1. Introduction

When light is emitted from a photo-excited sample, it may be phase-correlated to the incident light, such as in Rayleigh or stimulated Raman scattering, or fully incoherent, such as in photoluminescence (PL) or spontaneous Raman scattering [1]. A measurement of the degree of coherence, the first-order correlation [1,2] of incident and emitted light, distinguishes between these processes. It proved useful for understanding the optical properties of various systems, such as single ions [3], carbon nanotubes [4], gold nanostructures [5] or semiconductor quantum wells [6–9]. This distinction can be used to suppress incoherent background in, e.g., coherent anti-Stokes Raman scattering [10], stimulated Raman scattering [11,12] or second harmonic microscopy [13]. Also, in excitation-emission (EE) matrix spectroscopy, it is often necessary to remove scattering contributions, commonly done by a posteriori data interpolation or exclusion [14–18].

In experiments with narrowband excitation, spectral filtering using bandpasses [19,20], monochromators [21], or acousto-optical modulators [22] has been successfully employed to separate coherent scattering from incoherent emission. This, however, masks information about emission processes in the spectral vicinity of the excitation. In broadband spectroscopy, distinguishing between coherent scattering and incoherent emission is more challenging, requiring to record a phase-stable interferogram between the emission and a replica of the excitation light, acting as a local oscillator [23,24]. Quite often, sophisticated nonlinear optical spectroscopies, such as fluorescence up-conversion, are used to separate these emissions in the time domain, as demonstrated, e.g., by separating resonant Rayleigh scattering and incoherent luminescence from semiconductor quantum wells [8,9,24]. Also in nonlinear optics, coherent and incoherent emission channels are readily distinguished by using similar interferometry techniques [25].

Recently, such sensitive broadband interferometric experiments in the ultraviolet to near-infrared spectral range have been simplified by the introduction of a new intrinsically stable common-path interferometer, called TWINS [26,27]. In TWINS, birefringent wedges are used to introduce variable time delays between orthogonally polarized replicas of the incident light. The TWINS interferometer has recently successfully been used in a variety of linear and nonlinear experiments, e.g., in Fourier transform (FT)-based absorbance [28], EE [20,29] or time- and frequency-resolved fluorescence spectroscopy [30], hyper-spectral imaging [31], pump-probe [32] and two-dimensional electronic spectroscopy [33–35], and in stimulated Raman scattering microscopy [36].

Here we show that by using TWINS interferometers in both the excitation and detection paths of a broadband, white light EE spectrometer, we can distinguish and fully separate coherent and incoherent emission in linear light scattering spectroscopy. We demonstrate our approach for the suppression of the Rayleigh scattering signal from the weak fluorescence emission of the flavin adenine dinucleotide (FAD) chromophore in isolation and when it is bound inside the European Robin (Erithacus rubecula) Cryptochrome 4 protein (ErCry4) [37,38].

2. Experimental section

2.1. Main concept for the separation of coherent and incoherent emission signals

We use an interferometer to separate a spectrally broadband whitelight pulse into a sequence of two phase-locked pulses separated by a variable time delay ${\tau _1}$. This pulse pair optically excites the sample under investigation, as indicated in Fig. 1(a). We assume that this sample can either coherently scatter the incident light or emit incoherent PL. Coherent light scattering will lead to the emission of a secondary phase-locked pulse pair that is separated in time by ${\tau _1}$ (Fig. 1(a), blue pulses after the sample). In contrast, incoherent PL will result in the emission of intensity bursts of light (Fig. 1 (a), red) which are also time delayed by ${\tau _1}$. Both emission contributions are sent through a second interferometer, again creating time delayed replicas of the incident pulses that are now separated in time by a second variable time delay ${\tau _2}$. The total output signal from the second interferometer is sent onto a time-integrating photodiode. In the experiment, we record the photodiode intensity as a function of both time delays ${\tau _1}$ and ${\tau _2}$. The resulting correlation traces are fundamentally different for coherent and incoherent emission, as can be seen in the simulations in Figs. 1(b) and (d), respectively. While incoherent PL signals (Fig. 1(d)) are characterized by vertical and horizontal stripes only, coherent light scattering (Fig. 1(b)) shows additional correlation traces along the diagonals. Since these diagonal traces are absent in the incoherent signal map, they can be used to disentangle the two emission contributions. Figures 1(c) and (e) show the evaluated EE spectra corresponding to the time-domain interferograms in Figs. 1(b) and (d). Here the different structures of coherent light scattering and incoherent emission signals are clearly visible as well.

 

Fig. 1. Distinction between coherent scattering and incoherent emission in Fourier transform excitation-emission spectroscopy using two interferometers. (a) The first interferometer generates a phase-locked pulse pair with relative time delay ${\tau _1}$. This pulse pair excites the sample under investigation. The sample then coherently scatters the incident light (blue) and/or emits incoherent photoluminescence (red). Both emission signals are passed through a second interferometer after the sample that generates again phase-locked replicas of both components with a time delay ${\tau _2}$. Finally, the time-delayed replicas interfere at the photodiode, which records the total light intensity as a function of ${\tau _1}$ and ${\tau _2}$. (b), (d) Simulated excitation-emission interferograms of a sample that scatters coherently (b) and emits incoherent photoluminescence (d). The diagonal patterns in (b) are a distinct signature of coherent light scattering in these two-dimensional maps which allow for their isolation. (c), (e) Excitation-emission spectra of coherent scattering (c) and incoherent emission (e) as a function of excitation ${\lambda _{ex}}$ and detection ${\lambda _d}$ wavelength.

Download Full Size | PDF

These structural differences of the time-domain pattern can be understood as follows. For coherent light scattering processes, light emission occurs at the frequency of the excitation light and with a specific phase relation between excitation and emission. Thus, a pair of phase-locked pulses (P₁ and P₂ with relative delay ${\tau _1}$) from the excitation TWINS that interacts with the sample results in the emission of a secondary pulse pair with the same relative phase relation and time delay. This pair is then split into two pairs (P₁₁/P₁₂, P₂₁/P₂₂, each with relative delay ${\tau _2}$) in the detection interferometer. All features in the map in Fig. 1(b) can be understood based on the interference between these pulse pairs at the detector as a function of the relative time delays ${\tau _1}$ and ${\tau _2}$. These four pulses give rise to a total of six distinct interference pathways at the detector. In the following, we qualitatively discuss the effect of each of these pathways on the time-domain map. It is important that in these maps, only those pulses that are not separated by much more than the coherence time of each pulse can contribute to the interferogram. Therefore, around ${\tau _2} \approx 0$ mostly the interference of P₁₁ with P₁₂ and P₂₁ with P₂₂ contributes. Since it does not depend on ${\tau _1}$, this results in vertical stripes in the map (Fig. 1(b)). Along the diagonals, the interference of P₁₂ with P₂₁ dominates around ${\tau _2} \approx {\tau _1}$, whereas that of P₁₁ with P₂₂ is seen around ${\tau _2} \approx{-} {\tau _1}$. This forms the two diagonal stripe patterns. Finally, around ${\tau _1} \approx 0$, the interference of P₁₁ with P₂₁ and P₁₂ with P₂₂, which is independent of ${\tau _2}$, dominates. This contribution is visible as horizontal stripes in the map. Non-interfering pulses yield a constant background. For ${\tau _1} \approx {\tau _2} \approx 0$ all the interference patterns described above overlap. A detailed analysis of these four components is reported in Supplement 1, Sections 4 and 5.

In contrast to coherent light scattering, fluorescence emission is induced by vacuum field fluctuations. Thus, the phase coherence of the incident light that is absorbed in the excitation process of the sample is lost in the emitted light. A variation of ${\tau _1}$ changes the number of photons absorbed in the sample and thus also the number of re-emitted photons. This leads to a modulation along the excitation axis, but it does not affect the emitted spectrum. Therefore, the modulation along ${\tau _2}$ is independent of ${\tau _1}$ and only shows interferences between the two replicas of the fluorescence emission that are created in the detection interferometer. The product of the resulting interference patterns along the excitation and detection axes gives the incoherent map displayed in Fig. 1 (d).

Consequently, the incoherently emitted fluorescence does not interfere with the coherently scattered light. As such, the total map in the time domain arises from the superposition of the coherent and incoherent contributions.

2.2. Experimental implementation

The experimental setup of our Fourier-transform EE spectrometer, comprising two passively stabilized common-path interferometers (BI-TWINS), is schematically shown in Fig. 2(a). A supercontinuum whitelight source (Fianium SC400) delivers picosecond pulses with a spectral width from 400 nm to 2400 nm at a repetition rate of 40 MHz. Two shortpass filters (DM, Thorlabs DMSP805R and DMSP650) reduce their spectrum to the range from 400 nm to 680 nm to cover the excitation and emission spectrum of the sample. The power of the whitelight beam is set to 25-50 µW using a neutral density filter (NDF). The pulses first pass through an excitation TWINS interferometer, a passively stabilized common-path interferometer based on translating birefringent wedges made of α-barium borate (α-BBO) [26,27]. Its scheme is shown in the inset of Fig. 2(a). A first polarizer (P₁, set at 45°) creates a superposition of two pulses that are linearly polarized along the x and y directions, respectively. The birefringent block (B, vertical optical axis) and the wedges (W, horizontal optical axis) introduce a relative time delay τ between these two orthogonal components that is controlled by translating the position ${x_{ex}}$. The second polarizer (P₂, 45°) projects the two time-delayed, phase-locked pulses onto the same polarization state [26,27]. The polarization of this pulse pair is controlled using another polarizer (P₃). After the excitation TWINS, a small fraction of the light is split off by a beam splitter (BS) onto a photodiode (PD₁). This diode records the field autocorrelation [39] of the excitation pulses as a function of the excitation TWINS motor position ${x_{ex}}$. Fourier transform of the autocorrelation function provides the spectrum of the exciting light [28]. The pulse pairs are focused into the sample (S) using a microscope objective (O₁, 0.35 NA). A second photodiode (PD₂) records the autocorrelation of the transmitted light after the sample as function of ${x_{ex}}$, simultaneous to PD₁. It is used to obtain the absorption spectrum of the sample [28]. PL and scattered light from the sample are collected at a 90° angle with respect to the excitation path and are collimated using a microscope objective (O₂, 0.3 NA). The polarization of the collected light is selected using a polarizer (P₄). The emitted light then passes through a second, detection TWINS and is focused onto a single photon avalanche photodiode (SPAD, Micro Photon Devices PDM, 100 µm active sensing area diameter). Two irises (I₁ and I₂) are used to increase the spatial coherence of the light that passes through the detection TWINS. The latter creates two phase-locked replicas of the emitted light that interfere at the SPAD. The arrival time of the photons at the SPAD with respect to the excitation is stamped using a time correlated single photon counting unit (TCSPC, PicoQuant PicoHarp 300). A light intensity map is recorded as a function of arrival time and TWINS wedge positions ${x_{ex}}$ and ${x_d}$. Here, we do not make use of the time-resolved detection and focus on time-integrated EE spectra. For a single map, each TWINS is scanned stepwise over a range of 3 mm with a step size of 10 µm. The scan range is chosen to reach an average spectral resolution of 6.5 nm while the step size is set to fulfill the Nyquist theorem by recording at least 2 data points per spatial oscillation period, resulting in a BI-TWINS map with 300 × 300 pixels. To get a reasonable signal-to-noise ratio, we accumulate an average of 10⁵ photons per excitation/detection position. We avoid pile-up effects by keeping the total count rate below 10⁶ photons per second. Therefore, we set the integration time to 200 ms per pixel. We note that exploiting the symmetry properties of the BI-TWINS maps, it is possible to achieve a reduction in the data size by almost a factor of four. Moreover, the number of pixels per axis can be reduced substantially without degradation in resolution and signal-to-noise ratio by using undersampling concepts [36]. The integration time per pixel is limited by the specific type of detector, chosen to enable time-resolved measurements in future work, and could easily be reduced, e.g., by using time-integrating photodiodes and lock-in detection.

 

Fig. 2. (a) Fourier-transform excitation-emission spectroscopy using two passively stabilized common-path interferometers (BI-TWINS). A phase-locked pair of picosecond pulses from a supercontinuum whitelight source is created in a first, excitation TWINS interferometer. The time-delay $\tau $ between the pulses is controlled by translating the position ${x_{ex}}$ of the birefringent wedge W (inset). The pulses are focused into the sample using objective O₁. Coherent scattering and incoherent emission from the sample is collected and collimated using objective O₂. The light is then transmitted through a second, detection TWINS and focused onto a photon-counting avalanche photodiode (SPAD). The two phase-locked pulse pairs created in the detection TWINS interfere at the SPAD and their arrival time with respect to the excitation is mapped using time-correlated single photon counting (TCSPC). A map of the light intensity is recorded as a function of arrival time and wedge positions ${x_{ex}}$ and ${x_d}$ in the excitation and detection TWINS. A scheme of the TWINS setup [26] is shown in the inset, depicting the optical axes of the birefringent crystals by blue dots and arrows. Polarizer P₁ and P₂ are at 45° degree with the optical axis. (b) Absorbance spectrum of the chromophore flavin adenine dinucleotide (FAD) in a buffer solution (pH 8.0). The chemical structure of FAD is shown in the inset.

Download Full Size | PDF

2.3. Data analysis

To analyze the experimental data, we first remove the horizontal and vertical patterns from the maps, since these are included in the map for both coherent and incoherent contributions (Fig. 1(b) and (d)). Second, we realize that the autocorrelation traces recorded with the TWINS interferometers do not perfectly match those of an ideal, dispersion-free interferometer and that the two TWINS interferometers are not identical. This mainly results from the wavelength-dependent time delay introduced when moving the α-BBO wedges, and from manufacturing tolerances, respectively. We therefore correct, in the second step of the data analysis, the interferograms to those that would be obtained with two identical interferometers. This correction is described in detail in Supplement 1, Section 5. In the next step, we make use of the diagonals in the correlation maps to isolate the coherent scattering from incoherent PL emission. For this, we extrapolate the correlation signals in the region that is not overlapping with the PL correlation to the entire map. For both diagonals, we determine the cross-sections by a weighted average along the diagonals. Here, the weights are chosen such that the center of the interferogram, which is influenced by the PL contribution, is suppressed. Based on these cross-sections the diagonal patterns are reconstructed. The precise extrapolation routine is described in Supplement 1, Section 5. This allows us to isolate the coherent contribution to the EE map and to subtract it, leaving only the PL contribution. In the final step we perform a two-dimensional Fourier transform of the separated interferograms and transform the resulting frequency axes to wavelengths. This results in EE spectra that plot the emitted intensity as a function of excitation and detection wavelength (Fig. 1(c) and (e)). A detailed description of this transformation is given in Supplement 1, Sections 2 and 3. Briefly, the axis transformation requires to measure a calibration curve for each TWINS. To this end, we pass the broadband whitelight spectrum through the respective TWINS and use a spectrometer to record an interferogram at each TWINS motor position ${x_{ex,d}}$. From the Fourier transform of this interferogram along ${x_{ex,d}}$, we can then relate every color ${\lambda _{ex,d}}$ of the incident spectrum to the corresponding modulation wave number ${\tilde{k}_{ex,d}}$ of the TWINS. This measured relation yields the calibration curve $k_{ex,d}^c(\lambda )$. The TWINS calibration method has been described in detail earlier [26,27,33,40] and is summarized in Supplement 1, Section 2.

3. Results and discussion

To demonstrate the efficient separation of coherent light scattering from PL emission achieved by the BI-TWINS spectrometer, we use the weakly emissive protein ErCry4 located in the retina of the night-migratory songbird European Robin [37,38,41,42]. Photoexcitation of the FAD chromophore inside ErCry4 starts a series of ultrafast electron transfer reactions that lead to the formation of magnetic sensitive radical pairs [38]. These photoinduced electron transfers and the radical pairs generated as a result are currently discussed as a key mechanism underlying the magnetic field compass in migratory birds [41,42]. From a photophysical point of view, this rapid electron transfer results in a rapid repopulation of the ground state after optical excitation [43–45] and, thus, in inherently weak PL. This weak emission may be superimposed by spurious light scattering from, e.g., aggregated proteins in the solution or the solvent itself, hence further masking the weak PL. Also, inside proteins, the FAD molecules may be present in three different redox states: oxidized, as a one-electron reduced semiquinone or as the fully reduced hydroquinone, with distinct absorption and emission spectra [38,46–48]. Hence it is important to distinguish the weak PL signal from the scattering background. Here we show that this can be achieved by the interferometric excitation and detection scheme in BI-TWINS.

We first investigate free FAD without the protein. For this, we use molecular FAD (Sigma-Aldrich F6625) dissolved in a buffer solution at a concentration of 80 µM. The buffer solution (pH 8.0) consists of 20 mM tris(hydroxymethyl)aminomethane and 250 mM NaCl in 80% water and 20% glycerol, as also used for the protein sample. The measurements are recorded at room temperature. Under these conditions, free FAD is only found in its oxidized state. The chemical structure and absorbance spectrum of oxidized FAD are shown in Fig. 2(b). All measurements are performed using a low volume cuvette with 0.15 cm optical path length (Hellma, 105-252-15-40).

Figure 3(a) plots a BI-TWINS map of the time-integrated intensity of the emission from molecular FAD as a function of the motor positions of both the excitation (${x_{ex}}$) and the detection (${x_d}$) TWINS. At first sight, it contains mainly vertical and horizontal interference patterns induced by varying either the excitation or the detection TWINS delays. The crosscuts of the intensity map along ${x_{ex}}$ and ${x_d}$, plotted in Fig. 3(b), feature these interference patterns more clearly. The upper crosscut, taken along the horizontal ${x_d}$ axis at the position marked by a dashed line in Fig. 3(a), represents a field autocorrelation of the light that is emitted from the sample after excitation with a pulse pair. The lower crosscut is taken along the vertical dashed line in Fig. 3(a) and shows a field autocorrelation of that fraction of the excitation pulses that interacts with the sample. The fringe spacing in the (upper) detection crosscut is larger than that in the cross-section along the detection axis. This reflects the Stokes shift between the absorbed light and the light that is emitted from the sample. The fringe contrast in (b) is significantly lower than the ideal 100% contrast. This is mainly due to the reduced spatial coherence of the light that is emitted from the focal volume of the excitation beam in the cuvette. As the spectrum of the whitelight source used to excite the sample is highly structured (see Supplement 1, Section 6, Fig. S2), the oscillations along the ${x_{ex}}$ axis decay much slower than along ${x_d}$, as can be seen in the bottom crosscut (cf. Figure 3(b) upper and lower panels). Both cross sections are highly asymmetric around ${x_{ex,d}} = 0$. This asymmetry arises from the pronounced color ($\lambda $) dependence of the phase shift $\mathrm{\Delta }\varphi ({x,\lambda } )= \mathrm{\Delta }\tilde{\varphi }(\lambda )+ \tilde{k}(\lambda )x$ that is introduced in each interferometer (see Supplement 1, Section 1). Here, $\tilde{k}(\lambda ) = {{2\pi \mathrm{\Delta }n(\lambda )\sin (\vartheta )} / \lambda }$ is the modulation wave number introduced by the TWINS; $\Delta n(\lambda ) = {n_e}(\lambda ) - {n_o}(\lambda )$, the difference between extraordinary (${n_e}$) and ordinary (${n_o}$) refractive indices of α-BBO, denotes the birefringence introduced by the wedge pair W; $\vartheta $ is the opening angle of each wedge and $\mathrm{\Delta }\tilde{\varphi }(\lambda )$ is a position-independent phase offset introduced by the TWINS. For a quantitative analysis of the interference fringes, the one-dimensional cross sections in (b) need to be Fourier transformed. Afterwards, both resulting wave number axes need to be transformed to the corresponding wavelength axes, according to the respective calibration curve $k_{ex,d}^c(\lambda )$ of each TWINS. The one-dimensional cross sections in (b) therefore provide the spectra of the light that excites the sample and of the light that is emitted by the sample. These signals have been analyzed before [20,29,30] and are briefly discussed in Supplement 1, Section 4. However, they cannot distinguish between coherent scattering and incoherent emission from the sample. They can be measured separately by putting a single TWINS in either the excitation or detection path, as known from earlier work [20,30].

 

Fig. 3. Distinguishing between incoherent photoluminescence and coherent light scattering from FAD in buffer solution (pH 8.0) by BI-TWINS excitation-emission spectroscopy. (a) Measured interferogram (center part) of the temporally integrated FAD emission as a function of excitation (${x_{ex}}$) and detection (${x_d}$) wedge positions. (b) Cross sections along the horizontal (top) and vertical (bottom) black dashed lines in (a), showing characteristic interference structures recorded by the excitation and detection TWINS. (c) Interferogram from (a) after subtraction of the background signals. (d) incoherent contribution of (c) after subtraction of the diagonal structures. (e) coherent contribution to (c), magnified by a factor of 3. Excitation-emission spectra of (f) incoherent photoluminescence and (g) coherent light scattering from the FAD sample, deduced from the interference map in (a). (f) Spectrally averaged excitation and emission spectra of FAD_ox are displayed at the left and bottom side of the diagram. (g) Bottom: Light scattering spectrum, normalized to the excitation light spectrum (black) and the excitation spectrum itself (blue).

Download Full Size | PDF

Since these signals do not contribute directly to the EE spectrum that is of interest here, we subtract them from the raw data in Fig. 3(a). As the signals are constant along one of the coordinate axes, they are readily obtained by averaging along the TWINS position axes. After subtraction, the background-free EE interferogram ${\bar{I}_d}({{x_{ex}},{x_d}} )$, that is shown in Fig. 3(c), results. It consists of two distinct structures: a checkerboard-like pattern in the center originating solely from incoherent emission (Fig. 3(d)) and two diagonals that contain the information on the coherent scattering contribution (Fig. 3(e)). By separating these two structures, it is possible to distinguish between coherent and incoherent emission components.

The difference between the two structures can be understood intuitively. For an incoherent EE spectrum, excitation and emission color are uncorrelated and the spectrum is the product of an absorption and an emission (PL) spectrum (also see Eq. S16 in Supplement 1, Section 4). Since both spectra have a finite bandwidth, the signal will inevitably vanish for sufficiently large wedge positions ${x_{ex,d}}$. In contrast, excitation and emission colors are strictly the same in the case of coherent light scattering. Hence, finite correlations exist along the diagonals of the EE interferograms even for arbitrarily large wedge displacements. A detailed discussion of these two signal components is given in Supplement 1, Section 4.

To isolate coherent and incoherent EE interferograms, a two-dimensional Fourier transform technique is applied (see Supplement 1, Sections 4 and 5 for details). This method is based on the calibration of the two TWINS. Briefly, the data are first transformed to a new excitation axis ${\hat{x}_{ex}}$ that has the same frequency-dependent modulation wave number (calibration curve) $k_d^c(\lambda )$ as the detection axis. Then, the contribution of the incoherent emission to the signal along the diagonals is suppressed by applying an appropriately chosen windowing function, providing the coherent scattering interferogram ${\hat{I}^{sc}}({{{\hat{x}}_{ex}},{x_d}} )$ (Fig. 3(e)). Subtraction from the full interferogram gives the incoherent EE interferogram ${\hat{I}^{fl}}({{{\hat{x}}_{ex}},{x_d}} )$ (Fig. 3(d)). These interferograms are then transformed to wavelength axes, making use of the calibration curve of the detection TWINS, $k_d^c(\lambda )$.

Figure 3(f) shows the incoherent EE spectrum $I_0^{fl}({\lambda _{ex}},{\lambda _d})$, normalized to the excitation spectrum, of the FAD chromophore in buffer solution at pH 8.0 that results from this analysis. The insets in Fig. 3(f) display the retrieved absorption (left) and emission spectra (bottom) that are in excellent agreement with previously reported spectra of free, oxidized FAD in solution [46]. Figure 3(g) displays the coherent scattering contribution to the measured signal with the Fourier transform of the cross section of the diagonal structures showing the scattered spectrum (bottom). This map is obtained from the interferogram in Fig. 3(e). The strong scattering component arises mainly from the glycerol fraction in the buffer solution. Comparison of the two maps in Fig. 3(f-g) confirms the successful separation of coherent and incoherent components to the EE spectra of free FAD.

We now apply this approach to a weakly-emissive sample containing the chromophore FAD embedded in the ErCry4 protein. The ErCry4 protein (GenBank: ATE87950.1) was expressed and purified with FAD bound as described previously [38]. Here, we use the same concentration (80 µM) and the same buffer solution as for molecular FAD. Because these measurements are performed at 0°C, we increase the pH to 8.0 to ensure a similar acid-base balance to that in the bird’s eye at 40 - 43°C and pH 7.3 [49]. We centrifuge the protein sample before the measurement to sort out any previously aggregated proteins and thus to minimize the scattering. Inside the protein, the fluorescence yield is intrinsically lower than in free FAD, due to an electron transfer from the protein [38,41]. In the investigated sample, the fluorescence yield was reduced by a factor of about 30 in comparison with that of free FAD in buffer solution. Despite careful sample preparation, the solution unavoidably contains a finite amount of aggregated protein, which results in a non-negligible coherent scattering background. In addition, also the buffer solution causes such light scattering, as discussed for free FAD.

Figure 4(a) shows the complete EE spectrum of the sample retrieved from the BI-TWINS map using the method described above. It contains coherent light scattering (diagonal line) and PL (bottom) contributions of similar magnitude. The isolated incoherent EE spectrum in Fig. 4(b) shows mainly a single emission component with an absorption spectrum (Fig. 4(d), black solid line) that sets in for wavelengths below 480 nm, whereas the scattering component (spectrum shown in Fig. 4(c)) is efficiently suppressed. The unstructured PL emission spectrum (Fig. 4(d), blue solid line) peaks at around 530 nm and extends up to about 650 nm.

 

Fig. 4. Suppression of light scattering from the excitation-emission spectra of the weakly emissive FAD chromophore inside European Robin Cryptochrome 4 protein. (a) Excitation-emission spectrum recorded using BI-TWINS showing both fluorescence emission and coherent scattering. (b) Isolated incoherent excitation-emission spectrum. (c) Spectrum of the coherently scattered light from the protein sample. (d) Excitation (black solid) and incoherent emission (photoluminescence) spectrum (blue solid) of the protein sample. The data are compared to absorption and emission spectra of the FAD chromophore in its fully reduced state (FADH^-) in a Photolyase protein (red dashed) [46] and as a mutant in insect Cryptochrome (green dashed) [47]. All datasets have been normalized to the excitation light spectrum.

Download Full Size | PDF

Both the excitation and emission spectra are in good agreement with the ones reported for the flavin chromophore in its anionic, fully reduced (hydroquinone) form (FADH^-). For comparison, an absorption spectrum of FADH^- in Photolyase is shown in Fig. 4(d) as a red dashed line [46], while the fluorescence spectrum of an FADH^- mutant in insect Cryptochrome [47] is reported as a green dashed line. We therefore conclude from our retrieved emission spectrum in Fig. 4(d) that the PL in the investigated ErCry4 sample is dominated by emission from the FAD chromophore in its hydroquinone redox form. We find no sign of emission from the oxidized or semi-reduced forms. Since the redox equilibrium between the different redox states can readily be controlled by adding suitable oxidizing agents to the buffer solution, such EE spectra provide a quantitative means for monitoring the redox state. Moreover, analysis of the time dynamics of the EE spectra can give insight into the binding of the FAD chromophore to the protein environment.

4. Conclusion

We have described and demonstrated a FT-based approach for the acquisition of excitation-emission (EE) spectra using two passively phase-stabilized common-path interferometers (BI-TWINS). By introducing one interferometer in the excitation and one in the detection path, we are able to distinguish and efficiently separate coherent light scattering and incoherent photoluminescence contributions to the EE spectra. For this, we exploit the intrinsically different time structure of these two emission components in the measured interferograms in the time domain. This separation is demonstrated for the weakly emissive flavin chromophore both free in solution and inside the protein European Robin Cryptochrome 4. The optical excitation of this chromophore is the primary step in the formation of radical pairs inside this protein. The latter are long-lived intermediates with reaction dynamics that are believed to be sensitive to weak geomagnetic fields [38,41]. The time-domain interferometric acquisition of the EE maps allows us to unambiguously isolate the partly spectrally overlapping incoherent fluorescence and coherent scattering components. Since, inside proteins, this fluorescence depends on the redox state of flavin, the obtained EE spectra can monitor the redox state of the chromophore by unambiguously determining its EE spectrum and quantify protein binding.

In the present study, we have analyzed time-integrated EE maps. It is straightforward to apply the same separation approach to TCSPC time-resolved spectra [30]. This may prove particularly useful for fluorescence life-time analysis of weakly emissive and/or highly scattering samples. The interferometric EE scheme based on the BI-TWINS may be helpful also for other techniques where scattering suppression is critical, such as e.g., hyperspectral imaging [31], coherent Anti-Stokes Raman Scattering [10], or stimulated Raman Scattering [11,12]. Further miniaturization of the setup, e.g., by replacing free-space propagation of the excitation and collected light beams by optical fibers, will facilitate the design of portable spectrometers for simultaneous absorbance and (time-resolved) excitation-emission measurements [50]. Other interesting applications of the demonstrated method lie in probing the (nonclassical) statistics of the emitted light.