1. Introduction

Flow cytometry (FCM) is a well-known diagnostic method that allows multi-parametric quantitative and qualitative classification of cells. It can be applied with little or no enrichment or target isolation in-vitro [1], in-vivo [2,3], and ex-vivo [4]. Recently, advancements in FCM setups have led to novel discoveries in immunology, virology, cancer biology, and infectious disease diagnosis [5–8], among others.

In conventional ex-vivo FCMs, a coaxial flow channel facilitates hydrodynamic focusing of stained biological cells with fluorescent markers passing across a stimulating laser beam. Laser-induced fluorescent and/or forward and side scattered signals of cells or particles are then collected by highly sensitive photodetectors such as photomultiplier tubes (PMT). Polychromatic flow cytometry (PFC) [6], a derived type of FCM, enables the excitation of multiple markers with several visible laser lines matched to the absorption spectra of multiple fluorophores, resulting in an accurate intercellular and molecular identification and sorting. Despite the advances in FCM setups in both technological and biological aspects, their high cost, bulkiness and system complexity have so far restricted their utilization to centralized laboratories. The development of more compact and cost-effective components as well as robust methods for signal acquisition and analysis is thus of paramount importance to enable their use in multi-centric clinical trials.

Multi-color laser engines (MLEs) replacing several discrete light sources are such a key component for PFCs, as well as other advanced biophotonic analytical techniques such as fluorescent microscopy [9, 10]. MLEs simultaneously or sequentially supply several excitation wavelengths of commonly used fluorophores with tunable and high optical output power levels, typically ranging from 20 mW to 100 mW in the visible spectrum. Embedded fiber switching between multiple optical outputs allows the stimulation of dyes in both colinear and/or parallel laser arrangement at the interrogation point of FCMs [11]. Digital intensity modulation in the kHz to MHz range empowers MLEs to enhance the spatial resolution of fluorescent images in high-resolution confocal microscopy [10].

In previous work, we have miniaturized an MLE with such typical functionality by integrating the variable attenuation, multiplexing and switching in a silicon nitride (SiN) based photonic integrated circuit (PIC) supporting a broad range of wavelengths spanning the entire visible spectrum [11]. SiN waveguide platforms [12–14], with a transparency range from the ultra-violet (UV) to the mid-infrared (MIR), offer a versatile technology for implementing PICs at visible wavelengths for servicing a multitude of sensing and diagnosis applications in the life sciences, such as Raman sensing [15], fluorescent sensing [16,17], or refractive index sensing [18]. Their compatibility with standard complementary-metal-oxide semiconductor (CMOS) technology [16,19] combined with on-chip microfluidics [20] might be a major step towards low cost and reliable integrated diagnostic systems for point-of-care clinical trials.

In this article, we demonstrate an SiN PIC-based MLE that generates structured illumination patterns [21] in the flow channel and apply it to PFC. Patterned illumination enables identification of spatially overlapping excitation wavelengths since the fixed flow rate of the cells or particles then results in signals with clearly distinguishable signatures in the time domain. The excitation wavelengths of 405, 488, 561, and 640 nm are focused onto the flow channel at a common interrogation point, but each is spatially modulated with a different periodicity. As a consequence, resulting signals have a recognizable repetition rate in time as the particle flows through the channel.

This approach aims at addressing limitations of existing PFC setups. In these, different fluorescent markers excited by different wavelengths but emitting at wavelengths too close to be separated with sufficient extinction can be hard to discriminate. One possibility is to then opt for a parallel illumination pattern with multiple interrogation points. However, stray light might still reach adjacent interrogation points, in particular since the fluidic channel can result in some guiding of the light. Increasing the distance between interrogation points results in its own set of difficulties, as measurements done at different interrogation points can then be difficult to attribute to a single cell, in addition to increasing the required sample volume. Here, on the other hand, the emitted signals carry the signature of the excitation wavelength, which is used for further signal separation in a later processing step, and are simultaneously applied to a single interrogation point requiring a single photodetector to collect all the fluorescent signals.

The data acquisition method is verified by mounting the MLE onto a cytometer and performing experiments with well calibrated fluorescent reference beads. The miniaturization afforded by the PIC facilitates direct flanging of the MLE onto the cytometer without interposed fiber, which is essential to apply the interferometric pattern to the flow channel.

To implement the PIC, we rely on the TripleX SiN platform [12,13] offered by LioniX International, which has been adapted to cover the whole visible spectrum with a transparency window starting at 405 nm. In this platform, a thin stoichiometric Si₃N₄ film is deposited with a thickness of ∼26 nm onto an oxidized silicon wafer with 8 µm thick SiO₂. After defining the waveguide core by a full etch, the SiN layer is overclad by an additional 8 µm thick SiO₂ film. Compared to silicon-rich films, whose light absorption edge is reduced below the nominal bandgap of stochiometric SiN, deposition of high-quality SiN with low-pressure chemical vapor deposition (LPCVD) allows extension of the transparency window over the entire visible spectrum, resulting in waveguide losses of about 0.3 to 0.4 dB/cm at 405 nm. The low confinement SiN waveguides facilitate matching of the mode profiles at the edge couplers (ECs) to that of commercial visible wavelength single mode optical fibers [11].

Section 2 introduces the architecture of the PFC with structured illumination, compares it to that of a conventional PFC setup, and describes the PIC and beam forming concept used for its implementation. Section 3 is dedicated to the beam forming optics used to shape the emission of the PIC before the flow channel and to the characterization of the resulting structured illumination patterns. Section 4 describes the PFC setup and its characterization with actual flow cytometry experiments demonstrating the envisioned signal generation and analysis method. Results are compared with those obtained with a commercial flow cytometer.

2. Multi-color engine and flow cytometer concept

A general schematic of a multi-color flow cytometer is depicted in Fig. 1. It consists of a multi-color laser source, beam shaping optics, the fluidic flow channel, and the measurement of the optically filtered fluorescent signals and of the scattered laser light. The illuminating beams, shown here in colinear configuration, are focused onto the fluidic sample flow that contains a core stream of cells surrounded by a fluid sheath of saline solution, which creates a hydrodynamic focus of cells flowing in a single line. The cells flow with a constant speed of up to 20 m/s. Given the typical cell size in the range of 5–15 µm [5] and the variability of the cell position inside the flow channel in the order of ±15 µm, uniform illumination in a range of at least 50 µm across the flow channel is highly desirable.

 

Fig. 1. General schematic overview of a multi-color flow cytometer with colinear illumination.

Download Full Size | PDF

The concept of excitation with structured illumination is shown in Fig. 2. The light shaping is realized with a PIC in combination with a beam shaping lens system interposed between the PIC and the flow channel. The three wavelengths 405, 488 and 640 nm are directly sourced by commercial laser diodes. The 561 nm wavelength, on the other hand, is provided by a compact frequency doubled diode laser (FDDL) sourced with a commercial 1122 nm laser diode. For each wavelength, a pair of convex lenses serves to couple the light to the PIC via ECs. One-by-two multimode interferometers (MMI), individually optimized for each wavelength, split the light for each wavelength, as two output ECs are required for each wavelength to generate the interferometric patterns. Symmetric directional couplers are used as wavelength combiners (WLC) to combine the 405 nm with the 488 nm and the 561 nm with the 640 nm light in each of the two optical paths. Thus, two output ECs are used to generate the interferometric patterns for 405 and 488 nm, while another pair of output ECs are used for 561 and 640 nm. The corresponding beams are sent to the same spot on the flow channel, where they interfere and create a grating type pattern with a periodicity that is specific to each excitation wavelength. For wavelengths output by the same ECs, the differing wave vectors proved adequate to generate illumination patterns with sufficiently distinct periodicities. However, in order to sufficiently differentiate all four wavelengths from each other, it proved necessary to use a different pair of ECs for 405/488 nm and 561/640 nm. In this case, the additional change in periodicity results from the different relative angle with which the beams cross at the flow channel, as also illustrated in Fig. 2. By focusing the beams before the flow channel, they diverge again and intersect the latter over an extended length in the order of 300 µm along its flow direction. After their focus points, the paths of the diverging beams are also chosen such that they overlap at the flow channel, placed at the focal plane of the lens system, so that the interference pattern is formed at that position (see Fig. 2 for a schematic drawing and Fig. 5 in Section 3 for the actual lens system fulfilling this functionality).

 

Fig. 2. Architecture of the SiN PIC generating the structured illumination patterns. The inset illustrates the interferometric pattern inside the flow channel (the actual lens system is more complex than a simple lens, see Fig. 5). The beam shaping optics refocus the emitted light and shape the periodic illumination pattern at the sample flow. On-chip devices consist of ECs, 3-dB power splitters in the form of MMIs, and WLCs. Unused drop ports are recombined by Y-junctions and terminated at the edge of the chip.

Download Full Size | PDF

The length of the interference pattern determines the spread of Fourier components associated to each illumination pattern. The larger the spatial extent, the better defined the periodicity, but the larger the probability of having two cells transiting through the illumination pattern at the same time and the lower the light intensity given the available laser power. The latter is partially compensated for chemical noise limited processes by taking the signal over an entire event duration, defined by the time taken for the cell to transit through the illumination pattern, and bandpass filtering its spectrum over the signal bandwidth. However, the larger interrogation volume then still leads to a deterioration of the signal to noise ratio, since a larger number of unbound fluorophores dissolved in the solution are being illuminated.

By using a PIC to shape the light, the interference patterns can be precisely formed without requiring the precise alignment of a large number of beam routing optics (mirrors), that would be necessary in a discrete optics solution. As a drawback, a PIC requires precise alignment of the laser beams at its input ECs, which can also lead to some long-term reliability concerns for a prototype device as used here. In its current form, the beam forming lens system following the PIC output has also been assembled with several setscrews, allowing fine adjustment of spring-loaded alignment guides, but also compromising the long term stability of the setup. In a production part, these could be removed to increase the robustness of the assembly. The layout of the SiN PIC is shown in Fig. 3.

 

Fig. 3. PIC layout. Waveguides are colored according to the wavelength that they are transporting. Unused waveguides are terminated at the edge of the chip. The inset shows the details of the output EC array.

Download Full Size | PDF

Since the in-coupling lenses and on-chip devices are identical to the ones developed for the miniaturization of a conventional MLE reported in [11], their detailed description is not repeated here. Only the geometry of the output ECs, as shown in the inset of Fig. 3, have been modified. All are tapered to 3.6 µm at the output facet. The waveguides that carry wavelengths 405 and 488 nm are spaced by 38.58 µm from each other (edge to edge) at the output interface, while the waveguides that carry 561 and 640 nm are spaced by 74 µm. The PIC is optimized to work with the TM polarization.

Overall on-chip losses have been characterized by injecting light into the PIC with a polarization maintaining fiber (Nufern PM-S405-XP) that remains single mode across the relevant wavelength range and has a mode field diameter (MFD) ranging from 3.3 µm at 405 nm to 4.6 µm at 630 nm, very close to that of the focused in-coupled laser beams in the final assembly. Insertion losses of 7.5, 5.25, 3.25 and 3.9 dB have been determined for the wavelengths 405, 488, 561, and 640 nm, respectively. In the following, if not otherwise specified, data is provided for these wavelengths in that order.

The PIC is mounted onto the mechanical stage with dimensions 56 by 89.5 by 15 mm shown in Fig. 4(a), that includes the mounted laser sources and interposed coupling lenses. It is very similar to the one used in [11], with a slight modification allowing mounting of the beam shaping optics [Fig. 4(b)] where previously an output fiber array had been attached to the PIC. A detailed description of the beam forming optics is given in the next section.

 

Fig. 4. (a) Schematic of the mechanical stage with dimensions 56 mm, 89.5 mm by 15 mm, and (b) beam shaping optics with a total length of 65 mm between PIC and flow channel. (c) Photograph of the mounted SiN PIC and assembled beam shaping optics under operation.

Download Full Size | PDF

 

Fig. 5. (a) Top-view, and (b) side-view schematic of the implemented beam shaping optics mapping the waveguide outputs to the interferometric illumination pattern in the flow channel, nominally located 65 mm from the PIC. The results of the ray tracing simulation are exemplarily shown for the 640 nm wavelength.

Download Full Size | PDF

3. Beam forming optics and structured illumination pattern

The beam shaping optics shown in Fig. 2 could in principle be implemented by a single lens. However, independent shaping of the beam widths in the directions along and across the flow channel would then not be possible. Moreover, a much wider spacing between the output waveguides would be required to achieve structured illumination with the targeted 300 µm width along the direction of the flow channel. Thus, a set of three lenses is first used to magnify the output of the waveguides onto an image plane located 61.7 mm from the PIC edge with a 136× magnification resulting in a wider distance between the corresponding spots. In a second step, a cylindrical lens located a bit before this image plane deflects the beams along the axis of the flow channel and creates the interference patterns. Figure 5 shows a diagram of the beam shaping optics, with the distance between the edge of the PIC and the flow channel being 65 mm. It targets an illumination pattern with a full width at half maximum (FWHM) of 300 µm along the flow channel and 60 µm along the other directions to accommodate the variability in the cell positions inside the flow channel.

The spatial periodicity of the fringes depends on the angle of incidence of the rays onto the flow channel, which is a function of the distance between the imaged ECs and, to a lesser extent, to the distance of the cylindrical lens to the flow channel. The latter is a consequence of the fact that as the position of the flow channel is modified, a different set of rays interact with each other at the center of the illumination pattern, whose direction of propagation deviates somewhat from that of the central ray of each beam. The periodicity also depends on the wavelength, as it modifies the magnitude of the k-vector and thus of its projection onto the xy-plane in which the interference pattern is formed. Increasing the distance between the waveguides or decreasing the wavelength of the light both result in the interference pattern having a smaller periodicity. Utilizing different sets of waveguides at the output of the PIC has the advantage of enabling sufficiently distinct illumination patterns even for wavelengths that are very close to each other. For example, in a setup using 532 nm (not used here) and 561 nm, that are both commonly used excitation wavelengths, a different pair of output waveguides with sufficiently different waveguide-to-waveguide spacing should be used to generate sufficiently distinct interference patterns.

3.1 Ray-tracing modeling

The illumination profiles have been modeled with non-sequential ray tracing with Zemax OpticStudio (9×10⁶ rays per wavelength). The beams emitted by the ECs have been approximated as having Gaussian profiles, with 1/e² MFDs along the in-plane direction (x) and the normal (y) to the PIC extracted from prior finite element mode solves performed in the Lumerical finite difference element (FDE) package to which a far field transform has been applied. A summary of the launch conditions is given in Table 1. The modes are significantly non-Gaussian along the normal to the chip (y-direction), explaining the discrepancy between their far field diffraction angles, taken as a basis for further ray tracing, and their near-field MFD.

Table 1. Modeled far-field 1/e² MFDs and half-diffraction angles of 3.6 µm wide output ECs. θ_x corresponds to the in-plane and θ_y to the out-of-plane half diffraction angles.

View Table | View all tables in this article

The magnifier is built out of a sapphire ball lens with a 9.53 mm diameter, followed by a plano-concave and a bi-convex lens, that are co-axial and all in mechanical contact with each other. Finally, the cylindrical lens deflects the two beams for each wavelength such that they intersect in the flow channel nominally located 65 mm from the edge of the chip, creating the targeted interference pattern. By implementing the initial beam transformation with a triplet of lenses with different materials, we have been able to tailor the chromatic aberration in such a way that the beam sizes obtained at the flow channel were closer to target for all four wavelengths in the raytracing simulations, even though the EC diffraction angles feature a substantial wavelength dependence. The free-space optics utilized for implementing the imaging system are summarized in Table 2.

Table 2. Utilized free-space optics for implementation of the beam transformation. The position of the lenses is defined as their distance from the edge of the chip.

View Table | View all tables in this article

Figure 6(a) shows the optical intensities recorded for each of the four wavelengths at the xy-plane intersecting the center of the flow channel sampled with a 400 nm resolution, with x the flow direction and y the direction across the channel that is also perpendicular to the optical axis (z). A small transmission loss of 1.15 dB is expected due to cumulative losses at the dielectric interfaces, the ball lens in particular being left uncoated. The fringes generated for the longest wavelengths, 561 and 640 nm, have a higher spatial frequency than for the shorter wavelengths, 405 and 488 nm, as a consequence of their corresponding ECs being spaced further apart. 561 nm has a higher spatial frequency than 640 nm and 405 nm has a higher spatial frequency than 488 nm, as a consequence of their shorter wavelengths. The spatial periodicities have been determined to be 19.55, 22.5, 13.9, and 15.85 µm for the four wavelengths. The illumination patterns feature an extinction ratio (ER) of 14.3, 13.2, 13, and 13.5 dB at the center. Towards the edges of the illumination pattern, the extinction ratio decreases as a consequence of one of the interfering beams being weaker.

 

Fig. 6. Modeled illumination pattern. (a) 2D intensity profile in the xy plane intersecting the center of the flow channel and (b) intensity along the center axis of the flow. Curves are color coded according to their wavelength.

Download Full Size | PDF

The spatial frequencies of the four illumination patterns are sufficiently different from each other so that the corresponding spectral distributions are almost fully disjoint, even though the illumination patterns have a finite spatial extent of ∼300 µm, an important criterion for being able to separate the generated fluorescent or scattered signals later on in the cytometer [see Fig. 9(a) in the next section].

3.2 Experimental characterization

The generated illumination patterns were characterized for each wavelength with a CCD-based laser beam profiler with a 3.69 µm pixel size for distances ranging from 61 to 69 mm relative to the PIC facet. The recorded data at 65 mm from the chip is shown in Fig. 7. This corresponds to a distance of 9.75 mm from the surface of the last lens, that is sufficiently large to accommodate the wall of the cuvette (see Section IV). In addition to the expected primary illumination pattern, as predicted by ray tracing, we observe a weaker secondary pattern at lower y-ordinates. We attribute this to multiple reflections between lenses and possibly reflective surfaces on the mechanical holder or PIC facet. A slight misalignment of a lens in the form of a slight tilt could lead to the observed displacement of this spurious pattern. Since the secondary pattern is applied outside of the flow channel at a distance much larger than the variability of the cell position, it does not have detrimental consequences on the cytometry measurements.

 

Fig. 7. Measured illumination pattern. (a) 2D intensity profile in the xy plane intersecting the center of the flow channel and (b) intensity along the center axis of the flow. Curves are color coded according to the wavelength. In (a), BR stands for the back-reflection induced artefact in the illumination pattern.

Download Full Size | PDF

In addition, a distortion of the primary pattern is observed at 640 nm on one of its sides. This is due to an excessively tight bend implemented after the drop port of WLC 640 nm resulting in stray light at the output facet of the PIC. However, the resulting Fourier transform remains confined in a sufficiently narrow spatial frequency distribution well in line with the initial design [Fig. 9(a)].

The measured FWHM of the illumination patterns are 232, 199, 243, and 287 µm along the (x-)direction of the flow channel and 11, 25, 36, and 36 µm in the transverse y-direction. As seen in Fig. 8 that compares the intensity profiles in the y-direction between experiment and simulation, the consistency is quite good once the spurious pattern has been accounted for.

 

Fig. 8. Intensity profiles of the illumination patterns along the y-axis cutting through their center. The colored solid curves show the measured data and the black dashed curves the simulated data. All graphs are color-coded according to the laser wavelength.

Download Full Size | PDF

 

Fig. 9. (a) Comparison of the Fourier amplitudes of the modeled and measured illumination patterns at the nominal flow channel position, 65 mm from the chip. (b) Spatial frequencies of the illumination patterns as a function of the distance from the surface of the last lens. In (a), solid curves show the modeled and dashed curves the experimental data. In (b), solid curves show the modeled data and diamond markers the experimental data. The detector position in (b) corresponds to the distance from the last lens. The datapoints at 9.75 mm correspond to the nominal position at 65 mm from the PIC interface.

Download Full Size | PDF

The periodicity of the recorded illumination patterns is in close agreement with the simulation results and measured as 19.32, 22.05, 13.6, and 15.3 µm. Figure 9 compares the simulated and recorded spatial periodicities as a function of the distance from the surface of the last lens. As already mentioned above, different rays, propagating at slightly different angles, participate in the formation of the interferometric pattern depending on the position of the flow channel, explaining this dependency. The maximum ER for the recorded data is determined as 4.5, 6.7, 5.4, and 6.25 dB and is lower than what was modeled. This is attributed to asymmetric losses inside the PIC between the two optical paths for each of the wavelengths, leading to reduced extinction. The dependency of the spatial periodicity of the illumination patterns on the position of the cells inside the flow channel is a non-ideality that could limit the performance of the cytometer if too large, since it could lead to the generated signals to have Fourier components in the neighboring bins and to be attributed to another excitation wavelength. However, if the position of the cells varies in a realistic ±25 µm range around the nominal position, at 65 mm from the PIC interface the spatial frequency of the illumination patterns varies by less than 0.5×10⁻³ cycles/µm, while the difference between different illumination patterns is between 7.2×10⁻³ and 11.5×10⁻³ cycles/µm. Consequently, the incurred variability is rather small and does not significantly impact performance.

More compact illumination with a lower extent along the flow direction is possible, but would lead to wider Fourier distributions (see Fig. 9) and to a wider overlap between them, unless the center periodicities of the different illumination patterns are also further separated from each other. This could be accomplished by redesigning the beam forming optics to increase the angle between interfering beam pairs at a given wavelength and thus increase the spatial frequency of the corresponding interference pattern. A dedicated pair of ECs could also be provided for each wavelength, further differentiating the illumination pattern periodicities from each other. However, ultimately, the minimum extent of the illumination patterns, and thus their spatial frequency, are not limited by the optics, but rather by the size of the characterized particles or cells. In order for the generated signals to track the shape of the illumination patterns, the detected particles should be smaller than the interference fringes, so as not to smear them out. The illumination pattern periodicities of down to 13.6 µm (at 561 nm) used here have been chosen to be commensurate with the dimensions of typically detected particles. While this smallest periodicity can consequently not be shrunk further, the largest periodicity of 22 µm (at 488 nm) could have been chosen larger, improving the differentiation between the spatial frequencies and allowing further shrinking of the illumination patterns’ extent (by less than a factor 2).

Interrogation points are typically only 9 to 30 µm wide in commercial cytometers, but spaced by 40 to 75 µm, corresponding in a total spread of 129 to 255 µm comparable to the overall illumination pattern size utilized here (since the patterns for the four wavelengths are overlapping here rather than being disjoint). In experiments in which several tests have to be attributed to the same cell, the entire pattern size rather than the size of single interrogation points is of relevance in regard to the disambiguation when attributing data to individual cells. In its current configuration, with the largest pattern periodicity at 22 µm, the structured illumination light source also allows the addition of further excitation wavelengths without increasing of the pattern size, by going to even larger periods for those. For the conventional cytometer, on the other hand, the pattern size will necessarily grow further with the number of channels.

4. Flow cytometry experiments

The structured illumination pattern and the MLE generating it have been used to conduct flow cytometry experiments. In the following, we first describe the setup and then the experimental cytometry results.

4.1 Flow cytometer setup

The schematic of the fully assembled multi-color flow cytometer is depicted in Fig. 10. The optical subassembly (Fig. 4) is combined with electronics comprising the laser drivers and a management interface (that are taken from existing designs and could be straightforwardly miniaturized) and the sample flow subassembly. In the following experiments, the distance between the last lens of the MLE and the flow channel turned out to be ∼6.5 mm instead of the targeted 10 mm, so that the periodicity [Fig. 9(b)] and the lateral extent of the illumination patterns were reduced from target, as discussed below.

 

Fig. 10. Top (a) and side view (b) of the MLE assembled with the sample flow subassembly.

Download Full Size | PDF

The fluidic sub-assembly provides the interaction region between the cells and the laser beams. It combines a quartz glass cuvette containing the sample stream and an objective lens to collect the optical response of the passing cells. A pump system ensures a constant sheath flow inside the cuvette for the hydrodynamic focusing of the samples. A 0.125 mL syringe pump facilitates injection of the sample into the center of the sheath flow. The microfluidic channel has a cross-section of 200 by 250 µm. An adjustable objective lens (50X Nikon CFI60 TU Plan Epi ELWD Infinity Corrected) with a numerical aperture of 0.6 is attached to the cuvette in an orientation orthogonal to that of the excitation laser beams to collect fluorescent and side scatter (SSC) signals of passing cells. After reduction of the collected beam width with a pair of lenses (Thorlabs TTL165-A and Thorlabs AC254-045-A) the light is split by a 90:10 beam splitter (Thorlabs BS025). The reflected 10% are directly detected by a PMT (Hamamatsu H11900-20) for SSC detection, while the transmitted 90% are first filtered by a multiband filter (Semrock 446/523/600/677 HC Quadband Emitter) to remove light from the pump beams before being detected by a second PMT. This second PMT jointly records all the fluorescence signals, that are only distinguished by their time-domain signature. Finally, the PMT signals are processed using a field-programmable gate array (FPGA). The complete setup is shown in Fig. 11.

 

Fig. 11. Fully assembled flow cytometer with PIC-based structured illumination. The inset shows the positioning of the cylindrical lens versus the cuvette with a representation of the generated beams and the location of the interferometric pattern.

Download Full Size | PDF

4.2 Experimental results

The cytometer is first characterized with non-stained polystyrene particles with a diameter of 6.2 ± 0.1 µm (microParticles GmbH, PS-F-6.2), prior to performing fluorescence measurements with stained ones. Exemplary time-domain scattering signals obtained one at a time by sequentially turning on the excitation lasers are shown in Fig. 12 over the duration of a scattering event, i.e., the duration taken by the particle to flow through the illumination pattern at the dialed in flow rate of ∼3 m/s. Figure 13(a) shows the power spectral density (PSD) distribution of these signals, estimated from the squared magnitude of a fast Fourier transform (FFT) applied over the time domain signal shown in Fig. 12, with peak modulation frequencies at 278.7, 235.5, 380.5, and 340.6 kHz. It is apparent that here too the PSD distributions corresponding to the different pump lasers are mostly disjoint, as required for the signals to be discriminated from each other in the following experiments.

 

Fig. 12. Time-domain signals for scattering events with non-fluorescent particles recorded for each exciting laser wavelength. The curves are normalized to the peaks of the events and are color-coded according to the laser wavelength.

Download Full Size | PDF

 

Fig. 13. (a) PSD of the time domain signal recorded for non-fluorescent polystyrene beads at the flow speed of ∼ 3 m/s. (b) Histogram of the peak frequencies for illumination patterns at 561 and 640 nm for 10,000 events at the flow speed of ∼0.6 m/s. The curves are color-coded according to the wavelength of the illuminating laser.

Download Full Size | PDF

In a second experiment, the flow speed is reduced to ∼0.6 m/s for a higher exposure time and thus higher signal strength and to obtain a more stable flow velocity. 10,000 scattering events are recorded for both the 561 and 640 nm excitation wavelengths. The center frequencies are extracted for each event and plotted in the histogram shown in Fig. 13(b) in order to establish the repeatability of the measurements. The distributions corresponding to the two illumination patterns can be seen to be well separated. They have mean frequencies of 81.4 and 71.8 kHz for the wavelengths 561 and 640 nm, respectively, and a coefficient of variation (CV) of 2% attributed in both cases to fluctuations in the flow speed.

For the extraction of individual events, a filter with a passband from 65 to 95 kHz is first applied to the time domain signal, removing its DC component and reducing noise. A peak finding algorithm is then utilized to detect each event and extract the FWHM of its duration. Finally, an FFT is applied to each event over a duration of two FWHM, further referred to as the ‘event duration’.

For the fluorescent signal measurements, a mix of stained and unstained calibration particles with a diameter of 3.8 ± 0.2 µm (SPHERO^TM Ultra Rainbow Calibration Particle Kit, Cat. No. URCP-38-2K) is utilized with a concentration of 500 particles per µL. Particles are each stained with the four fluorophores Coumarin 30, fluorescein isothiocyanate (FITC), Nile Red, and allophycocyanin (APC), at five intensity levels to assess all common fluorescence channels at different staining levels. These have their main absorption peaks at 405, 488, 550, and 651 nm, close to the utilized excitation wavelengths. An exemplary section of the recorded time-domain signal for the SCC and fluorescence channels, with the 561 and 640 nm lasers turned on, is shown in Fig. 14(a). Figure 14(b) shows a magnified view of a single fluorescent event containing the response to both excitation wavelengths. The PSD of the selected scattered and fluorescent signals is depicted in Fig. 14(c) and features two distributions centered on the expected frequencies 85.3 and 76 kHz. For both the SSC and the fluorescence channel, it is apparent that the signal resulting from 640 nm excitation is much weaker than the 561 nm one, which was not the case in earlier experiments [see Fig. 13(a)] and was a result of the MLE being damaged. The output emission at 405 nm and 488 nm was also significantly weakened due to the misalignment of the built-in micro-optics and already suffered from a weaker extinction ratio [see Fig. 12], so that it proved difficult to reliably collect and analyze data at these wavelengths. Consequently, the quantitative analysis done in the following focuses on the signal corresponding to the 561 nm wavelength, selected by choosing the proper frequency domain integration bounds, with a nominal laser output power level of 20 mW.

 

Fig. 14. (a) Recorded PMT signals over an exemplary time-duration comprising two events with the SSC (black) and fluorescent (orange) signals. The first event corresponds to an unstained particle, the second to a stained one. (b) Detail of the fluorescent signal resulting from both 561 and 640 nm excitation. (c) PSD of the modulated SSC (solid curve) and fluorescent signal (dashed curve) with peak frequencies at 76 and 85.3 kHz.

Download Full Size | PDF

For a systematic analysis of the data, a similar methodology as above is used. After bandpass filtering, identification of individual events and of their duration, the scattering and fluorescence signals corresponding to excitation with the green (561 nm) wavelength are estimated by integrating the corresponding PSDs, estimated from the squared magnitude of the FFT, over the frequency range 77.5 (${f_{min}}$) – 95 kHz (${f_{max}}$). They are given by

For comparison purposes, a reference measurement was also performed with a commercial MACSQuant Analyzer 16 flow cytometer, in which forward scattering data (FSC) is recorded in addition to the side scattering data (SSC). The combination of these two datasets allows to distinguish events corresponding to single particles from clustered ones. It is thus a common practice to subselect the main cluster in the scattering data for further analysis of the fluorescent data. The selected datapoints are shown in green in Fig. 15(a) and correspond to the region with the highest density of datapoints accounting for 79% of the overall recorded data.

 

Fig. 15. Analysis of the flow cytometry data for (a) and (b) the commercial MACSQuant, (c) and (d) the investigated cytometer. (a) SSC vs. FSC. (b) Nile Red fluorescence vs. SSC. (c) Total SSC vs. event duration. The dashed red lines correspond to the mean value of the event durations (470 µs) ± its std. dev. (43 µs). (d) Normalized FL_G vs. SSC_G. The selected data is color coded in green for both datasets. In the MACSQuant datasets, the selected data includes 79% of the overall data. In the investigated cytometer, 94% of data points fall between the red lines and are further down-selected to correspond to the strongest total SSC signals, representing 27.8% of the overall data points.

Download Full Size | PDF

For exciting the Nile Red dye, a laser at 488 nm falling inside its absorption spectrum was utilized with a maximum power of 50 mW. The fluorescent signal is plotted as a function of SSC in Fig. 15(b). The green colored points correspond to the selected particles as shown in Fig. 15(a), while the black-colored ones correspond to the rejected ones that are excluded in the further analysis. Six distinct populations representing unstained particles as well as particles with five distinct staining levels are visible.

A similar selection has been done for the data collected from the experimental cytometry setup investigated in this study. In this case the Nile Red was excited with the 561 nm laser. It is labeled as “green” fluorescence accordingly, as this is the information available to the cytometer (as opposed to the emission wavelength, that is not filtered). Since FSC data was not available here, the total SSC was plotted against the determined event duration to perform the selection [Fig. 15(c)]. The first criterion was for the event duration to fall within its mean value (470 µs) ± its standard deviation (43 µs) [shown as red dashed lines in Fig. 15(c)], which results in 94% of the events being included (due to outliers increasing the std. dev. relative to a Gaussian distribution) with a CV of 9%. We further subselect the datapoints with the strongest total SSC signal, resulting in 27.8% of the overall population shown in green in Figs. 15(c) and 15(d), for the reasons explained below:

Upon closer inspection of the data, when plotting FL_G versus SSC_G, a strong correlation is seen that results from variations of the illumination pattern strength across the flow channel. As a consequence of the distance between the last lens and the flow channel being shorter than initially envisioned, the MFD of the illumination pattern in the y-direction orthogonal to the flow direction is in the range of 18 to 20 µm, which is small given the variability of the bead position. In comparison, the commercial MACSQuant cytometer has an MFD of 700 and 115 µm at 488 and 642 nm, that are much larger than the bead position variability. However, since the SSC signal reflects the illumination pattern strength independently on how strongly the bead is stained and all the particles used here have the same scattering cross-section, it can be used to normalize FL_G in regard to the illumination strength as FL_G/SSC_G, with results shown in Fig. 15(d). It is apparent that this results in distinct clusters, for which the normalized green fluorescence FL_G/SSC_G is tightly clustered around an average value, but for the weakest signal cluster. In the lowest cluster a slope appears as a consequence of FL_G being submitted to a noise floor not yet seen in the stronger SSC_G signal. By selecting the datapoints with the strongest SSC signals, this problem can be circumvented so that the weakest cluster can also be clearly separated out based on the normalized green fluorescence signal. This measurement floor also leads to the weakest staining level to be indistinguishable from the unstained particles, so that we obtain 5 instead of 6 clusters.

It should be noted that in a measurement involving several different kinds of particles or cells, this strategy would not work as straightforwardly, as the scattering signal would also change as a function of the type of detected object. However, this is an aspect that can be significantly improved with a redesign of the beam shaping optics to create a more homogeneous illumination pattern strength, which could be done with more resources, for example in the framework of a commercial development.

The distribution of the normalized FL_G signal for the selected datapoints is represented as a histogram in Fig. 16. The relative populations for each of the clusters are summarized in Table 3 for both cytometers. There is overall a good agreement between the two measurement sets. The discrepancies at Blank + Intensity Level 1 and Intensity Level 3 are a bit larger, but remain within the estimated confidence intervals derived from the number of taken data points (that do not otherwise reflect on the accuracy of the two measurement systems).

 

Fig. 16. Histogram of the normalized FL_G signals for the investigated cytometer.

Download Full Size | PDF

Table 3. Relative populations. Confidence intervals are given as two times the estimated std. dev., based on the number of analyzed gated datapoints (14400 and 2500, respectively for the commercial and the experimental PFC).

View Table | View all tables in this article

5. Conclusions

We have demonstrated a new concept for the generation and discrimination of signals in polychromatic flow cytometers relying on structured illumination. Other than in conventional PFCs, signals can be discriminated based on the exciting wavelength without requiring spatial separation between the interrogation points and without being subjected to stray light. We have shown that signals can be generated in four commonly used exciting wavelength channels that could be clearly separated in scattering signals obtained from non-fluorescent beads. Quantitative fluorescent measurements have been performed with 561 nm excitation, resulting in a categorization of the test beads according to staining levels within expected error margins based on the number of acquired data points. As a current performance limitation, the investigated cytometer shows a large variability of the illumination strength across the flow channel, that requires normalization of the fluorescence with the scattering data and should be improved before particles of different size and composition can be reliably analyzed. The noise floor in the experimental device also prevented resolving the most weakly stained beads resolved by the commercial device with which it was compared, showing that some improvement would still be required to match the performance of mature, commercial devices. Nonetheless, these results are already very promising given the amount of development that has gone into the latter.