A roll-to-roll fabricated lab-on-foil platform for point-of-care genetic detection of lactose intolerance: system design, integration and analytical validation

Introduction

Lactose intolerance is a common gastrointestinal disorder caused by reduced activity of lactase, the enzyme responsible for hydrolyzing lactose into glucose and galactose (Olds and Sibley, 2003). While secondary deficiency may result from intestinal disease, the predominant form is primary lactose intolerance, reflecting a genetically programmed decline in lactase expression after early childhood (Swallow, 2003; Misselwitz et al., 2019). This phenotype, termed lactase non-persistence, affects approximately two-thirds of the global adult population.

At the molecular level, lactase expression is regulated by cis-acting elements upstream of the LCT gene. The single nucleotide polymorphism (SNP) −13910C>T (rs4988235) is the primary genetic marker for lactase persistence in European populations, located in intron 13 of the neighboring MCM6 gene (Enattah et al., 2002; Alves et al., 2021). The T allele enables continued lactase expression, whereas the ancestral C allele leads to post-weaning downregulation (Olds and Sibley, 2003; Lewinsky et al., 2006). Additional variants in other populations highlight the genetic heterogeneity of lactase persistence (Tishkoff et al., 2007; Itan et al., 2010), with strong selection signals reflecting gene–culture co-evolution linked to dairy consumption (Bersaglieri et al., 2004; Gerbault et al., 2011). Recent studies further suggest selective advantages under conditions where lactose malabsorption increased health risks (Evershed et al., 2022).

From a diagnostic perspective, lactose intolerance is commonly assessed by hydrogen breath or tolerance tests, which are influenced by physiological and external factors (Misselwitz et al., 2013). In contrast, genetic testing provides a stable, non-invasive marker, with SNP genotyping enabling reliable identification of primary lactose intolerance (e.g., LCT −13910C>T) (Rasinperä et al., 2004; Anguita-Ruiz et al., 2020). Conventional approaches rely on amplification-based methods such as Polymerase Chain Reaction (PCR), including real-time PCR and sequencing-based assays. While highly sensitive and specific, these techniques require thermal cycling, specialized laboratory infrastructure, and trained personnel, resulting in turnaround times of several hours and limiting their applicability in point-of-care settings (Niemz et al., 2011; Mattar et al., 2012).

To overcome these limitations, isothermal amplification methods in combination with microfluidic platforms have emerged as promising alternatives for genetic testing. (Notomi et al., 2000; Fang et al., 2010; Tanner et al., 2014; Zhang et al., 2019; Conceição et al., 2024).

Among these approaches, loop-mediated isothermal amplification (LAMP) enables rapid and efficient DNA amplification under constant temperature conditions without the need for complex thermal cycling, as required in PCR (Notomi et al., 2000) and is particularly well suited for integration directly within microfluidic chips. This on-chip implementation supports low-volume, closed-system amplification, reducing contamination risks, shortening time-to-result, and allowing versatile detection through colorimetric, electrochemical, or fluorescence-based readouts (Tanner et al., 2014; Gill and Amree, 2020). Combined with automated reagent flow and precise temperature control, on-chip LAMP paves the way for a compact and user-friendly platform for POC nucleic acid testing. Such systems enable rapid, near-patient genetic analysis with minimal user intervention, facilitating timely diagnosis and improving accessibility in decentralized and resource-limited settings.

To translate point-of-care concepts into widely accessible solutions, scalable manufacturing approaches are required. In this context, R2R manufacturing of lab-on-foil (LOF) microfluidic chips represents a scalable manufacturing approach for point-of-care genetic diagnostics. Disposable devices can be continuously fabricated in roll-to-roll using UV nanoimprint lithography to define channel structures, microarray spotting for biomolecule immobilization, and UV-assisted lamination for sealing, enabling reproducible, high-throughput mass production of low-cost chips (Khumwan et al., 2025).

In this work, we implemented a roll-to-roll (R2R) fabrication strategy for the production of a foil-based microfluidic chip integrating an on-chip loop-mediated isothermal amplification (LAMP) chamber for the genetic diagnosis of lactose intolerance. Detection was achieved via chemiluminescence using a portable reader with automated reagent handling. The system enabled discrimination between lactase non-persistent and lactase-persistent genotypes from both genomic DNA extracted from whole blood and synthetic DNA samples within approximately 50 min. These results demonstrate the feasibility of rapid, accessible genotyping beyond centralized laboratory settings, supporting deployment in resource-limited, near-patient environments. Furthermore, the platform establishes a versatile foundation for expanded applications, including multi-marker analysis and personalized medicine.

Materials and methods

Chemicals and equipment

Supplementary Table S1: Supplementary Material.

Chip production on R2R pilot line

Chip design

The LOF chip is a compact, capillary-driven system integrating all assay steps within a 75 mm × 25 mm format and a channel depth of ∼85 μm, resulting in a total volume of ∼85 µL. The footprint (75 mm × 25 mm) was defined by compatibility with the modified GENSPEED reader platform, while the channel depth of approximately 85 µm represented a compromise between reliable capillary filling, sufficient reaction volume and reproducible large-area R2R fabrication. The chip dimensions were selected based on a combination of reader compatibility, roll-to-roll UV nanoimprint lithography (R2R UV-NIL) design rules, capillary-flow requirements and the volume requirements of the adapted on-chip LAMP assay. The chip consists of two reagent inlets (∼30 µL each), an integrated LAMP chamber (∼11.7 µL), a capillary-driven transport channel (∼6.7 µL) with four sequence-specific detection zones comprising allele-specific probes together with positive and negative controls, a waste reservoir (∼65 µL) and integrated vent structures (Figure 1B). Each functional element was designed to support fully automated sequential processing from amplification to hybridisation and chemiluminescence-based detection. The reaction chamber volume of approximately 12 µL was selected to ensure reliable chamber filling, robust amplification and sufficient transfer of amplified product to the downstream hybridisation channel. Lower reaction volumes were not systematically investigated and therefore remain a potential route for further assay miniaturisation.

Roll-to-roll (R2R) UV nanoimprint lithography of microfluidic chips

Microfluidic lab-on-foil (LOF) chips were fabricated using continuous R2R UV nanoimprint lithography (UV-NIL). The imprinting process was carried out on a customized pilot-scale R2R system (JOANNEUM RESEARCH and Coatema) enabling high-throughput replication of microfluidic structures on flexible polymer foil substrates (Figure 1). A UV-curable imprint resin was applied to the polymer foil via slot-die coating. The coated foil was then brought into conformal contact with a rotating cylindrical nickel shim (28 cm × 60 cm, containing ∼30 chip layouts) carrying the negative microfluidic pattern. During contact, the resin was partially cured by UV irradiation before demoulding, yielding microfluidic channel networks with a characteristic depth of approximately 85 µm. This depth was selected to balance channel volume and reliable capillary-driven flow. Imprinting was performed at a web speed of 1 m min⁻¹, corresponding to a throughput of several thousand chips per hour. After imprinting, the semi-cured structured foil was rewound for subsequent processing steps.

Bio-functionalisation by R2R micro-array spotting

Biofunctionalisation of the detection regions was performed using R2R-compatible microarray spotting on a sciFLEXARRAYER S12 system (Scienion AG, Germany). The structured foil was processed in stop-and-go mode to enable precise non-contact piezo-driven pulsed microdispensing of the probes from glass microcapillaries into predefined regions of the detection channels. Allele-specific oligonucleotide probes targeting the LCT −13910C and LCT −13910T alleles, as well as a negative control probe consisting of a nonspecific nonsense sequence, and a positive control probe complementary to a biotin labelled control oligonucleotide. All spotting probes were diluted in a UV-curable spotting buffer consisting of 25 µL SciPoly SOLD1 (Scienion AG), 10 µL SciPoly 3D (Scienion AG) and 10 µL PCR water. The UV-curable spotting buffer served both as carrier medium for the oligonucleotide probes and as immobilisation matrix. UV curing fixed the probes to the polymer surface while preserving accessibility for subsequent hybridisation. The probes were dispensed using piezoelectric forces and immobilised as rectangular detection zones (Fields 1–4 in Figure 1). A droplet volume of 4 nL was selected during process optimisation as it provided reproducible droplet formation and merging into homogeneous detection spots while maintaining precise positioning within the detection channel. Larger droplet volumes were not used because they may increase spot spreading, alter probe density and reduce spatial signal definition. The UV-curable spotting buffer was employed to enable probe fixation by subsequent UV exposure (30 s at 254 nm). Each chip contained four detection zones aligned with the photodiode array of the reader (Genspeed Biotech, Austria). Spotting accuracy was ensured by fiducial-based optical alignment, while droplet formation and positioning were monitored in-line with a camera. The probe immobilisation was achieved by UV curing (30 s at 254 nm) in a downstream module integrated into the R2R process. The UV-curable spotting buffer and the UV imprint resin were selected to ensure compatibility with the LAMP reaction, while maintaining probe accessibility for subsequent hybridisation. Material formulations were evaluated during assay development to avoid inhibition of amplification reactions or interference with probe functionality.

UV-assisted lamination and chip sealing

Following biofunctionalisation, the microfluidic structures were sealed by UV-assisted lamination using a second film with a partially cured 20 µm thin resin layer (Götz et al., 2022). The film was laminated onto the structured foil with a laminator (Renz, HT 330 P). The curing process was performed in four sequential steps. During the first two steps, only the sealing layer was processed. To prepare the sealing layer, the same UV-curable resin used in the R2R process was distributed between a polyethylene terephthalate (PET Melinex ST505, 125 µm thickness) foil and a polymethyl methacrylate (PMMA) foil (used as a release film) using a bar coater. The resin was initially cured for 10 s at 10 mW/cm², after which the PMMA foil was removed. The second curing step was carried out in a nitrogen (N₂) atmosphere, where the sealing layer was held for 2 min without UV exposure, followed by irradiation for 25 s at 10 mW/cm². The sealing film was then laminated onto the structured layer, and the resulting assembly was subjected to a third curing step under identical N₂ conditions (2 min without UV exposure, followed by 25 s UV irradiation at 10 mW/cm²). The curing sequence was experimentally optimized to balance three competing requirements: sufficient polymer conversion to prevent swelling after liquid filling, preservation of residual reactive groups for interfacial bonding, and maintenance of immobilised probe functionality. Nitrogen incubation minimized oxygen inhibition during polymerisation, while the selected UV exposure times and intensities provided reproducible sealing without deformation of the microfluidic structures.

Laser-based inlet cutting and chip singulation

Inlet ports, vent holes, and chip outlines were defined by laser-based post-processing using a Trotec Speedy 300™ flex system. This step enabled precise formation of fluidic access points and separation of individual chips. After processing, the chips were visually inspected prior to functional testing. Laser processing parameters (laser power (10 W), scan speed (28 inch/sec) and pulse frequency (1000Hz)) were optimised to obtain reproducible cutting of inlet ports and vent holes while avoiding thermal damage to adjacent microfluidic structures.

Reader instrumentation

To enable on-chip DNA amplification (LAMP) and to support largely automated assay execution and analysis, substantial functional upgrades were implemented on a commercial point-of-care (POC) reader (GENSPEED R2 platform, Genspeed Biotech, Austria) (Figure 2A). Key modifications included the integration of a closed-loop-controlled chip heating element and extended firmware capabilities for automated reagent handling. In particular, the system was adapted to support controlled fluid delivery and drawer positioning for addressing two chip inlets instead of the single inlet used in the standard configuration.

In addition, the hardware was extended with modules and spectral filters for time-gated fluorescence detection to allow for future compatibility with fluorescent labels.

The resistive heating element, aligned with the on-chip LAMP chamber, provides precise temperature control for DNA amplification and subsequent denaturation of amplicons into single-stranded DNA, which is required for hybridisation with probes in the detection zones. Chemiluminescence signals are detected using an integrated 35-pixel photodiode array (35 mm × 6 mm; 32 pixels are addressable by the readout electronics), aligned with the detection channel. Since the assay relies on controlled heating during both LAMP amplification and thermal denaturation, the thermal properties of the flexible foil substrate are of particular importance. The selected substrate thickness provided sufficient mechanical stability while enabling efficient heat transfer from the integrated heater. Nevertheless, local temperature gradients across the chip cannot be excluded, particularly between the heater, foil substrate and reaction chamber and may have contributed to assay variability observed during prototype validation.

LAMP and detection on chip–workflow

DNA samples

Genomic DNA from whole blood with known LCT −13910C>T genotypes (C/C, C/T, T/T), as well as synthetic DNA constructs (gBlocks, Integrated DNA Technologies, United States), were used. Here, LCT denotes the lactase gene, −13910 indicates the relative genomic position upstream of the gene, and C or T refer to the nucleotide present at this position. The notation C>T describes a single nucleotide polymorphism in which cytosine (C) is replaced by thymine (T). Initial validation employed 12 genomic samples and two gBlocks, followed by analysis of 67 genomic DNA samples, 55 of which came from the Coriell Institute (United States) and 12 from ViennaLab Diagnostics. DNA concentrations were adjusted to 1–10 ng/μL. All human samples were collected with informed consent. Whole blood was not introduced directly into the microfluidic chip. Genomic DNA was extracted off-chip using the GENXTRACT™ Blood DNA Extraction System or obtained as purified genomic DNA before loading onto the chip.

Oligonucleotide capture probe design and LAMP

Allele-specific oligonucleotide probes (15–30 nucleotides) targeting the LCT −13910C>T SNP were designed based on the NCBI Reference Sequence (RefSeq ID: NG_008104.2). Separate probes specific for the −13910C (lactase non-persistence) and −13910T (lactase persistence) alleles were synthesized as 4× or 6× repeats without spacers and immobilized as separated detection zones on the microfluidic chip surface. Initial selection and optimization were performed on commercial injection moulded chips to improve signal-to-background ratios and to achieve balanced signal intensities between both alleles. Selected probes were subsequently validated on the imprinted microfluidic chip using on-chip amplification and chemiluminescence-based detection.

LAMP was employed for target DNA amplification; the assay procedure is depicted schematically in Figure 2B. The LAMP reaction mixture consisted of an isothermal amplification buffer, deoxynucleotide triphosphates (dNTPs), a LAMP primer mix (partially containing biotin-labels at the 5′-end), and Bst 2.0 WarmStart DNA polymerase (New England Biolabs, United States), prepared according to optimised assay conditions. Prior to on-chip processing, the reaction mixture was prepared in a tube, and genomic or synthetic DNA was added immediately before testing. A total volume of 12 µL reaction mix and 2 µL glycerol was introduced via inlet 1 of the microfluidic chip, filling the LAMP chamber and incubated at 60 °C ± 1 °C for 40 min, with thermal control provided by an integrated resistive heater in the reader device (Figure 2). Following isothermal amplification, thermal denaturation of the LAMP products was achieved by increasing the temperature to 80 °C for 90 s, enabling strand separation without the use of chemical denaturation reagents. Subsequently, 12 µL of wash buffer was added via inlet 1 to drive capillary-force assisted transport of the denatured LAMP products into the detection channel. Following a 5-min incubation period, the biotin-labelled LAMP amplicons were hybridised to the immobilised, allele-specific capture probes by adding 10 µL of hybridisation buffer and allowing the reaction to proceed for 3 min at room temperature. Identification of hybridised variant sequences was accomplished by binding a horseradish peroxidase (HRP)–streptavidin conjugate to the biotinylated amplicons (20 μL, 90 s), followed by removal of non-specifically bound DNA through the addition of 20 µL wash buffer for 2 min, and a subsequent enzymatic reaction with 30 µL of a luminol-based substrate solution. These reagents were delivered via inlet 2.

The resulting chemiluminescence signals were detected using the photodiode array of the Genspeed reader allowing for a site- and thus probe-specific differentiation of the signals.

The complete workflow was completed within approximately 50 min. All assay steps following sample loading were executed automatically by the modified reader system, allowing fully integrated and largely automated POC operation. Sequence-specific hybridisation and chemiluminescence detection were intentionally performed as two separate steps. Following thermal denaturation, the single-stranded biotin-labelled LAMP products were first allowed to hybridise selectively to the immobilised allele-specific capture probes. Only after completion of this sequence-specific binding step were HRP–streptavidin and the chemiluminescent substrate introduced, thereby reducing non-specific signal generation and improving genotype discrimination.

Data analysis

Genotype discrimination was achieved by calculating the ratio of signals obtained from allele-specific probe zones.

Decision thresholds were determined using receiver operating characteristic (ROC) curve analysis and statistical optimisation based on maximization of the Youden index (J), defined as:

J = Sensitivity + Specificity − 1

The Youden index represents a global measure of diagnostic accuracy and corresponds to the maximum vertical distance between the ROC curve and the diagonal line of no discrimination, thereby identifying the threshold that optimally balances sensitivity and specificity under equal weighting assumptions (Fluss et al., 2005).

Testing of the final chip design and analytical validation

Final chip design and preliminary testing

The final microfluidic chip design comprised allele-specific probes targeting the LCT −13910C and LCT −13910T alleles, as well as a negative and a positive control probe integrated into the linear detection channel (Figure 2). To assess probe performance, hybridisation efficiency, and signal detection independently from on-chip amplification, initial validation experiments were performed using pre-amplified LAMP products generated under controlled conditions (40 min at 60 °C) in a real-time isothermal amplification instrument (Genie® III, OptiGene, UK). Subsequently, preliminary on-chip tests including LAMP were conducted using 12 pre-typed human genomic DNA samples and two synthetic DNA templates following the same acceptance criteria applied during probe selection. In total, 49 test runs were performed across three reader instruments to verify functional performance of the final chip design.

Analytical performance evaluation

Analytical performance of the integrated assay on the foil-based chip was assessed using 67 human genomic DNA samples with known LCT −13910C>T genotypes. For improved genotype discrimination, a slightly modified hybridisation protocol was applied, incorporating an additional temperature-controlled step using pre-warmed (45 °C) hybridisation buffer instead of buffer at room temperature. Measurement runs not reaching the minimum signal intensity were classified as technically invalid according to the predefined QC criteria. Acceptance criteria for valid measurements were refined to ensure robust genotype calling based on the allele-specific probes targeting the −13910C and −13910T alleles, each designed to selectively detect the corresponding allele. Valid measurement required a signal intensity of ≥250 arbitrary units (AU) for the correct genotype (−13910T and/or −13910C ≥ 250 AU), while background and control thresholds were defined as NEG ≤350 AU and POS ≥300 AU. DNA samples were re-tested up to two additional times if acceptance criteria were not met. These acceptance criteria served as predefined quality-control (QC) criteria for technically successful assay execution. Only measurement runs fulfilling all QC criteria were considered technically valid and were thus included in the subsequent diagnostic performance evaluation. In total, 167 measurement runs were performed on four different reader instruments. The microfluidic chips were designed for single-use applications and were discarded after each measurement. Consequently, low-cost fabrication and compatibility with high-throughput mass production were essential design criteria.

Repeatability assessment

Assay repeatability was evaluated using a heterozygous (C/T) genomic DNA sample tested on three different reader instruments. A single master mix containing polymerase, primers, probes, and DNA template was prepared and used for 30 consecutive measurements (10 per instrument). Testing conditions and acceptance criteria were identical to those applied during analytical performance evaluation, enabling assessment of inter-instrument repeatability under representative operating conditions.

Statistical analysis and cut-off determination

Signals recorded at each probe position were expressed as AU. Measurement runs were considered valid if acceptance criteria were met for all allele-specific and control probes.

Genotype classification was based on the ratio of signals obtained from the −13910T and −13910C probes (AU_T/AU_C), rendering the analysis independent of absolute signals.

Descriptive statistics, including mean values and standard deviations, were calculated for valid measurements and visualised using box plots. Ambiguous or discrepant results were excluded from initial cut-off determination and analysed separately. In a subsequent analysis, all valid signals exceeding predefined intensity thresholds were included to optimise diagnostic decision thresholds.

The medically relevant homozygous C/C genotype associated with lactose intolerance was defined as the positive class. Diagnostic performance was evaluated by calculating sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) from contingency tables, using the reference genotyping method as the comparator. Each sample was classified as a true positive (TP), true negative (TN), false positive (FP), or false negative (FN). Positive predictive value (PPV) and negative predictive value (NPV) therefore reflect agreement with the reference genotyping assay rather than prediction of disease status. Optimal cut-off values were determined by maximising Youden’s index. Receiver operating characteristic (ROC) analysis was performed to evaluate diagnostic performance and to visualise the relationship between sensitivity and specificity across a range of decision thresholds.

Results

Fabrication and functional integrity of R2R lab-on-foil chips

R2R-fabricated lab-on-foil (LOF) microfluidic chips were successfully produced and used throughout assay development and validation, demonstrating the feasibility of scalable manufacturing for integrated point-of-care genetic diagnostics. The chips exhibited robust mechanical stability during handling and assay execution, with no delamination or leakage observed under assay-relevant thermal conditions.

Capillary-driven fluid transport was generally reproducible under controlled operating conditions, and full compatibility with on-chip isothermal amplification at 60 °C confirmed the functional integrity of the R2R-imprinted and UV-laminated microfluidic architecture (R2R UV NIL process Supplementary Video). While minor variability was observed during prototype handling—primarily due to mechanical bending of the foil substrates—this effect is attributed to the early-stage fabrication workflow and is expected to be mitigated in a fully continuous R2R process. From a process perspective, biofunctionalisation by R2R-compatible microarray spotting and subsequent UV-assisted lamination emerged as critical steps. Optimisation of spotting parameters significantly improved signal uniformity and reproducibility, while precise alignment of probe zones with the photodiode array ensured consistent signal acquisition. Similarly, chip sealing required careful balancing between complete curing and preservation of probe functionality. Insufficiently optimised UV-assisted lamination and chip sealing led to swelling of the UV resin upon liquid filling (Figures 3, panels A1 and A2), which can compromise channel integrity and fluidic performance. After optimisation (Materials and Methods: UV-Assisted Lamination and Chip Sealing), the chips exhibited no swelling or leakage and enabled stable, well-controlled capillary-driven filling of all functional structures (Figures 3, panels B1–B4). Together, these results define critical process parameters for enabling improved device reproducibility and functional stability.

Performance of probe immobilisation and signal uniformity

To enable allele-specific detection, oligonucleotide probes were immobilised in spatially defined detection zones using a roll-to-roll compatible microarray spotting process. During assay development, different spotting layouts comprising allele-specific probes, positive controls, and negative controls were evaluated. Initial chip designs contained up to seven detection zones, which were subsequently reduced to five and finally four zones.

This optimisation became necessary following modifications of the optical reader configuration. In particular, the introduction of additional optical spacer elements increased the distance between the microfluidic chip and the photodiode array from near-contact conditions to approximately 1 mm. As a consequence, broader signal profiles and partial overlap between neighbouring peaks were observed. To reduce optical crosstalk and improve signal assignment, the final chip layout employed four detection zones with increased spatial separation.

Signal acquisition was performed using the integrated photodiode array, with 32 pixels addressable by the readout electronics. Signal evaluation was carried out using an in-house MATLAB routine that extracted and integrated the peak profiles corresponding to the individual detection zones. The resulting signal distributions showed clearly distinguishable spatial maxima for the respective probe regions.

Under stable assay conditions, the spotting and immobilisation procedure enabled reproducible formation of spatially confined probe regions with specific target binding. However, signal reproducibility was strongly influenced by the fluidic behaviour within the chip. Incomplete wetting, irregular capillary flow, or trapped air bubbles occasionally resulted in altered peak intensities, reduced hybridisation signals, or unspecific adsorption effects.

Homozygous samples of the LCT –13910C>T variant (T/T and C/C) generally produced characteristic and distinguishable signal patterns at the corresponding allele-specific probe positions. In contrast, heterozygous samples (C/T) generated intermediate signal ratios with partial overlap towards neighbouring genotype clusters, making discrimination less robust compared to the homozygous states (Figure 4).

On-chip amplification and allele-specific detection

On-chip LAMP amplification was successfully performed using genomic DNA extracted from whole blood. Stable amplification was achieved at 60 °C (±1 °C) under analytically valid assay conditions, confirming compatibility of the chip materials and surface chemistry with the enzymatic reaction.

Following amplification and thermal denaturation, labelled amplicons were hybridised to the immobilised allele-specific probes, resulting in distinguishable allele-specific signal patterns within the detection channel Figure 4). Homozygous samples (C/C and T/T) generated dominant signals at the respective probe positions, while heterozygous samples (C/T) produced characteristic dual-signal patterns.

The resulting AU_T/AU_C ratios were 0.54 (SD = 0.27), 0.95 (SD = 0.18), and 3.98 (SD = 2.65) for C/C, C/T, and T/T genotypes, respectively (Figure 5). With only a single exception, the data formed three well-separated clusters, enabling reliable discrimination of the three genotype classes under analytically valid conditions. No pronounced systematic cross-hybridisation patterns were observed under the selected assay conditions, demonstrating robust allele-specific hybridisation under on-chip conditions and supporting simplified POC operation without stringent assay control requirements.

Analytical validation using human DNA samples

To evaluate clinical performance, the final assay was tested using 67 genomic DNA samples. In total, 167 measurement runs were performed across four reader instruments, including repeated runs following failure to meet predefined quality-control (QC) acceptance criteria. Of these, 49 runs fulfilled all predefined QC acceptance criteria and were included in the diagnostic performance analysis (49/167 runs; 29.3%). The remaining 118 runs (70.7%) were classified as technically invalid according to the predefined QC criteria and were therefore excluded from genotype evaluation. Importantly, these invalid runs were not considered genotype discordances but reflected technically unsuccessful assay execution (e.g., incomplete reagent delivery, air entrapment or insufficient signal generation), resulting in failure to meet the predefined QC acceptance criteria and preventing reliable genotype calling. The analytical performance evaluation was therefore based on 49 valid measurements, comprising 13 C/C, 11 C/T and 19 T/T genotypes. The AU_T/AU_C ratios showed clear separation between genotype groups, with mean values of 0.37 (SD = 0.20), 1.31 (SD = 0.55), and 4.40 (SD = 3.46), for C/C, C/T and T/T respectively (Figure 5). Most measurements (69.8%) clustered around three genotype-associated regions, while 30.2% fell into overlapping regions between adjacent genotypes (Figure 5B), indicating the need for optimised decision thresholds.

Initial cut-off values derived from the C/T distribution (mean ±1 SD) resulted in thresholds of 0.76 (C/C vs. C/T) and 1.86 (C/T vs. T/T). Using these cut-offs, 83.7% of genotyping results were concordant with reference methods. Sensitivity and specificity for detection of the clinically relevant C/C genotype were 84.6% and 93.3%, respectively, resulting in a Youden’s Index J = 0.779. To further improve classification performance, the lower cut-off was optimised using Youden’s index, incorporating ambiguous results and applying a signal threshold of ≥250 AU. This analysis identified an optimal cut-off range of 0.81–0.89. Combining both approaches resulted in final thresholds of 0.81 (C/C vs. C/T) and 1.9 (C/T vs. T/T) (see red lines in Figure 5), yielding a sensitivity of 100.0% (95% CI: 75.3%–100.0%) and a specificity of 80.6% (95% CI: 64.0%–91.8%), corresponding to an improved Youden index of J = 0.806. Using the final decision thresholds, the assay achieved a positive predictive value (PPV) of 65.0% (95% CI: 48.9%–78.3%) and a negative predictive value (NPV) of 100% (95% CI: 88.1%–100.0%) relative to the reference genotyping method (Supplementary Table S2). The corresponding ROC curve is shown in Figure 5. Despite the improved sensitivity, 24.5% of valid results remained discordant with the reference method (see Supplementary Table S2 for detailed information). These discrepancies were primarily associated with misclassification of heterozygous samples or overlap between C/T and T/T clusters.

Repeatability and system robustness

Repeatability was assessed across three reader instruments using 30 consecutive measurements. Valid results were obtained in 18 cases (60%), with a mean AU_T/AU_C ratio of 2.01 (SD = 0.84). Variability was predominantly observed for borderline samples, particularly those with the heterozygous C/T genotype, where nine valid measurements were classified as T/T when applying the upper decision threshold (Supplementary Table S4). The observed AU_T/AU_C ratios ranged from 0.74 to 4.14, reflecting the broad signal distribution of heterozygous samples and the resulting overlap with neighbouring genotype clusters. The results of the device to device repeatability are depicted in Figure 5D.

Of the 167 measurement runs performed during analytical validation, 49 (29.3%) fulfilled the predefined acceptance criteria and were included in the diagnostic performance analysis, whereas 118 runs (70.7%) were classified as technically invalid and excluded. These results indicate that, while reliable genotype discrimination could be achieved for analytically valid measurements, further optimisation of the integrated prototype system is required to improve operational robustness. The underlying causes of the reduced valid-run rate are discussed in the Discussion section.

Discussion

The presented platform integrates LAMP-based isothermal amplification, allele-specific hybridisation, and on-chip chemiluminescence detection within a scalable lab-on-foil microfluidic architecture. The use of continuous R2R fabrication enables industrially relevant, cost-efficient production, representing a key prerequisite for the widespread implementation of genetic point-of-care diagnostics.

Multiple validated molecular methods are currently available for LCT genotyping, with real-time PCR representing the established laboratory standard because of its high analytical accuracy and widespread clinical implementation. LAMP-based approaches have attracted increasing interest for point-of-care applications by eliminating the need for thermal cycling while maintaining rapid nucleic acid amplification. In contrast to previously reported LAMP-based microfluidic systems, the present work demonstrates the integration of scalable roll-to-roll cartridge manufacturing with automated reagent handling, on-chip amplification, allele-specific hybridisation and chemiluminescence detection within a single disposable platform. Although the current prototype has not yet reached the operational maturity required for routine clinical use, it demonstrates the feasibility of scalable decentralized SNP genotyping and provides a promising foundation for future point-of-care genetic diagnostics. Unlike conventional laboratory-based workflows, the presented platform integrates amplification, fluid handling and allele-specific detection within a disposable cartridge fabricated by scalable roll-to-roll processing.

While the analytical validation yielded a comparatively low proportion of valid assay runs (29.3%), this finding should be interpreted in the context of the current prototype implementation rather than as a limitation of the underlying allele-specific assay chemistry. The majority of technically invalid runs originated from engineering-related aspects associated with the integration of a newly developed dual-inlet roll-to-roll fabricated foil cartridge into a modified prototype reader that had originally been designed for injection-moulded cartridges with a single inlet and substantially larger reagent volumes. Consequently, the combination of small reagent volumes, dual-inlet operation and the prototype reader architecture increased the system’s sensitivity to dispensing inaccuracies, fluid handling and capillary transport. In addition, the surface wetting characteristics of the current foil cartridge were not yet fully optimised, occasionally resulting in excessive capillary uptake, air entrapment and incomplete reagent transport through the detection channel. These engineering-related effects led to insufficient or non-uniform signal generation and failure to meet the predefined quality-control criteria. Importantly, successful genotype discrimination was consistently achieved for analytically valid measurements, indicating that the principal limitation of the current system lies in its operational maturity rather than in the analytical capability of the integrated biochemical assay. Future development will therefore focus on improving reader–cartridge integration, automated reagent dispensing, fluidic reliability and temperature control.

Analytically, the integrated platform was able to identify the LCT −13910C>T C/C genotype (lactase non-persistence) and distinguish it from the C/T and T/T genotypes (lactase persistence), as reflected by the observed true-positive and false-positive rates. While the positive predictive value (PPV) was moderate (65.0%), the negative predictive value (NPV) reached 100%. The combination of moderate PPV and excellent NPV suggests that the assay may be particularly useful in clinical settings where reliable exclusion of lactase non-persistence is required. However, these findings should be interpreted in the context of the present proof-of-concept study and require confirmation in larger prospective clinical cohorts. The total assay time of approximately 50 min compares favourably with previously reported genetic testing approaches.

The selected classification strategy was intentionally optimized to maximize sensitivity for identifying the clinically relevant C/C genotype, thereby prioritizing sensitive identification of lactase non-persistence. Consequently, no false-negative C/C classifications were observed among the analytically valid measurements, resulting in a negative predictive value of 100%. Misclassification was observed predominantly among heterozygous (C/T) samples. This behaviour is expected because heterozygous samples inherently generate intermediate AU_T/AU_C ratios located close to the decision thresholds, making them substantially more susceptible to analytical variation than homozygous samples. The observed misclassification therefore most likely reflects the expected overlap of intermediate signal ratios near the decision threshold rather than inadequate allele-specific probe hybridisation. The lower specificity (80.6%) and positive predictive value (65.0%) indicate that some lactase-persistent individuals (C/T or T/T) would be incorrectly classified as C/C. From a clinical perspective, these false-positive classifications would primarily lead to unnecessary confirmatory testing, whereas individuals with genetically determined lactase non-persistence would not be missed. Nevertheless, further optimisation of the assay chemistry, signal evaluation algorithms and overall system robustness will be required to improve specificity before routine clinical application.

The remaining limitations—particularly the incomplete separation of all three genotypes—reflect both the current operational maturity of the integrated reader–cartridge platform and the need for further optimisation of assay classification. A formal analytical limit of detection (LOD) was not determined in the present study because the primary objective was demonstration of integrated genotype discrimination using clinically relevant DNA concentrations. Determination of the LOD will therefore be part of future analytical validation studies.

Importantly, the demonstrated compatibility with chemiluminescence-based readout provides a solid foundation for future multiplexed SNP detection within the same chip and reader architecture, thereby expanding the diagnostic versatility of the platform.

Conclusion

In this work, we demonstrated the successful design and roll-to-roll fabrication of lab-on-foil microfluidic chips for single-nucleotide polymorphism (SNP) diagnostics, combined with their integration into a compact reader platform enabling fully automated LAMP amplification and on-chip chemiluminescence detection. The analytical performance of the system was evaluated using 67 DNA samples, demonstrating on-chip identification of the clinically relevant C/C genotype of the LCT −13910C>T variant under analytically valid operating conditions. Importantly, the presented manufacturing and system architecture provide a clear pathway toward scalable product development, clinical validation, and the implementation of multiplexed diagnostic panels. Beyond the specific application of lactose intolerance testing, the platform establishes a versatile technological foundation for decentralized, rapid molecular diagnostics at the point of care.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The authors used generative AI tool (e.g., ChatGPT, Gemini) as a productivity and editing assistant for minor tasks, including checking for spelling and grammar, style consistency, and improving sentence clarity. The AI tool did not perform any core tasks such as data analysis, interpretation, or the generation of original content or arguments. All intellectual contributions and creative works are the sole responsibility of the authors.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/frlct.2026.1886537/full#supplementary-material

Supplementary material Supplementary material at the source

Supplementary material Supplementary material at the source

Footnotes

Edited by: Xiangchun Xuan, Clemson University, United States

Reviewed by: Mohammad Ameen Qasaimeh, New York University Abu Dhabi, United Arab Emirates

Siamak Salami, Shahid Beheshti University of Medical Sciences, Iran