The core insight is that UCNPs do not respond to light the way ordinary fluorescent dyes do. Conventional organic fluorophores are essentially linear emitters: when you modulate the excitation intensity, the fluorescence follows the same waveform, containing exactly the same frequency components as the input. Upconversion nanoparticles, by contrast, rely on a sequential, multi-step energy transfer process. A sensitizer ion, typically ytterbium, absorbs an infrared photon and hands its energy to an activator ion such as thulium or erbium, which must receive a second energy packet before it can emit. This inherent nonlinearity means the emission does not simply mirror the excitation. When the team ran numerical simulations based on rate equations for a two-photon energy-transfer system, they found that sinusoidally modulated excitation at a frequency f produced upconversion emission containing a second-harmonic component at 2f, a frequency entirely absent from the excitation light itself.

The experiments confirmed the predictions in striking fashion. Using NaYF4 nanoparticles doped with 20 percent ytterbium and 0.5 percent thulium and coated with an inert shell, the researchers applied square-wave excitation at 980 nanometers with a 50 percent duty cycle. The Fourier transform of the excitation trace showed only odd harmonics of the 20 hertz base frequency, as expected for a square wave. But the Fourier spectrum of the resulting upconversion luminescence, collected between 430 and 850 nanometers, revealed prominent even harmonics at 40, 80, and 120 hertz and beyond. These components were generated by the nanoparticles themselves, not by any instrumental imperfection, since the laser light monitored throughout the experiments never displayed comparable even-harmonic content. The strength of this second-harmonic signal, quantified as a weight relative to neighboring frequency components, depended on excitation intensity and on the base modulation frequency, exactly as the nonlinear photophysics would predict.

Second harmonics, however, come with a practical catch. Because they sit at twice the modulation frequency, they can be too fast for ordinary scientific cameras, which are limited in frame rate by the Nyquist theorem. The researchers realized that the same nonlinear physics could be pushed further. If the nanoparticles are excited by two laser beams modulated at two different base frequencies, f1 and f2, the nonlinear response mixes these frequencies, generating a beat-frequency signal at the difference, f2 minus f1, as well as a sum-frequency component. In effect, the nanoparticles behave as optical frequency mixers, analogous to the mixers used in radio engineering. When the team superimposed two 980-nanometer beams modulated at 21 and 23 hertz, the upconversion emission spectrum clearly contained a beat-frequency signal at 2 hertz, alongside the base frequencies, their second harmonics at 42 and 46 hertz, and a sum-frequency peak at 44 hertz. Crucially, none of these extra components existed in the excitation light.

The beauty of the beat-frequency channel is that it can be placed at an arbitrarily low frequency, well within the reach of a slow camera, while remaining completely free of residual excitation light, which carries only the base frequencies and their harmonics. The researchers then showed that the strength of this signal could be tuned through nanoparticle engineering. By inserting an intermediate shell doped with ytterbium between the core and the outer inert shell, creating a core-multishell architecture, they substantially boosted the beat-frequency weight of the 800-nanometer thulium emission. Erbium-activated particles with neodymium-containing outer shells, excitable at both 980 and 808 nanometers, also produced robust beat-frequency signals. The team was careful to note that the relationship between architecture and performance is not simply that more complex means better; rather, it is the way specific design elements modify energy transfer rates, saturation behavior, and ion kinetics that governs how efficiently dual-frequency excitation couples into the emission.

The imaging demonstrations are where the method reveals its full power. In a surface-imaging experiment, the team imaged a fluted cross-shaped pattern, a screw cap coated with engineered nanoparticles, under dual-frequency excitation at 21 and 23 hertz. The time-averaged image was dominated by reflected laser light that leaked past a short-pass filter, obscuring the pattern. But when each pixel of the image sequence was filtered at the 2-hertz beat frequency, the residual excitation light vanished almost entirely while the nanoparticle pattern remained crisp. Filtering at the base frequency of 21 hertz, the conventional lock-in approach, failed to achieve the same result, because the leaked excitation light is modulated at exactly that frequency. The researchers then flooded the scene with strong, nonuniform ambient light that completely washed out both the signal and the reflection in the averaged image. The beat-frequency filter still recovered the pattern cleanly, demonstrating immunity to ambient interference as long as the detector is not saturated.

Deep-tissue imaging posed an even sterner test, and here the safety constraints of real biomedicine come into play. A capillary filled with thulium-doped core-shell nanoparticles was buried behind a five-millimeter slab of chicken breast tissue, and the excitation intensity was kept at roughly 72 milliwatts per square centimeter, nearly an order of magnitude below the ANSI safety limit for continuous-wave 980-nanometer skin exposure. Two coaxial beams modulated at 22 and 23 hertz illuminated the tissue at an oblique angle, and a scientific CMOS camera recorded sequences with 400-millisecond exposures. Individual frames and even frame averages were murky, contaminated by excitation leakage that persisted despite a combined optical density of 18 in the detection filters. Spectral measurements confirmed that the leakage was concentrated at the excitation wavelength. Yet Fourier analysis of the integrated image intensity revealed the 1-hertz beat-frequency component, and pixel-wise filtering at that frequency suppressed the leakage and other background contributions, improving the signal-to-background ratio by nearly an order of magnitude.

The method proved equally resilient under dynamically varying ambient light and at drastically reduced nanoparticle concentrations. When the researchers diluted the nanoparticle suspension a hundredfold, to 100 micrograms per milliliter, they simply shifted the beat frequency down to 0.1 hertz by modulating the lasers at 22 and 22.1 hertz and lengthened the camera exposure. Even through two millimeters of chicken tissue, the inclusion remained clearly detectable with a high signal-to-background ratio. The authors emphasize that PP-LID improves contrast and background rejection but does not circumvent the fundamental spatial resolution limits imposed by diffusive light transport in tissue, which remain on the millimeter scale in wide-field luminescence diffuse optical tomography geometries. What it does provide is a cleaner signal that enables less aggressive regularization and better discrimination during image reconstruction.

What makes PP-LID particularly attractive for practical deployment is its modest technical demands. Unlike conventional lock-in detection, it requires no precise synchronization between the excitation source and the detector, tolerates ambient light provided detector saturation is avoided, and imposes no stringent requirements on detector time resolution. It is also fundamentally distinct from previously reported frequency-encoding techniques for UCNPs, which modulated and detected at the same frequency and therefore remained vulnerable to excitation leakage. The approach can be combined with established background-suppression methods such as time-gated detection, and it does not even require two separate laser beams; a single beam carrying a superposition of two modulation waveforms would suffice. The authors point to several directions for future work, including quantitative models that explicitly account for ion distributions and interionic interactions, strategies to boost nanoparticle brightness, and solutions to the temporal fluctuations introduced by physiological motion such as breathing and blood flow in living subjects.

If the promise holds, the implications could reach well beyond the laboratory. Background-free imaging at safe excitation powers, with inexpensive cameras and minimal synchronization hardware, could sharpen small-animal studies, deepen the reach of optical biosensing, and bring upconversion-based diagnostics closer to clinical reality. In a field where progress has often meant building ever more elaborate filters and detectors, the KTH-led team has taken a different path: they let the physics of the nanoparticles do the filtering. By coaxing lanthanide-doped nanocrystals to act as tiny frequency mixers, they have opened a spectral channel that no stray laser light or ambient glare can follow, and in doing so they may have rewritten the rules of what low-light bioimaging can achieve.

Subject of Research: Photophysical lock-in detection of upconversion nanoparticles for background-free bioimaging

Article Title: Photophysical lock-in detection enables background-free upconversion emission imaging

Article References: Bagheri, N., Wang, C., Guo, D., Lakshmanan, A., Zhu, Q., Chen, X., Ghazyani, N., Zhan, Q., Sotiriou, G. A., Liu, H., & Widengren, J. (2026). Photophysical lock-in detection enables background-free upconversion emission imaging. Light: Science & Applications, 15(1), Article 371. https://doi.org/10.1038/s41377-026-02414-2

Image Credits: AI Generated

DOI: 10.1038/s41377-026-02414-2

Keywords: upconversion nanoparticles, lock-in detection, bioimaging, lanthanide, frequency mixing, beat frequency, deep-tissue imaging, signal-to-background ratio, nanoparticle engineering, fluorescence, optical filters, sCMOS camera

Virus-like particles, often abbreviated as VLPs, are engineered or natural assemblies that mimic the architecture of true viruses but lack any genetic cargo and therefore cannot replicate or cause infection. They occupy a special place in modern biotechnology. Because they present the same repetitive protein lattice that the immune system encounters on a genuine viral surface, they are exceptionally potent platforms for vaccines, and several of the most successful immunizations in current use are built on VLP scaffolds. They are also widely used as delivery vehicles in gene therapy and as programmable nanocontainers in materials science. Understanding how these shells assemble is therefore not a purely academic question; it determines how efficiently such particles can be manufactured, how stable they are once formed, and how their surfaces can be modified for medical applications.

The central difficulty in studying self-assembly has always been one of scale and speed. A complete viral shell, or capsid, is typically built from sixty to several hundred copies of a single capsid protein, and the assembly reaction can pass through fleeting intermediates that exist for microseconds or less. Conventional structural biology techniques excel at determining the static structure of the finished particle. X-ray crystallography and cryo-electron microscopy can render the final capsid at near-atomic resolution, revealing the precise contacts that hold the shell together. But these methods average over enormous numbers of particles and freeze the ensemble at one point in its history. What they cannot easily show is the route by which the components travel from a disorganized solution of subunits to the finished, closed shell.

The new study addresses that gap by following the assembly reaction at the level of individual molecules. The researchers combined advanced single-particle imaging with time-resolved structural analysis, allowing them to observe the same population of virus-like particles as they progressed through distinct assembly stages. Rather than reconstructing a single averaged picture, the approach preserved the heterogeneity of the reaction, exposing the coexistence of half-formed intermediates, partially closed shells, and completed particles at any given moment. This heterogeneity is not noise; it is the physical signature of the assembly pathway itself, and capturing it is what makes a mechanistic reading of the process possible.

From these observations, a coherent picture of the assembly mechanism emerges. The capsid protein does not appear to be a passive brick that simply clicks into place wherever it collides with a growing shell. Instead, the experiments indicate that the subunit adopts distinct conformational states as it participates in the reaction, and that transitions between these states are coupled to the binding events that extend the shell. Early in assembly, small clusters of subunits nucleate the process, overcoming an energetic barrier that must be crossed before growth becomes favorable. Once a stable nucleus exists, the addition of further subunits proceeds rapidly, with each incoming protein locking into the lattice and, in doing so, preorganizing the binding surface for the next arrival.

This coupling between structure and binding is a hallmark of what biophysicists call a nucleation-and-growth mechanism, and it explains both the speed and the specificity of viral assembly. If subunits could bind indiscriminately, misassembled and malformed particles would dominate the reaction. Instead, the observed pathway funnels the components toward the correct geometry. The intermediate states captured in the study show that incorrect associations are either short-lived or structurally primed to convert into productive arrangements, so that the reaction is continually steered back onto the correct trajectory. The result is a yield of properly formed particles that would be the envy of any synthetic chemist, achieved without any external template or instruction beyond the information encoded in the protein sequence itself.

The energetic logic of the process is as important as its structural choreography. Capsid assembly must balance two opposing forces: the favorable contacts between subunits that drive the shell to grow, and the cost of conformational changes and electrostatic interactions that must be paid along the way. Too weak an attraction, and the reaction stalls before a nucleus forms. Too strong, and subunits aggregate irreversibly into useless clumps. The observations reported in Nature suggest that the virus-like particle sits at a finely tuned point between these extremes, with subunit binding affinities and conformational transitions calibrated so that assembly is both efficient and reversible enough to correct local errors. This principle of kinetically regulated, error-correcting self-assembly is one that researchers in nanotechnology have long tried to emulate in synthetic systems, and the direct structural evidence for how a biological assembly achieves it is likely to inform those efforts.

Beyond its fundamental interest, the study carries practical weight for biomedicine. Vaccine manufacturers producing VLP-based immunizations depend on assembly reactions that proceed with high yield and high uniformity, since misassembled particles can compromise both the potency and the safety profile of a product. A mechanistic understanding of the assembly pathway provides rational levers for optimization: adjusting solution conditions, protein sequence, or the presence of cofactors can shift the reaction toward faster nucleation, more stable intermediates, or improved final yield, depending on what the manufacturing process requires. Similarly, in gene therapy and drug delivery, where VLPs and related particles are engineered to encapsulate therapeutic cargo, the timing and coupling of assembly relative to cargo loading is a critical design parameter. The ability to see which intermediates form, and when, turns what has been a largely empirical optimization exercise into an evidence-guided engineering problem.

The work also illustrates a broader shift in structural biology. The field has traditionally been dominated by the determination of static, high-resolution structures of purified, stable states. Increasingly, however, the most pressing questions concern dynamics: how molecular machines move, how signaling proteins switch states, and how supramolecular assemblies build themselves. Methods that can resolve conformational heterogeneity and connect it to reaction progress are transforming those questions from matters of inference into matters of direct observation. In the case of virus-like particles, the application of such methods bridges a long-standing divide between virology, which has emphasized the architecture of mature viruses, and biophysics, which has modeled assembly largely through theory and simulation. Direct structural data on intermediates now provide the empirical anchor that theoretical models of self-assembly have needed.

The demonstration that a virus-like particle can be watched as it assembles, stage by stage, at molecular resolution closes one of the enduring gaps in the understanding of biological self-organization. The finished capsid, so elegant in its symmetry, is revealed not as a structure that simply exists but as the endpoint of a tightly controlled kinetic journey, one in which each subunit both responds to and shapes the assembly around it. For virologists, the findings deepen the picture of how the simplest biological entities achieve such remarkable reliability with so few components. For biotechnologists, they supply a mechanistic foundation for designing and producing the next generation of VLP-based vaccines, delivery vehicles, and nanomaterials. And for the wider study of molecular self-assembly, they offer a vivid reminder that the most sophisticated construction projects in the world are carried out, continuously and invisibly, by molecules following rules that science is only now learning to observe directly.

Subject of Research: Molecular-level observation of the self-assembly pathway of a virus-like particle

Article Title: Molecular-level observation of the self-assembly of a virus-like particle

Article References: Asor, R., Loewenthal, D., Melnyk, D., Tan, T. K., & Kukura, P. (2026). Molecular-level observation of the self-assembly of a virus-like particle. Nature, 657(8132), 653-660. https://doi.org/10.1038/s41586-026-10948-z

Image Credits: AI Generated

DOI: 10.1038/s41586-026-10948-z

Keywords: virus-like particle, self-assembly, capsid, virology, structural biology, biophysics, nucleation and growth, vaccines, nanotechnology, molecular imaging, Molecular-level, observation