Multimodal sensing technologies for HPAI biosurveillance in poultry production systems

Abstract

Highly pathogenic avian influenza (HPAI), with a current focus on the emergent H5N1 clade 2.3.4.4b, remains a substantial and evolving threat to animal health, food security, and zoonotic safety. Since early 2024, novel genotypic variants within this clade have driven widespread epizootics across US poultry and dairy production systems, along with zoonotic transmission events. With the depopulation of over 168 million birds and economic losses exceeding 1.4 billion USD since 2022, recent outbreaks highlight the urgent need for complementary, decentralized, real-time biosurveillance strategies. This review outlines the molecular pathobiology and transmission kinetics of contemporary HPAI strains and evaluates diagnostic bottlenecks. Then, we explore how molecular amplification, electrochemical detection, and acoustic anomaly analysis can be combined into a single approach for in situ disease recognition. Finally, we describe how behavioral and physiological signal integration can enhance biosensor accuracy and support adaptive One Health biosurveillance systems for anticipatory and scalable field responses.

Introduction

Since early 2022, the re-emergence of highly pathogenic avian influenza (HPAI), particularly H5N1 strains within clade 2.3.4.4b, has escalated into a complex biosurveillance and agrosecurity crisis across the United States^1,2. Vertically integrated poultry production systems and backyard flocks experienced 1689 confirmed outbreaks across 681 counties with depopulation exceeding 168 million birds^3,4. The associated macroeconomic impact has exceeded 1.4 billion USD, driven by costs associated with culling mandates, disruption of intra- and interregional trade, implementation of biocontainment protocols, and reduced productivity across the poultry sector^3,5. A 3% contraction in the national laying hen population and a corresponding 4% reduction in table egg output triggered a 60% surge in consumer egg prices between January 2022⁶ and December 2024, underscoring the fragility of supply elasticity in the face of sustained biological disruption. Projections for H5N1-associated pandemics estimate economic liabilities between 500 and 800 billion USD, accounting for combined losses across agriculture (e.g., culling, trade restrictions), labor markets (e.g., worker absenteeism), and public health systems (e.g., hospitalization and treatment costs)⁷. H5N1’s virulence shows flock mortality rates approaching 100% within 48 h post exposure⁸. Compliance with regulatory containment frameworks necessitates immediate whole-flock depopulation upon viral confirmation, leading to abrupt supply chain disruptions, lost breeder stock, and prolonged recovery periods in poultry production. Although essential to controlling outbreaks, current biosecurity measures—particularly immediate whole-flock depopulation—exacerbate operational volatility and supply chain fragility. Combined with increasing input costs and heightened zoonotic risks disproportionately faced by socioeconomically marginalized agricultural laborers, these challenges underscore the urgent need for scalable, real-time diagnostic infrastructure and integrated risk-mitigation strategies.

The H5N1 virus has infiltrated other agricultural sectors beyond the poultry industry, most notably the bovine dairy industry^1,2. Since early 2024, infections in over 1020 dairy herds spanned 17 states, reflecting an expansion of host range and ecological plasticity^9,10. Although clinical manifestations in cattle are generally sublethal¹¹, the virus has been linked to subclinical mastitis and reduced milk production, resulting in economic losses of up to 950 USD per head in high-impact regions¹¹. These production issues carry implications for the resilience of the dairy industry, particularly under concurrent stressors such as heat stress, feeding volatility, and labor shortages¹². The virus’s ability to infect a broad range of mammalian hosts, including swine, caprines, camelids, companion animals, and both terrestrial and marine wildlife, highlights its cross-species transmissibility and zoonotic risk¹³. This promiscuity gives the virus more opportunities to mix with other influenza strains and accumulate mutations that could enhance infectivity, alter tissue tropism, or improve its capacity for human-to-human transmission.

As of 2025, there have been over 70 confirmed human cases of H5N1 avian influenza in the United States, including one death in Louisiana¹⁴. Most cases occurred in people who work closely with animals: 41 were linked to contact with infected dairy cattle, 26 to exposure to poultry, and 3 remain under investigation¹⁵. Notably, 53 of these cases were identified in 2024 alone, indicating a recent surge in zoonotic spillover events at the livestock–human interface, although the number of reported cases in 2025 to date appears lower¹⁶. A substantial proportion of affected individuals are low-wage, underserved agricultural workers with limited access to healthcare, compounding morbidity risk and hindering timely clinical response¹⁷. Such cascading consequences highlight the need to close gaps in intersectoral coordination—ensuring that early warning systems not only detect cross-species transmission but also trigger rapid, equitable responses across public health and agriculture. Effective interventions would include automated, multilingual alert systems triggered by sensor thresholds to inform workers of potential exposure, or dynamic personal protective equipment (PPE) protocols linked to real-time environmental monitoring data.

On-farm biosecurity constitutes the foundational barrier for excluding and containing pathogens such as HPAI, by preventing pathogen ingress, limiting intra- and interspecies transmission, and thus safeguarding both livestock and human health¹⁸. Standard biosecurity protocols rely on fixed procedures such as controlled access points, disinfectant footbaths, equipment sanitation, and visual health inspections. While these measures are essential, questions remain about their ability to detect asymptomatic or rapidly evolving infections, as several outbreaks have occurred despite strict adherence to standard protocols. If prevention fails, early detection is central to effective containment. Yet conventional surveillance approaches, which rely on clinical signs or periodic laboratory sampling, are hampered by significant logistical and statistical limitations. Sampling a statistically representative subset of a commercial flock of tens of thousands of birds is logistically challenging, and the inherent delays in sample transport and laboratory processing (often > 24 h) create a critical window for viral spread¹⁹. In addition, existing surveillance often fails to capture rapidly fluctuating infection dynamics or low-prevalence carriers, meaning early-stage outbreaks can be missed entirely. Environmental persistence of pathogens, such as virus contamination on surfaces or in dust, further complicates detection, as these reservoirs are rarely sampled in routine protocols. Given the increasing complexity of disease emergence patterns at the animal–human–environment interface, there is a need to move beyond static, protocol-based models toward dynamic, real-time monitoring systems capable of complementing standard biosecurity measures.

Multimodal biosensing platforms, which can integrate molecular, physiological, and behavioral signals, offer a promising approach for earlier detection and rapid response prior to the exponential spread of HPAI²⁰. These integrated systems will be uniquely powerful because they combine two complementary strategies: targeted pathogen identification and nontargeted anomaly detection. Targeted biosensing diagnostics directly identify specific viral biomarkers, such as viral RNA²¹ or antigens²², using high-sensitivity, high-specificity tools like impedance²³ or surface plasmonic resonance (SPR) biosensors⁵. In parallel, nontargeted modalities monitor for general indicators of disease without needing a specific biomarker receptor-based probe. Examples of nontargeted methods include label-free surface-enhanced Raman spectroscopy (SERS) molecular fingerprinting of biological samples²⁴ thermal imaging for detecting febrile responses²⁵, and acoustic surveillance for identifying respiratory anomalies and changes in vocalization rates^26,27. While unimodal systems remain relevant for validating individual sensing modalities, their performance can be limited by sampling frequency, environmental noise, or measurement variability in commercial poultry operations²⁸. Although this review focuses on HPAI, the proposed architectural framework—integrating nontargeted behavioral and molecular fingerprinting with pathogen-specific molecular tests and biosensors—is broadly applicable across poultry health surveillance. For instance, similar multimodal strategies could be deployed for other high-consequence respiratory diseases such as Newcastle Disease Virus (NDV) and Infectious Bronchitis Virus (IBV), where early detection is equally critical for outbreak prevention and flock protection. This transferability underscores the potential of a unified sensing architecture to strengthen resilience against a spectrum of avian pathogens, rather than being confined to a single disease context. As an example, a multimodal H5N1 biosensor that combined fluorescence, UV-Vis, and colorimetric modes achieved an ultra-low detection limit (LoD) of 11.6 fM and recovery rates of 93–118% in spiked serum, suggesting that cross-validated readouts may help reduce false negatives and enable earlier detection²⁰.

In this review, we introduce foundational and recent literature to describe Avian influenza viruses (AIVs) characteristics and delineate the challenges of current on-farm biosecurity practices during the ongoing HPAI pandemic, before turning to opportunities for integrated surveillance systems. Specifically, we conducted database searches using keywords related to ‘avian influenza,’ ‘biosecurity,’ and ‘sensing technologies,’ restricted to English-language peer-reviewed articles published between 2020 and 2025. Our primary emphasis was on recent literature to capture the latest developments, though a limited number of older, high-impact references were also included to provide historical context and improve the comprehensiveness of the review. Studies were included if they reported experimental or field-deployable sensing approaches, diagnostic assays, or surveillance strategies relevant to poultry production, while purely theoretical, non-peer-reviewed, or non-avian studies were excluded. We focus on how integrating these two complementary strategies—targeted viral pathogen detection (e.g., molecular assays, biosensors) and nontargeted anomaly detection (e.g., SERS molecular fingerprinting, acoustic phenotyping, and activity tracking)—can create a connected early-warning network. We will discuss how the integration has the potential to transform poultry production biosecurity systems to become more resilient, responsive, and aligned with One Health principles.

Spatiotemporal co-occurrence of HPAI in animals and influenza in humans

Surveillance data highlight the concurrent trends of HPAI and seasonal influenza circulation in the United States. As shown in Fig. 1a, monthly reports from the U.S. Department of Agriculture Animal and Plant Health Inspection Service (APHIS) indicate recurring waves of HPAI detections in both poultry and dairy cattle, with major peaks occurring in parallel to seasonal influenza activity. Human influenza-like illness (ILI) burden, captured through the U.S. Outpatient Influenza-like Illness Surveillance Network (ILINet), also shows age-stratified seasonal peaks in outpatient visits during the 2024–2025 influenza season (Fig. 1b). These parallel epidemics emphasize the overlapping risk landscape where zoonotic influenza outbreaks coincide with human respiratory illness activity.

Fig. 1: Overview of HPAI activity in U.S. livestock and poultry, alongside national influenza surveillance indicators.

a Monthly summary of confirmed HPAI detections in dairy cattle and poultry across the United States, based on APHIS (U.S. Department of Agriculture) data as of August 2025. b Percentage of outpatient visits for respiratory illness by age group, reported through the U.S. Outpatient Influenza-like Illness Surveillance Network (ILINet) weekly national summary for the 2024–2025 influenza season through the week ending August 2025. c State-level distribution of confirmed HPAI detections in poultry as of August 2025. d State-level distribution of confirmed HPAI detections in livestock as of August 2025. e Influenza A and B positive tests reported to CDC by clinical laboratories as of August 2025. f Influenza-positive tests reported to CDC by public health laboratories, showing the proportion of circulating viruses by type, subtype, and lineage¹⁶.

Spatial patterns of HPAI detections further demonstrate the heterogeneity of outbreaks across species. Confirmed poultry detections as of August 2025 were concentrated in high-production states such as California, Iowa, and Ohio (Fig. 1c), while livestock detections, including dairy cattle, were most frequently reported in western and central states (Fig. 1d). In parallel, influenza testing data from both clinical (Fig. 1e) and public health laboratories (Fig. 1f) show the predominance of influenza A viruses, with subtype and lineage analysis revealing circulation of multiple viral lineages, including H1N1pdm09, H3N2, and influenza B Victoria lineage. Together, these findings illustrate the multi-layered burden of influenza viruses at the human–animal interface and underscore the importance of integrating veterinary and public health surveillance for timely risk assessment.

AIV structure, pathogenicity, and host range

A clear understanding of the structural and genetic features of HPAI viruses is essential for selecting diagnostic targets and designing effective sensing strategies. AIVs are members of the Influenza A genus within the family Orthomyxoviridae and are classified as either low pathogenic (LPAI) or HPAI based on molecular features and disease severity in birds (Table 1). Structurally, AIVs are spherical, enveloped viruses with a segmented, negative-sense single-stranded RNA genome approximately 13.5 kb in length²⁹. The genome comprises 8 RNA segments that collectively encode 12–14 proteins, including structural proteins such as hemagglutinin (HA), which mediates viral entry, and neuraminidase (NA), which facilitates viral release (Fig. 2)^30,31. These glycoproteins define viral subtypes (e.g., H5N1, H3N2) and are central to host specificity and infectivity⁷. The genome also encodes matrix proteins (M1, M2), nucleoprotein (NP), three polymerase subunits (PB1, PB2, PA), and nonstructural proteins including NS1, NS2/NEP, and PB1-F2^32,33. In Fig. 2, the segments 1–3 encode the polymerase subunits (PB2, PB1, PA) along with accessory proteins (PB2-S1, N40, PB1-F2, PA-N155, PA-N182, and PA-X), segments 4–6 encode HA, NP, and NA, and segments 7–8 encode the matrix proteins (M1, M2, and M42) and nonstructural proteins (NS1, NEP). The segmented genome architecture enables high mutation rates and frequent genetic reassortment, facilitating antigenic drift and shift. This genetic flexibility drives rapid viral evolution, promotes cross-species transmission, and underpins the zoonotic and pandemic potential of AIVs. Understanding the genomic organization of AIV, as illustrated in Fig. 2b, directly informs the design and optimization of sensing technologies described later in this review. The HA and NA genes define subtype identity and serve as the primary targets for RT-qPCR and CRISPR–Cas12a assays through subtype-specific primer or guide-RNA design. The matrix (M) and NP genes are highly conserved across influenza A lineages and thus provide universal diagnostic markers for broad-spectrum molecular assays and electrochemical immunosensors³⁴ using anti-M1 or anti-NP antibodies. In parallel, sequence and structural variations within HA cleavage sites or NA stalk regions, highlighted in the figure, guide the selection of aptamers used in optical biosensors³⁵. Together, these genomic insights bridge molecular virology and biosensor engineering, enabling rational target selection and improved diagnostic specificity for emerging H5N1 clade 2.3.4.4b variants.

Fig. 2: Genomic organization of the segmented ssRNA(−) influenza virus.

a Genomic structure of Avian Influenza A. b The virus possesses a segmented, negative-sense ssRNA(−) genome encapsidated by NP. It comprises eight segments that encode 12–14 proteins, depending on the strain. Segment lengths range from 890 to 2341 nucleotides (nt), with a total genome size of approximately 13.5 kilobases (kb).

Table 1 Comparative features of LPAI or HPAI avian influenza viruses^125,126

As shown in Fig. 3, HPAI avian influenza viruses spread through interconnected environmental interfaces such as air (aerosols, dust), water (drinking sources and wetlands), fomites (equipment, clothing, surfaces), and feed contamination, each shaping the choice of detection strategy and biosensor type. The figure links these transmission routes to corresponding sensing modalities, showing how molecular assays (RT-qPCR, CRISPR Cas12a), electrochemical sensors (impedance or capacitive), optical methods (SERS, SPR), and behavioral approaches (acoustic monitoring) target distinct sample types and detection ranges. This schematic highlights the ecological context of viral circulation in poultry systems and the complementary roles of diverse sensing technologies in advancing a multimodal One Health aligned surveillance framework.

Fig. 3: Environmental transmission routes and corresponding biosensing modalities for HPAI detection.

Arrows depict primary pathways through air, water, fomites, and feed, each linked to suitable sensing approaches including molecular, electrochemical, optical, and behavioral methods.

The fundamental molecular difference between LPAI and HPAI lies in the structure of the HA cleavage site³⁶. LPAI strains contain a monobasic cleavage site, limiting HA activation to trypsin-like proteases in the respiratory and gastrointestinal tracts³⁷. As a result, LPAI infections in poultry are typically localized and either mild or asymptomatic³⁸. These viruses circulate endemically in wild aquatic birds and can occasionally spill over into domestic poultry, where they may cause reduced egg production or growth³⁹. In contrast, HPAI viruses such as H5N1, H5N8, and certain H7N9 strains harbor a polybasic HA cleavage site, enabling cleavage by ubiquitous intracellular proteases and facilitating systemic replication⁴⁰. HPAI-infected poultry often experience acute, multisystemic disease with mortality rates exceeding 90% within 48 h. The virus invades multiple organs, including the heart, brain, spleen, and kidneys, necessitating rapid culling to prevent outbreak escalation⁴¹.

Influenza A virus primarily infects humans, birds, pigs, horses, and marine mammals. However, recent spillover events in 2023–2025 have confirmed that H5N1(clade 2.3.4.4b) can also infect dairy cattle, marking host expansion. In humans, the Influenza A virus is transmitted primarily via respiratory droplets and can cause a range of illnesses from mild symptoms to severe complications such as pneumonia and acute respiratory distress syndrome (ARDS), particularly in immunocompromised individuals or those with preexisting conditions. Human-to-human transmission has not been observed. In poultry, H5N1 spreads rapidly through fecal-oral transmission, aerosols, and contaminated fomites⁴². High viral shedding, dense housing conditions, and environmental stability facilitate outbreaks, especially in settings with inadequate biosecurity⁴³. Preventive measures, including vaccination and antiviral therapies (e.g., NA inhibitors), remain the primary defenses, though ongoing viral evolution continues to challenge vaccine efficacy⁴⁴.

Transmission modes and host range differ between LPAI and HPAI⁴⁵. Both LPAI and HPAI are transmitted through fecal–oral contact, aerosols, droplets, and contaminated fomites, with wild aquatic birds serving as the primary natural reservoir⁴⁵. While LPAI typically spreads more slowly and causes mild or subclinical infections, HPAI replicates systemically, producing higher viral loads and enabling rapid flock-to-flock transmission⁴⁵. Certain HPAI strains also cross avian species barriers and infect mammals, raising zoonotic concerns.

The comparison between H5N1 outbreaks in poultry and seasonal influenza activity in humans (Fig. 1b) highlights synchronized epidemic peaks that are likely driven by shared ecological and climatic factors. Both avian and human influenza viruses show strong seasonality, with peaks during colder months when lower temperatures and humidity favor viral stability and transmission. Migratory waterfowl movements further synchronize H5N1 circulation across regions, introducing new viral clades during autumn and winter flyway overlaps. Concurrently, increased human indoor activity and intensified poultry handling during winter contribute to higher contact rates and cross-species exposure potential. Together, these factors explain the co-seasonality observed between H5N1 in birds and seasonal influenza in humans, indicating the need for integrated One Health surveillance during peak transmission periods.

The comparison between H5N1 outbreaks in poultry and seasonal influenza activity in humans (Fig. 2b) highlights synchronized epidemic peaks that are likely driven by shared ecological and climatic factors. Both avian and human influenza viruses show strong seasonality, with peaks during colder months when lower temperatures and humidity favor viral stability and transmission. Migratory waterfowl movements further synchronize H5N1 circulation across regions, introducing new viral clades during autumn and winter flyway overlaps. Concurrently, increased human indoor activity and intensified poultry handling during winter contribute to higher contact rates and cross-species exposure potential. Together, these factors explain the co-seasonality observed between H5N1 in birds and seasonal influenza in humans, underscoring the need for integrated One Health surveillance during peak transmission periods.

Biosecurity and diagnostic challenges

Current on-farm biosecurity approaches

Commercial poultry farms implement a set of well-established biosecurity practices designed to prevent pathogen entry and minimize farm-to-farm spread^46,47. These include restricted access to facilities, vehicle and equipment disinfection, use of PPE, controlled entry points with footbaths, and protocols such as all-in/all-out flock management and downtime between production cycle⁴⁸. Collectively, these measures form the foundation of AIV prevention and are reinforced by regulatory audits in many regions⁵. Despite their importance, these approaches face significant limitations. HPAI outbreaks have repeatedly occurred in farms that were fully compliant with biosecurity standards outbreak report⁴⁹, indicating that while these practices reduce risk, they cannot eliminate it. Dense housing conditions, long supply chains, and human-mediated fomite transmission undermine effectiveness⁵⁰. Critically, biosecurity relies on the visible detection of clinical illness, which fails to capture asymptomatic or subclinical infections that may silently spread within and between flocks⁵¹. Moreover, compliance can be inconsistent due to economic or logistical constraints, particularly in smallholder or resource-limited settings⁵². These factors highlight the gap between protocol design and real-world performance, underscoring the need for complementary surveillance systems that enable earlier and more reliable detection of HPAI.

Current laboratory approaches

A number of laboratories based diagnostic strategies for H5N1 surveillance in poultry production focus on molecular amplification, immunoassay technologies, and sensor-integrated platforms (Fig. 3 and Table 2). Conventional diagnostics, such as quantitative PCR (qPCR), enzyme-linked immunosorbent assays (ELISAs), virus isolation, isothermal amplification, and CRISPR-cas12a-based assays, are typically performed by accredited state or provincial veterinary diagnostic laboratories, national reference laboratories (e.g., USDA National Veterinary Services Laboratories in the U.S.), or regional university-affiliated veterinary labs. Samples are usually oropharyngeal or cloacal swabs, and in some cases, blood for serology, which are collected on-farm by poultry veterinarians, trained farm personnel, or animal health officers. These are then transported under cold-chain conditions to the testing facility.

Table 2 Diagnostic strategies for avian influenza

Real-time PCR (qPCR)

Quantitative qPCR remains the benchmark methodology for high-fidelity detection and molecular subtyping of AIVs under laboratory conditions. This technique quantifies viral RNA in real time while concurrently enabling genotypic differentiation (e.g., H5 vs H7) and detection of pathogenicity-associated motifs such as polybasic insertions at the HA0 cleavage site—hallmarks of HPAI strains. Modern multiplex qPCR platforms allow simultaneous interrogation of multiple genomic targets (HA, NA, NP, and M), while internal amplification controls are routinely incorporated to ensure assay robustness and minimize false negatives. With analytical sensitivity often below 10 RNA copies per reaction and high-throughput formats (96–384 well plates), qPCR is highly suitable for confirmatory testing, molecular epidemiology, and surveillance workflows, including certain Differentiating Infected from Vaccinated Animals (DIVA)-compatible designs that target genes absent from vaccine strains. Despite these advantages, qPCR remains resource-intensive. Each test requires specialized thermal cycling instrumentation, RNA purification reagents, and trained technical personnel, and per-sample costs remain high relative to on-farm assays^7,53. Moreover, reliance on sample shipment, cold chain logistics, and biosafety protocols exacerbates turnaround delays⁵⁴. During the 2022 U.S. outbreak, confirmatory qPCR results frequently required 48–72 h after collection, by which time the virus had already spread within or between flocks; similar bottlenecks were documented in Europe during peak outbreak periods⁵⁵. Given that HPAI can cause flock mortality reaching 100% within 48 h of infection, such delays severely constrain outbreak response. These limitations highlight the urgent need for rapid, low-cost, and portable diagnostic methods—such as isothermal amplification or CRISPR-based assays—that can complement qPCR by enabling decentralized, on-site detection during fast-moving outbreaks.

Isothermal amplification

Isothermal nucleic acid amplification techniques, such as reverse transcription loop-mediated isothermal amplification (RT-LAMP) and reverse transcription recombinase polymerase amplification (RT-RPA), provide thermocycler-independent alternatives to qPCR by enabling nucleic acid amplification at constant temperatures (37–65 °C)⁵⁶. These assays target influenza gene segments (e.g., HA, NP, M) and support diverse detection modalities, including colorimetric, turbidity, fluorescence, and even lateral flow-based or smartphone-compatible readouts. With reaction times of 30–60 min and detection thresholds ranging from 10 to 100 RNA copies per reaction, they offer rapid, portable diagnostics well-suited for on-farm deployment, often using minimal equipment such as heat blocks or handheld devices with lyophilized reagents⁵⁷. Relative to qPCR, isothermal assays generally demonstrate reduced sensitivity, limited quantitative resolution, and higher susceptibility to nonspecific amplification or contamination, though they may tolerate crude samples with less processing. Integration with CRISPR–Cas systems, microfluidic cartridges, and multiplexed detection strategies is being actively pursued to overcome these limitations and enhance specificity⁵⁸. While isothermal platforms are highly promising for on-farm use, their most effective role may be as a complementary component within a multimodal surveillance framework that balances speed, accessibility, and diagnostic robustness during HPAI outbreaks.

CRISPR-cas12a-based assays

CRISPR-Cas12a–based diagnostics represent an emergent class of programmable nucleic acid detection systems with exceptional analytical specificity and on-farm use⁵⁹. Functionally coupled with isothermal amplification platforms such as RT-RPA or RT-LAMP, these assays enable sequence-guided detection of AIV gene segments (e.g., HA, NP) with sensitivities approaching qPCR⁶⁰. Upon target recognition, Cas12a undergoes conformational activation and triggers collateral trans-cleavage of labeled reporter oligonucleotides, producing signals detectable by fluorescence, paper-based lateral flow strips, or smartphone-integrated readouts⁵⁹. With time-to-result typically under 60 min and detection thresholds as low as 2 copies/µL, CRISPR platforms combine rapid turnaround with near-gold-standard sensitivity. Key advantages include modular programmability—allowing rapid reconfiguration to track emergent genotypes, point mutations, or antiviral resistance loci—and compatibility with low-cost, decentralized testing formats⁶¹. However, current implementations still require pre-amplification, careful contamination control, and stable reagent formulations, which limit widespread field deployment. Ongoing advances in one-pot chemistries, lyophilized reagents, and cartridge-based integration are expected to overcome these barriers. Within an HPAI surveillance framework, CRISPR-Cas12a offers a promising bridge between laboratory-grade sensitivity and true on-farm usability.

Serological tools

Serological assays are essential components of avian influenza virus (AIV) surveillance, enabling retrospective exposure assessment, vaccine efficacy evaluation, and DIVA (Differentiating Infected from Vaccinated Animals) strategies^22,62. While they do not detect active infection, these tools measure host humoral responses, primarily IgM (acute phase) and IgG (long-term), against viral antigens such as HA, NA, and NP. ELISAs are widely recognized as the reference method for high-throughput, quantitative antibody detection. Strain-specific or DIVA-compatible ELISAs⁶³ targeting non-vaccine antigens (e.g., NP, M1) facilitate differentiation between vaccinated and naturally infected birds. IgM-specific ELISAs identify recent infections, while IgG-based assays inform exposure history or vaccine-induced immunity^22,64. Despite moderate turnaround times (2–3 h), their reliance on centralized laboratories and trained personnel introduces delays in sample transport and result reporting, reduces accessibility in resource-limited regions, and creates diagnostic bottlenecks during outbreak surges⁶⁵.

The hemagglutination inhibition (HI) assay remains a subtype-specific reference test, endorsed by international agencies, and quantifies neutralizing antibodies that block HA-mediated red blood cell agglutination. While inexpensive and rapid (~30 min), HI has lower sensitivity than ELISA and limited DIVA compatibility unless paired with updated antigen panels²². Lateral flow immunoassays (LFAs), in contrast, are immunochromatographic strip-based assays that provide portable, low-cost, and rapid (10–20 min) detection of antibodies or viral antigens. For AIV surveillance, LFAs commonly target conserved NP-specific IgG or detect viral proteins directly from swabs, making them well-suited for on-farm testing and wildlife monitoring^66,67. However, they are qualitative, lack subtyping capability, and cannot differentiate infection timing. Antigen detection assays, including ELISA and LFA formats, enable direct identification of viral proteins (e.g., HA, NP) from poultry specimens such as oropharyngeal or cloacal swabs. Though less sensitive than molecular methods, they deliver results within 15–30 min and exhibit specificity > 90%, supporting rapid triage during outbreaks. Antigen detection platforms such as LFAs⁶⁸ provide rapid results but can experience variable sensitivity⁶⁹.

Collectively, serological and antigen-based assays form a layered diagnostic toolkit: ELISAs provide scalable laboratory surveillance, HI continues to support subtype-specific antibody profiling despite limitations, and LFAs extend access to decentralized, rapid screening. Another key limitation is that most validated assays are species-specific, optimized for chickens or turkeys, with reduced sensitivity in ducks, swine, or cattle⁷⁰. This constraint is particularly concerning given recent H5N1 spillover events into mammals, including mink and dairy cattle, underscoring the need for cross-species diagnostic platforms to trace viral reservoirs and transmission pathways²².

Virus isolation

Virus isolation remains the definitive reference method for detecting and characterizing AIVs, including HPAI subtypes such as H5N1⁶⁶. By confirming replication-competent virus, this technique enables comprehensive phenotypic and genotypic analyses—spanning assessments of pathogenicity, antigenic drift, receptor-binding affinity, and molecular markers of host adaptation. In practice, clarified swab or tissue homogenates are inoculated into the allantoic cavity of 9–11-day-old embryonated chicken eggs or onto confluent Madin–Darby Canine Kidney (MDCK)²² cell monolayers under biosafety-level-3 (BSL-3) containment. The eggs are incubated at 37 °C for 48–72 h and candled daily; allantoic fluid from dead or harvested embryos is tested by HA using 0.5% chicken red blood cells to confirm viral replication. Positive isolates are subsequently typed by hemagglutination-inhibition (HI) and neuraminidase-inhibition (NI) assays or by RT-PCR confirmation. Despite its precision, virus isolation is inherently slow—typically requiring 2–7 days from inoculation to confirmation and demands specialized containment for HPAI strains^71,72. These operational demands, coupled with variable sensitivity in field samples with low viral titers, constrain its role in rapid outbreak response. As a result, virus isolation has shifted from frontline diagnosis to confirmatory and archival use—providing reference strains for assay validation, surveillance, and vaccine development pipelines. Once isolated, viruses can be phenotyped using standardized assays such as the intravenous pathogenicity index (IVPI)⁷³, which differentiates HPAI from LPAI strains based on morbidity and mortality outcomes. Thus, virus isolation remains indispensable for reference laboratories and research networks, where it underpins molecular surveillance and benchmarks the performance of emerging diagnostic platforms. However, its time- and resource-intensive nature underscores the need for complementary rapid methods (qPCR, isothermal amplification, CRISPR-based assays, and biosensors) to enable timely detection and intervention during outbreaks.

Sample type and processing critically influence detection efficiency and assay selection. As summarized in Table 3, commonly used matrices including swabs, blood, feces, aerosols, water, and feed, require specific pretreatments such as filtration, lysis, or centrifugal enrichment to optimize viral recovery and reduce matrix interference. Aligning each sample type with the most suitable detection modality ensures reliable performance and practical implementation across laboratories and on-farm surveillance settings.

Table 3 Representative sample matrices, pretreatment protocols, and compatible detection modalities for avian influenza virus detection^128,129

Opportunities in multimodal avian influenza sensing

Advanced sensing methods for AIV detection are emerging for both on-farm and laboratory use (Fig. 3). To align with our integrated architecture, we organize these approaches into two complementary tracks that converge on a shared triage step: Targeted Biosensing & Diagnostics (high-specificity confirmation) and Nontargeted Anomaly Detection (broad, early-stage screening).

Targeted biosensing and diagnostics

Among targeted molecular techniques, qPCR and CRISPR-based assays represent the backbone and frontier of avian influenza diagnostics, respectively⁷⁴. RT-qPCR remains the gold standard for confirming AIV infection because of its high sensitivity, subtype specificity, and compatibility with established reference protocols. In parallel, CRISPR-Cas–based methods have emerged as the next generation of molecular diagnostics, enabling rapid nucleic acid detection through programmable guide RNAs and collateral cleavage-based readouts⁷⁴. Together, these molecular platforms form the foundation of both laboratory and field-level detection strategies and complement emerging biosensors and acoustic modalities within a multimodal surveillance framework.

This diagnostic category provides sequence or affinity-based confirmation and subtyping. It includes isothermal nucleic-acid amplification [reverse-transcription loop-mediated isothermal amplification (RT-LAMP), reverse-transcription recombinase polymerase amplification (RT-RPA)], CRISPR-Cas assays, reverse-transcription quantitative PCR (RT-qPCR), and targeted biosensors (electrochemical impedance/capacitance and surface plasmon resonance (SPR)/optical platforms). RT-LAMP and RT-RPA can be performed by trained personnel using portable incubators, returning results in ~30–60 min from oropharyngeal or cloacal swabs⁷⁵. CRISPR-Cas12a paired with RT-RPA achieved limits of detection as low as 6.7 copies/µL with >97% concordance to rRT-PCR and virus isolation in poultry samples, supporting use as a rapid confirmatory channel^76,77,78.

Building on these molecular advances, CRISPR-Cas-based diagnostics have been rapidly engineered for miniaturization and on-site deployment⁷⁹. Recent one-pot RT-RPA–Cas12a and Cas13a platforms integrate amplification and detection within enclosed cartridges, removing transfer steps and reducing contamination risk. Lyophilized reagents extend shelf stability under ambient conditions, while fluorescence, colorimetric, and lateral-flow readouts enable visual interpretation without specialized equipment⁸⁰. Smartphone-assisted readers and portable fluorimeters further enhance usability for non-laboratory settings⁸¹. Systems such as SHERLOCK, DETECTR, and AIOD-CRISPR achieve sub-10-copy-per-microliter sensitivity within 45–60 min, rivaling benchtop RT-qPCR^82,83. Collectively, these developments establish CRISPR diagnostics as field-ready, rapid, and scalable tools that complement biosensor-based surveillance in poultry production environments.

Field-deployable CRISPR–Cas12a systems integrated with isothermal amplification require only simple equipment—a portable incubator for RT-RPA or RT-LAMP and a handheld fluorimeter or smartphone-based reader for visual detection^84,85. The total assay time, including RNA extraction and amplification, is typically 45–60 min, with reagent and consumable costs around USD 3–7 per test. Training requirements are minimal and limited to basic sample handling and pipetting, enabling use by veterinary or extension personnel. Field studies have reported >95% agreement with RT-qPCR, confirming that CRISPR-based diagnostics are both cost-effective and operationally feasible for on-farm AIV screening as part of a multimodal surveillance system.

Quantitative surveillance data across these environmental matrices help contextualize sensor performance and deployment feasibility. As summarized in Table 4, viral concentrations during early infection often fall below 10³ copies mL⁻¹ in water and aerosols^28,47,48,72, ranges that challenge conventional ELISA or lateral-flow formats^66,69 but remain within the detection capability of optimized RT-qPCR, CRISPR–Cas12a, and impedance-based biosensors^28,76,77,86. In contrast, high-load matrices such as feces and swabs (10⁴–10⁸ copies mL⁻¹)^72,77,86 are compatible with both molecular and electrochemical approaches but may require dilution or extraction to reduce matrix inhibition. Optical methods, including SPR and fluorescence, perform well at moderate concentrations (10³–10⁵ copies mL⁻¹) under clean laboratory conditions^77,86, but suffer signal instability in turbid samples. Acoustic and SERS-based systems, while indirect, are particularly valuable at the lowest concentration regimes because they provide continuous anomaly detection that can trigger confirmatory molecular testing when viral titers remain below direct-assay thresholds^{24,87,88,89,90,91}. Together, these quantitative relationships define realistic operational windows for each sensing modality across heterogeneous on-farm and environmental contexts. Electrochemical devices detect viral RNA or antigens directly from swabbed or environmental matrices in <1 h⁷⁷, and SPR aptasensors can resolve broad viral-load ranges in ~1.5 h—tools that complement molecular assays within the targeted track⁸⁶. Handheld nanopore platforms operated by lab staff or mobile veterinary teams can produce same-day genomic data for detection and molecular epidemiology⁹². As biosecurity layers, a capacitive airborne biosensor reported H5N1 within ~5 min, illustrating a path to real-time environmental confirmation in an integrated workflow²⁸.

Table 4 Reported AIV concentrations across environmental reservoirs and detection compatibility of major sensing modalities

Electrochemical impedance and capacitive sensors can exhibit small baseline drift when exposed to temperature fluctuations, humidity, changes in background conductivity, or gradual electrode aging^62,93,94. These factors alter charge-transfer resistance and double-layer capacitance, but their influence is usually mitigated through temperature logging and software compensation, buffered sample solutions of fixed ionic strength, antifouling or passivation layers, and periodic calibration using internal standards or disposable reference electrodes. By contrast, optical techniques such as SPR and fluorescence are largely immune to conductivity changes yet remain sensitive to refractive-index variation, mechanical alignment, and thermal drift of optical components⁹⁵. In practice, both systems can perform reliably under on-farm or field conditions when such environmental variables are monitored and compensated, although electrochemical sensors are generally more tolerant of mechanical vibration and dust, whereas optical sensors demand tighter control of alignment and refractive-index stability.

Electrochemical impedance biosensors^62,93,94 for AIV are generally designed as single-use cartridges, since regeneration by acid or detergent can detach capture probes and cause baseline drift. This single-use format enhances reproducibility and prevents cross-contamination when testing environmental or swab samples. The fabrication cost of screen-printed or 3D-printed electrodes is typically USD 1–5 per test, with a 30–60 min total assay time including incubation and readout. Laboratory prototypes can be regenerated up to 3–5 times with less than 10% signal loss, but disposable use remains preferred for field deployment. In comparison with molecular and optical assays, EIS sensors offer similar sensitivity (10¹–10³ copies mL⁻¹), faster turnaround than RT-qPCR or SPR, and lower per-test cost than isothermal or CRISPR platforms, supporting their practicality for rapid, on-farm AIV surveillance.

Nontargeted anomaly detection

This track flags infection-related perturbations without prior knowledge of the agent, which can include label-free SERS molecular fingerprinting of air/dust/condensate^24,87 and acoustic monitoring of respiratory/vocal anomalies⁹⁶. In challenge and field contexts, acoustic features have indicated respiratory distress up to ~48 h before overt clinical signs, enabling earlier isolation and targeted follow-up testing⁹⁷. SERS adds a probe-free molecular fingerprint that can be classified with machine learning to cue targeted assays when broad anomalies are detected.

Fusion and triage

Combining tracks yields real-time, noninvasive, species-agnostic surveillance that detects spatiotemporal anomalies in both molecular signatures and flock-level behavior⁹⁸. In practice, nontargeted alerts (SERS/acoustic) trigger targeted confirmation (isothermal ± CRISPR, RT-qPCR, electrochemical/SPR), with concordant positives escalating actions (isolation, personal protective equipment (PPE) changes, confirmatory sampling). Embedding this logic within a unified One Health surveillance architecture strengthens anticipatory and adaptive biosecurity by enabling earlier detection and coordinated response across animal, human, and environmental health domains⁹⁹.

Targeted biosensing diagnostics

Targeted biosensors translate specific biomolecular binding events (antigen–antibody or nucleic-acid hybridization) into quantifiable electrochemical, optical, or colorimetric signals, enabling rapid, decentralized detection of H5N1 (Fig. 4)²⁸. These platforms support sensitive, near-real-time analysis without centralized laboratory infrastructure^100,101. Within the targeted stack, biosensors complement RT-qPCR and CRISPR by extending coverage to environmental matrices (air, water, and dust) and providing on-farm confirmation that triggers timely escalation when nontargeted screens flag anomalies. Materials-based performance enhancements (e.g., gold nanoparticles, carbon nanotubes, graphene) are discussed later, with evidence for their impact on sensitivity and robustness¹⁰².

Fig. 4: Targeted diagnostics and nontargeted anomaly detection within a multimodal AIV surveillance stack.

a Targeted biosensing and molecular diagnostics: Electrochemical and optical biosensors, along with isothermal amplification ± CRISPR and RT-qPCR, provide high-specificity pathogen detection and subtyping. Inputs include swabs, water, surfaces, and targeted air concentrates. b Nontargeted anomaly detection: label-free SERS molecular fingerprinting of air, dust, or condensate, and acoustic monitoring of coughs, sneezes, and vocal anomalies, generate broad alerts without requiring prior knowledge of etiology. Inputs include environmental samplers and distributed microphones. c Fusion and triage: a decision node integrates both targeted and nontargeted outputs using rule sets (e.g., anomaly alert → triggers confirmatory testing). This orthogonal design leverages broad anomaly detection with high-specificity confirmation to reduce false alarms and shorten time-to-action. Resulting actions may include cohort isolation, escalation of PPE, and confirmatory sampling. Together, this framework illustrates a synergistic multimodal surveillance system for early and reliable AIV detection.

Electrochemical biosensors

Electrochemical biosensors typically employ a three-electrode configuration—working, reference, and counter (auxiliary) electrodes—enabling stable potentials and accurate current measurements^102,103. Within this domain, EIS is widely used for pathogen detection because it is label-free (no secondary tags or enzymes), low-power, and highly sensitive to small changes at the sensor surface^93,104. Assays are run on common poultry samples such as oropharyngeal/cloacal swabs, tracheal swabs, drinking water, and surface wipes. The working electrode is functionalized with monoclonal antibodies or nucleic-acid aptamers against conserved AIV proteins (e.g., HA, NA). Target binding alters the electrical double layer and, consequently, the measured impedance across a range of frequencies—providing a quantitative signal of presence and approximate concentration⁹⁴. A capacitive biosensor integrated with a wet-cyclone bioaerosol sampler enabled near real-time detection of airborne H5N1 (Fig. 5)²⁸. The Prussian-blue/graphene-oxide (PB/GO) sensing interface was functionalized with aptamers or antibodies to enable molecular recognition beyond bulk dielectric perturbations; lab-generated H5N1 aerosols were introduced and captured by the sampler, and results were cross-checked by PCR²⁸. The system produced readouts in <5 min, 25 exhibited strong linearity across serial dilutions (R² ≥ 0.95), and supported quasi-quantification of H5N1 directly from air samples; replicate measurements were consistent, and agreement with PCR supported analytical accuracy. Experiments used lab-generated aerosols (no on-farm poultry-house air), and environmental factors (particle size, humidity, and co-collected particulates) can shift baseline capacitance; therefore, orthogonal molecular confirmation (e.g., qPCR or CRISPR) is recommended for decision-making.

Fig. 5: Overview of the capacitive biosensor system for airborne H5N1 detection.

a Aerosols containing pathogens released from infected animals are captured using a wet cyclone particle sampler. The collected sample is then processed by the capacitive biosensing unit. b Fabrication of the Prussian blue/graphene oxide (PB/GO)-based sensor, where PB and GO are co-deposited to form a redox-active capacitive layer (Cr) on a screen-printed carbon electrode (SPCE). Capture probes—aptamers or antibodies—are immobilized via glutaraldehyde cross-linking. At the biosensor–sample interface, double-layer capacitance (Cdl) forms, while diffusion capacitance (Cdf) arises from the ion diffusion layer. Binding of target pathogens to the capture probes changes the total capacitance (Ctot), enabling detection. c Capacitive biosensor responses for varying concentrations of H5N1. d Sensitivity assessment of the H5N1 biosensor using serial dilutions in PBS, showing a strong linear correlation (R² ≥ 0.95) between normalized ΔC% and log concentration, with significant differences between concentrations (t-test, p < 0.01). e Decision workflow for quasi-quantification of airborne H5N1, where samples are sequentially tested without dilution, 10× dilution, and 100× dilution to classify concentrations relative to the limit of detection (LoD). f Quasi-quantification of H5N1 aerosols collected via the wet cyclone sampler, with results classified as positive (Ο) or negative (X) based on normalized ΔC% relative to the H5N1 limit of detection (LoD)²⁸. This publication is licensed under CC BY-NC-ND 4.0 (https://creativecommons.org/licenses/by-nc-nd/4.0/).

A maskless, laser-patterned PMMA microfluidic aptasensor with integrated electrodes has been reported for quantitative detection of H5N1-target DNA sequences, using an Fe₃O₄@TMU-8 nanocomposite to drive catalytic redox-recycling of [Ru(NH₃)₆]³⁺ (RuHex) and amplify the electrochemical signal¹⁰⁵. The positively charged RuHex associates with the nucleic-acid phosphate backbone, and hybridization at the aptamer interface modulates the redox-recycling current to yield a quantitative readout. Under optimized conditions, the device achieved a linear range of 1 fM–1 nM with an LOD of 0.16 fM, and demonstrated good reproducibility and long-term stability; applicability was shown in spiked food matrices. Because H5N1 is an RNA virus, deployment on clinical or environmental samples would require prior reverse transcription and multistep processing (extraction → thermal denaturation → hybridization), which limits real-time or aerosol-phase use without substantial automation of sample preparation¹⁰⁵. Accordingly, this platform is best positioned as a high-sensitivity confirmatory channel within a multimodal AIV workflow rather than a frontline on-farm screen.

Optical biosensors

Optical biosensors, most commonly surface plasmon resonance (SPR)¹⁰⁶ and fluorescence-based assays¹⁰⁷, convert target binding at a sensor surface into changes in light. In SPR¹⁰⁶, binding alters the refractive index at a thin metal film, shifting the resonance and enabling real-time, label-free measurement of association and dissociation kinetics. In fluorescence formats¹⁰⁷, targets are reported by dye or nanoprobe emission; binding events change intensity or lifetime to yield very high sensitivity, down to the attomolar range, although performance can be affected by quenching, spectral overlap, and autofluorescence in complex matrices⁸⁶. Within a multimodal surveillance framework, optical readouts provide an orthogonal physical principle to electrochemical and nucleic acid amplification methods, enable rapid confirmation with compact readers, and support wavelength-based multiplexing, which together improve overall robustness and time to decision for AIV detection¹⁰⁸. A miniaturized 3D printed SPR biosensor was able to detect H5N1 avian influenza virus in under 12 min, achieving a LoD of 4.3 × 10⁴ copies/mL, demonstrating a 37-fold enhancement in sensitivity over conventional LFAs¹⁰⁹. In this study, the authors used PBS-diluted culture medium (cell-culture fluid) spiked with H5N1 AIV; specificity controls used PBS-diluted culture medium containing H1N1, H3N2, and H7N9. They coupled the biosensor with a near-infrared (NIR) organic photodetector. The integration of an organic photodetector optimized for 980 nm wavelength enhanced responsivity and signal-to-noise performance, positioning the system as a promising candidate for portable optical biosensing. However, the platform’s reliance on stable optical path alignment, angular precision, and refractive index uniformity imposes constraints when used on-farm. Moreover, SPR-based detection requires specialized readout instrumentation, often involving angular interrogation or spectroscopic components, which may hinder use on the farm unless future iterations address this need.

Despite strong analytical performance, biosensors face practical constraints for decentralized AIV surveillance, including environmental instability, matrix-induced interference, the need for sample preprocessing, and dependence on external instruments. EIS-based devices are particularly susceptible to nonspecific impedance drifting from temperature and humidity, biointerface instability and fouling, and limited capacity for on-farm multiplex calibration. Optical platforms require stable alignment and refractive index control, while fluorescence readouts can be affected by quenching and autofluorescence; colorimetric formats may suffer subjective thresholds in dusty or complex matrices. Mitigation strategies include antifouling and passivation layers, rugged solid state reference electrodes, periodic baseline normalization and temperature compensation, disposable cartridges with on-board filtration or enrichment, and built-in positive and negative controls with internal standards. At the protocol level, calibration transfer and drift correction, blinded benchmarking against quantitative PCR and CRISPR on both spiked and field poultry house samples, and fusion with orthogonal channels are essential to raise confidence and reduce false alarms. Positioned within a multimodal stack, biosensors provide fast triage and environmental coverage while molecular assays deliver definitive confirmation, yielding a system that is more robust than any single modality.

Nontargeted anomaly detection

Early, probe-free screening is essential because targeted diagnostics—while specific—can miss presymptomatic infection or emerging variants. Label-free SERS generates a holistic vibrational fingerprint from barn aerosols/dust, and acoustic monitoring quantifies respiratory/vocal changes; both yield classifier scores that prompt targeted tests when thresholds are exceeded. Integrating these nontargeted signals with targeted confirmation increases robustness, reduces unnecessary disruption, and preserves a critical lead time for intervention.

Label-free SERS molecular fingerprinting

SERS combines vibrational spectroscopy’s molecular fingerprint specificity with plasmonic nanostructures’ hotspot sensitivity, offering an ultrasensitive non-targeted fingerprinting strategy for detecting infection-induced biochemical signatures without specific probes¹¹⁰. In SERS, laser interrogation of aerosol or dust samples yields complex Raman spectral patterns amplified by plasmonic nanostructures. These spectra serve as unique “fingerprints” of the sample’s molecular composition. Importantly, the presence of virus particles or host metabolic changes due to infection can alter the spectral profile in detectable ways. As shown in Fig. 6, a recent work has demonstrated that a single SERS platform can distinguish multiple respiratory viruses (including influenza A viruses) based on intrinsic spectral features, when coupled with machine learning algorithms⁸⁷. This label-free detection can achieve high accuracy (often >90% in classification tasks) without any virus-specific antibodies or primers, highlighting its potential as a broad surveillance tool in animal populations. A newly reported methodology exemplifies how SERS can be adapted for environmental virus monitoring²⁴, which involves virion disintegration – using a gentle lysis protocol to break intact virions in an aerosol/dust sample into their constituent proteins, lipids, and nucleic acids – thereby enhancing the richness of the Raman signal. The released viral components are then concentrated into densely packed “hotspots” on a nanostructured SERS substrate, boosting sensitivity to trace levels of virus. Using multivariate analysis, they could reliably identify virus-specific spectral patterns even against the complex backdrop of environmental dust. Notably, this method distinguished SARS-CoV-2, influenza A, and Zika virus signatures within a mixed dust matrix with over 86% classification accuracy. Such results underscore that SERS, combined with machine learning, can extract subtle virus-derived signals from noisy field samples.

Fig. 6: Label-free SERS detection of pathogenic enveloped viruses using machine learning data analytics.

The process is demonstrated in two scenarios: a clean cell supernatant and a complex environmental dust matrix. A–D show that average SERS spectra and principal component analysis-linear discriminant analysis (PCA-LDA) can differentiate SARS-CoV-2, influenza A, and Zika virus in the clean matrix. A schematic (E) outlines the workflow. F–J Confirm that even within a noisy dust matrix, the viruses are successfully identified, evidenced by distinct spectral fingerprints, PCA-LDA scores, and a confusion matrix showing high classification accuracy²⁴. Reproduced with permission from Garg et al., Biosensors and Bioelectronics (2024), Elsevier. License Number: 6153910838293.

Analyzing poultry house aerosols via SERS poses challenges due to the complex matrix: poultry dust contains dander, feed, microbes, and chemical fumes that all contribute to the Raman background. However, advanced computational approaches can deconvolve these spectra. Pattern recognition algorithms (e.g., PCA-LDA or more advanced random forests and neural networks) are adept at differentiating virus-positive spectra from baseline barn air. For instance, a recent report demonstrated a “single-shot” SERS sensor with machine learning could classify multiple respiratory viruses, including influenza, in patient swab samples with ~97% accuracy¹¹¹. This research indicates that, even in complex biological media, SERS spectral differences can be learned and used for robust identification. The key to success is building spectral libraries of known healthy vs infected aerosols and using algorithms to discern the minute but consistent Raman peaks associated with the virus or infection metabolites, while filtering out confounding signals. Therefore, a portable SERS-based aerosol sensor—bolstered by machine learning—could serve as an early warning system for avian influenza in poultry farms. Such a system would continuously sample air in chicken houses and flag abnormal spectral fingerprints in near real-time, enabling pen-side biosurveillance. The SERS approach is inherently complementary to targeted tests: it casts a wide net to catch unexpected or novel pathogens (or variants) via their chemical fingerprints, after which traditional qPCR or antigen assays can confirm the pathogen identity. Ongoing advances in SERS substrate fabrication and miniaturization of Raman devices are bringing this vision closer to reality. With detection times on the order of minutes and no need for elaborate sample processing, machine-learning-enabled SERS platforms hold promise for real-time, on-site screening of poultry flocks. Ultimately, this nontargeted SERS surveillance can significantly augment existing biosecurity measures by providing early alerts to incipient viral outbreaks, buying valuable time to implement containment strategies.

SERS can be applied to a range of biological and environmental matrices¹¹², including aerosols collected on filters, surface wipes, fecal suspensions, and liquid extracts from oropharyngeal or cloacal swabs. Depending on the sampling method, the total analysis—from specimen collection to result—typically requires 10–30 min, enabling near real-time assessment when used for Tier 1 early warning. Portable Raman spectrometers and flexible nanostructured substrates now allow on-farm operation with minute- to hour-level response times. SERS signal stability is influenced by humidity, dust, and matrix turbidity, but reproducible performance has been demonstrated under moderate field conditions when using gold- or silver-nanostructured substrates enclosed in sealed cartridges. Accuracy improves further when spectral fingerprints are classified by chemometric or machine-learning models, which help distinguish true viral signatures from environmental background.

Acoustic monitoring

Acoustic monitoring is a noninvasive tool that analyzes flock soundscapes to detect early respiratory and behavioral changes associated with infection. It decomposes bioacoustic signals—vocal emissions and respiration-related sounds—into time–frequency features and classifies health states with machine learning using distributed microphone arrays^90,91,113. Respiratory infections, including HPAI, alter pulmonary and syrinx biomechanics and produce measurable changes in vocal patterns (paroxysmal coughing, stertorous respiration, inspiratory gasping, reduced vocalization rate, increased entropy)¹¹⁴. These anomalies can be captured and classified in real time. In a controlled AIV challenge, a chicken sound CNN (CSCNN) distinguished infected from healthy birds as early as day 2 post-inoculation, with accuracies of 93.01%, 95.05%, and 97.43% on days 2, 4, and 6, respectively, using spectrogram inputs⁸⁹. In barn-scale recordings from two commercial flocks, a compact CNN (light VGG11) trained on 5336 one-second segments achieved ~95% accuracy (precision 94.58%, recall 94.89%) for automatic distress-call detection under real farm noise⁸⁸. Group-level sneeze and cough detection has also been demonstrated in noisy multi-bird conditions¹¹⁵.

Acoustic monitoring can advance detection by one to two days before visible morbidity, enabling earlier isolation, sanitation, and targeted molecular confirmation; it reduces handling and labor, scales across houses with low-cost microphones and edge computing, and remains species agnostic across production systems.

Key limitations of acoustic monitoring include environmental noise, variability in acoustic environments, and the need for extensive labeled datasets to train generalized classifiers¹¹⁶. In high-noise poultry houses (ventilation fans, augers, human activity), background signals can mask low-amplitude coughs and inspiratory sounds, dropping sensitivity to about 60–90% depending on infection stage, sensor fidelity, and environmental entropy¹¹⁷; mitigation includes spectral subtraction or Wiener filtering, adaptive beamforming with small microphone arrays, source separation to isolate respiratory transients, and noise-aware training using augmented soundscapes that include barn machinery and human speech^118,119. Acoustic profiles also vary across barns, breeds, ages, stocking densities, and building geometries, while microphone placement and array topology (height, spacing, near field vs far field) influence signal to noise ratio and class balance; mitigations include standardized array layouts with calibration tones per barn, domain adaptation and transfer learning to retune models across facilities, continuous recalibration triggered by drift detectors, and adding environmental covariates such as temperature, humidity, and fan duty cycle^88,120.

Robust deployment further requires large, diverse labeled datasets for coughs, sneezes, and respiratory distress across ages, genotypes, and co-infections; to ease labeling burdens and improve generalization, use self-supervised pretraining on unlabeled barn audio, semi-supervised and active learning, federated learning across farms, and synthetic augmentation with room impulse responses and crowd-noise overlays. Additional operational constraints include etiologic non-specificity—respiratory sound patterns overlap among HPAI, Mycoplasma gallisepticum, and Newcastle disease, so laboratory confirmation, such as RT qPCR, is required—and flock-level biological variance that necessitates periodic model recalibration^121,122. Looking ahead, priorities are noise-robust front ends on edge devices (beamforming, adaptive denoising, source separation), data-efficient learning and domain adaptation, open and versioned benchmark datasets with common labeling protocols and reporting standards (per bird and per flock sensitivity, false alarm rate per hour, time to alert), reference microphones with periodic calibration tones and recommended array topologies by house size, low power edge accelerators for real time inference, and prospective multi-site validation with concurrent gold-standard testing to quantify lead time, specificity, and operational false alarm rates; place any general statement about integrating acoustics with complementary biosensing and molecular modalities in the overall conclusion, since that benefit applies across all modalities rather than to acoustics alone.

Looking ahead, the integration of acoustic sensing with wearable ultrasound or flexible acoustic patch technologies could extend surveillance from flock-level soundscapes to individual-level continuous monitoring, enabling personalized health tracking and early detection of respiratory distress within precision poultry systems.

Table 5 summarizes the analytical sensitivity, operational characteristics, and technology readiness levels (TRLs) of major avian-influenza sensing modalities discussed in this review. Molecular assays such as RT-qPCR and isothermal amplification remain the most mature technologies (TRL 8–9), offering high sensitivity but requiring trained personnel and moderate instrumentation. Emerging CRISPR-Cas platforms combine sub-10-copy sensitivity with simplified reagent formats and growing portability (TRL 6–8). Electrochemical biosensors demonstrate rapid response and low reagent demand, supporting near-real-time environmental monitoring at comparable TRL 6–7. Optical and SERS systems deliver strong analytical performance under laboratory conditions, yet still face challenges in alignment stability and field adaptation (TRL 5–6). Behavioral sensing through acoustic monitoring has reached operational deployment (TRL 8–9) and provides continuous, non-invasive early warning that complements molecular diagnostics. Together, these comparisons highlight a convergence of maturing technologies toward integrated, multimodal AIV surveillance capable of laboratory-grade accuracy and on-farm practicality.

Table 5 Comparative analytical performance, operational characteristics, and TRLs of major sensing modalities for avian influenza detection

Conclusions and outlook

Current biosecurity diagnostics often detect HPAI too late because early clinical signs are nonspecific, sample collection and transport add delay, testing capacity is concentrated in designated laboratories, and routine swabbing is invasive and hard to scale. We recommend a modular, tiered multimodal stack that draws from the six candidate modalities in Fig. 4 rather than deploying all six at once.

In practice, farm staff oversee continuous acoustic monitoring and simple environmental biosensors through a dashboard. When an anomaly is detected, a tiered response is triggered. For example, a continuous acoustic system might flag a potential respiratory anomaly with a 70% confidence score. Instead of immediate flock-wide action, this Tier 1 alert would trigger a targeted environmental air sample for a 1-h on-site CRISPR assay (Tier 2). A negative CRISPR result de-escalates the alert, while a positive result triggers immediate notification for confirmatory RT-qPCR sampling by veterinary personnel (Tier 3). This two-stage logic could drastically reduce false alarms compared to acting on nontargeted data alone, while still preserving a critical 24–48 h lead time over waiting for overt clinical signs.

Successfully fusing these disparate data streams requires sophisticated analytical approaches far beyond straightforward machine learning. The challenge lies in integrating heterogeneous data types (e.g., continuous acoustic time-series, discrete electrochemical readings, and binary molecular test results) across different time scales and noise profiles. Robust predictive algorithms must be developed to maximize sensitivity while minimizing the costly false alarms that can lead to alert fatigue. This direction may involve deep learning architectures, such as transformers, to model long-range temporal dependencies in acoustic or behavioral data¹²³. Additionally, Bayesian networks could be employed to probabilistically integrate evidence from multiple, independent sensor modalities and quantify uncertainty¹²⁴. Furthermore, effective system integration demands clear sensor fusion logic and protocols for handling discordant results. For instance, a persistent, low-grade acoustic anomaly with consistently negative molecular tests might indicate a different co-circulating pathogen rather than HPAI, requiring a distinct diagnostic response. Reliability across diverse farm environments also necessitates robust calibration-transfer strategies, such as incorporating per-house acoustic calibration tones or embedding internal standards within biosensor cartridges to account for site-to-site variability.

As detailed in the main text, limitations remain: acoustic systems are affected by noise and site variability; biosensors by matrix effects and sample prep; and optical platforms by alignment stability. Key gaps include multi-site validation, improved discrimination of co-circulating agents, integrated on-site sample prep for nucleic acid assays, and standardized benchmarks for lead time, false alarms, and flock-level sensitivity.

While integrators and cooperatives are the natural payers for fixed hardware, widespread adoption also hinges on overcoming non-technical barriers. A viable technology readiness pathway must be established, moving from benchtop validation to pilot studies and multi-farm trials under the guidance of national veterinary laboratories. In parallel, clear data governance policies are needed to address on-farm data security and privacy, and regulatory frameworks for real-time animal health diagnostics with trade implications require proactive engagement between technology developers, producers, and agencies. Addressing adoption barriers will also require stakeholder engagement and capacity-building, including multilingual alerts, paid worker training, and transparent data use policies that protect the privacy of agricultural workers, who are often from minority and immigrant communities. More broadly, aligning technological advances with social, ethical, and policy considerations will be essential to ensure trust, equitable deployment, and long-term sustainability of multimodal surveillance systems.

Figure 7 presents the three-tier workflow for on-farm HPAI surveillance. Tier 1 integrates continuous behavioral and environmental sensing—acoustic distress signatures and airborne biosensor data—to generate early anomaly alerts within minutes to hours. When deviations exceed adaptive thresholds, the system advances to Tier 2, where rapid molecular (RT-LAMP or CRISPR-Cas12a) and electrochemical impedance assays provide confirmatory results within 30–60 min. Confirmed positives then progress to Tier 3 laboratory validation via RT-qPCR, virus isolation, or sequencing within one to two days. False alarms are minimized through cross-validation across sensing modalities prior to escalation. This structured workflow demonstrates how multimodal fusion enables a timely, evidence-based response compared with conventional single-channel surveillance systems.

Fig. 7: Tiered surveillance framework for rapid detection of HPAI integrating behavioral, molecular, and electrochemical sensing.

Tier 2 activation occurs upon molecular positivity (cycle threshold, Ct < 35) or impedance change, followed by laboratory confirmation and sensor recalibration.

In summary, to effectively combat the ongoing H5N1 pandemic, multimodal sensing platforms that integrate molecular assays, biosensors, and behavioral monitoring offer more than just earlier detection. They can enhance biosurveillance by capturing both molecular signatures and flock-level anomalies in real time, enabling rapid containment before widespread transmission. They can also inform vaccine development by generating high-resolution epidemiological data on circulating strains and host responses, thereby accelerating the design and validation of targeted immunization strategies. Furthermore, they promise economic resilience, as early detection reduces emergency culling, trade restrictions, and downstream supply-chain disruptions that currently cost the poultry industry billions. Finally, by linking farm-level monitoring with laboratory confirmation and One Health data streams, such systems can support broader pandemic control, mitigating risks to both animal and human populations. While technical challenges remain—such as matrix interference in biosensors, acoustic variability across farms, and the need for robust sample preparation in molecular assays—a modular multimodal approach offers a path toward responsive, equitable, and economically sustainable avian influenza management at scale.