Cancer-associated fibroblasts can constitute up to 70% of the tumor microenvironment, according to the study. They restructure the extracellular matrix, promote epithelial-to-mesenchymal transition, and interact with cancer cells and other stromal cells by secreting cytokines, chemokines, growth factors and extracellular components. Their effects on tumor growth vary by subtype: the authors describe myCAFs as pro-tumorigenic and apCAFs as anti-tumorigenic, with apCAFs associated with T-cell activation and anti-tumor immunity. Flow cytometry and gene-expression profiling are established approaches for identifying CAF subtypes, but the paper notes that these methods can be time-consuming, expensive and affected by the lack of uniform identification standards. A label-free physical method could provide an additional way to distinguish these populations.

The researchers designed the sensor as a metasurface made from contact-coupled asymmetric double-arm split-ring resonators. Each resonator functions as a patterned optical element, with the displacement parameter set to 20.5 micrometers to balance the amplitude, quality factor and figure of merit of the two operating modes. With x-polarized terahertz illumination, the metasurface supported a sharp quasi-bound state in the continuum, or q-BIC, resonance near 0.92 terahertz. Under y-polarized illumination, it generated an electromagnetically induced transparency, or EIT, window near 0.82 terahertz. Field and surface-current simulations indicated that the q-BIC arose from interference between a magnetic dipole and four L-shaped electric dipoles. The EIT response instead came from destructive interference between two oppositely circulating magnetic dipoles.

For the experiments, the team sorted and immobilized normal intestinal fibroblasts, myCAFs and apCAFs on identical chemically functionalized metasurfaces. Each cell population was prepared at a concentration of 1 million cells per milliliter before measurement. The functionalized sensors were characterized under ambient conditions by terahertz time-domain spectroscopy, with the incident pulse traveling along the z-axis, perpendicular to the metasurface. Cell-free functionalized surfaces served as controls. Keeping the surface density of the three populations at the same level allowed the researchers to treat differences in cell number as negligible when interpreting the resonance shifts. The subtype-specific biochemical and material environment near the resonators was therefore the principal experimental variable.

The q-BIC channel gave a clear three-part pattern. The cell-free control had a resonance center at 0.92 terahertz, while normal intestinal fibroblasts shifted it slightly downward to 0.91 terahertz, a 10-gigahertz red shift. A much larger red shift appeared with myCAFs: the center moved to 0.86 terahertz, 60 gigahertz below the control. apCAFs moved the resonance in the opposite direction to 0.94 terahertz, representing a 20-gigahertz blue shift. The Fano-resonance linewidth, denoted by γ, remained close to 0.1 terahertz for the control and all three cell populations. Corresponding quality factors ranged from 8.6 for myCAFs to 10.1 for normal fibroblasts and were not described as reliably distinguishing the subtypes. The useful separation in this channel came from the position of the resonance center rather than its linewidth.

The EIT channel reproduced the same overall direction of change while adding an independent spectral parameter. The control EIT window was centered at 0.82 terahertz. Normal intestinal fibroblasts shifted it to 0.79 terahertz, myCAFs produced a further red shift to 0.76 terahertz, and apCAFs shifted it to 0.81 terahertz. Relative to normal fibroblasts, that amounted to an additional 30-gigahertz red shift for myCAFs and a 20-gigahertz blue shift for apCAFs. The group delay within the EIT window also differed: 1.57 picoseconds for normal fibroblasts, 1.65 picoseconds for myCAFs and 1.80 picoseconds for apCAFs, compared with 1.58 picoseconds for the control. Group delay is the time displacement of a terahertz wave packet associated with the phase response across the transparency window.

The physical interpretation centers on changes in the dielectric environment immediately surrounding the resonators. Under perturbation theory, the frequency shift induced by a localized change in permittivity is proportional to an integral of the squared electric-field intensity over the region where the permittivity changes. The researchers therefore extracted the real and imaginary components of the cells’ complex permittivity from the measured transmission spectra. Because the biological materials and their surface markers were treated as nonmagnetic, changes in magnetic permeability were neglected. Effective medium theory was used to relate the dielectric constant of the cell layer to cell number and type. With density held constant, the authors attributed the observed shifts primarily to biochemical differences among the fibroblast subtypes, including their differing extracellular matrices and surface compositions.

That interpretation was consistent with molecular measurements performed on the same fibroblast categories. Podoplanin, abbreviated PDPN, and CD74 were selected as hallmark markers for distinguishing the populations. apCAFs were double-positive for PDPN and CD74, myCAFs were PDPN-positive but CD74-negative, and normal intestinal fibroblasts were negative for both markers in the reported analysis. The study also described higher expression by myCAFs of α-SMA, PDPN, COL1A1, FAP, IL-6, TGF-β and VEGF, together with chemokines including COL8A1, COL10A1, COL11A1 and COL12A1. apCAFs expressed CD74, PDPN, VEGF and IL-6 along with chemokines including CXCL13, CCL3, CCL4 and CCL5. Flow-cytometry and gene-expression measurements thus agreed with the subtype assignments used in the optical experiment.

The differences in surface markers and extracellular-matrix composition provide a possible link between the cellular measurements and the terahertz spectra. The extracellular matrix contains collagens, elastin, laminin and proteoglycans, whose relative abundance can differ among CAF subtypes and influence the dielectric properties of the layer next to the metasurface. The authors also noted that myCAFs and apCAFs differ in the levels of genes and key surface proteins, including cytokines, chemokines and other signaling molecules. The study did not quantify the separate contribution of each macromolecule to the measured frequency shifts, so the evidence supports a link between subtype-specific composition and optical response without resolving that link at the level of individual molecules. The agreement among spectroscopy, flow cytometry and gene-expression analysis supports the classification of the tested cultures; it does not by itself establish performance in mixed human tumors.

Several limitations keep the findings at the proof-of-concept stage. The spectral resolution of the terahertz time-domain system was limited, leaving some subtype-specific shifts closely spaced. The paper states that more distinct discrimination might require higher-precision instrumentation or higher cell concentrations, but neither modification was established in these experiments. Biologically, the experiments used isolated, cultured populations at a standardized density of 1 million cells per milliliter, whereas human colorectal tumor tissue contains multiple CAF subtypes with heterogeneous concentrations. It therefore remains challenging to identify a particular subtype directly within patient-derived tissue. The experimental concentration and surrounding biological conditions were also relatively idealized, and the study did not compare the sensor with pathological datasets or calculate clinical diagnostic accuracy. The demonstrated finding is narrower: two terahertz parameters separated normal intestinal fibroblasts, myCAFs and apCAFs in controlled cell cultures without molecular labels. Substantial additional biological and clinical validation would be required before this strategy could be considered for routine detection.

Subject of Research: Terahertz metasurface sensing of colorectal cancer-associated fibroblast subtypes

Article Title: Screening colorectal tumor-associated fibroblasts using dual-channel THz biosensor

Article References: Xu, H., Gu, R., Serville Torvund, A. G., Grønlien, I., Xomalis, A., Nguyen, H. S., Yu, Z., & Zhao, Z. (2026). Screening colorectal tumor-associated fibroblasts using dual-channel THz biosensor. iScience, 29(11), Article 117735. https://doi.org/10.1016/j.isci.2026.117735

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117735

Keywords: terahertz biosensor, colorectal cancer, cancer-associated fibroblasts, quasi-BIC resonance, electromagnetically induced transparency, label-free sensing, tumor microenvironment, dielectric response, metasurface, myCAFs, apCAFs, Screening

Chikungunya virus is an enveloped, positive-sense single-stranded RNA alphavirus transmitted primarily by Aedes aegypti and Aedes albopictus mosquitoes. While mortality remains low, generally below 0.1 percent, the broader public health burden arises from severe polyarthralgia that can persist for months or even years in a substantial fraction of patients across tropical and subtropical regions. With no approved antiviral therapy or licensed vaccine currently available, diagnostic accuracy sits at the center of clinical management and transmission control. Yet existing diagnostic tools face significant practical constraints, particularly in point-of-care and field settings.

Serological methods such as enzyme-linked immunosorbent assay suffer from cross-reactivity with related alphaviruses, including Ross River and O’nyong-nyong viruses, reducing diagnostic specificity. Real-time reverse transcription polymerase chain reaction offers high sensitivity during the acute phase but depends on laboratory infrastructure, trained personnel, and controlled sample handling. Virus isolation in cell culture provides confirmatory results but requires biosafety level-3 facilities and extended processing time. These limitations have motivated the search for rapid, label-free sensing approaches capable of operating at clinically relevant concentrations in decentralized environments.

The new work turns to the terahertz band, spanning roughly 0.1 to 10 THz, which lies between microwave and infrared regions of the electromagnetic spectrum. Terahertz radiation is non-ionising, with photon energies in the millielectronvolt range, allowing interaction with biological samples without causing ionising damage. Crucially, terahertz waves couple to low-frequency vibrational and rotational modes of biomolecules such as proteins, nucleic acids, and polysaccharides, producing measurable spectral features that support label-free detection. Because biological media including blood plasma and viral suspensions exhibit refractive-index variations in this domain, differences on the order of 0.01 to 0.02 refractive-index units can shift resonance frequency and field distribution when coupled to resonant sensing structures.

The proposed sensor is built around a periodically repeated square unit cell containing two identical square resonators, 2000 nanometers on a side, coated with tungsten disulfide—a two-dimensional transition metal dichalcogenide with strong light–matter interaction, high carrier mobility, and excellent dielectric confinement. This layer localizes the electric field in the sensing region and increases the interaction volume between the resonant field and the surrounding analyte. A pair of concentric circular rings, 3200 and 3800 nanometers in diameter and functionalized with MXene, surrounds the resonators, enhancing electromagnetic coupling, promoting efficient charge transport, and strengthening plasmonic interactions between resonator elements. An outer gold ring, 6000 nanometers in diameter, serves as the main plasmonic resonator supporting strong localized surface plasmon resonances with low optical loss. The entire multi-resonant assembly is embedded within a monolayer graphene sheet measuring 14,000 by 14,000 nanometers, which acts as a broadband absorber and a high-mobility charge transport medium whose electrically tunable chemical potential allows dynamic, active control of the resonance characteristics. Copper elements woven into the architecture provide supplementary conductive pathways that reinforce current distribution and coupling among the hybrid resonators. The structure rests on a silicon dioxide substrate, chosen for its low dielectric loss and mechanical robustness, which preserves the resonance quality factor.

The authors stress that the multilayer design is compatible with established microfabrication techniques. Fabrication would begin with a cleaned silicon dioxide substrate, followed by polymer-assisted transfer of monolayer graphene and thermal annealing. Gold and copper resonators can be deposited through electron-beam evaporation or magnetron sputtering with lift-off patterning, MXene rings through solution-based deposition, and the tungsten disulfide layer through chemical vapor deposition or thin-film transfer. With a unit-cell dimension of approximately 14 micrometers, the required feature sizes fall comfortably within conventional photolithography capabilities, and because fabrication imperfections remain far smaller than the operating terahertz wavelength, moderate tolerances are expected to have only limited influence on the electromagnetic response. For biosensing, the graphene surface could subsequently be functionalized with chikungunya-specific antibodies or aptamers using EDC/NHS or pyrene-based linker chemistries.

The electromagnetic modeling, performed in COMSOL Multiphysics, reveals a pronounced resonance at 0.355 THz, where electric-field simulations show the strongest and most spatially extensive enhancement. Varying the graphene chemical potential from 0.1 to 0.9 electronvolts produced dramatic modulation of the transmission contrast, which increased roughly 36-fold at the highest bias, from less than one percentage point to more than 27 percentage points, while peak transmission remained essentially unchanged above 98.6 percent. This asymmetric response—stable maxima paired with strongly modulated minima—arises because increasing chemical potential enhances graphene’s intraband conductivity, deepening resonant absorption through stronger light–matter interaction. The sensor also demonstrated remarkable angular robustness, maintaining peak transmittance above 97.8 percent across incident angles from 0 to 80 degrees, an important tolerance for realistic deployment where normal incidence cannot always be guaranteed.

Geometric optimization identified an outer circular resonator radius of 2500 nanometers and a square resonator side length of 2500 nanometers as the dimensions that maximize resonance strength, driving minimum transmittance down to approximately 65 to 67 percent and signifying optimal electromagnetic confinement within the active sensing region. When evaluated across refractive indices from 1.33 to 1.39, a range encompassing reported values for healthy plasma (approximately 1.35 RIU), healthy platelets (1.39 RIU), chikungunya-infected plasma (1.33 RIU), and infected platelets (1.38 RIU), the optimized configuration achieved a maximum sensitivity of 907 GHz/RIU, a figure of merit of 18.519 RIU⁻¹, a minimum detection limit of 0.061 RIU, a sensor resolution of 55.178, and a refractive-index resolution of 0.001 RIU. In comparative benchmarks against recently reported graphene- and hybrid-material-based terahertz biosensors—including graphene/copper/aluminum, graphene/silicon dioxide, MXene–graphene–black phosphorus, and graphene-integrated copper–silver designs—the proposed architecture’s sensitivity exceeded most published values, reflecting the synergistic field confinement produced by the five-material stack.

The machine-learning component addresses one of the central bottlenecks in metasurface development: the computational cost of repeated full-wave finite-element simulations during structural optimization. The team trained an XGBoost regressor, an ensemble method based on gradient-boosted decision trees, on simulation data covering incidence-angle variation and circular-ring dimension variation, using stratified sampling with 80 percent of data for training and 20 percent held out for independent testing, alongside five-fold cross-validation and grid-search hyperparameter optimization. The model captured more than 99.7 percent of response variance on unseen data, with mean absolute percentage errors below 0.162 percent and prediction accuracies from 99.838 to 99.959 percent. Acting as a computational surrogate, the trained model enables rapid interpolation across the design space without sacrificing the fidelity of the underlying electromagnetic model.

The authors are careful to frame the work as a computational proof-of-concept rather than an experimentally validated diagnostic. The sensor detects changes in the local dielectric environment surrounding the resonators, not viral particles or nucleic acids directly, and refractive-index sensing alone cannot intrinsically distinguish chikungunya from other viruses with similar dielectric properties. Analytical specificity in practice will depend on surface biofunctionalization with chikungunya-specific recognition molecules immobilized on the graphene or gold sensing surface. Future work will focus on device fabrication, biofunctionalization, calibration with reference refractive-index standards, experimental validation with clinically relevant plasma and platelet samples, and systematic evaluation of selectivity against related arboviruses—steps that would transform this numerically validated platform into a practical tool for rapid, label-free infectious disease diagnostics in the field.