Rapid screening of animal coronaviruses in livestock and wildlife using a silicon photonics biosensor

Abstract

Coronaviruses (CoVs) pose a persistent global health threat, with animal reservoirs such as bats, rodents, and poultry driving their emergence. Rapid, field-deployable detection tools are urgently needed to monitor these viruses at the animal-human interface, where spillover events are most likely. We present a nanophotonic interferometric biosensor for label-free, amplification-free detection of alpha/beta- and gamma-CoVs in animal reservoirs. The biosensor achieves limits of detection below 10 copies/μL for in vitro transcribed (IVT) RNA samples, strongly correlating with PCR assays. Field validation using RNA extracts from bats and chickens yielded sensitivities of 72–90% and specificities of 83–90%, confirming robust performance across diagnostic relevant ranges. Fabricated on silicon microchips via standard techniques, the biosensor platform is scalable, cost-effective, and compatible with miniaturized components suitable for point-of-care and decentralized use. Our biosensor device addresses a critical gap in veterinary diagnostics, supporting early surveillance and pandemic preparedness under a One Health framework.

Introduction

Globalization, intensified trade, and the expansion of industrial livestock production have markedly increased interactions between humans, domestic animals, and wildlife. Currently, these interactions are changing at an unprecedented rate, modifying the structure and functionality of our ecosystems. This has consequences not only on biodiversity loss but also on human health. The interconnected ecosystem dynamics have facilitated the emergence and spread of zoonotic pathogens, posing significant risks to global public health. Intensive farming practices, characterized by high animal densities due to confined conditions, contribute to accelerate viral amplification and cross-species transmission. Simultaneously, habitat encroachment and wildlife trade create novel interfaces where pathogens can spill over into human populations. A key example of zoonotic threat is posed by coronaviruses, known for their adaptability and capacity to spark large-scale human epidemics^1,2.

Coronaviruses (CoVs) are a diverse group of enveloped, positive-sense single-stranded RNA viruses that infect a broad range of avian and mammalian hosts. They are classified into four genera: Alphacoronavirus and Betacoronavirus, which predominantly infect mammals, and Gammacoronavirus and Deltacoronavirus, which primarily infect birds but can also affect some mammals³. Several coronaviruses have demonstrated zoonotic potential, notably severe acute respiratory syndrome coronavirus (SARS-CoV), Middle East respiratory syndrome coronavirus (MERS-CoV), and the recently emerged SARS-CoV-2, all of which originated in animal reservoirs and subsequently crossed into human populations. These viruses possess high recombination rates and genetic plasticity, which facilitate adaptation to new hosts and environments, enhancing their potential for interspecies transmission⁴. Animal reservoirs, particularly bats, civets, camels, and various livestock species, have been identified as sources or intermediate hosts for potentially zoonotic coronaviruses, underscoring the complex ecology and transmission dynamics of these pathogens⁵. Given these threats, systematic surveillance of coronaviruses in both livestock and wildlife reservoirs is critical to early detection and prevention of potential epidemics. Reliable, rapid, and field-deployable diagnostic tools are urgently needed to monitor reservoir animal populations and mitigate the risk of future zoonotic spillover events.

Conventional nucleic acid amplification tests, such as reverse transcription-polymerase chain reaction (RT-PCR), remain the gold standard for coronavirus genome detection due to their high sensitivity, specificity, and ability to accurately identify viral genetic material⁶. However, despite their analytical robustness, these methods present notable limitations when applied to time-critical surveillance and large-scale screening of animal populations. RT-PCR assays rely on multistep workflows, including RNA extraction reverse transcription, amplification, and downstream data analysis, each contributing to cumulative processing time and requiring skilled personnel and rigorous quality control. As a result, even under optimal laboratory conditions, RT-PCR typically entails turnaround times of several hours from sample receipt to result availability. Moreover, these assays necessitate well-equipped, centralized laboratory infrastructure with stable power supply, specific high-grade reagents, and sophisticated instrumentation, which are often unavailable in field or resource-limited settings. Sample transport to laboratories further introduces time delays, potential degradation of viral RNA, and additional logistical burdens. These factors collectively limit the feasibility of RT-PCR for rapid, high-throughput, and on-site surveillance of coronaviruses in livestock and wildlife populations, where timely detection is critical for outbreak prevention and containment⁷.

Consequently, there is a pressing need for alternative diagnostic approaches that combine high sensitivity and specificity with operational simplicity, rapid turnaround times, and ideally, suitability for decentralized or field deployment. Label-free biosensing technologies, particularly those based on photonics, offer a promising solution by enabling direct, real-time detection of viral targets without the need for amplifications or complex sample processing. In this work, we introduce a nanophotonic biosensor technology for rapid, accurate detection of alpha/beta-CoVs and gamma-CoVs directly from animal samples, delivering results in under 20 min and with potential for decentralized operation. Our biosensor is based on proprietary bimodal waveguide (BiMW) interferometric technology, a silicon photonics microchip fabricated through scalable cost-effective techniques, enabling parallel analysis of different samples with outstanding sensitivities (Fig. 1)^8,9. The BiMW microchip contains an array of 20 silicon nitride (Si₃N₄) rib waveguides acting as independent sensors. Briefly, each waveguide supports the propagation of two optical modes - the fundamental mode and the first-order mode - that travel along the same waveguide structure but with different profiles. When a biological recognition event occurs on the sensor surface, such as the specific binding of target molecules to immobilized capture probes, it induces a local refractive index change in the waveguide surface. This alteration differentially affects the effective refractive indices of the two guided modes, generating a measurable phase shift between them. The interference pattern resulting from the superposition of the modes is highly sensitive to minute variations in the surface refractive index, allowing for real-time, label-free detection of biomolecular interactions with high precision. By monitoring shifts in the interference signal, the BiMW biosensor can quantitatively assess target analyte concentrations across multiple independent channels, enabling multiplexed and rapid diagnostics suitable for large-scale screening applications¹⁰. The BiMW biosensor has already established itself as a state-of-the-art technology for amplification-free detection of proteins and nucleic acids, reaching sensitivities in the aM-fM range^11,12,13,14, and demonstrating clinical relevance in both bacterial and viral diagnostics. Previous studies validated its ability to identify bacterial pathogens at ultralow concentrations and to directly detect antimicrobial resistance genes without amplification^15,16, as well as to quantify SARS-CoV-2 viral loads with high precision¹⁷. While these works consolidated BiMW as a mature and powerful photonic sensing platform, they were predominantly oriented toward human health applications, reflecting a broader trend in the biosensing field.

Fig. 1: Conceptual illustration.

Schematic illustration of the silicon photonics interferometric biosensor for coronavirus identification in livestock and wildlife animals.

In contrast, the present study represents a substantive conceptual expansion of BiMW technology, extending photonic biosensing into the largely unexplored domain of animal health surveillance. Despite the critical role of livestock and wildlife as reservoirs and amplifiers of zoonotic viruses, biosensor-based diagnostics have been overwhelmingly developed for human pathogens, leaving animal virology underrepresented in advanced sensing strategies. Addressing this gap, we introduce for the first time a photonic-based approach specifically designed for direct genomic detection of animal coronavirus, encompassing alpha-, beta-, and gamma-CoV genera. Given the high mutation rates inherent to CoV RNA genomes, particularly in genes encoding external structural proteins such as the spike (S) protein¹⁸, we targeted conserved genomic regions to ensure reliable detection. Specifically, the BiMW biosensor was designed to recognize sequences within the RNA-dependent RNA polymerase (RdRp) and nucleocapsid (N) genes, which display lower variability across different CoV strains¹⁹. This strategy overcomes the limitations of antibody-based biosensors, whose performance can be compromised by antigenic drift, and enables sensitive and specific genomic detection of alpha-, beta-, and gamma-CoVs. To our knowledge, this study constitutes the first demonstration of a photonic biosensor expressly developed for the direct detection of animal coronavirus. By translating high-performance nanophotonic biosensing into the context of animal health, this work establishes a new application space for the technology and highlights its potential as a transformative tool for early disease detection at the human-animal interface, with direct implications for outbreak prevention, food security, and pandemic preparedness.

Results

Sensor biofunctionalization and DNA/RNA assay optimization

To enable broad coronavirus detection across different host species, we designed two parallel BiMW biosensor systems, each targeting conserved genomic regions specific to mammalian and avian CoVs. On one hand, we designed a BiMW biosensor targeting the RdRp region, specifically the non-structural protein 12 (nsp12), for the detection of alpha-/beta-CoVs (i.e., α/β-CoV assay). This approach aligns with the semi-nested PCR protocol described by Gouilh et al.²⁰ widely used for detection and monitoring of coronaviruses in mammals. For the gamma (γ)-CoV assay, on the other hand, we targeted the 5’ untranslated region (5’UTR) as described by Callison et al.²¹ that enables identification of most common coronaviruses in birds. A 50 nucleotide (nt) RNA sequence target was defined for each case, together with the corresponding complementary single-stranded DNA probes for direct hybridization assay development (Table 1). The DNA probes were synthesized by incorporating a vertical spacer (C₆) and a thiol (-SH) reactive group in the 5’ terminal region, which enables the chemical attachment to the sensor surface. The vertical spacer is intended to enhance the mobility of immobilized probes and improve their accessibility to complementary target sequences²². It also distances the DNA probe from the sensor surface, minimizing non-specific adsorptions and steric hindrance.

Table 1 Target RNA and DNA probe sequences

For BiMW sensor biofunctionalization, the silicon-based BiMW chips were initially modified by silanization with (3-aminopropyl) triethoxysilane (APTES) using a vapor-phase deposition protocol²³. This process generated a uniform and stable monolayer of primary amine (NH₂) groups on the sensor surface, providing reactive sites for subsequent probe immobilization. Covalent attachment of the thiolated single-stranded DNA (ssDNA) probes was achieved through the application of the homobifunctional crosslinker 1,4-phenylene diisothiocyanate (PDITC), which forms stable thiourea linkages between the amine-functionalized surface and the thiol-modified probes (Fig. 2a)¹⁴. The DNA probe immobilization step was monitored through real-time BiMW sensorgrams, confirming an efficient attachment of the oligonucleotides to the sensor surface, with a significant increase of the signal, Δϕ = 52.2 ± 3.8 rad (Fig. 2b). The biofunctionalization strategy proved reproducible and repeatable across different waveguide sensors and chips, with a coefficient of variation (CV) of 7.20%.

Fig. 2: Optimization of sensor biofunctionalization.

a Schematic illustration of the sensor surface biofunctionalization with APTES silanization, PDITC crosslinking, and DNA probe immobilization mixed with lateral spacer (OH-PEG-SH); b Representative real-time BiMW sensorgrams of DNA probe immobilization in different waveguides of the BiMW sensor chip over 2500 s (40 min). The dotted line represents the 0 value of the sensorgram. The Δɸ indicates the phase shift in radians (rad) after the immobilization onto the sensor surface. Biosensor assay optimization: c RNA target signals obtained with different DNA probe concentration and spacer ratio; d Synthetic RNA target for α/β-CoV assay (blue histograms) and negative control (miR-180, light gray histograms) signals obtained with different formamide percentage (FA at 1, 5, and 10%); e Representative sensorgram of RNA target detection (50 nM concentration) followed by biosurface regeneration using FA at 35%.

During the development of the DNA/RNA hybridization biosensor assay, critical conditions were systematically optimized to ensure maximal hybridization efficiency and enhanced sensor performance. Key studied parameters included the concentration of the immobilized DNA probes and the introduction of lateral spacer molecules, which together influence probe density, accessibility, and sensor surface passivation. To optimize probe density and enhance hybridization efficiency, polyethylene glycol (PEG)-based spacers (SH-PEG₅₀₀-OH) were co-immobilized alongside the DNA probes. The incorporation of these large, hydrophilic molecules improves the spatial distribution of the probes, reduces steric hindrance, and minimizes non-specific adsorptions, thereby increasing the accessibility of the complementary target sequences¹⁴. This mixed monolayer strategy promotes a more favorable orientation and conformational flexibility of the biorecognition probes, ultimately enhancing biosensor performance. To evaluate the impact of probe density on the hybridization efficiency of the α/β-CoV assay, the DNA probe concentration and lateral spacer ratios were varied. A range of probe concentrations (1–10 µM) were tested, and mixed at different ratios (1:0, 1:1, 1:2, and 1:5) with the lateral spacer. As can be observed in Fig. 2c at a fixed concentration of 1 µM, the 1:0 and 1:1 (probe:spacer) ratio showed a low hybridization efficiency, likely due to the lack of accessibility to the RNA target sequence. Increasing the concentration to 2 µM and the spacer content to a 1:1 or 1:2 ratio greatly improved the hybridization response, suggesting an enhanced probe accessibility and an optimal surface coverage. However, further increasing the concentration or the spacing led to a reduction in the sensor signal, indicating that excessive spacing between probes decreased the probability of target binding. A probe/spacer ratio of 1:2 with a probe concentration of 2 µM was found to provide the optimal balance between probe coverage and spatial separation, enhancing target binding and minimizing steric hindrance.

Based on previous studies, we selected 5× SSC (50 mM saline sodium citrate) as the hybridization buffer²⁴. To further enhance hybridization efficiency, we evaluated the addition of formamide (FA), a well-established additive known to reduce the formation of secondary structures in target RNA¹². By destabilizing hydrogen bonds, formamide increases the accessibility of the target sequence, thereby facilitating more efficient hybridization. Incorporating 1% FA into the SSC buffer resulted in an increase of the sensor signal up to 52.38% compared to SSC alone (Fig. 2d). However, at higher FA concentrations, signal intensity slightly decreased, and non-specific adsorption of a negative control (i.e., a 22 nucleotide non-complementary RNA sequence, see Methods section) was observed to increase. In addition, a sensor surface regeneration protocol was implemented by injecting a 35% FA solution after each detection cycle. The high concentration of FA effectively disrupted the DNA/RNA duplexes and removed bound target molecules, enabling the reuse of the biosensor (Fig. 2e).

Alpha/beta-CoV and gamma-CoV biosensors characterization

Once the biosensor hybridization assay was optimized, standard calibration curves were generated for the direct detection of both α/β-CoV and γ-CoV RNA targets. Serial dilutions of synthetic RNA targets, spanning a concentration range of 0.5–100 nM, were sequentially introduced to the corresponding DNA-functionalized biosensor surfaces. Sensor signals were recorded in real-time, with the assay exhibiting a total turnaround time of 25 min (Fig. 3). As expected, increasing concentrations of the target RNA resulted in a proportional increase in the sensor response, demonstrating the capability of the BiMW biosensor for precise RNA quantification. From these standard calibration curves, we calculated the limit of detection (LOD) and limit of quantification (LOQ) for both biosensor assays. The α/β-CoV biosensor showed an LOD of 38 pM and a LOQ of 186 pM (R² = 0.90, Kd = 31.2, CI 95%), with a coefficient of variability around 23%, while the γ-CoV biosensor showed a LOD of 91 pM and a LOQ of 305 pM (R² = 0.96, Kd = 72.0, CI 95%), with a coefficient of variability ranging from 1.59% for lower concentrations to 27% for higher concentrations.

Fig. 3: BiMW characterization with synthetic RNA targets.

a Standard calibration curve using a synthetic RNA target at concentrations 0.5, 5, 10, 25, 50 and 100 nM specific for the α/β-CoV biosensor assay. Dots represent the mean and standard deviation of sensor signals of triplicates; b Representative sensorgrams of the synthetic RNA targets for the α/β-CoV assay; c Standard calibration curve using a synthetic RNA target at concentrations 0.5, 5, 10, 25, 50 and 100 nM specific for the γ-CoV biosensor assay. Dots represent the mean and standard deviation (barely visible) of sensor signals of triplicates; d Representative sensorgrams of the synthetic RNA targets for the γ-CoV assay. Note: The top portions of the curves in (b) and (d) are truncated to provide a zoomed view of the sensor response.

While the results obtained were promising, further work was required to fully evaluate the BiMW biosensor performance with more complex RNA targets. The experiments already conducted have utilized synthetic 50 nt target sequences, which are relatively short compared to real RNA samples, typically around 700 nt in length²⁵. This difference in target size can influence both the sensitivity and the hybridization efficiency of the biosensor. Larger RNA targets offer a larger change in refractive index upon hybridization, potentially leading to enhanced sensor sensitivity. However, the increased size also introduces steric hindrance, which may affect hybridization efficiency by limiting the accessibility of target sequences to the immobilized probes.

To assess the performance of the BiMW biosensor with longer RNA targets, we performed an additional experiment using in vitro transcribed (IVT) RNA for the γ-CoV assay. Serially 10-fold diluted samples (from 10⁴ to 10⁻² cp/µL) were characterized by real-time (rt) RT-PCR²¹ to check their positive/negative status and to confirm the linearity of the relationship between RNA concentrations and the measured Ct values.

Same samples were then evaluated with the BiMW biosensor and the obtained responses were compared to the viral load concentrations (Fig. 4a). As can be observed, no significant response is observed from 10⁻² to 10 copies/μL, corresponding to rtRT-PCR negative samples, whereas a clear hybridization response appears from 10 copies/μL onward. The biosensor signal then increases progressively up to 10⁴ copies/μL, demonstrating a good correlation with the IVT RNA concentration and the rtRT-PCR results. At higher concentrations, the biosensor response becomes inconsistent and deviates from linearity (data not shown), likely due to surface saturation or steric hindrance effects. To further investigate this behavior, the sensor signals were plotted as a function of IVT RNA concentration and fitted to a classical one-site specific binding model. As shown in Fig. 4b, the signal clearly begins to plateau at 10⁴ copies/μL, confirming partial saturation of the sensor surface, a phenomenon commonly observed in biosensors and immunoassays, where high target concentrations reduce the measurable signal by limiting the accessibility of bioreceptor sites²⁶. From the calibration curve, the detection limit of the biosensor for the γ-CoV assay with IVT RNA samples was determined to be 1.52 copies/µL, with a limit of quantification of 9.4 copies/µL, corresponding to the onset of a clearly detectable and quantifiable biosensor signal. Finally, to further validate the analytical performance of the biosensor, the sensor responses were plotted against the corresponding rtRT-PCR Ct values for the same IVT RNA samples. A linear regression analysis of these data yielded a strong inverse correlation, with a coefficient of determination (R²) of 0.95 (Fig. 4c).

Fig. 4: BiMW characterization with IVT RNA samples.

a BiMW biosensor responses (orange) to serial 10-fold dilutions of γ-CoV IVT RNA (from 10⁻² to 10⁴ copies/μL) compared with rtRT-PCR results (purple); b Calibration curve of biosensor signal versus RNA concentration fitted to a one-site specific binding model; c Benchmarking of the biosensor against rtRT-PCR. All sensor signals are measured in duplicate.

Finally, to further validate the BiMW biosensor in biological samples, a pooled swab sample from chickens infected with a common γ-CoV (i.e., infectious bronchitis virus, IBV) was quantified by rtRT-PCR and serially diluted for analysis. Sensor responses at 10³ and 10¹ cp/μL, corresponding to the estimated LOQ, were measured and compared with the IVT RNA samples at the same concentration (Fig. 5). The results show that RNA extracted from swabs produces sensorgrams comparable to those of IVT samples, with no evidence of significant matrix effects or interferences. These data confirm that the biosensor can reliably detect low concentrations of CoV RNA in extracted animal samples, supporting the validity of the reported sensitivity estimates.

Fig. 5: Validation of biosensor sensitivity.

a BiMW biosensor responses to IVT RNA sample (purple) and IBV-infected chicken swab RNA sample (orange) at around 10 cp/μL; b responses to the same samples at around 10³ cp/μL. Note: The overall downward shift of the sensorgrams, in contrast to the upward response observed in Fig. 3, is due to refractive index differences between the CoV RNA sample matrices and the running buffer.

Biosensor validation with animal samples

For the final biosensor validation of both CoV assays, a panel of 74 animal samples, including specimens from wild and domestic animals, was collected and characterized through RNA extraction and PCR analysis. The α/β-CoV biosensor was applied to evaluate 46 bat feces samples from the panel (Fig. 6a), a species recognized as primary reservoirs of alpha- and beta-coronavirus with zoonotic potential. The semi-nested RT-PCR assay, developed by Gouilh et al.²⁰, identified 34 positive and 12 negative specimens, while the biosensor identified 25 positive and 19 negative specimens, and 2 indeterminate samples, yielding 2 false-positive and 9 false-negative results. Using the predefined gray zone, indeterminate outcomes represented 4.30% of the analyzed bat samples. These discrepancies may be attributed to multiple factors. The semi-nested RT-PCR assay, while highly sensitive, is not quantitative and involves two rounds of amplifications, which can lead borderline samples being classified as positive even when the absolute target copy number is very low. Besides, the α/β-CoV biosensor had not been previously characterized or fully optimized with IVT RNA samples, and its sensitivity may be limited for samples with low viral loads. Nonetheless, biosensor measurements showed a statistically significant difference between positive and negative samples (Kruskal–Wallis test, p < 0.0001) (Fig. 6b). From these data, diagnostic performance metrics were determined, yielding a sensitivity (SE) of 72%, specificity (SP) of 83%, positive predictive value (PPV) of 92%, and negative predictive value (NPV) of 53%. Recalculation of diagnostic metrics under extreme scenarios in which all indeterminate samples were classified either as positive or as negative resulted in only minimal changes in sensitivity and specificity (data not shown). Receiver operating characteristic (ROC) analysis further demonstrated the discriminatory capability of the method, with an area under the curve (AUC) of 0.76 (95% CI: 0.64–0.89; Fig. 6c).

Fig. 6: Validation of the BiMW biosensors with animal samples.

a Sensor signals for alpha/beta-CoV RNA extracted from bat feces samples (34 positive (blue) and 12 negative (gray), according to RT-PCR), b signal distribution with Kruskal–Wallis test, and c statistical ROC analysis; d Sensor signals for gamma-CoV RNA extracted from chicken swab samples (17 positive (orange) and 9 negative (gray), according to rtRT-PCR), e signal distribution with Kruskal–Wallis test, and f statistical ROC analysis.

On the other hand, the γ-CoV biosensor was validated using chicken samples, as representative species for coronavirus-related animal health concerns in livestock. A set of 26 swab samples, including tracheal and cloacal specimens, were analyzed with the real-time RT-PCR developed by Callison et al.²¹, which identified 17 positives and 9 negatives. Subsequent evaluation with the biosensor resulted in 15 positive and 10 negative specimens, and 1 indeterminate sample, yielding 1 false-positive and 2 false-negatives (Fig. 6d). This indeterminate result corresponded to 3.85% of the tested chicken samples and similarly had a negligible impact on performance metrics when classified as positive or negative. In general, samples with higher concentrations (cp/µL), characterized by rtRT-PCR generated higher sensor signals, with the response saturating at around 1 × 10⁵ cp/µL. The biosensor successfully detected positive samples across a wide concentration range of 1 × 10² to 1 × 10⁶ cp/µL. The 2 false negative results can be explained by the fact that these samples presented a concentration around 1 × 10⁵ copies/µL, where the biosensor signal reaches saturation. These results showed strong overall agreement, with statistically significant differences (Kruskal–Wallis test, p < 0.0001) (Fig. 6e) and an AUC of 0.92 (95% CI: 0.8029–1.000) (Fig. 6f). The diagnostic performance for γ-CoV detection showed a sensitivity of 88%, specificity of 89%, PPV of 93%, and NPV of 80%. These results demonstrate that the BiMW biosensors provide robust detection capabilities for both mammalian and avian coronaviruses directly from animal samples, without the need for nucleic acid amplification.

Discussion

This study presents a novel application of the BiMW biosensor platform for the rapid, label-free detection of alpha/beta- and gamma-coronaviruses in animal reservoirs, including bats and chickens.

To adapt the biosensor for coronavirus detection, we developed a surface biofunctionalization protocol and optimized the hybridization assay to enhance sensitivity. These improvements enabled the detection of viral RNA targets with excellent limits of detection: 38 pM for α/β-CoV and 91 pM for γ-CoV, respectively. The analytical sensitivity achieved by the BiMW biosensor represents a significant advancement, surpassing the scarce state-of-the-art label-free optical biosensors recently reported in the literature for genomic-based viral detection²⁷. For instance, Courtney et al. developed a waveguide-based optical biosensor for direct PCR-free detection of influenza RNA. Their technology achieved a LOD of 50 nM, although they required the use of molecular beacons for signal amplification²⁸. In another example, Koo et al. utilized a microring resonator biosensor for multiplexed genomic identification of different respiratory viruses from nasopharyngeal swab samples. Their methodology proved higher sensitivity than conventional RT-PCR in only 20 min biosensor assay, but requiring previous isothermal RNA amplification⁶. Chen et al. developed a plasmonic biosensor integrating CRISPR/Cas12a-based detection for SARS-CoV-2 Omicron RNA. Thiolated ssDNA probes were immobilized on the gold surface, and the target binding activates Cas12a inducing a collateral cleavage of ssDNA reporters, originating a shift in resonance angle. The assay achieved a LoD of 15 fM in 38 min, without requiring amplification²⁹. However, the methodology is biochemically complex and slow, which may limit its robustness and applicability at point-of-care settings. Our BiMW biosensor thereby distinguishes itself by enabling direct, label-free detection of viral RNA with high sensitivity, without the need for pre- or post-amplification steps, enzymatic reactions, or signal labeling, making it more suitable for decentralized point-of-care testing.

To further characterize the performance of our biosensor, validation studies were conducted using in vitro transcribed (IVT) γ-CoV RNA, showing strong correlation between BiMW signals and quantitative RT-PCR results. The estimated LOD of 1.52 copies/µL and LOQ of 9.4 copies/µL for the γ-CoV biosensor demonstrate that the biosensor achieves a sensitivity comparable to conventional molecular methods, while providing the advantages of rapid and label-free detection³⁰. This validation also highlights the platform versatility across a broad range of RNA targets, reliably detecting sequences as short as 50 nt (synthetic targets) and up to 700 nt (IVT RNA). Signal quantification becomes less reliable at very high RNA concentrations due to steric hindrance and surface saturation, an inherent limitation of surface-based biosensors, though this does not affect the binary detection of positive versus negative samples. These findings underscore the potential of the BiMW biosensor not only for sensitive viral detection but also for flexible applications in RNA analysis, offering a promising alternative or complementary approach to amplification-based techniques in both research and diagnostic settings.

Finally, a comprehensive field validation was performed on 72 RNA extracts from bat feces and chicken cloacal and tracheal swabs. The α/β-CoV biosensor assay yielded a diagnostic sensitivity of 72% and specificity of 83% in bat samples, while the γ-CoV biosensor assay showed 88% sensitivity and 89% specificity in chicken swabs. Indeterminate results for both cases were infrequent (<5%) and had negligible impact on diagnostic performance when classified as positive or negative. The performance observed for the α/β-CoV bat panel is modest relative to ideal screening benchmarks and reflects the intrinsic difficulty of detecting highly diverse and low-abundance bat CoVs in wildlife specimens. This lower performance can be attributed to the more stringent benchmarking against a semi-nested PCR assay and the larger number of samples tested compared with the γ-CoV biosensor. Importantly, bat CoV surveillance frequently relies on nested or semi-nested PCR protocols precisely because standard single-round RT-PCR often lacks sufficient sensitivity in this context. In this regard, the results obtained with the BiMW biosensor are consistent with the known limitations of molecular diagnostics applied to complex wildlife samples and support its role as a complementary analytical tool for surveillance and monitoring of these viruses.

Designed for high-precision diagnostics, the BiMW biosensor provides a reproducible and scalable nanophotonic platform for rapid, sensitive, and reproducible detection of viral nucleic acids. A key advantage of this technology lies in its short analytical turnaround time, completing the detection assay within 10–15 min once the RNA is available, in contrast to conventional RT-PCR workflows that typically require several hours from sample processing to result generation. The BiMW sensor is fabricated on silicon-based microchips using standard microelectronics manufacturing techniques, ensuring high reproducibility and compatibility with scalable, cost-effective production. These characteristics support its suitability for high-throughput laboratory-based screening and surveillance applications, where rapid result delivery is essential for timely decision-making. At present, the BiMW biosensor requires operation within a laboratory environment, as upstream sample preparation steps and optical instrumentation require controlled conditions. Nevertheless, its fast readout and label-free operation already provide substantial gains in analytical responsiveness compared to amplification-based methods. By decoupling analytical sensitivity from enzymatic amplification, the BiMW platform reduces assay complexity and processing time at the detection stage, positioning it as a complementary technology to RT-PCR for time-sensitive surveillance applications. Importantly, ongoing developments within our research group are directed toward the integration and automation of both the multiplexed biosensor chip and the associated optical and fluidic components. These efforts aim to consolidate the full detection workflow into a compact and user-friendly system, paving the way for future implementations that could operate closer to the point of need while preserving the analytical performance demonstrated in the current laboratory-based configuration, with future studies planned to evaluate performance directly in complex matrices such as spiked feces or swabs.

This work also addresses a significant gap in the current landscape of diagnostic biosensors. Although the importance of animal health within the One Health framework is increasingly recognized, most biosensor development remains heavily focused on human clinical applications. Only a limited number of biosensor platforms have been tailored or validated for use in veterinary settings, despite the critical role that animal reservoirs play in the emergence and transmission of zoonotic diseases. In this context, our biosensor provides a much-needed technological advance for the early detection and monitoring of viral pathogens, including those at the animal-human interface.

Overall, the BiMW biosensor represents a powerful tool for global health diagnostics. Its speed, portability, and cost-effectiveness make it ideally suited for routine surveillance of livestock and wildlife populations. Implementing such technologies in farms, live animal markets, and other high-risk environments as complementary diagnostic techniques could significantly enhance our capacity to prevent future outbreaks by enabling earlier intervention. As part of an integrated One Health surveillance strategy, this biosensor platform could help close the current diagnostic gap between human and animal health, ultimately contributing to pandemic preparedness and improved biosecurity worldwide.

Methods

Chemical and biological reagents

Organic solvents, like acetone and ethanol, were obtained from PanReac AppliChem (Barcelona, Spain). (3-Aminopropyl)triethoxysilane (APTES), N,N’-phenylenedimaleimide (PDITC), dimethylformamide (DMF), pyridine, formamide (FA), reagents and salts used for the preparation of phosphate-buffered saline (PBS, 50 mM (750 mM NaCl, 33 mM Na₂HPO₄, 17 mM NaH₂PO₄, and 2 mM EDTA), pH 7.4) and saline-sodium citrate (SSC, 5× (75 mM SSC, 750 mM NaCl, 4 mM EDTA), pH 7.4) were purchased from Merck (Darmstadt, Germany). Thiolated DNA probes (SH-modified) and negative control (miR-180: CAG CAG CAA TTC ATG TTT TGA A) were synthesized and supplied by Ibian Technologies (Zaragoza, Spain), while synthetic RNA target sequences (α/β-CoV and γ-CoV) were obtained from Integrated DNA Technologies, IDT (Newark, New Jersey, USA). Diethyl pyrocarbonate (DEPC)-treated water was used for the preparation of all buffers to ensure RNase-free conditions. Lateral spacers SH-PEG-OH (MW 2000 g/mol) were purchased from Laysan Bio (Alabama, USA). Bond-Breaker™ TCEP Solution (Tris(2-carboxyethyl)phosphine hydrochloride solution) was purchased from ThermoFisher (Waltham, Massachusetts, USA). The in vitro transcribed (IVT) RNA was produced from Mass41 strain (belonging to the infectious bronchitis virus (IBV), a gamma coronavirus) to quantify RNA using a rtRT-PCR specific to poultry gamma coronaviruses²¹. The IVT RNA is about 700 bases long and was produced at 7 × 10¹¹ cp/µL using RiboMAX^TM Large Scale RNA Production Systems T7 kit (Promega). Residual DNA was treated by several DNase I (Qiagen) treatments. The production has been validated from 10⁷ cp/µL for further use. Then, 10-fold diluted samples from 10⁷ to 10⁻² cp/µL were prepared in RNase free water and used as standards.

Animal samples collection and characterization

A complete panel of 72 animal samples were collected and characterized for biosensor validation. The panel includes 46 bat fecal samples (34 positive and 12 negative) and 26 tracheal and cloacal samples from chickens (17 positive and 9 negative).

Bat samples were obtained from Spanish colonies in which alpha and beta CoVs were circulating. Total RNA was extracted using a commercial kit consisting of lysis buffer, ethanol, proteinase K and a silica membrane, following the protocol described by Monchatre-Leroy et al.³¹ RNA was eluted in 50 mL RNase-free water and subsequently analyzed by RT-PCR targeting RdRp CoV gene. The assay follows the protocol described by Gouilh et al.²⁰, with primers designed based on conserved regions of reference sequences obtained from GenBank (Betacoronavirus MG772933; Alphacoronavirus MG772934, MG772935, MG772936). The semi-nested design involves two amplification rounds: the first uses an outer primer pair to amplify the target region, and the second uses one original and one internal (“nested”) primer to increase specificity and sensitivity. Some positive samples were analyzed two times to confirm the result. The animal study followed the ethical recommendations of the Spanish Legislation (Law 42/2007). Bat sampling was authorized by Department of Climate Action, Food and Rural Agenda of the Generalitat de Catalunya, the Conselh Generau d’Aran and the Conselleria de Medi Ambient i Territori of the Govern de les Illes Balears. All manipulations were performed avoiding pain and in such a way as to minimize animal discomfort. The study was conducted in accordance with the ARRIVE guidelines (https://arriveguidelines.org).

Chicken samples were obtained from animal trials. They were conducted in animal facilities approved for animal experiments (C-22-745-1), in accordance with EU and French regulations on animal welfare, and followed a protocol (permit number APAFIS #38242-2022081917217949 v6) that was approved by the ethical committee ANSES/ENVA/UPEC CNREEA16 and authorized by the French Ministry for higher education. Two animal trials were conducted on specific pathogen-free (SPF) White Leghorn chickens (ANSES Ploufragan-Plouzané-Niort laboratory) to produce anti-IBV sera while monitoring the viral excretion over time. SPF chickens were inoculated by the intranasal route with a different IBV strain per each animal trial, at 33 days of age (0-day post-inoculation (dpi)). Tracheal and cloacal swabs were then collected individually and at regular intervals until the end of the protocol. RNA from swabs was extracted using NucleoMag VET kit (Macherey-Nagel) and the KingFisher^TM Flex 96 Purification System (ThermoFisher Scientific) following an adapted protocol previously described³². The presence of IBV was finally quantified by the poultry gamma-CoV-specific RT-PCR²¹. This assay is a real-time TaqMan RT-PCR targeting the 5′-UTR of the IBV genome, designed to amplify a 143-bp product. Primers and probe were based on the IBV reference sequence with GenBank accession number AY851295. For the purposes of this study, 10 samples (5 tracheal swabs and 5 cloacal swabs) from one animal trial were selected, all at 0 dpi, before the IBV inoculation. Their coronavirus-negative status was confirmed. Twenty samples (10 each) from a second animal trial were selected at 3, 7 and 14 dpi (n = 5, 8 and 7 respectively), all coronavirus-positive, with varying quantification levels (from 19.4 to 31.7 Ct, corresponding to 4.1E + 05 to 1.7E + 02 cp/µL).

BiMW technology and set-up

The BiMW sensors are fabricated at wafer-scale using standard microelectronics processes in the ICTS cleanroom facilities of the National Microelectronics Center (IMB-CNM-CSIC, Barcelona, Spain) as previously described^8,9. Each BiMW sensor chip (measuring 3 × 1 cm) incorporates an array of 20 independent silicon nitride (Si₃N₄) straight rib waveguides. These waveguides have a width of 3 μm, a rib height ranging from 1 to 3 nm, and are spaced 250 μm apart. The structure begins with a single-mode section featuring a core thickness of 150 nm, which supports propagation exclusively of the fundamental optical mode. This is followed by a step transition that increases the core thickness to 340 nm, enabling simultaneous excitation of both the fundamental and first-order modes in the bimodal region. A sensing window measuring 15 mm in length and 0.05 mm in width is precisely opened within this bimodal section to allow interaction with the surrounding medium for detection purposes³³.

The sensor chip is integrated into a five-channel polydimethylsiloxane (PDMS) microfluidic module, where each channel measures 1.25 mm in width and 500 μm in height. This configuration enables continuous flow of samples and automated washing procedures. A syringe pump provides a constant and controlled delivery of buffer solution, while a six-port injection valve, equipped with a 150 μL sample loop, facilitates the sequential introduction of samples and reagents with high precision.

The BiMW interferometric sensor system utilizes a low-power, polarized diode laser operating at a wavelength of 660 nm (HL6545MG, Thorlabs). Optical coupling is achieved through edge injection of the laser light into the waveguide, while interferometric signal readout is performed using a dual-section photodiode detector (S4349, Hamamatsu Photonics). The sensor holder is mounted on a Peltier thermoelectric element (TEC3-2.5, Thorlabs), which maintains a highly stable operating temperature with a precision of ±0.01 °C. Control electronics comprised a benchtop photodiode amplifier (PDA200C, Thorlabs), a benchtop LD current controller (LDC205C, Thorlabs), a benchtop temperature controller (TED200C, Thorlabs), and a data acquisition card (USB-6361, National Instruments). Data processing is implemented through custom-developed software incorporating an advanced optical phase modulation technique. This approach converts complex interferometric outputs into a linear phase shift (Δφ), which is quantitatively correlated with variations in the local refractive index at the waveguide sensor surface. This mechanism enables real-time, label-free detection of biomolecular interactions with high sensitivity.

Sensor biofunctionalization and biosensor assay protocol

BiMW sensor chips were sequentially sonicated for 10 min each in acetone, isopropanol and water. Surface activation was performed by 5 min of oxygen plasma treatment, followed by a 20-min immersion in piranha solution (3H₂SO₄:1H₂O₂). Chips were thoroughly rinsed with water, dried under nitrogen, and placed in a vacuum desiccator with APTES for 5 min under vacuum, then incubated for 1 h. A curing step was performed at 120 °C for 20 min. Subsequently, the amine-functionalized sensor surfaces were incubated with PDITC in DMF for 2 h in the dark. For in-flow probe immobilization, a 2 μM DNA probe solution mixed with a lateral spacer (HO-PEG-S) was prepared in 50 mM PBS containing lateral spacers, NaCl, and EDTA at pH 7.4. The solution was incubated with 100 nM TCEP at 37 °C for 20 min to reduce disulfide bonds and ensure optimal probe orientation and surface attachment. The sensor chip was mounted in the experimental setup, and the immobilization solution was flowed in DEPC-treated water at 10 μL/min. To optimize the surface coverage and hybridization efficiency, different DNA probe concentrations (1, 2, 5 and 10 μM) and mixed with lateral spacer (HO-PEG-SH) in different ratios (1:0, 1:1, 1:2, and 1:5) were tested. The resulting solutions were used for in-flow immobilization following the procedure described above. Based on these experiments, the probe concentration and probe:spacer ratio previously described were identified as the optimal conditions and were used in all subsequent assays.

For the biosensor assays optimization, the synthetic RNA targets at concentrations ranging from 0.5 to 100 nM were injected over the biosensor surface at a flow rate of 10 μL/min, using 5× SSC buffer supplemented with 1% FA as the running buffer. This buffer, characterized by its high ionic strength, promotes the stabilization of probe–target duplexes by shielding electrostatic repulsion between negatively charged nucleic acid strands. Calibration curves were performed by measuring these synthetic RNA samples, analyzed in triplicate. For the negative control, a non-complementary synthetic RNA sequence was used. The in vitro transcribed (IVT) RNAs ranging from 10⁻² to 10⁵ cp/µL were injected, with the samples at 10⁻², 10⁻¹, and 10 cp/µL considered as negative controls based on real-time RT-PCR results (LOD: 10 cp/μL). The sensor loop was filled with the previously employed hybridization buffer followed by the sample injection. Samples were flowed for 20 min on the biofunctionalized sensor surface. To regenerate the sensor biosurface between experiments, a 35% FA solution was flowed for 3 min. This regeneration strategy permitted up to 14 consecutive assay cycles without observable loss in the sensor performance or sensitivity. For the biosensor assays validation, we used the RNAs extracted from the animal samples without any prior treatment.

Data analysis

Biosensor data analysis was conducted using Origin 8.0 software (OriginLab, Northampton, MA, USA). Statistical analyses were performed with GraphPad Prism (GraphPad Software, Inc., CA, USA). The mean sensor response and corresponding standard deviation (SD) were plotted as a function of synthetic RNA target concentration. Non-linear regression was applied using the specific binding model with Hill slope. The limit of detection (LOD) is defined as the lowest target concentration that can be reliably distinguished from the noise, while the limit of quantification (LOQ) corresponds to the lowest concentration that can be accurately measured. Both parameters are expressed in the same units as the standards: pM for synthetic RNA and copies/μL for IVT RNA. The LOD is calculated as the mean signal of the blank plus three times its standard deviation (mean +3 SD), and the LOQ as ten times the standard deviation (10× SD). To calculate the limit of detection expressed in copies/µL, the real-time RT-PCR results of IVT RNAs were correlated to BiMW biosensor signal using the non-regression line fit model, previously employed for the synthetic RNA detection assay. To calculate the accuracy of the BiMW biosensor, a linear regression was used, comparing the RT-PCR results in Ct values with the biosensor signals.

For biosensor validation with animal samples, measurements were performed blinded to PCR results, which were disclosed only after all experiments were completed. Repeated measurements were treated as technical replicates and not as independent observations; each biological sample was included only once in statistical analyses. Group comparisons were evaluated using the Kruskal–Wallis test, with statistical significance defined as p < 0.0001. Threshold (cut-off) values for determining positive samples were established as the median response of negative controls plus two times the median of the absolute deviations (median + 2MAD). Samples yielding responses below 0.9× (median + 2MAD) were classified as negative, those exceeding 1.1× (median + 2MAD) were considered positive, and responses falling within the intermediate range, defined as values between 0.9× (median + 2MAD) and 1.1× (median + 2MAD), were categorized as indeterminate³⁴. Diagnostic performance parameters, including sensitivity (SE), defined as true positives divided by the sum of true positives and false negatives; specificity (SP), defined as true negatives divided by the sum of true negatives and false positives; positive predictive value (PPV), defined as true positives divided by the sum of true positives and false positives; and negative predictive value (NPV), defined as true negatives divided by the sum of true negatives and false negatives, were calculated based on true positive (TP), true negative (TN), false positive (FP), and false negative (FN) results, as previously described¹⁴. ROC curves were generated by plotting sensitivity against 1-specificity across decision thresholds, and diagnostic accuracy was quantified by calculating the AUC, with values closer to 1 indicating higher discriminatory performance.