Compact mid-infrared fiber probe for in vivo multi-compound monitoring demonstrated using ex vivo human skin

Abstract

Simultaneous monitoring of multiple biomarkers in tissues is critical for biomedical applications. However, few existing platforms enable concurrent in vivo detection. This study presents a compact mid-infrared transflection optical fiber probe for label-free, simultaneous monitoring of three physiologically relevant biomarkers – ethanol, glucose, and lactate. The probe comprises two silver halide fibers - one with an angled tip and one gold-coated as mirror - housed in polyetheretherketone tubing and surrounded by a semi-permeable membrane. With an outer diameter of only 1.1 mm, this is the smallest mid-infrared transflection probe reported to date. Coupled with a quantum cascade laser, the probe achieves ~1 mM detection limits for the three compounds. Peak deconvolution was deployed to resolve overlapping spectral features, enabling quantification of individual compounds in mixtures. Validation was performed in ex vivo human skin against microdialysis. Additionally, monitoring of the concentration changes for all three compounds in the skin was demonstrated.

This study reports a 1.1-mm diameter mid-infrared transflection fiber probe, the smallest reported to date, enabling label-free, simultaneous detection of ethanol, glucose, and lactate, with validation in ex vivo human skin and comparison to microdialysis.

Introduction

Real-time monitoring or detection of small-molecule metabolites such as ethanol, glucose, and lactate is crucial for advancing precision medicine, early disease detection, managing metabolic disorders, guiding therapeutic decisions, and improving patient outcomes in critical care settings^1–4. These molecules serve as important biomarkers across diverse physiological and pathological conditions; glucose and lactate levels are used to assess brain metabolic dysfunction following traumatic brain injury (TBI), where elevated lactate and reduced glucose are associated with poor outcomes⁵. Continuous glucose monitoring is fundamental in diabetes management⁶, while elevated lactate serves as a critical prognostic marker in sepsis⁷. Ethanol monitoring is important in contexts such as alcohol intoxication and addiction treatment⁸. Monitoring changes in concentration over time can help with better understanding of the metabolic pathway, which is important for developing new treatments and therapeutics. Ethanol metabolism also influences glucose and lactate dynamics due to hepatic metabolic competition and mitochondrial effects^9,10. In scenarios such as ethanol-induced liver injury¹¹, simultaneous monitoring of ethanol, glucose, and lactate provides a more complete picture of the metabolic status. Table 1 summarizes clinical relevance and applications of glucose, lactate and ethanol that underscore the importance of developing a sensing platform capable of measuring them simultaneously.

Table 1.

Physiological levels, clinical relevance, importance of simultaneous monitoring and applications for ethanol, glucose and lactate

Compound	Typical physiological concentration in tissue	Clinical and physiological significance	Importance of multi-analyte monitoring	Applications
Ethanol	Negligible under normal conditions; can reach 100 mM during intoxication46.	Relevant in intoxication, addiction, brain injury47; influences glucose and lactate metabolism9; significant for understanding neural effects and systemic toxicity48.	Ethanol alters glucose and lactate levels due to hepatic metabolic competition and mitochondrial effects9,10; combined monitoring provides insights into metabolic disruption during alcohol exposure or injury.	Addiction treatment, forensic toxicology, ICU monitoring, alcohol-related brain injury research.
Glucose	~3–9 mM in blood and in interstitial fluid49.	Essential source of energy50; critical for brain metabolism50; primary target in diabetes management6; changes indicate metabolic dysfunction after traumatic brain injury51.	Glucose metabolism is linked to lactate production via glycolysis52; ethanol metabolism competes with glucose utilization and affects blood glucose regulation10.	Diabetes monitoring, metabolic studies, traumatic brain injury assessment.
Lactate	~0.3–1.3 mM under normal conditions53; elevated ( > 5 mM) in hypoxia, sepsis, trauma53.	Indicator of anaerobic metabolism54; elevated levels signal tissue hypoxia, sepsis, or metabolic stress55; serves as a prognostic marker in critical care56.	Ethanol metabolism alters NAD⁺/NADH balance, impacting lactate levels57; lactate rises when glucose metabolism is impaired58.	Sepsis monitoring, trauma evaluation, sports medicine, metabolic research.

Conventional approaches, such as microdialysis, are widely used for in vivo and ex vivo monitoring of biochemical markers^12,13. Microdialysis allows localized sampling of interstitial fluid for subsequent offline analysis^14,15, and thus inherently does not provide real-time continuous measurement and suffers from low temporal resolution due to slow perfusion rates and long sampling intervals. Additionally, accurate quantification of concentration in tissues requires tedious calibration procedures. Furthermore, this extraction-based technique can deplete molecules locally in the tissue surrounding the microdialysis probe, resulting in changes of the chemical environment and leading to lower concentrations in the collected sample. Electrochemical sensing, particularly through enzyme-based detection, is the primary approach for monitoring metabolites such as glucose and lactate, providing rapid, real-time measurements and integration into wearable and implantable devices^16–21. Although effective, electrochemical sensors are limited by fouling and the need for immobilized enzymes with finite lifespans and degradation of the electrode²². Raman-based probes enable label-free detection with low interference from water; however, they typically require high excitation power and are susceptible to background fluorescence in biological media²³. Although recent studies, such as an implantable near-infrared spectroscopy sensor²⁴, have demonstrated simultaneous monitoring of ethanol, glucose and lactate in biological tissue, these systems involve surgical implantation which increases the overall cost and the complexity of the procedure. Moreover, the relatively large device size, requiring an incision of approximately 3.5 cm for implantation, restricts its use to specific tissue sites and limits broader medical applicability.

Mid-infrared (IR) spectroscopy enables molecular detection in tissues by directly probing the fundamental vibrational fingerprints of molecular bonds. This approach allows chemical identification without the need for additional targeting markers and provides distinct spectral features that are not captured in the overtone-dominated near-IR region. Each compound shows unique absorbance peaks in the mid-IR region (wavelengths 2.5–20 μm, corresponding to wavenumbers 4000–500 cm⁻¹), which result from characteristic bond vibrations. This enables direct molecular identification and quantification. Several mid-IR sensing modalities have been explored for chemical sensing purpose; Attenuated total reflectance (ATR) spectroscopy, which measures evanescent-wave absorption from a sample in contact with a high-refractive index crystal, has been used for analyzing glucose^25,26, and lactate²⁷. Due to the limited light-matter interaction, the sensitivity is generally lower than that of other approaches. Evanescent-wave fiber sensors utilize uncladded, bent or tapered mid-IR fibers - typically made from silver halide or chalcogenide materials - where the evanescent field extends beyond the fiber surface to interact with surrounding medium. Bending or tapering increases the extent of the evanescent field interacting with the sample, which enhances sensitivity. In a previous study²⁸, glucose was measured using a triple-coiled silver halide fiber sensor with a bend radius of 2.5 mm. Other studies have demonstrated the use of evanescent-wave sensors for detecting volatile organic compounds in aqueous environments²⁹, as well as for keratin quantification³⁰ and for distinguishing between healthy and cancerous skin tissue³¹. Fiber-optic sensors operating in transmission mode, where light passes directly through the sample between two aligned fibers, have been used to measure various concentration of glucose in different synthetic mixtures³². However, the implementation of evanescent-wave and transmission fiber sensors in tissues remains challenging, as their size and configuration can cause tissue damage (e.g., transmission measurements require alignment of two fibers within the tissue). To minimize the invasive impact, alternative design for fiber probes, such as a transflection configuration, is preferable³³.

In this study, a transflection fiber probe operating in the mid-IR region with an average pathlength of 63 µm was developed. The mid-IR transflection fiber probe, which uses a quantum cascade laser (QCL) as its light source, has an overall diameter of 1.1 mm. The sensing area of the probe is surrounded with a semi-permeable membrane having pores of 2-5 nm to prevent direct contact of the silver halide fiber with the tissue to enhance biocompatibility. Additionally, the membrane excludes large particles such as cells and proteins from the sensing region, thus preventing biofouling and reducing potential interference in the IR measurement. The probe was first tested by measuring ethanol, glucose and lactate in aqueous solutions. The temporal response of the fiber probe was investigated in a glucose solution, comparing probes with and without a membrane. The performance of the probe was validated in a tissue by measuring ethanol concentration over time in ex vivo human skin against a standard technique for sampling small molecules in tissues – microdialysis. Additionally, ethanol, glucose and lactate concentrations were measured in the skin sample simultaneously to demonstrate the multi-compound sensing capability. Peak deconvolution of the measured spectra was performed to resolve overlapping spectral features to quantify the concentrations of each compound. A rapid injection–dilution experiment was performed to show that the probe is capable of capturing concentration changes in tissue with a temporal resolution of 30 seconds. To estimate chemical stability and biocompatibility of the fiber probe, the dissolution of silver halide fiber was measured in ultrapure water over 7 days. The results showed that the maximum silver ion concentration remained below 26 ppb. This design demonstrates a versatile platform for multi-compound sensing in a biologically relevant environment.

Results and Discussion

FTIR spectroscopy of ethanol, glucose, and lactate in aCSF

To establish a reference for evaluating the performance of the fiber probe, mid-infrared absorbance spectra of ethanol, glucose, and lactate in artificial cerebrospinal fluid (aCSF) were measured using a benchtop Fourier-transform infrared (FTIR) spectrometer equipped with a 50 μm pathlength liquid transmission cell. aCSF was selected as the solvent to simulate the ionic environment of biological fluids.

Figure 1a–c show the absorbance spectra of ethanol, glucose, and lactate, respectively, measured over the spectral range of 1000–1180 cm⁻¹. All compounds were measured at concentrations between 2.5 and 40 mM. The spectrum of aCSF was used as the background for absorbance calculation. Ethanol has two prominent C–O stretching bands at 1046 and 1088 cm⁻¹, consistent with previous literature³⁴. Glucose shows multiple peaks, resulting from C–O stretching vibrations, as expected, the peak positions are centered at 1036, 1066, 1082, 1110, and 1154 cm⁻¹³⁵. Lactate presents three distinguishable peaks centered around 1042, 1086, and 1124 cm⁻¹. The peak at 1042 cm^-1 corresponds to C-C stretching, and the peaks at 1086 and 1124 are corresponding to C–O stretching vibrations³⁶.

Fig. 1. FTIR spectroscopy and calibration of ethanol, glucose and lactate.

FTIR absorbance spectra of a ethanol, b glucose and c lactate in aCSF at various concentrations between 2.5 and 40 mM measured with a pathlength of 50 µm. d Calibration curve showing peak heights versus concentration for each compound. Error bars represent the standard deviation (N = 3).

For quantification, a single absorbance peak was selected for each compound based on the signal intensity and linear response to concentration. Specifically, 1046 cm⁻¹ was used for ethanol, 1036 cm⁻¹ for glucose, and 1124 cm⁻¹ for lactate as indicated by the red shaded regions. The peak heights were determined by fitting Gaussian functions to the experimental spectra using nonlinear least-squares curve fitting, as described in the Methods section. The calibration curves shown in Fig. 1d demonstrate strong linearity across the tested concentration range, with correlation coefficients (R²) exceeding 0.99 for all three compounds. The sensitivities, defined as the slope of the calibration curves, for ethanol, glucose and lactate are 6.3 × 10^-4 1/mM ± 5.7%, 2.0 × 10^-3 1/mM ± 3.7%, 6.8 × 10^-4 1/mM ± 4.9%, respectively. The uncertainties correspond to the relative standard error calculated from three measurements. Among the three, glucose has the highest slope in the calibration curve, reflecting its stronger mid-IR absorption due to its higher number of interacting chemical bonds. Lactate showed a slightly steeper slope than ethanol.

The limits of detection (LoD), calculated as three times the standard deviation of the blank divided by the calibration slope, were 1.8 mM ± 5.4% for ethanol, 1.3 mM ± 4.0% for glucose, and 1.2 mM ± 3.7% for lactate. These results confirm the capability of mid-IR spectroscopy to detect these biologically relevant molecules in aqueous environments at physiologically meaningful concentrations, as summarized in Table 1.

Measurement of ethanol, glucose, and lactate solutions using the transflection fiber probe

To evaluate the performance of the developed mid-infrared transflection fiber probe for chemical sensing in aqueous biological environments, ethanol, glucose, and lactate dissolved in artificial cerebrospinal fluid (aCSF) were measured and directly compared with the FTIR results.

Figure 2a–c present the absorbance spectra of ethanol, glucose, and lactate, measured using the fiber probe with an average optical pathlength of 63 μm. The same characteristic peaks identified in the FTIR reference measurements were used for quantification: 1046 cm⁻¹ for ethanol (C–O stretch), 1036 cm⁻¹ for glucose (C–O stretch), and 1124 cm⁻¹ for lactate (C–O stretch), as indicated by the red shaded regions. The peak heights were extracted by Gaussian fitting, as in the FTIR measurements. Figure 2d shows the corresponding calibration curves for all three compounds, each showing linearity with R² values exceeding 0.99. The sensitivities for ethanol, glucose and lactate are 4.3 × 10^-4 1/mM ± 3.3%, 9.4 × 10^-4 1/mM ± 2.8% and 4.5 × 10^-4 1/mM ± 4.0%, respectively. The uncertainties correspond to the relative standard error calculated from three measurements. Notably, glucose again has the highest sensitivity among the three compounds. Lactate showed a slightly higher sensitivity than ethanol, with a difference of approximately 5%. Although the fiber probe has a longer geometric pathlength than the FTIR flow cell, it has lower sensitivities for ethanol, glucose and lactate measurements. The reported 63 µm optical path length corresponds to the geometric spacing between the angled fiber tip and the gold-coated fiber. However, the effective optical path length may be shorter. In addition, refractive-index differences among the analyte solutions can slightly alter refraction and reflection at the interfaces, resulting in analyte-dependent variations in sensitivity. These effects are fully accounted for through the calibration curves.

Fig. 2. Transflection fiber probe measurements and calibration of ethanol, glucose and lactate.

Absorbance of a ethanol, b glucose, and c lactate in aCSF measured using the mid-IR transflection fiber probe with an averaged optical pathlength of 63 μm. d Calibration curve showing peak height versus concentration for each compound. Error bars represent the standard deviation (N = 3).

The limits of detection (LoD), determined as three times the standard deviation of the blank divided by the calibration slope, were 1.0 mM ± 3.1% for ethanol, 0.81 mM ± 2.7% for glucose, and 1.1 mM ± 3.7% for lactate. These values are lower than those obtained from the benchtop FTIR setup described in the previous section. Despite lower sensitivity, the fiber probe achieved better limits of detection than the FTIR system, which may be attributed to the high intensity QCL light source, which enables better signal-to-noise ratios. To identify the main source of noise in the fiber-probe system, 10 repeated background spectra with the probe in air and 10 spectra with the laser output measured directly at the detector were recorded. The standard deviation at each wavenumber was calculated and normalized by the corresponding mean intensity to obtain the relative noise. The average relative noise over the 1000–1180 cm⁻¹ region was 0.025% for the direct laser measurement and 0.078% for the fiber-probe. The result indicates that, compared to detector and laser noise, propagation and coupling in the silver halide fiber contribute the majority of the overall noise in the system.

One key advantage of the transflection probe is its suitability for in vivo environment. Unlike FTIR transmission cells, which require samples to be contained within a flow cell, the fiber probe interacts directly with the surrounding medium, making it ideal for in situ measurements in tissue.

These results confirm that the mid-IR transflection fiber probe offers sensitive detection of small biomolecules in aqueous environment, and provides a promising platform for biochemical sensing, especially in settings where traditional FTIR setups are impractical.

Simultaneous quantification of mixture of ethanol, glucose, and lactate using peak deconvolution

To demonstrate the capability of simultaneously measuring multiple compounds, mixtures of ethanol, glucose, and lactate in aCSF were measured and analyzed. The same fiber probe with an average optical pathlength of 63 μm was used. Three spectral scans were averaged for each measurement, resulting in a total acquisition time of 1.5 minutes. Peak deconvolution was applied to resolve overlapping absorbance features and quantify each component.

Figure 3a shows the mid-IR absorbance spectrum of a mixture containing 20 mM of ethanol, 20 mM of glucose, and 20 mM of lactate and the Gaussian fits for each compound where the position of peaks were identified based on prior FTIR calibration. In the spectra of mixture, Gaussian functions were fitted at 1046 and 1088 cm⁻¹ for ethanol; at 1036, 1066, 1082, 1110, and 1154 cm⁻¹ for glucose; and at 1042, 1086, and 1124 cm⁻¹ for lactate. To quantify each compound in the mixture, the specific peak heights of the Gaussian functions at 1046, 1036, and 1124 cm^-1 were extracted for ethanol, glucose, and lactate, respectively. The concentrations were determined using the calibration curves in Fig. 2d. The combined fit achieved an R² of 0.995, indicating good agreement between the measured spectrum and the sum of individual spectral contributions.

Fig. 3. Quantification of mixture of ethanol, glucose and lactate using peak deconvolution.

a Absorbance spectrum of a mixture containing 20 mM ethanol, 20 mM glucose, and 20 mM lactate measured using the fiber probe with an averaged optical pathlength of 63 μm, overlaid with the peak deconvolution fit. b Absorbance spectra of three different mixtures with varying concentrations of glucose, lactate and ethanol used to validate the deconvolution method.

To validate the peak deconvolution method, three standard mixtures containing different known concentrations of each compound were tested as summarized in Table 2. Figure 3b shows the corresponding infrared spectral features of these standard mixtures. The average relative prediction errors, defined as the percentage difference between the predicted and actual concentrations relative to actual concentration, for ethanol, glucose and lactate in the three standard mixtures were 10.8%, 5.2%, and 3.3%, respectively, demonstrating that quantitative multicomponent analysis is feasible with the developed fiber probe. Additionally, the accuracy of ethanol quantification was slightly more variable, which may be due to peak overlap with both lactate and glucose. Lactate showed the highest accuracy, likely due to its characteristic peak being separated from those of other compounds, resulting in less spectral interference. Glucose quantification was also reliable, with deviations typically below 1 mM across all conditions.

Table 2.

Prediction of concentration of each compound in the mixture. Values are presented as mean ± relative standard error

Compound	Mixture 1		Mixture 2		Mixture 3
	Actual concentration (mM)	Predicted concentration (mM)	Actual concentration (mM)	Predicted concentration (mM)	Actual concentration (mM)	Predicted concentration (mM)
Ethanol	20	22.1 ± 5.8%	10	9.0 ± 6.7%	10	11.2 ± 4.5%
Glucose	10	9.2 ± 3.3%	20	19.3 ± 4.7%	10	9.6 ± 4.2%
Lactate	10	10.4 ± 2.9%	10	10.3 ± 3.9%	20	20.4 ± 3.4%

However, due to spectral overlap, the use of peak deconvolution introduces greater uncertainty into the concentration measurements compared to single-component analysis. In single-compound measurements, the signal originates from a single compound, and the quantification relies on direct peak fitting. In contrast, when multiple compounds are present, their spectra overlap in the mid-IR region, leading to shared spectral features and baseline shifts. Deconvolution algorithm resolves these overlapping peaks by fitting multiple parameters, such as peak height and width, for each compound simultaneously. This process increases the degrees of freedom in the fitting model, which in turn propagates fitting errors and increases the variability of predicted concentrations.

Table 3 shows the relative standard error (RSE) of the peak heights used for quantifying ethanol, glucose, and lactate. The peak heights used to calculate the RSEs were measured both individually and within mixtures at concentrations of 10 and 20 mM. For all three compounds, the RSE is consistently higher in the multi-compound measurements compared to single-component analysis. For instance, ethanol shows an RSE increase from 3.4 to 8.9% at 10 mM and from 2.3 to 5.9% at 20 mM when comparing single to multi-compound detection. This trend is observed for glucose and lactate as well. The magnitude increase is smaller for lactate, likely due to its more isolated spectral peaks. This increase in RSE directly affects the LoD for each compound in mixtures. The LoD is defined as three times the standard deviation of the measurement divided by the slope of the calibration curve. As the standard deviation increases due to uncertainties in deconvolution, the LoD for multi-compound measurements becomes higher than that for single-compound measurements, even if the calibration slope remains similar. As a result, while the fiber probe demonstrates feasibility for simultaneous multi-compound sensing, the achievable detection limits in mixtures are higher than those in single-component analyses (1.0, 0.8, and 1.1 mM for glucose, ethanol and lactate) due to the uncertainty introduced by peak overlap and deconvolution fitting.

Table 3.

Relative standard error (RSE) of the peak heights used for quantifying ethanol, glucose and lactate

	RSE of ethanol peak height		RSE of glucose peak height		RSE of lactate peak height
	Individual (%)	In mixture (%)	Individual (%)	In mixture (%)	Individual (%)	In mixture (%)
10 mM	3.4	8.9	3.0	4.0	2.2	3.1
20 mM	2.3	5.9	2.7	4.6	3.4	3.5

RSE was calculated using peak heights measured as individual components and multi-compound mixtures at 10 and 20 mM concentrations.

These findings highlight the potential of the fiber probe for simultaneous multi-compound quantification, without the need for prior sample processing or the use of markers to target the molecules. This capability is particularly valuable in biological contexts where monitoring of metabolic markers such as glucose and lactate, along with ethanol, may offer insights into tissue health and metabolic state. However, this capability comes with the trade-off of higher detection limits when resolving complex mixtures through peak deconvolution. Advanced chemometric and machine-learning approaches such as partial least squares (PLS) or principal component regression could be implemented in future studies to improve sensitivity and reduce analytical variance in multi-compound measurements.

Probe’s performance with and without a semi-permeable membrane

In addition to considerations related to sensor dimensions and ability to quantify concentrations using IR, biofouling - the attachment of cell debris or adsorption of proteins on the fiber - can deteriorate the performance of the sensor. Additionally, previous literature have noted potential biocompatibility concerns for bare silver halide fibers in direct contact with cells³⁷. This effect can be mitigated through including a semi-permeable membrane or surface coating³⁷. Additionally, a semi-permeable membrane or coating can prevent direct contact of the silver halide fiber to the tissue to prevent potential risk of biocompatibility. A mPES membrane was chosen for this purpose. Its hydrophilic nature helps prevent bubble formation, and its 10 kDa MWCO allows diffusion of small molecules but blocks large particles such as proteins. However, with the membrane, the diffusion dynamics of ethanol, glucose and lactate may change. Two experiments were conducted to investigate the effect of membrane on the sensitivity (slope of the calibration curve) and the response time of the fiber probe, respectively.

In the first experiment, glucose calibration curve was measured following the procedure in “Measurement of ethanol, glucose, and lactate solutions using the transflection fiber probe” section using the same probe with and without a semi-permeable membrane around the sensing region. Measurements were started at least 5 minutes after each sample change to ensure that steady-state concentrations were reached. Figure 4a shows the calibration curves for glucose in aCSF, measured both with and without the membrane. The sensitivities of the probe with and without a membrane were 9.4 × 10^-4 1/mM ± 2.0% and 9.4 × 10^-4 1/mM ± 2.2%, respectively. The uncertainties correspond to the relative standard error calculated from three measurements. Difference between the two curves is within 1%. Similar sensitivity with and without membrane is expected, as 5 min is sufficient to reach steady-state concentration, as verified in the following section discussing the temporal response of the probe. This also applies to ethanol and lactate, since these smaller molecules diffuse faster than glucose. Consequently, the results demonstrate that the probe performs effectively, despite the presence of the membrane.

Fig. 4. Performance comparison with and without a semi-permeable membrane.

a Calibration curves for glucose in aCSF obtained using the probe with and without the semi-permeable membrane. Time-course infrared peak height at 1036 cm^-1 of 20 mM glucose measured using the fiber probe b without and c with a semi-permeable membrane. The shaded region indicates the period when the probe was immersed in the glucose solution. Error bars represent the standard deviation (N = 3).

In the second experiment, temporal response studies were conducted to examine the effect of a semi-permeable membrane on the fiber probe. Absorbance of 20 mM glucose was monitored over time with a single-scan acquisition time of 25 s, both in the presence and absence of the membrane. These experiments evaluate the sensor’s capacity to monitor dynamic concentration changes.

In both conditions, fiber probe with an average optical pathlength of 63 μm was used. The fiber was initially immersed in a vial of aCSF to record baseline signals (t < 0). At t = 0, the vial was changed to another vial containing 20 mM glucose. After a defined period, the vial was changed back to aCSF to observe the signal decay over time. The response curves are shown in Fig. 4b and Fig. 4c for the probe without a membrane and the membrane-protected probe, respectively. The response was quantified by calculating the time constant, defined as the time required for the signal to reach 90% of its maximum (rise).

For the probe without a membrane, the response was faster, with a time constant of 33.4 s ± 5.4%. In contrast, the membrane-protected probe showed a slower response, with a time constant of 59.1 s ± 5.9%. The increase in response time reflects the added diffusion resistance introduced by the membrane. The semi-permeable membrane will help with filtering out large macromolecules and particles, potentially reducing biofouling and signal interference, but it slows the transport of molecules to the sensing region. Despite the slower response, the membrane-protected probe still reached 90% of the steady-state signal within one minute, indicating that real-time monitoring remains feasible for many applications.

The decay curves after switching back to aCSF also support these findings. The decay time constants, defined as the times required for the signal to decrease to 10% of the maximum (when immersed in glucose solution), were 61.7 s ± 4.7% and 36.1 s ± 5.5% for probe with and without a membrane, respectively. The unprotected probe showed a sharp drop in signal, returning near baseline within 1 min. In comparison, the membrane-equipped probe displayed a more gradual signal decay, consistent with residual glucose diffusing out of the membrane-protected sensing zone.

Overall, these results demonstrate that while the membrane reduces the rate of mass transfer, and thus slightly compromising the temporal response, the concentration changes of ethanol, glucose and lactate in the human body typically occur slowly enough to be captured with a temporal resolution of 1 min^38–40. The choice of using a membrane should depend on the specific application requirements, whether fast response or selective filtering is prioritized.

Ex vivo skin measurement using the fiber probe and comparison with microdialysis

To compare the performance of the fiber probe and microdialysis, concentrations in ex vivo human skin were monitored by both techniques following ethanol spiking of the medium under the skin. The mid-IR fiber probe (with a semi-permeable membrane and an averaged optical pathlength of 63 µm) and the microdialysis probe were inserted into the same skin sample, positioned at depths of 2.7 and 2.9 mm, respectively. Ethanol was added to the culture medium at t = 0 to achieve a final concentration of 20 mM. Figure 5a shows the ethanol concentrations measured by both techniques over time. The concentrations increased after ethanol addition, with the rate of increasing slowing approximately 30 min later. The temporal concentration profiles reflect the gradual penetration of ethanol into the skin. The fiber probe measured a maximum tissue concentration of approximately 8 mM, which is lower than the spiked medium concentration, as expected. This discrepancy is due to diffusion barriers within the skin and underlying nourishing matrix, which slows down the transport process of ethanol to the region where the probes were embedded. Compared to microdialysis, which operates at a low perfusion rate (0.5 μL/min) and requires a 10 min sampling interval, the fiber probe provides better temporal resolution, allowing faster tracking of concentration changes over time. Additionally, the concentration measured using microdialysis was slightly lower than those obtained with the fiber probe. This may be attributed to the evaporation losses of ethanol during sample collection and handling required for the microdialysis process. Moreover, because the molecules diffused into the microdialysis probe were continuously carried to the sample collection vial at outlet, local depletion of molecules in the surrounding tissue can occur. This is a limitation not encountered with the non-extractive, in situ sensing mechanism of the fiber probe.

Fig. 5. Fiber probe measurements in ex vivo skin and comparison with microdialysis.

a Time-course comparison of ethanol measurements in the skin samples using fiber probe and microdialysis. Ethanol was spiked into the medium at t = 0 min to achieve a final concentration of 20 mM. b Simultaneous monitoring of ethanol, glucose, and lactate in the skin sample, with the medium spiked at 100 mM for each compound. c Monitoring of ethanol dynamics in the skin sample following the injection of 50 μL of 200 mM ethanol at t = 0 min. A rapid increase in ethanol concentration was observed, followed by dilution at t = 30 min with the addition of 20 μL of medium.

To demonstrate the fiber probe’s capability for measuring mixtures in ex vivo human skin, the culture medium beneath the skin sample was spiked with 100 mM ethanol, 100 mM glucose and 100 mM lactate and their concentrations were measured over time. As shown in Fig. 5b, all three compounds were simultaneously detected using peak deconvolution. Each compound has a distinct temporal concentration profile, even though their initial spiked concentrations were the same. Ethanol showed the highest concentration, while glucose and lactate showed lower levels. These differences result from their distinct diffusion coefficients of each molecule within skin tissue. Ethanol, due to its small size, diffuses more easily through both the nourishing matrix and the skin tissue above it compared to glucose and lactate. As a result, the steady-state concentration of ethanol is higher than that of glucose and lactate. Among the three, glucose is the largest molecule, leading to the slowest diffusion rate and the lowest steady-state concentration. Additionally, the skin cells in the samples remain metabolically active and use glucose for energy. However, the metabolic demand of the skin tissue is relatively small compared to the large glucose reservoir provided in the medium. As a result, any glucose consumption during the 90 min measurement period would be too small to detect, which is consistent with the experimental results shown in Fig. 5b.

To evaluate the temporal response of the fiber probe in capturing dynamic concentration changes, ethanol spectra were collected every 30 seconds using single-scan acquisition to minimize measurement time. Figure 5c shows measurements of ethanol concentration over time following the injection of 50 μL of 200 mM ethanol into the skin near the probe at t = 0 min. A quick increase in measured concentration was observed within the first 10 minutes. At t = 30 min, 20 μL of culture medium was added directly on the skin near the probe to quickly dilute the local ethanol concentration. The corresponding decrease was detected by the fiber probe. The results underscore the capability of the fiber probe to capture concentration variations on sub-minute timescales.

Overall, these findings highlight two main advantages of the mid-IR fiber probe over microdialysis: (1) in situ measurement without the need for sample extraction or processing, (2) high temporal resolution suitable for capturing dynamic changes. These features make the developed probe a promising tool for in vivo molecular monitoring in tissues. Future work will focus on further miniaturing both the probe and the overall measurement system, enhancing the probe’s sensitivity and validating its performance in in vivo models to advance toward clinical applications. A key challenge is replacing the current large infrared laser system and optical components with portable alternatives. Identifying suitable light sources and detectors that balance size, sufficient power, and cost-effectiveness will be critical for translating this technology into practical, handheld devices.

Evaluation of silver ion release and biocompatibility considerations

To assess the potential risk of the release of ions from the silver halide fiber in the probe for biological applications, the dissolution of silver halide in aqueous environments was investigated. The concentration of silver ions was measured as an indicator of the extent of fiber dissolution. This was quantified by immersing a fiber probe with a semi-permeable membrane in ultrapure water for 7 days. Samples were collected daily and analyzed using an ICP-MS.

As shown in Fig. 6, the measured silver ion concentration remained low throughout the test period, with a maximum concentration below 26 ppb. The control sample, prepared before fiber immersion, showed an undetectable silver ion concentration (0 ppb), confirming minimal background contamination. The highest concentration of silver ions is well below known cytotoxic thresholds and is considered safe for biomedical applications. Toxicity effects are generally reported at levels around 1000 ppb for human fibroblasts⁴¹. In addition, the amount of dissolved chloride and bromide ions can be estimated based on the measured silver ion concentration, as these halide ions are released in stoichiometric proportions during the dissolution of the silver halide material. Therefore, the concentrations of chloride and bromide ions are expected to be lower than the silver ion levels detected in the solution. Notably, the physiological concentrations of chloride and bromide ions in human body are substantially higher than these levels^42,43, further indicating that the small quantities potentially released from the fiber probe are unlikely to pose a health risk.

Fig. 6.

Silver ion concentration released from the silver halide fiber probe with a semi-permeable membrane immersed in ultrapure water over seven days.

In the proposed probe design, a semi-permeable PES membrane encloses the entire sensing region, creating a physical barrier that further prevents direct tissue contact with the silver halide fiber, which can enhance the overall biocompatibility of the device. Based on the low silver ion concentrations measured and the use of the membrane, the results support the potential use of the probe in short-term ex vivo and in vivo applications Future work will focus on long-term evaluations of the effect of silver halide fiber probe on tissue, to ensure safe use in vivo.

Methods

Optical fiber

Uncladded polycrystalline silver halide fibers (PIR 500, Art Photonics, Germany), composed of a solid solution of silver chloride and silver bromide, were used to fabricate the fiber probes. The fiber has a diameter of 500 μm, a numerical aperture (N.A.) of 0.3, and is transparent in the wavelength range of 3 to 17 μm, with transmission rate above 55% in the 8–10 µm range for a 1 m length and an optical loss of 0.2–0.3 dB/m in the 9–13 µm range.

Reagents

An artificial cerebrospinal fluid (aCSF) was prepared to simulate the chemical and ionic composition of the interstitial fluid, according to a previous study⁴⁴. The concentration of each component was as follows: NaCl at 149 mM (S9888, Sigma), KCl at 2.8 mM (409316, Fluka), CaCl₂·2H₂O at 1.2 mM (21097, Sigma), MgSO₄·6H₂O at 1.2 mM (63068, Sigma), ascorbic acid at 0.25 mM (95209, Fluka), glucose at 5.4 mM (G7021, Sigma). 200 proof ethanol (111000200, PHARMCO), glucose (G7021, Sigma) and lactate (L7022, Sigma) were used to prepare standard solutions in aCSF through serial dilution.

Skin tissue models

Abdominal skin (NativeSkin, Genoskin) collected from a 31-year-old Hispanic female donor after plastic surgery was split into multiple samples used for testing probe performance ex vivo. The skin samples have a diameter of 20 mm and a thickness of approximately 4.6 mm. Each skin biopsy is contained in a cell culture insert designed to fit standard six-well plates. The insert has a porous membrane at its base, allowing nutrient exchange between the culture medium and the skin tissue. A nourishing gel-like matrix of proprietary composition is positioned between the membrane and the skin, providing physical support and keeping the tissue chemically stable. A silicone ring circles the exposed surface of the skin, preventing the applied liquids from leaking into the culture medium. As a result, the working area of the skin surface is reduced to 1.76 cm². The skin sample was received one day after production. Upon delivery, the skin samples were refreshed with new medium and kept in culture at least 3 h prior to experiment. Experiments were performed in accordance with the manufacturer’s instructions and before the expiration date of the tissue. This study is not classified as human subject research, and no Institutional Review Board approval was required. Product certification from the manufacturer can be provided upon request.

Fourier-transform infrared spectroscopy (FTIR)

A Fourier-transform infrared (FTIR) spectrometer (Nicolet is50, Thermo Scientific), equipped with a liquid cell (Oyster, Specac) operating in transmission mode, was used to measure solutions of ethanol, glucose and lactate as a reference method. The liquid cell has an optical pathlength of 50 µm and CaF₂ windows, requiring a sample volume of 10 µL. The number of scans was 300, resulting in spectra collection time of 187 s, and the spectral resolution was 0.482 cm^-1. The spectrum of aCSF was used as the background for absorbance calculations in the experiments.

Headspace gas chromatography with flame ionization detector (GC-FID)

A gas chromatograph (GC, 430-GC, Bruker) equipped with a flame ionization detector (FID) and an autosampler (COMBI PAL-xt) was used to analyze the concentration of ethanol in the dialysate samples. CompassCDS (Bruker, Netherlands) software was used to record and analyze all chromatograms. A capillary column (30 m x 0.53 mm×1 µm film thickness) with nitrogen as the mobile phase, at a flow rate of 8.5 mL/min, was used. A carboxen/PDMS coated solid phase micro-extraction (SPME) fiber (Supelco Analytical, Bellefonte, PA) was used for extraction and desorption. The extraction time was set to 3 minutes, followed by a desorption period of 1 minute. The oven and injection temperature were maintained at 65 °C and 220 °C, respectively.

Fabrication of microdialysis probes

Microdialysis probes were assembled in a side-by-side configuration, following the procedure described in a prior study⁴⁴. Briefly, two pieces of fused silica tubing used as the inlet and outlet lines and were inserted into a segment of regenerated cellulose membrane (132280, Spectra/Por) with a 13 kDa molecular weight cutoff (MWCO), which corresponds to an effective pore size of 2–5 nm⁴⁵. The active dialysis region was determined by adjusting the relative insertion lengths of the tubing inside the membrane and was set to 2 mm. Epoxy was applied externally to coat the region that is not used for dialysis.

Fabrication of optical fiber probes

Figure 7a illustrates the schematic design of the mid-infrared transflection fiber probe. The probe consists of two aligned silver halide optical fibers. One fiber is used for light delivery and collection, and the other serves as a reflective mirror. The fiber mirror was fabricated by coating the end-face of a short segment of silver halide fiber with a gold layer of approximately 267 nm thickness, using a sputter coater (6002, Ted Pella, Inc.). Before gold sputtering, the fiber end was polished using aluminum oxide lapping film (50-20060, Allied High Tech Products, Inc.). The other fiber, used for delivery and collection of the IR light, was polished at a 7° angle on its end face facing the fiber mirror using a custom-made fiber holder. This angled fiber tip helps reduce Fresnel back reflections by redirecting the reflected light away from the fiber core.

Fig. 7. Schematic of the transflection fiber probe and experimental setup.

a Configuration of the transflection fiber probe. b Photograph of the probe. c Schematic of the optical setup for the transflection probe measurements. d Schematic of the experimental setup using ex vivo skin. Both the transflection fiber probe and lab-made microdialysis probe were inserted into the dermis and secured in place, with the setup maintained at 37 °C. The photograph on the right shows the fiber probe (left) and microdialysis probe (right) inserted in the human skin sample inside the cell culture insert. e Schematic of the injection experiment, where ethanol was injected into the skin near the probe.

Both fibers were held and aligned face-to-face inside a custom-fabricated connector made from a 6 mm-long hollow polyetheretherketone (PEEK) tube (1532 L, IDEX Health and Science) with an inner diameter of 510 μm and an outer diameter of 1588 μm. The outer surface of the PEEK tube was machined down to an outer diameter of approximately 850 μm to fit within a semi-permeable membrane (modified polyethersulfone (mPES), 10 kDa MWCO corresponding to effective pore size of around 3 nm). The alignment of the two fibers was adjusted under a microscope to control the optical pathlength – twice the distance between the two fibers. The average optical pathlength of the fiber probe was 63 μm, which is twice the average distance between the fiber tip and the fiber mirror. This average distance was determined by measuring the shortest and longest distances between the angled fiber tip and the fiber mirror under a microscope and calculating their mean. The space between the fiber tip and the fiber mirror defines the sensing region, which requires a sample volume of 13 nL to fill completely. To enable molecules to reach the sensing region, a 2 mm-wide groove was created in the PEEK tube along the wall using a hand drill. The fiber probe was inserted into a 3 mm-long piece of the semi-permeable membrane enclosing the tube and covering the sensing window and secured using epoxy at both ends. This configuration allows small molecules to diffuse through the membrane into the sensing region while excluding large particles such as proteins. The overall outer diameter of the final probe is 1.1 mm. The photograph of the final assembled probe is shown in Fig. 7b. Before measurement, the probe was pre-wetted with a 25% ethanol solution, which wets the narrow sensing gap due to its lower surface tension and displaces any trapped air. The probe was then immersed in water for at least 15 min to allow the ethanol to diffuse out prior to use. This procedure ensures that the sensing region is fully filled with liquid and free of air bubbles during experiments.

Optical setup

Figure 7c shows the schematics of the experimental setup. A mid-infrared quantum cascade laser (QCL, MIRcat, DRS Daylight Solutions) was used as the light source, delivering a spectral range of 800–1800 cm⁻¹ (5.56–12.5 μm). A focusing coupler (Art Photonics GmbH), consisting of a gold mirror (39-193, Edmund Optics) and an off-axis parabolic mirror (35-488, Edmund Optics), was installed at the laser output, as shown in the schematics. The mirror mounted inside the focusing coupler features three adjustment screws for fine-tuning the x, y, and z positions, enabling signal optimization at the FC-PC connector output.

The output of the focusing coupler was connected to one input of a fiber splitter (Art Photonics GmbH), which combined two silver halide fibers (500 µm diameter) for delivering the laser light and transmitting the return signal to the detector, as illustrated in the schematics. The two fibers of the splitter were bundled together into a single fiber cable, which was connected to the transflection probe through an SMA connector. The fiber lengths were 23 cm for both the source-to-SMA connector and detector-to-SMA connector paths, and 5 cm from the SMA connector to the sensing region. The fiber splitter provides an approximately 50/50 division of the optical signal between the two fibers. To connect the transflection probe to the fiber splitter, a 3D-printed hollow cylinder was fabricated with an inner diameter matching that of the fiber probe for secure insertion, and an outer diameter designed to tightly fit the SMA connector port. The signal carrying spectral information was directed to a thermoelectrically cooled infrared detector (PVI-4TE-10.6, VIGO Photonics) using a dual-mirror fiber coupler (Art Photonics GmbH). This coupler consists of two off-axis parabolic mirrors (MPD229-P01, ThorLabs) that collect and focus the light onto the detector, as illustrated in the schematic.

Ex vivo skin measurements

To assess the performance of the mid-IR transflection fiber probe, ex vivo experiments were conducted using human abdominal skin samples. For direct comparison of the fiber probe and microdialysis, both the optical fiber probe and a lab-made microdialysis probe were inserted into the dermis of the same skin sample, positioned approximately 2.7 mm and 2.9 mm below the top skin surface, respectively, as shown in Fig. 7d. Insertion of both probes was performed by first making a small incision in the dermis using a scalpel, followed by piercing the tissue with a 27-gauge needle to create a tract for the probes. The fiber and microdialysis probes were then carefully inserted along this path.

The fiber probe was used to acquire mid-IR spectra in situ, while the microdialysis probe served as a reference method for sampling extracellular molecules. The culture insert, which contains skin tissue, placed in a six-well plate containing medium, was kept inside a humidified incubator at 37 °C for the duration of the experiments. The microdialysis system included a syringe pump delivering artificial cerebrospinal fluid (aCSF) at a constant perfusion rate of 0.5 μL/min. Dialysate was collected in vials at 10 min intervals. A 2 μL aliquot from each dialysate sample was transferred to a glass vial for subsequent analysis using GC-FID.

To directly compare the measurement results of the fiber probe and the microdialysis probe, both were inserted in the skin sample, and ethanol concentration measurements were initiated simultaneously upon spiking the culture medium with ethanol to a concentration of 20 mM. During this experiment, spectra were acquired every 1.5 min, corresponding to an average of 3 scans, which provided sufficient noise reduction while maintaining the ability to resolve concentration changes on the timescale of diffusion through the skin tissue. The total experiment duration was 90 min, allowing both the initial rapid increase in analyte concentration and the subsequent approach to steady state within the tissue to be captured. The results obtained with both techniques are presented and compared in the first paragraph of “Ex vivo skin measurement using the fiber probe and comparison with microdialysis” section.

For simultaneous multi-compound quantification, only the optical fiber was inserted in the skin sample. The medium was spiked with a mixture of 100 mM ethanol, 100 mM glucose, and 100 mM lactate. These high concentrations in the medium were used in order to achieve physiologically relevant concentrations in the dermis of the skin. Spectra were acquired every 1.5 min, and the acquired spectra were analyzed using peak deconvolution to resolve and quantify the overlapping absorbance features. The results of simultaneous multi-compound quantification are presented in the second paragraph of “Ex vivo skin measurement using the fiber probe and compison with microdialysis” section.

To evaluate the fiber probe’s capability for capturing dynamic changes in concentration, an injection experiment was performed with the fiber probe in the tissue, as shown in Fig. 7e. A volume of 50 μL of 200 mM ethanol was injected into the skin at a site approximately 3 mm from the probe, followed by a dilution step in which 20 μL of culture medium was added directly at the insertion site of the probe to quickly reduce the local ethanol concentration. During this experiment, spectra were acquired every 30 s using single-scan acquisition to minimize measurement time. The results of tracking ethanol concentration over time are presented in the third paragraph of “Ex vivo skin measurement using the fiber probe and comparison with microdialysis” section.

Spectral processing and peak deconvolution

All infrared absorbance spectra were processed and analyzed using MATLAB (R2023a, The MathWorks, Inc.). The spectra were first smoothed using a moving average method with a window size of 9 points. This window size was chosen to preserve peak shape and intensity while reducing high-frequency noise. Following smoothing, baseline correction was performed. An asymmetric least squares algorithm was used to estimate and subtract the baseline. This method iteratively minimizes a cost function that balances how closely the baseline follows the data and the smoothness of the estimated baseline. The two parameters, penalization and asymmetry parameters, were tuned empirically to achieve effective baseline correction across the spectral range of 1000–1180 cm⁻¹.

To resolve overlapping spectral features and quantify individual compounds, peak deconvolution was performed using nonlinear least-squares curve fitting with Gaussian functions. Each compound (ethanol, glucose, and lactate) was characterized by a predefined set of characteristic peaks, identified from measured infrared absorbance spectra of individual compounds using the fiber probe and validated with the reference FTIR spectra. The positions of the peaks were fixed based on experimental calibration using fiber probe, while the amplitude and width were allowed to vary during fitting. The peak height of each fitted Gaussian function was used for concentration quantification. Calibration curves relating peak height to concentration were generated for each compound.

To predict the concentration of each compound in the mixture, peak deconvolution was performed using all the characteristic peaks from the absorbance spectra of the pure compounds. The concentrations were then estimated using the corresponding calibration curve. Goodness-of-fit was assessed by calculating the coefficient of determination (R²) between the experimental and fitted spectra.

Investigation of silver halide fiber dissolution

To evaluate the potential biocompatibility risks associated with the silver halide fiber probe, a dissolution study to measure the release of silver ions over time was conducted. A fully assembled fiber probe was submerged in 12 mL of ultrapure water and stored at room temperature for 7 days. 1 mL of the solution was withdrawn every day. Prior to analysis, each collected sample was diluted to a final volume of 5 mL with ultrapure water and acidified with nitric acid (HNO₃) to a final concentration of 2% (v/v), as required for inductively coupled plasma–mass spectrometry (ICP-MS) measurements.

The concentration of silver ions in the samples was quantified using an ICP-MS (Agilent Technologies) in solution mode. A baseline control sample (ultrapure water before the insertion of probe) was included to confirm background levels.

Author contributions

T.-A.L: Conceptualization, data curation, formal analysis, investigation, methodology, visualization, writing – original draft; T.H.: conceptualization, funding acquisition, project administration, resources, supervision, writing – review and editing.

Peer review

Peer review information

Nature Communications thanks Markus W. Sigrist, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon request.

Competing interests

The authors declare that Tanya Hutter and Tse-Ang Lee are inventors on a pending and unpublished U.S. provisional patent application (No. 63/749,262) related to the work described in the manuscript, and it is owned by The University of Texas at Austin.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-70300-x.