Multimodal optical system for quantitative discrimination of bilirubin species

Abstract

Bilirubin occurs in human serum in multiple molecular forms: free unconjugated, albumin-bound, and conjugated species, and distinguishing between them is clinically relevant for assessing liver function and bilirubin metabolism. However, current routine assays do not resolve these subtypes individually. Here, we present a compact, multimodal optical platform that combines absorbance spectroscopy, fluorescence intensity, fluorescence anisotropy, and photobleaching analysis to investigate the optical signatures of bilirubin species in solution. Using well-defined synthetic standards representing free, bound, and conjugated bilirubin, we characterize each modality's discriminative capability. Absorbance measurements enable total bilirubin determination, fluorescence provides high sensitivity to albumin-bound bilirubin, and anisotropy measurements, done with a plate reader, reveal different responses to each species through their differing rotational mobilities. Photobleaching kinetics further highlight species-dependent photostability under controlled irradiance conditions. Taken together, these complementary, label-free optical readouts enable improved resolution of bilirubin subtypes in synthetic samples and establish a foundation for future translational studies in biologically complex matrices.

1. Introduction

Bilirubin is a linear tetrapyrrole formed during the degradation of heme, which is released primarily from senescent red blood cells. The breakdown process begins in macrophages of the reticuloendothelial system, where heme is enzymatically converted to biliverdin by heme oxygenase [1,2]. Biliverdin is then reduced to bilirubin by biliverdin reductase. The resulting bilirubin is hydrophobic and poorly soluble in plasma; to enable safe transport, it binds tightly (but non-covalently) to serum albumin [3,4]. This circulating form, known as unconjugated bound bilirubin (Bb), constitutes the majority of bilirubin in the bloodstream under physiological conditions. A very small fraction may exist unbound (free bilirubin (Bf)), which is biologically active and potentially neurotoxic due to its ability to cross the blood–brain barrier [5].

In hepatocytes, unconjugated bilirubin undergoes enzymatic conjugation by UDP-glucurono-syltransferase (UGT1A1), forming first bilirubin monoglucuronide and then bilirubin diglucuronide. These conjugated forms (Bc) are water-soluble and actively excreted into bile and eliminated via the kidneys. In pathological conditions such as cholestasis or hepatocellular dysfunction, conjugated bilirubin can reflux into the bloodstream. There, it may form stable covalent bonds with albumin, resulting in delta bilirubin, a long-circulating form with clinical significance [6,7]. The chemical structures of the bilirubin species relevant to this study are illustrated in Figure 1.

Fig. 1.

Chemical structures of key bilirubin species used in this study. (a) Unconjugated bilirubin: hydrophobic and albumin-binding; used to model both free and bound forms. (b) Bilirubin monoglucuronide: an intermediate hepatic conjugate not evaluated experimentally. (c) Bilirubin diglucuronide: fully conjugated and water-soluble; structurally similar to the model compound used for Bc.

In clinical applications, bilirubin is typically tested using the diazo method which resolves direct and indirect bilirubin. Direct bilirubin corresponds primarily to the conjugated fraction, while indirect bilirubin is calculated by subtracting direct from total and reflects the unconjugated pool. While this two-part classification supports routine clinical interpretation, it does not resolve the individual species, such as free vs. bound unconjugated bilirubin, or the intermediate monoglucuronide. More detailed profiling could improve diagnostic accuracy for disorders involving bilirubin metabolism, liver function, or red blood cell turnover.

In neonates, where bilirubin-induced neurologic dysfunction is a serious concern, total serum bilirubin (TSB) values above 15–20 mg/dL are typically used as intervention thresholds [8]. However, the concentration of free (unbound) bilirubin is more closely associated with neurotoxicity and can pose a risk even at lower TSB levels. Free bilirubin concentrations exceeding 10–20 nM have been linked to increased risk of kernicterus, particularly in preterm or low-birth-weight infants [4,9,10]. Since free bilirubin is not routinely measured in clinical practice, improved methods to resolve this fraction could enhance neonatal risk assessment.

The current reference method for total serum bilirubin quantification relies on diazo-coupling chemistry [11]. While reliable, it is limited to centralized laboratory settings and requires venous blood draws. Non-invasive transcutaneous bilirubinometer, such as the Dräger JM series (Drägerwerk AG & Co. KGaA, Lübeck, Germany) and BiliChek (Philips Children’s Medical Ventures, Monroeville, PA, USA), are widely used for neonatal screening [12], but they are also limited to reporting total bilirubin and are susceptible to interference from factors such as skin pigmentation, hematocrit, ambient light, and gestational age [13].

The limitations of current clinical methods have prompted research into optical sensing technologies aimed at improving bilirubin measurement [14]. Most emerging systems still focus on total bilirubin, seeking greater portability or reduced invasiveness. For example, a wearable photometric sensor developed by Inamori et al. [15,16] enables continuous, real-time monitoring of total bilirubin in neonates and demonstrated strong correlation with serum levels (r≃0.91). Similarly, a handheld diffuse reflectance spectroscopy (DRS) device by Cheng et al. [17] achieved even higher correlation (r≃0.95) and showed improved robustness to confounding factors such as skin pigmentation and hematocrit. Recent advances in mobile multispectral imaging, such as the SpeCamX system by He et al. [18], demonstrate that even unmodified smartphones can be adapted for bilirubin estimation via reflectance analysis in phantom models. While these approaches support rapid total bilirubin assessment, their inability to distinguish among bilirubin subtypes significantly constrains their clinical significance.

However, few optical systems currently aim to resolve bilirubin subtypes. Examples include microfluidic capillary electrophoresis systems developed by Nie & Fung, which separates free and albumin-bound bilirubin based on their electrophoretic mobility, with detection performed via UV absorbance at 450 nm [19]. In contrast, electrochemical sensors have also shown promise for selectively detecting free bilirubin without prior separation. For instance, Kamel et al. developed a potentiometric sensor based on a screen-printed electrode modified with a nickel–bilirubin ion-association complex and ordered mesoporous carbon, achieving nanomolar sensitivity in serum [20]. Similarly, Thangamuthu et al. reported a graphene- and carbon nanotube–based electrochemical sensor with detection limits around 0.1 nM and successful testing in human serum samples [21].

To date, label-free optical approaches for selectively quantifying different bilirubin types in vitro remain at an early exploratory stage. While photobleaching-based methods have been investigated for transcutaneous bilirubin assessment [22], these efforts are still research-oriented and have not yet resulted in clinically established assays. In this work, a prof of concept benchtop system was build that integrates absorbance, fluorescence, fluorescence anisotropy, and photobleaching modalities to characterize and differentiate bilirubin species.

The measurement modalities used in this work were selected because each has previously been applied to bilirubin detection, although they have not been combined to discriminate between bilirubin subtypes. Absorbance spectroscopy is the most widely used method in clinical practice, as it enables fast and reliable quantification of total serum bilirubin [19]. Fluorescence measurements have been shown to depend strongly on bilirubin–albumin interactions [23], providing a modality sensitive to binding dynamics, together with anisotropy measurements. Photobleaching behaviour has further been explored as a potential means of distinguishing bilirubin species by exploiting differences in photostability, as demonstrated in prior work [22]. Together, these established optical techniques motivate the multimodal approach implemented here to enhance species-level discrimination.

The system uses discrete-wavelength laser excitation at 450 nm and 635 nm with photodiode-based detection. We evaluate its performance using synthetic samples representing free unconjugated bilirubin (Bf), albumin-bound unconjugated bilirubin (Bb), and conjugated bilirubin (Bc, modelled using bilirubin ditaurate). Our results demonstrate that each optical modality captures distinct physicochemical characteristics, and that their combination could enable differentiation of bilirubin species. These findings support future development of subtype-resolving bilirubin sensing platforms.

2. Materials and methods

2.1. System design and optical setup

The custom-built optical system employs an L-shaped geometry optimized for multimodal bilirubin sensing (Fig. 2). The system combines dual-wavelength excitation acquisition of absorbance and fluorescence, sequentially followed by fluorescence anisotropy and photobleaching data. This modular architecture minimizes crosstalk between modalities and enables a compact benchtop footprint.

Fig. 2.

Sectional view of the optical measurement system showing the internal layout and light path (outer dimensions: 145 mm × 95 mm). Red indicates the matrix-correction light, blue the excitation light, purple their combination, and green the emission light. The excitation power can be digitally adjusted from $0.5\mathrm{mW}$ to $10\mathrm{mW}$ . The reference and absorbance sensors have a transimpedance of $10\mathrm{k}\mathrm{\Omega }$ , while the integrating transimpedance has a feedback capacitance of $10\mathrm{pF}$ .

Excitation is provided by two laser diodes: a 450 nm laser (4) for bilirubin excitation and bleaching (SHD4580MG, Roithner Laser), and a 635 nm reference laser (1) to correct for matrix effects (QL63D5SA, Roithner Laser). The light from both lasers is collimated using aspherical lenses (21-211, Edmund Optics) (2,5). A mirror (32-945, Edmund Optics) (3) and a dichroic mirror (66-245, Edmund Optics) (6) combine the beams, which then pass through a linear polarizer (71-035, Edmund Optics) (7) before entering the sample module.

The beam is focused using a small lens (45-236, Edmund Optics) (8) and then split using a 25/75 beam splitter (46-664, Edmund Optics) (9). Two mirrors (45-610, Edmund Optics) (10,11) direct a portion of the excitation light to a reference photodiode (BPW-34, Vishay) (12) before the sample (13), while the transmitted light is collected by a second photodiode of the same type (14) behind the cuvette. Absorbance is calculated via ratiometric comparison of the two signals. For calibration, neutral density filters of different optical densities were temporarily placed in the sample path to perform a three-point calibration.

Fluorescence is detected orthogonally to the excitation path, minimizing stray light and back-scattering artifacts. The emission is collimated with a high NA lens (63-484, Edmund Optics) (15), passes through a long-pass filter (3 mm SCHOTT glass) (16) to increase SNR by suppressing non-collimated and scattered excitation light, and then through a 535 nm band-pass filter (65-095, Edmund Optics) (17). The light is focused using a lens (32-853, Edmund Optics) to maximize sensitivity. A motorized linear polarizer (71-035, Edmund Optics) (19) is placed after the focusing lens to enable polarization-resolved fluorescence detection. The photodiode (BPW-34, Vishay) (20) collects the filtered emission while the rotating polarizer alternates between parallel and perpendicular orientations, allowing computation of fluorescence anisotropy in a time-resolved manner.

All optical elements are mounted in a modular, 3D-printed enclosure (FDM, PLA, Stealth Black, Formfutura) with integrated alignment features for reproducible assembly. Light-tight compartments and adjustable lens mounts ensure consistent measurement geometry.

The system is controlled by a central mainboard holding power supply and a microcontroller (RP2040, Raspberry Pi), which synchronizes excitation and acquisition across three dedicated electronic modules (Fig. 3): (1) the Laser Diode Module, regulating excitation power via a digitally adjustable potentiometer (AD5270, Analog Devices) and a laser driver (iC-NZN, iC-Haus), with onboard temperature monitoring (ADT77301, Analog Devices); (2) the TIA Module, which digitizes signals from the reference and transmission photodiodes using a two-stage transimpedance amplifier (LTC6268 and ADA4807, Analog Devices) for high linearity and low noise; and (3) the Integrating TIA Module, based on an IVC102 integrator (Texas Instruments), which accumulates low-level fluorescence currents over programmable integration times. All signals are digitized via 16-bit ADCs (AD4008, Analog Devices) and communicated to the central microcontroller via SPI. For time consistent measurements, a state-machine was created using the programmable input output module of the RP2040. This module works independent of the CPU allow for precise timing. In combination with DMA buffering, this enables a time accuracy in the nanosecond range, enabling great repeatability over the measurements.

Fig. 3.

Schematic overview of the system’s electronic modules. In reading order: laser diode module, integrating transimpedance (IVC) module, transimpedance (TIA) module and main board. All modules interface with the central main board via SPI.

The entire system is controlled via custom Python software, enabling real-time parameter tuning, automated acquisition, and post-processing routines including baseline subtraction, calibration, unit conversion, and anisotropy calculation.

2.2. Sample preparation

Bilirubin samples were prepared to represent the three clinically relevant forms: free unconjugated bilirubin, bound unconjugated bilirubin, and conjugated bilirubin. All stock solutions were freshly prepared under low-light conditions to prevent photodegradation and were stored in opaque containers covered with aluminium foil.

To prepare the Bf and Bc stock solutions, 2 mg of bilirubin (NIST916B, Sigma-Aldrich) or bilirubin ditaurate disodium salt (201102, Sigma-Aldrich) were transferred into 100 mL volumetric flasks. Dimethyl sulfoxide (DMSO) (23500.297, VWR) was added dropwise to dissolve the powder, and the flasks were subsequently filled to volume with DMSO, yielding a final concentration of 20 mg/L. Complete dissolution was typically achieved before reaching the final volume, indicating that higher concentrations would also be feasible. Attempts to dissolve Bc directly in aqueous buffer resulted in aggregation, confirming the need for initial dissolution in DMSO.

For Bb, an albumin–PBS stock solution was prepared by dissolving 20 mg of recombinant human serum albumin (A9731, Sigma-Aldrich) in 10 mL of phosphate-buffered saline (PBS) (10010-015, Gibco), resulting in a final concentration of 2 g/L. The solution was ultrasonicated for 15 minutes to ensure complete dissolution.

Dilution series were prepared by pipetting the appropriate diluent (PBS for Bf and Bc; albumin–PBS for Bb) into 1 cm pathlength cuvettes, followed by the addition of bilirubin stock solution. The final sample volume was 400 μL per cuvette, which corresponds to the minimum required for measurement in the custom optical system. For Bb, the bilirubin-to-albumin ratio was kept well below the known binding capacity of four bilirubin molecules per albumin [23] to avoid non-linear fluorescence effects.

All samples were prepared in a dark environment and measured immediately (within 2 h) after preparation to minimize artifacts from photodegradation and pigment aggregation. As previously observed, Bf exhibits significant aggregation within 24 hours in aqueous media, underlining the importance of minimizing sample ageing prior to measurement.

2.3. Measurement protocol

Each sample was measured for absorbance and fluorescence intensity followed by fluorescence anisotropy and photobleaching. All measurements were conducted in standard 1 cm pathlength cuvettes with a sample volume of 400 μL. Each concentration was measured three times to ensure reproducibility.

Absorbance was measured using a ratiometric approach, comparing the transmitted light intensity through the sample to a reference beam intensity recorded upstream of the cuvette:

$$A=-{\log }_{10}\left(\frac{I_{\text{trans}}}{I_{\text{ref}}}\right)$$ (1)

Excitation was provided by a 450 nm laser diode. An additional measurement at 635 nm, where bilirubin has minimal absorbance, was used to account for background absorption from the sample matrix and correct for optical artifacts.

Fluorescence emission was excited at 450 nm and detected at 535 nm using an integrating transimpedance amplifier circuit (IVC102). The integration time was set to 100 ms unless otherwise specified. The fluorescence signal was normalized to the reference detector signal to correct for fluctuations in excitation power. All measurements were background-subtracted using dark and blank controls. Signal values were internally converted to optical power units using calibrated circuit parameters and prior calibration. Enabling better understanding of the bleaching effects knowing the excitation power.

For anisotropy analysis, fluorescence emission was sequentially recorded through parallel and perpendicular polarization filters relative to the excitation polarization. A motorized filter wheel enabled automated switching between polarization states. Anisotropy was calculated as:

$$r=\frac{F_{\parallel }-F_{\perp }}{F_{\parallel }+2F_{\perp }}$$ (2)

where $F_{\parallel }$ and $F_{\perp }$ denote the fluorescence intensities measured in the parallel and perpendicular channels, respectively.

A laser power of 7 mW was selected to ensure a sufficient SNR for low-concentration bilirubin samples, which exhibit inherently weak fluorescence. Although this excitation level is not optimal for photobleaching studies, it represents the lowest power at which reliable fluorescence signals could be obtained across all concentrations with the current optical configuration. The relatively high irradiance contributed to the rapid bleaching observed for some species, which limited the extraction of stable bleaching kinetics. Photobleaching dynamics were recorded by repeatedly exciting the sample over a 9 s period, with each excitation cycle lasting 140 ms at 7 mW, corresponding to a peak intensity of approximately 1.1 MW/m² at the focal point. Fluorescence traces $F(t_n)$ were normalized to the initial signal $F(t_0)$ to compute a bleaching index.

$$\text{BL}=\frac{F(t_n)}{F(t_0)}$$ (3)

2.4. Data evaluation

The data from the device was stored in custom text files. A Python file was then used to interpret the text files and stored them in numpy arrays. Since the variations between samples was expected to be normal distributed, the measurements of the same type at the same concentrations were averaged over three samples, resulting in one point per concentration and type. For showcasing and interpreting the measurements, a linear fit was applied to the points using scipy’s linregress function. This function applies a linear least-squares regression using the normal equation. The function returns a slope and an intercept together with the R² and the p-value. The p-value is calculated using the null hypothesis that the slope is zero using the Wald test with t-distribution for the test statistics.

3. Results

3.1. Absorbance measurements

Absorbance measurements (Fig. 4) were performed across a concentration range of 0–10 mg/L for Bf, Bb, and Bc. The absorbance increased linearly with concentration at 450 nm, consistent with Beer–Lambert behaviour for all species. Among the three species, Bf exhibited the steepest slope, followed by Bb, while Bc showed the lowest absorbance response at equivalent concentrations, aligning with Babin et al. [24], which shows different absorbance peaks for free and conjugated bilirubin.

Fig. 4.

Absorbance of free (Bf), bound (Bb), and conjugated (Bc) bilirubin across increasing concentrations at 450 nm. Absorbance increased linearly with concentration for Bf and Bc, while Bb showed significantly lower absorbance values at the same nominal concentrations. Linear regression fits are shown as dashed lines. Error bars represent the standard deviation of triplicate measurements. The matrix effects where removed using the 630 nm laser measurement.

The fitted calibration curves yielded slope values of 0.05121 1/(mg/L) for Bf, 0.04981 1/(mg/L) for Bb, and 0.04173 1/(mg/L) for Bc, with corresponding $R^2$ values above 0.98 for all three fits. The coefficient of variation (CV) across triplicate measurements remained below 5% for all concentrations and bilirubin types.

3.2. Fluorescence measurements

Fluorescence emission was recorded at 535 nm upon 450 nm excitation following the multispectral evaluation of Croce et al. [25]. As presented in Fig. 5 Bf exhibited an approximately linear increase in fluorescence intensity with concentration across the tested range (0–10 mg/L), consistent with expectations for an unbound, monomeric fluorophore in dilute solution. The sensitivity of Bf and Bc was almost the same with 2.79 and 2.61 1/(mg/L). While Bb showed a higher sensitivity at 27.9 1/(mg/L).

Fig. 5.

Fluorescence intensity of free (Bf), bound (Bb), and conjugated (Bc) bilirubin as a function of concentration. Bf and Bc showed a linear increase in fluorescence intensity with concentration across the measured range. Bb exhibited a non-linear response consistent with albumin binding site saturation, plateauing at 5 mg/L. The linear fit for Bb was done using the measurements until 3 mg/L to show the expected theoretical trend without saturation.

Bb showed a non-linear fluorescence trend, with a decrease in intensity at concentrations above 6 mg/L (Fig. 5). This behaviour can be explained by albumin binding dynamics: at low to moderate concentrations, bilirubin binds to available albumin sites, resulting in increased fluorescence. As the bilirubin concentration increases, the available albumin binding sites become saturated, leaving excess unbound bilirubin that tends to aggregate in aqueous solution [23]. In addition, the non-negligible DMSO content in the samples may further alter albumin’s binding capacity, as preferential solvation effects have been reported by Grigoryan [26]. It was tested to dissolve Bc in aqueous solvent but, in contrast to the data sheet this was not possible without aggregation.

Aggregation can lead to self-quenching and a reduction in the observed fluorescence signal, consistent with the saturation effects described by Thaer et al. [23]. For this reason, only measurements up to 3 mg/L were used for fitting the concentration-dependent trend. Measurement variability for Bb was higher than for Bf and Bc, likely reflecting both sample preparation challenges and the dynamic nature of albumin–bilirubin binding at higher concentrations.

3.3. Fluorescence anisotropy

Device-based anisotropy measurements were omitted in this work because they exhibited large deviations across the entire concentration range and therefore did not yield reliable information. To describe the underlying theoretical approach, anisotropy was additionally measured using a plate reader (Fig. 6).

Fig. 6.

Fluorescence anisotropy of free (Bf), bound (Bb), and conjugated (Bc) bilirubin across increasing concentrations. Bb exhibited consistently higher anisotropy values than Bf, reflecting reduced rotational mobility due to albumin binding. Bf showed low and concentration-dependent anisotropy, while Bc displayed concentration-dependent intermediate values. For this measurements no replicates were done, since they were initially planed to only be used as quality control. Due to the bad measurement results of the device this measurements had to be used.

As expected from steady-state anisotropy theory [27], Bb shows a concentration-independent anisotropy of approximately 0.26 with only a slight increase at higher concentrations, consistent with normalization. In contrast, Bf and Bc display an increasing anisotropy with increasing concentration, which likely reflects aggregation effects caused by their limited solubility in aqueous solutions.

This interpretation cannot be fully verified using fluorescence intensity alone, as the fluorescence signals show a strong linear correlation with concentration and therefore do not reveal aggregation-related deviations.

3.4. Bleaching dynamics

Photobleaching measurements were conducted over a 9-second period, using repeated 140 ms excitation pulses at 7 mW.

As can be seen in the Fig. 7, 8, Bf and Bc showed low to no bleaching. Since one measurement induces 15 J/m² it is assumed, that the majority of the Bilirubin is already bleached after the first measurement. As Ndabakuranye et al. [28] show the major bleaching happens until 3 J/m². Due to the robustness of Bb to bleaching, the measurements of Bb (Fig. 9) showed more bleaching dynamics for low concentrations, compared to Bf and Bc.

Fig. 7.

Photobleaching behaviour of free unconjugated bilirubin (Bf) over a measurement time of 9 s with repeated 140 ms excitation pulses at 7 mW. It can be seen, that low concentrated Bf does not show a change in fluorescence over time, while higher concentrated samples show an increase, followed by a recovery to the initial value. The samples were measured once due to photodegradation during the measurements.

Fig. 8.

Photobleaching behaviour of conjugated bilirubin (Bc) over a measurement time of 9 s with repeated 140 ms excitation pulses at 7 mW. It can be seen, that low concentrated Bc does not show a change in fluorescence over time, while higher concentrated samples show an increase, followed by a recovery to the initial value. The samples were measured once due to photodegradation during the measurements.

Fig. 9.

Photobleaching behaviour of bound bilirubin (Bb) over a measurement time of 9 s with repeated 140 ms excitation pulses at 7 mW. It can be seen, that low concentrated Bb shows a bleaching behaviour, while higher concentrated samples show an increase, followed by a recovery to the initial value. The samples were measured once due to photodegradation during the measurements.

It was observed, that for higher concentrations all forms show an increase in fluorescence. This dynamics could show the effect of diffusion in the samples over time or could be another impact of the aggregation of the sample, due to bad solubility.

3.5. Modality performance

A summary of sensitivity, accuracy, and resolution across absorbance and fluorescence is presented in Table 1. Absorbance can be used as an absolute measurement method, while fluorescence can be used for differentiate between bound and unbound bilirubin, based on the signal strength.

Table 1. Analytical performance of absorbance and fluorescence for detecting different bilirubin species. Sensitivity is the slope of the calibration curve. Accuracy is the maximum deviation from the fit within the tested range (0–10 mg/L). Resolution was calculated from signal noise and slope.

Measurement	Bilirubin Form	Sensitivity $(1/(\mathrm{mg}/\mathrm{L}))$	Accuracy (mg/L)	Resolution (mg/L)
Absorbance	Free (Bf)	0.051 a	0.588	0.167
	Bound (Bb)	0.049 a	0.963	0.163
	Conjugated (Bc)	0.041 a	0.362	0.115
Fluorescence	Free (Bf)	2.791 a	3.465	1.538
	Bound (Bb) b	27.917 a	0.843	0.565
	Conjugated (Bc)	2.614 a	2.741	1.453

^a High statistical significance p-value $(<0.001)$ .

^b Only measurements up to 3 mg/L were included in the linear fit for bound bilirubin due to saturation effects.

4. Discussion

This study evaluated a compact, multimodal optical system for differentiating artificial bilirubin species in solution. Each optical modality provided distinct and complementary strengths for quantification and characterization. Absorbance measurements were robust and reproducible for all species, with linear concentration–response behaviour and low variability. This makes absorbance particularly suited for total bilirubin determination.

Fluorescence measurements showed highest sensitivity for Bb, with a tenfold higher slope than Bf. However, this advantage was limited by binding site saturation beyond 3 mg/L, which caused the signal to plateau. Bf and Bc, in contrast, showed a lower absolute sensitivity but maintained a linear response across the full tested range.

Fluorescence anisotropy, while not reliable with the build device, enabled differentiation between both free bilirubin forms and Bb based on molecular mobility. Bb consistently exhibited higher anisotropy values across all concentrations. Bf showed low and concentration-dependent values, consistent with its free rotational diffusion. Bc displayed intermediate and also concentration-dependent values. This variability is likely due to aggregation or photo-instability under prolonged illumination and highlights the sensitivity of anisotropy to sample preparation quality. To further investigate these effects, complementary techniques such as dynamic light scattering, size exclusion chromatography, or microscopy should be employed to quantify the extent of aggregation.

Photobleaching measurements showed a time-resolved decay only for Bb, whereas Bf and Bc appeared comparatively stable over the recording period. This likely indicates that the latter species bleach much faster and may already be significantly photodegraded before the first measurement point. Such rapid early bleaching may also contribute to the large variability observed in anisotropy, since the two polarization channels are acquired sequentially rather than simultaneously. To address this issue, the sensitivity of the device would need to be improved so that lower excitation power can be used.

While these data suggest species-dependent photostability, alternative explanations cannot be fully excluded. The relatively high excitation intensity used to obtain sufficient SNR at low concentrations may introduce localized heating, which could affect species with different solubility or aggregation tendencies differently. Moreover, photobleaching is sensitive to oxygen availability, and variations in oxygen diffusion or micro-aggregation between species could mimic true photostability differences. Micro-environmental factors such as solvation effects or partial precipitation may also play a role.

Therefore, although the bleaching curves provide qualitative insight, their current utility for species discrimination is limited. Improved control of excitation power, sample temperature, oxygenation, and measurement timing may help clarify these effects in future work.

The combination of optical modalities can be exploited in two complementary ways. The first approach aims to understand and interpret the underlying photophysical mechanisms of bilirubin species. In this scheme, the absorbance measurement provides an estimate of the total bilirubin concentration. The fluorescence signal then reports on the fraction of albumin-bound bilirubin relative to free and conjugated forms, since the emission intensity depends strongly on binding state. Anisotropy measurements add an additional discriminative dimension, as the rotational mobility differs between free and conjugated bilirubin. Photobleaching behaviour can serve as another independent variable, further increasing robustness, because the different species show distinct photostability and aggregation tendencies.

Although the individual modalities can be interpreted from a physical perspective, the combined dataset might also benefit from multivariate analysis. Approaches such as PCA followed by LDA, logistic regression, or related classifiers may offer additional discriminatory power by integrating absorbance, fluorescence intensity, anisotropy, and photobleaching–derived parameters into a shared feature space.

In this work, such multivariate modelling was not carried out, mainly because the available dataset was limited and not consistently acquired across all modalities. It is therefore difficult to draw strong conclusions about the potential performance of these classifiers at this stage. Nevertheless, the presented platform was designed to generate multiple complementary observables, and a more systematic multivariate evaluation could be explored in future studies.

Taken together, these findings demonstrate that combining absorbance, fluorescence, and anisotropy could enable quantification and discrimination of bilirubin species. While each modality has its limitations, their complementary characteristics support the development of integrated, label-free sensing strategies for subtype-specific bilirubin analysis.

5. Conclusion

We developed and evaluated a compact, multimodal optical system capable of resolving synthetic free/conjugated and bound bilirubin species in solution. By integrating absorbance, fluorescence, fluorescence anisotropy, and photobleaching measurements, the platform captures complementary physicochemical features including concentration, binding state, molecular mobility, and photostability.

Absorbance and fluorescence provided robust quantification performance, with fluorescence achieving the highest sensitivity for bound bilirubin at low concentrations and absorbance offering consistent linearity across species. Anisotropy, in theory, enables discrimination between free and albumin-bound species, while bleaching measurements were reproducible only for the bound form. Showing the need for improvement of the device.

These results highlight the value of combining multiple label-free optical readouts to enhance specificity and performance in bilirubin detection. The system provides a foundation for future development of subtype-resolving bilirubin sensors and supports broader efforts toward diagnostics in liver function assessment and neonatal care.

Disclosures

The authors declare no conflicts of interest.

Data availability

All data supporting the findings of this study are available within the article in graphical form. Underlying numerical data used to generate the plots can be provided by the corresponding author upon request.