performing sensors, YFP−31 and GFP−10 (Supplementary Fig. 9), demonstrated an approximately 2.5-fold or greater increase in their signal response to Tb3+, Tm3+, and Dy3+ across a dose-dependent 10 µM − 5 mM range, consistent with typical ranges of total REEs found in highly polluted streams (Fig. 4b, c). Additional testing with the lanthanides Eu3+, Sm3+, and Yb3+ revealed comparable response ratios and an apparent Kd of 25−30 µM for YFP−31 and 210−500 µM for GFP−10 across all six lanthanide species (Fig. 4b, c and Supplementary Table 1). In summary, our findings suggest that negatively supercharging YFP and GFP enables robust lanthanide binding and antennae signaling via (i) specific and non-specific sensitizing interactions within the electronically manipulable framework of the protein interior and (ii) a near-UV range stimulus resulting in the excitation of lanthanide cations and subsequent energy transfer to the protein chromophore.

Fig. 4: Lanthanide sensitivities of supercharged protein biosensors.
figure 4

a Ratiometric excitation responses of supernegative and superpositive protein sensors in the presence of 1 mM Tb3+, Tm3+, or Dy3+ in 50 mM Tris-HCl at pH 7. Fluorescent protein concentration of 0.1 mg/mL (~3.7 µM). Error bars denote S.D. +/− the mean. b GFP−10 biosensor ratiometric responses to Tb3+, Tm3+, Dy3+, Eu3+, Sm3+, and Yb3+ in the 1 nM–10 mM range. GFP−10 concentration of 0.1 mg/mL (~3.7 µM) in 50 mM Tris-HCl at pH of 7. c YFP−31 biosensor ratiometric responses to Tb3+, Tm3+, Dy3+, Eu3+, Sm3+, and Yb3+ in the 1 nM–10 mM range. YFP−31 concentration of 0.1 mg/mL (~3.7 µM) in 50 mM Tris-HCl at pH of 7. Experiments were performed in biological triplicate. Source data are provided as a Source Data file.

Full size image

Notably, the CFP series was non-performing, irrespective of the net charge (Fig. 4a and Supplementary Fig. 7). The inability of the CFP variants to detect lanthanides could be attributed to the absence of spectral overlap with the radiative energy levels of Tb3+ and Tm3+, as well as the excess overlap with the emission of Dy3+ (Fig. 1b)45. In the absence of overlap, the energy received by the lanthanide antenna cannot be transmitted to the acceptor fluorophore, even if the oppositely charged parties are positioned within range for LRET signaling18. Likewise, excess overlap prevents biosensor readout, but in this case, the antenna excitation outputs a signal that is not readily distinguishable from the signal arising from the biosensor-specific excitation37. Conversely, the excitation ranges of YFP and GFP appear to favorably overlap with potentially radiative configurations of trivalent terbium, thulium, and dysprosium in their excited states (Fig. 1b), thereby enabling the successful transmission of energy from the lanthanide antenna to the fluorescent biosensor.

Supercharged fluorescent protein biosensors do not respond to common interferents

Lanthanide-containing waste streams and deposits, such as mine drainage and electronic scrap, contain a surplus of potentially interfering, soluble metal compounds, including, most commonly, the chloride salts of aluminum and iron43,46,47,48,49. A biomining operation would therefore benefit from a system capable of distinguishing between the capture of lanthanides versus non-target metals.

To evaluate the response of our biosensor platform to interfering metals, we measured the fluorescence output of our top performing sensors, GFP−10 and YFP−31, in the presence of Al3+ and Fe2+, first as separate, non-interacting entities (Fig. 5a and Supplementary Fig. 10a), and second, as species competing for binding and detection (Supplementary Fig. 10b). Upon UV excitation, the YFP−31 sensor in combination with 1 mM of Tb3+ exhibited a nearly twofold increase in fluorescence output compared to YFP−31 in combination with 1 mM of non-target Al3+ (Fig. 5a). Notably, the Al3+:YFP−31 condition produced an emission spectrum that was only slightly perturbed compared to the YFP−31-only condition, indicating that the YFP−31 sensor can readily discriminate between Tb3+ and a non-lanthanide, non-luminescent metal species of the same ionic charge (Fig. 5a).

Fig. 5: Responsivities of biosensors in the presence of interferents.
figure 5

a Supernegative YFP series emission spectrum (ex. 280 nm) in the presence of equimolar (1 mM) amounts of Tb3+ or Al3+. Yellow and gray shading denote the fraction of YFP fluorescence attributable to Tb3+ placement and aluminum noise, respectively. b GFP−10 and YFP−31 biosensor responses to 1 mM of Tb3+ versus a 10-fold excess of contaminants (10 mM) found in mine drainage from the Virginia Canyon. 50 mM Tris-HCl at pH of 7. Error bars denote S.D. +/− the mean. Green and yellow shading represent the fraction of GFP and YFP fluorescence attributable to Tb3+ placement and aluminum noise, respectively. c Virginia Canyon (VC) and modified Virginia Canyon (mVC) interference of the GFP−10 ratiometric signal in the presence of 1 mM Tb3+. GFP−10 concentration of 0.1 mg/mL (~3.7 µM) in 50 mM Tris-HCl at pH of 7. d Virginia Canyon (VC) and modified Virginia Canyon (mVC) interference of the YFP−31 ratiometric signal in the presence of 1 mM Tb3+. YFP−31 concentration of 0.1 mg/mL (~3.7 µM) in 50 mM Tris-HCl at pH of 7. Experiments were performed in biological triplicate. Source data are provided as a Source Data file.

Full size image

Next, we established a competitive interaction between Tb3+ and Al3+ or Fe2+, holding the signal transmission between 1 mM of Tb3+ and the GFP−10 sensor constant as we introduced 1 nM–1 mM of Al3+ or Fe2+ in separate microplate wells (Supplementary Figs. 10b and  11). As we titrated up the concentration of Al3+, the excitation ratio remained constant, with a scale increase in the concentration of six orders of magnitude producing an unperturbed excitation ratio of 0.013 ± 0.001 (Supplementary Fig. 10b). Likewise, the presence of 10 µM of Fe2+ did not impact LRET signaling between GFP−10 and Tb3+, although concentrations greater than 10 µM resulted in a modest diminution of the signal response. At 1 mM of Fe2+, the signal weakened even more, indicating that an equimolar concentration of Fe2+ in the environment could disrupt the detection of lanthanides (Supplementary Fig. 10b). It should be noted that under the reaction conditions presented herein, Fe2+ is potentially readily oxidized to Fe3+. Irrespective of this detail, iron removal via pH-controllable precipitation presents a viable pre-processing strategy to minimize signal interference prior to biosensor readout47.

Importantly, the GFP−10 and YFP−31 biosensors retained functionality for detecting total lanthanides when subjected to mixed ion solutions resembling actual mining outflows (Fig. 5b−d and Supplementary Fig. 12a, b). First, to isolate any potential inhibitory effects of the contaminant metals, we measured the dose responses of Mg2+, Ca2+, Mn2+, Zn2+, and Cu2+ as single species in a competitive assay with GFP−10:Tb3+ and YFP−31:Tb3+. Apart from moderate perturbations in the fluorescence signal at 1 mM of Zn2+ and 100 µM of Cu2+, the contaminants interfered minimally with Tb3+ detection, even at a 10-fold excess of contaminant relative to the lanthanide (Supplementary Fig. 13a, b). Interestingly, Cu2+ appeared to interfere more significantly with Tb3+ detection by the GFP sensor compared to the YFP sensor (Supplementary Fig. 13). Additionally, our analysis yielded no false positive indications; the contaminants did not elicit biosensor fluorescence on their own (i.e., in the absence of Tb3+) as expected (Supplementary Fig. 14). To examine the GFP−10 and YFP−31 response to a true-to-nature mixed ion solution, we approximated the composition of acid mine drainage found in the Virginia Canyon47. With fidelity to the ratio of contaminants, we subjected GFP−10:Tb3+ and YFP−31:Tb3+ to mixtures of Al3+, Fe2+, Mn2+, Zn2+, and Cu2+, termed ‘VC’, as well as Mn2+, Zn2+, and Cu2+, termed ‘mVC’ (Supplementary Table 2). At an adjusted pH of 7, a 5-fold excess of the VC mixture elicited minimal perturbation of Tb3+ detection (Fig. 5c, d). Of note, an approximately equimolar amount of the mVC mixture produced modest interference with GFP−10 detection of Tb3+ (Fig. 5c), and at an adjusted pH of 5, the contaminant mixtures disrupted the detection of 1 mM of Tb3+ at approximately 10 µM and 1 mM of contaminants for GFP−10:Tb3+ and YFP−31:Tb3+, respectively (Supplementary Fig. 12a, b). In addition to transition metal contaminants, REE-rich process streams often contain a high concentration of the non-luminescent lanthanide lanthanum (La3+)47. Upon measuring the dose response of La3+ in a competitive assay with GFP−10:Tb3+ and YFP−31:Tb3+, we observed no significant perturbation of Tb3+ detection at a 5-fold excess of La3+ relative to Tb3+ (Supplementary Fig. 15a, b).

At concentrations of lanthanides high enough to elicit a strong biosensor response, we hypothesized that aggregates could form due to the interaction of near equimolar amounts of oppositely charged species. To evaluate the effect of potential aggregation on fluorescence signal detection, we incubated mixtures of supercharged GFP variants and 1 mM of Al3+ for 30 minutes prior to fluorescence readout. The Tb3+:GFP pairs exhibited excitation ratios on par with an immediate readout, whereas the Al3+:GFP pairs produced slightly increased excitation ratios compared to the GFP-only conditions. Still, the GFP−10 biosensor, as well as the other GFP variants, can discriminate between Tb3+ and Al3+ following the extended incubation period, highlighting the versatility of our system (Supplementary Fig. 16).

To further validate the functionality of our biosensors in real-world conditions, we assessed other potentially interfering factors, such as ionic strength, temperature, and pH changes (Supplementary Figs. 17–19). Of note, concentrations of NaCl from 0 to 800 mM were found to minimally perturb GFP−10 and YFP−31 detection of 1 mM of Tb3+ (Supplementary Fig. 17). Additionally, the temperature range from 0 °C to 70 °C was conducive to stability in Tb3+:GFP−10 signal transduction while the temperature range from 0 °C to 60 °C was conducive to stability in Tb3+:YFP−31 signal transduction (Supplementary Fig. 18). Finally, the pH dependence of Tb3+ detection was such that below a pH of 5, Tb3+ detection was rendered unlikely and virtually indistinguishable from the absence of Tb3+ (Supplementary Fig. 19).

Higher order structures can also detect lanthanides

Symmetry in protein oligomers gives rise to remarkable complexity in biological systems50,51,52. Given the ubiquity of multimeric protein assemblies and their capacity for diverse structural and transport functions within living organisms, we set out to create a higher order protein architecture for detecting lanthanides53,54,55,56. Using a method termed the ‘supercharged protein assembly’ (SuPrA), we selected two oppositely charged fluorescent proteins from our set of biosensors, CFP+32 and GFP−31, for assembly into a globular CFP+32/GFP−31 protomer (Supplementary Fig. 20)32. After combining equimolar volumes of CFP+32 and GFP−31, we observed a unique FRET interaction on par with the signal manifested in the assembly of a 16-unit CFP+32/GFP−17 protomer with eight-fold symmetry (Fig. 6a)32,57.

Fig. 6: Supercharged protomer fluorescence in the presence of lanthanides.
figure 6

a Schema of the posited LRET interaction between the lanthanide and the CFP+32/GFP−31 protomer (PDB: 6MDR). Created in BioRender. Huang, K. (2023) BioRender.com/f23r502. b Lanthanide and aluminum (1 mM) interference of the protomer signal (ex. 433 nm). An increased signal relative to the unperturbed protomer FRET is observed when exciting at 340 nm. Protomer concentration of 3.7 µM in 50 mM Tris-HCl at pH of 7. Created in BioRender. Huang, K. (2023) BioRender.com/f23r502. c The FRET signal is ‘restored’ in varying degrees when exciting at 340 nm versus 433 nm. Error bars denote S.D. +/− the mean. Experiments were performed in biological triplicate. Source data are provided as a Source Data file.

Full size image

Upon introducing 1 mM of Tb3+, Tm3+, Dy3+, or Al3+, we disrupted the FRET output of GFP−31 in response to excitation at 433 nm, the optimal excitation wavelength of the donor CFP+32. The pattern of disruption in FRET signaling between CFP+32 and GFP−31, in which each of the lanthanides exhibited a stronger effect than aluminum, implies some combined effect of competitive binding, protomer disassembly, and fluorescence modulation. Notably, upon exciting the protomer system in the UV range, following our protocol for transmitting a signal from a lanthanide antenna to its acceptor fluorophore, the fluorescence modulation appeared to reverse course, with each Ln3+:protomer pair manifesting some degree of signal restoration (Fig. 6b). The UV-excited GFP−31 still responded to the lanthanides with a decrease in fluorescence output, although this time, the disruption in FRET between CFP+32 and GFP−31 appeared to be masking a response potentially arising from signaling between the CFP+32/GFP−31 protomer and the lanthanide antennae. To quantitate this masked signal, we calculated the percentage of GFP−31 fluorescence relative to the unperturbed signal. Dy3+ produced the greatest masked signal, a factor change of 1.01 ± 0.09 in the fluorescence output of UV-excited GFP−31 compared to a factor change of 0.77 ± 0.04 for GFP−31 excited at 433 nm. Similarly, Tm3+ induced a factor change of 0.94 ± 0.05 in the fluorescence of UV-excited GFP−31 compared to a factor change of 0.72 ± 0.11 for GFP−31 excited at 433 nm. Tb3+ appeared to demonstrate a diminutive increase in the UV-excited signal, albeit not significantly (Fig. 6c). Hypothetically, the formation of the supramolecular assembly of fluorescent proteins could have induced a spectral shift toward a less favorable overlap with the radiative transitions of Tb3+ and, in turn, decreased the signal transmission.

Discussion

The lanthanide series elements are in high global demand, though a green and scalable path forward in the quantifiable capture of lanthanides from the environment remains elusive. While previous methods of lanthanide sensing and luminescence modulation have utilized sophisticated forms of chelation, sensitization, and semi-synthetic modification, such systems are not scalable, difficult to utilize in living systems, or lack an appropriate sensitivity range for detecting lanthanides in highly concentrated streams16,18,23,58. By demonstrating that free lanthanide cations can bind to charge-engineered protein surfaces and act directly as excitable antennae in the absence of a specific luminescence sensitizer, organic or otherwise, we instantiate an underutilized model for energy transfer and notably shift the paradigm for lanthanide luminescence signaling. Future computational modeling of the interactions between lanthanides and monomeric and multimeric protein architectures should reveal more precise appositions in which the interactions between lanthanides, surface charge, and electronically dense organic scaffolding give rise to enhanced radiative transfers.

Unlike existing biosensors for lanthanides, the supercharged protein interface exhibits dual functionality in binding and detecting lanthanides in the millimolar range, thereby implicating its utility in monitoring the development and efficiencies of high load lanthanide processing streams. Even amid the suite of ultra-selective LanM-based sensors, the supercharged fluorescent biosensor offers several unique advantages. Particularly, in environments where the amount of total lanthanides exceeds the upper bound for detection afforded by previously developed LanM derivatives, such as certain acidic discharges and highly concentrated filtration streams, the supercharged series of biosensors retains high capacity for quantitatively reporting on lanthanides25,26,27,28. At present, the platform’s operational scope encompasses the environmentally and industrially significant micromolar to millimolar range, though future machine learning guided predictions could enable the broad, structure-based modulation of lanthanide sensitivities to meet specific processing needs. In contrast, the lanthanide biosensor LaMP1, composed of a FRET donor–acceptor pair joined by a collapsible LanM arm, has a considerably higher affinity for lanthanides, potentially predisposing it to oversaturation in more concentrated processing streams20. Likewise, the tryptophan-substituted LanM biosensor boasts an impressive picomolar sensitivity for terbium but meets its upper bound for a dose-dependent fluorescence readout in the nanomolar range, again rendering it prone to saturation23. It should be noted, however, that the upper bound of detection will be condition-dependent; the tryptophan-substituted sensor indeed achieves a maximum sensitivity in 1–10 µM range, approaching that of the YFP–31 sensor, when subjected to conditions like those employed in this study (e.g., ~3.7 µM of protein).

Even in the presence of common interferents that plague other sensor platforms, such as aluminum and iron, the supercharged series of biosensors retains its micromolar to millimolar sensitivity for lanthanides, pointing to its utility in environments mirroring not only this broad range of lanthanide concentrations but also the presence of equimolar amounts of competing metals. Even then, the removal of potentially problematic interferents, such as iron and aluminum, could be readily achieved via a pH-based pre-processing strategy, wherein the competing metal ions are precipitated out of the filtration stream in a minimally obtrusive manner47. More complex conditions, such as the presence of a greater than 10-fold excess of multiple competing metals, will necessitate additional study, and may require more aggressive refinement to be made compatible with the present sensors.

Importantly, the adaptation of the sensing paradigm to organized nano-assemblies suggests the intriguing possibility of utilizing these sensors in engineered materials, or even just protein precipitates that could be readily applied to detectors. Future insights into cross-linking, self-assembly, and modular design could reveal broad opportunities in the development of optically active nanomaterials and macroscale filtration systems alike. In addition, the possibilities for utilizing these simple, genetically encoded biosensors in organisms that could concentrate and detect surface lanthanides are manifest.

Methods

Cloning and purification of supercharged fluorescent proteins

Synthetic genes encoding supercharged variants of YFP, GFP, and CFP were obtained from Twist Bioscience and Integrated DNA Technologies (Supplementary Tables 3 − 5). Escherichia coli DH10B and BL21 (New England Biolabs) were used for cloning and expression, respectively. Expression plasmids based on the pET system were either constructed through Golden Gate assembly or acquired directly from Twist Bioscience.

Starter cultures for expression were prepared by picking single colonies of E. coli for overnight growth in Luria-Bertani broth at 37 °C and 250 r.p.m. The E. coli were then inoculated into 250 mL of broth and grown at 37 °C and 250 r.p.m. until an OD600 of 0.6 was achieved. The cultures were cooled at 4 °C for 45 minutes before 1 mM isopropyl-β-D-1-thiogalactopyranoside was added to induce protein expression. Following 16 hours of incubation at 18 °C and 250 r.p.m., the cells were harvested by centrifugation (6,000 g, 4 °C, 30 minutes). The pellets were then either stored at −20 °C for later use or resuspended in 30 mL of phosphate buffer (2 M NaCl, 20 mM imidazole, 5 mM MgSO4, pH 7.5). Once resuspended, the cells were subjected to a sonication routine for cell lysis, and the supernatant containing the soluble proteins was obtained by centrifugation (35,000 g, 4 °C, 30 minutes). The proteins were subsequently purified using HisPur Ni-NTA Resin (Thermo Fisher Scientific) and custom wash (50 mM phosphate buffer, 2 M NaCl, 20 mM imidazole, pH 7.5) and elution buffers (50 mM phosphate buffer, 2 M NaCl, 500 mM imidazole, pH 7.5). Desalting and buffer exchange were performed using Amicon Ultra Centrifugal Filter Units (MilliporeSigma), and the purified proteins were concentrated to a volume of 0.4 mL in 50 mM Tris-HCl (pH 7.0). The concentration of the proteins was measured using the Pierce Coomassie Plus Bradford Assay (Thermo Fisher Scientific), and the purity was confirmed through sodium dodecyl sulfate–polyacrylamide gel electrophoresis.

Assaying fluorescence in the presence of lanthanides and interferents

Fluorescence measurements were obtained at room temperature using a Cytation 5 microplate reader (BioTek). The sample volume was 0.1 mL unless otherwise specified, and the experiments were conducted using 96-well clear bottom polystyrene assay plates (Corning). At least three independent measurements were obtained for each of the conditions tested. Solutions of TbCl3, TmCl3, DyCl3, EuCl3, SmCl3, and YbCl3 were prepared by dissolving the compounds in 50 mM Tris-HCl. Likewise, the competing metal salts AlCl3, FeCl2, MgCl2, CaCl2, MnCl2, ZnCl2, and CuCl2 were dissolved in 50 mM Tris-HCl. To minimize aggregation, the proteins were carefully agitated by pipetting up and down before diluting the concentrated stock solutions to 0.1 mg/mL in 50 mM Tris-HCl. The pH was buffered at 7.0 unless otherwise indicated.

Time-resolved luminescence responses of GFP−10 and YFP−31 were measured at 0 µs, 100 µs, and 300 µs in the absence and the presence of 1 mM of Tb3+. The ratio of the protein biosensor’s fluorescence upon excitation at a wavelength outside of its normal excitation range versus at a wavelength within its normal excitation range was obtained as a proxy for the magnitude of the lanthanide-based energy transfer from the lanthanide to the biosensor. This quantity, termed the ‘excitation ratio’, normalizes the signal observed when exciting the lanthanide antenna to the signal observed when exciting the fluorescent protein, thereby enabling comparisons of energy transfer across reaction conditions. Excitation ratios were calculated for GFP−10 and YFP−31 in the presence of 1 nM to 10 mM of Tb3+, Tm3+, Dy3+, Eu3+, Sm3+, and Yb3+. Subsequently, the factor changes in the excitation ratio were calculated to compare the performance of charge variants of GFP, YFP, and CFP in the presence of 1 mM of Tb3+, Tm3+, and Dy3+.

To evaluate luminescence responses to interfering metals, the fluorescence output of GFP−10 and YFP−31 were measured in the presence of Al3+, Fe2+, Mg2+, Ca2+, Mn2+, Zn2+, and Cu2+ via non-competitive and competitive assays. In the non-competitive setup, GFP−10 and YFP−31 responses were assayed in the absence and presence of 1 mM of the competing metal. The response in the presence of 1 mM of Tb3+ served as the positive control. In the competitive setup, the biosensors were incubated with 1 mM of Tb3+ prior to introducing 1 nM–10 mM of the competing metal. To confirm that the results were not dependent on the sequence in which the reagents were added, the biosensors were also incubated with 1 nM–10 mM of the competing metal prior to adding 1 mM of Tb3+. Mixed solutions of Al3+, Fe2+, Mn2+, Zn2+, and Cu2+, as well as Mn2+, Zn2+, and Cu2+, were evaluated via the non-competitive and competitive setup. To assemble the CFP+32/GFP−31 protomer, equimolar volumes of CFP+32 and GFP−31 suspended in 50 mM Tris-HCl were combined and incubated for 10 minutes at room temperature. The fluorescence output of the CFP+32/GFP−31 protomer was measured via the method employed for the individual fluorescent proteins, in the presence of 1 mM of Tb3+, Tm3+, Dy3+ and Al3+.

Statistics and reproducibility

The results are presented as the mean  + /− S.D. unless otherwise indicated. No statistical method was used to predetermine sample size, and no data were excluded from the analyses. The experiments were not randomized, and the investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.