3X multiplexed detection of antibiotic resistant plasmids with single molecule sensitivity

Abstract

Bacterial pathogens resistant to antibiotics have become a serious health threat. Those species which have developed resistance against multiple drugs such as the carbapenems, are more lethal as these are last line therapy antibiotics. Current diagnostic tests for these resistance traits are based on singleplex target amplification techniques which can be time consuming and prone to errors. Here, we demonstrate a chip based optofluidic system with single molecule sensitivity for amplification-free, multiplexed detection of plasmids with genes corresponding to antibiotic resistance, within one hour. Rotating disks and microfluidic chips with functionalized polymer monoliths provided the upstream sample preparation steps to selectively extract these plasmids from blood spiked with E. coli DH5α cells. Waveguide-based spatial multiplexing using a multi-mode interference waveguide on an optofluidic chip was used for parallel detection of three different carbapenem resistance genes. These results point the way towards rapid, amplification-free, multiplex analysis of antibiotic-resistant pathogens.

Graphical abstract

This work presents a rapid sample-to-answer system enabling multiplexed detection of three carbapenem antibiotic resistance plasmids with single molecule sensitivity.

1.0. Introduction

Some pathogenic bacteria have acquired genes or developed mutations that make them resistant to antibiotics. They have emerged as a global threat with several epidemics involving these pathogens already observed in the past decade ^1–3. Widespread use of antibiotics in animal production has contributed to this problem and has made antibiotic resistant bacteria a zoonotic disease ^4,5. Bacteria can quickly transfer antibiotic resistance genes via mobile elements such as plasmids ⁶. A new class of resistant genes such as Klebsiella pneumoniae carbapenemase (KPC), New Delhi metallo-β-lactamase (NDM) and Verona integron metallo-β-lactamase (VIM) have recently emerged ^7–9. These genes have spread among bacteria and confer antibiotic resistance against higher-level β-lactam antibiotics such as the carbapenems ¹⁰. Ineffectiveness of tertiary level antibiotics against these bacteria makes it difficult for physicians to combat and contain infections, especially among weak patients who have gone through complicated medical treatments and are at risk for secondary infections. This scenario has led to an urgent need to develop not only new antibiotics, but also new technologies for rapid, sensitive and early stage diagnosis of these resistance traits ¹¹.

Currently recommended diagnostic tests for detection of antibiotic resistant bacteria are mainly phenotypic or molecular based procedures ^12,13. Examples of phenotypic methods for testing antibiotic resistance in bacterial cells are disk diffusion test¹⁴ and Hodge test¹⁵. These tests involve culturing bacterial isolates in a growth medium plate with the presence of antibiotics. Assessment of antibiotic resistance in disk diffusion tests is based on measuring the zone diameter of the grown bacterial culture and the Hodge test is based on observing the shape of the bacterial colony (clover leaf). Since screening of antibiotic resistance using phenotypic tests is done by over-night incubation of cells and measuring the growth patterns, these tests are time consuming and neither highly specific nor sensitive ^16,17. Molecular based diagnostic tests such as direct gene sequencing or PCR give far more accurate and sensitive results ^18–20, but these methods rely on target amplification and require laborious sample preparation steps to avoid false results due to contamination from blood or serum ²¹. Microfluidics has been used to build chip scale devices which can perform such complex tasks. Droplet microfluidics^22,23 and micro actuator based sample processing chips are a few example of such devices that have demonstrated very precise handling of fluids and combining reagents. But they all rely on an external sensing device for analysis. Also, due to the emergence of multiple genes coding for resistance against an antibiotic, it is important to develop detection platforms with multiplexing capabilities, a feature many of the current diagnostic tests lack.

Here, we demonstrate a complete sample-to-answer system which consists of three different components (two sample preparation units and one sensor) for amplification-free, 3X multiplexed detection of carbapenem resistance genes (KPC, NDM and VIM) from whole blood samples with single nucleic acid sensitivity. Each of the components in the system was selected to minimize analysis time with the goal of keeping sample preparation and detection time to 60 minutes or less. Fig. 1 shows a schematic overview of the system. The first step in the analysis process is rapid separation of bacterial cells from red (RBC) and white (WBC) blood cells. This task is accomplished with a disk device that uses centrifugal force to manipulate fluids and extract target cells with 60% efficiency ^24,25. Another critical sample preparation step is target-specific capture and enrichment of DNA molecules. This is implemented on a polypropylene microfluidic chip with acrylate based porous monolith columns polymerized inside the fluidic channel ^26,27. The porous nature of the monolith provides a large contact surface area which allows for fast extraction and enrichment of biomolecules. These devices have been used to demonstrate capture and elution of bacterial DNA by functionalizing the monoliths with target-specific oligonucleotides ²⁸. Finally, amplification-free, multiplex detection of individual bacterial nucleic acids is enabled by a waveguide-based optofluidic sensor ²⁹. This optofluidic analysis chip consists of silicon dioxide based antiresonant reflecting optical waveguides (ARROW) that orthogonally intersect liquid-core (LC) ARROW waveguide channels ³⁰. They have been used to detect viral targets with sensitivity down to single nucleic acids and have shown detection limits below clinically relevant concentrations ^31–33. This platform has also been integrated with multi-mode interference (MMI) waveguides which generate wavelength specific well-defined multi-spot excitation patterns in the LC waveguide ^34,35. MMI excitation waveguides were used to demonstrate up to 6X multiplexed detection of single viruses and simultaneous detection of DNA and protein biomarkers using spectral and spatial multiplexing ^36,37.

Figure 1.

Cartoon depicting the process flow of the entire analysis system, (a) Sample: Whole human blood spiked with E. coli cells having antibiotic resistant plasmids. (b) Sample separation hollow disk device which separates bacterial cells from WBC and RBC. (c) Microfluidic chip which elutes selectively captured antibiotic resistant plasmids using a functionalized polymer monolith column. (d) ARROW bio-sensor chip detects the three antibiotic resistant plasmids (KPC, VIM, NDM) simultaneously using an MMI waveguide with single molecule sensitivity.

The system was validated with human blood samples spiked with E. coli cells that have pUC19 plasmids cloned with antibiotic resistant mutations. Three different antibiotic resistant plasmids pUC19-KPC, pUC19-NDM and pUC19-VIM were extracted, fluorescently labelled, and detected simultaneously using the ARROW based optofluidic biosensor platform. The amplification-free multiplexed detection of single antibiotic resistance plasmids from a sample set starting from whole blood points towards development of an autonomous sample-to-answer diagnostic system.

2.0. Principle and methods

2.1. Target Extraction from whole blood

i). Bacterial cell separation using spinning hollow disk:

When starting the analysis process with clinical samples such as blood, it is critical to separate the potential antibiotic-resistant bacterial cells from the blood before extracting the DNA. This greatly reduces contamination of the sample with other non-specific targets such as human DNA and blood cells, and reduces errors during target-specific plasmid extraction performed later. The sample separation hollow disk device uses quick application of centrifugal force to separate particles by their contrasting sedimentation velocities, which is based on both density and size differences. This device is 12 cm in diameter and can separate bacteria from 7 mL of whole blood per run, which is necessary for processing clinical blood samples which usually have very low bacterial concentrations. The disk, which spins around the central axis using a motor, has a bowl-shaped surface which slopes downwards to the center of the disk and has an annular fluid collection vestibule at the edge ³⁸. The vestibule has an upper chamber bounded by with a partial lid and a trough below this chamber. The trough is separated from the bowl by a retaining weir (see Supplemental material S2: Fig.1). 7 mL of blood spiked with E. coli DH5α cells containing pUC19-KPC plasmids (S.1: Antibiotic resistance plasmid-containing E. coli DH5α stock preparation) (final concentration of ≈5x10⁶ CFU/mL) and 1.5 mL of buffer were placed in the bowl and rapidly spun (S2(i): Sample separation disk device operation). All experiments involved with the collection and use of human blood from human volunteers were performed in compliance with the regulations of the United States Office of Human Research Protections, and in compliance with and under the regulation of the Brigham Young University Institutional Review Board for Human Subjects, approval IRB# X18-340. Informed consent was obtained from all volunteers. The sample is pushed outward and thrown into the vestibule, spreading it over a larger surface area, forming the blood into a very thin layer (about 2.5 mm thick and 9 mm wide) which is much smaller than the length that blood cells must travel during centrifugation techniques using test tubes. This enables quick separation of the cells with a run time of about two minutes which is important for developing a rapid sample processing system. The large discoidal blood cells, which have a higher sedimentation coefficient (24-27 times larger), sediment to the back wall of the vestibule much faster than the rod-shaped and smaller bacterial cells, thus forming a dense cell pack. The plasma is pushed out towards the central axis of the disk as the blood cells settle pack at the wall, creating a two-layer system (S2: Fig. 1a). The smaller bacterial cells get caught in this fluid flow and migrate towards the center of the disk, giving some concentration in the plasma. At the time when 99% of the blood cells have reached the cell pack at the wall, about 20% of the bacteria are also trapped in the cell pack and are not recoverable. The disk is then decelerated carefully to avoid mixing of the clear plasma with the blood cell pack ³⁹. The dimensions of the trough and thickness of the weir were designed ²⁵ so that the blood cell pack at the wall slumps down into the trough and is retained by the weir while the plasma with the bacterial cells flows over the weir and down into the bowl when the disk comes to a stop (S2: Fig. 1b).

Four mL of plasma with an average of about 5.25±0.3 x10⁶ CFU/mL (12 sets of blood separation experiments were done) of bacteria were recovered after separation (which is about 60% recovery). Any remaining blood cells were lysed and the dsDNA was extracted from the cells using standard protocol (S2(ii): Blood lysing and buffer exchange). 50 μL of 0.12ng/μL of DNA was obtained after the extraction procedure (S2(iii): DNA extraction). The dsDNA samples were sonicated to randomly break the circular plasmids. Fragmented DNA with size varying roughly from about 200 bp to 1,200 bp was obtained after sonication. Similar extraction procedures were applied to a blood sample spiked with E. coli cells having pUC19-NDM plasmids and a sample with E. coli cells having pUC19-VIM plasmids. The total time for the spinning sedimentation process was three minutes. When combined with the lysis and extraction steps, a total of 40 minutes was needed for each target.

ii). Microfluidic-chip based target specific elution of plasmids:

Target specificity is a crucial element of a diagnostic platform in order to give quick and accurate test results with few false negatives. This is achieved here using a polypropylene based microfluidic sample processing device which has a DNA-functionalized porous monolith polymerized inside the fluidic channel. The large contact surface area of the monolith enables sequence-specific capturing and elution of antibiotic-resistance plasmids with short incubation times, which would be good for point of care applications. The fragmented dsDNA extracted from the E. coli cells (Section 2.1.i) is flowed through the inlet of the chip which has a 500 μm wide X 500 μm tall fluidic channel. A heater kept below the serpentine channel of the chip denatures the dsDNA. The next section of the chip has a 600 μm long polymer monolith column (Fig. 2a) functionalized with commercially synthesized (Eurofins) acrydite-modified capture-oligonucleotides, designed with sequence complementary to a particular resistance gene sequence. The monolith is fabricated in the microfluidic channel in a single-step through photo-activated polymerization of acrylate based monomers ^22,23 and is porous in nature (S3(i): Chip Fabrication, S3(ii):Capture-oligonucleotides). The acrydite group⁴⁰ in the oligos polymerizes with acrylate-based monomers and is thus used for solid phase attachment of the capture-oligonucleotides with the monolith. Since bulk monoliths have different properties than ones prepared in smaller capillary or microchannel devices, most likely due to edge effects, BET measurements may provide inaccurate results and so SEM analysis for monoliths in microchannels was done ⁴¹. Fig. 2b and 2c show the porosity and surface area of the synthesized monoliths. Measurements (six measurements form SEM images) from the SEM data indicate an average nodule size of 1.2 ± 0.2 μm and pore size of 2.0 ± 0.4 μm. This ensures that the denatured target DNA follows a tortuous path through the monolith with more contact surface area ²⁸, increasing the likelihood of capture via hybridization. This is absent in other microfluidic chips which uses functionalized flat substrates or particles to capture targets, requiring them to have hour long incubation time periods. The monolith is also positioned close to the serpentine channel to make transit times after the heating step short to minimize renaturing of the DNA. After capturing, the monolith column goes through multiple rinse steps to completely remove non-specific DNA strands. The antibiotic resistance plasmid fragments are then thermally released (S3(iii): Heater temperatures) and eluted at the outlet of the chip in fractions of 35 μL. Three separate chips were fabricated to selectively capture and elute pUC19-KPC, pUC19-NDM and pUC19-VIM plasmid fragments. Successful elution of captured plasmids was confirmed via qPCR (measurements from 7 experiments), by comparing Ct values for the two rinse aliquots immediately preceding the subsequent two elution aliquots (S3(iv): qPCR reaction confirmations). For the seven qPCR confirmed runs, the mean difference (± standard deviation) in Ct values before and after elution was 2.3 (± 1.2). The remaining of the thermally released plasmids eluted form the monolith is fluorescently labelled by mixing with 1μM of POPO-3 nucleic acid staining dye off-chip and detected using the ARROW based optofluidic bio-sensor chip. The staining dye has high affinity for nucleic acids and upon binding to the plasmids, becomes fluorescent. The porous nature of the monolith column allows good flow and low incubation times enabling denaturing, capturing and elution of target plasmids with a quick run time of 12 mins. Since sample preparation is done in-flow using the monolith, this device doesn’t need microvalves for fluidic handling, compared to other valve-based microfluidic chips.

Figure 2.

(a) Top down image of the microfluidic chip used for target specific elution of the antibiotic resistant plasmids (scale: 1 cm). The chip has a serpentine channel in the bottom where a heater denatures the dsDNA. The white shaded part is the polymer monolith column functionalized with target capture probe oligos. (b) SEM image of the cross section of the microfluidic channel with the monolith (scale: 500 μm). (c) SEM image with increased magnification of the polymer monolith shows them to be porous with average nodule size of 1.2 ± 0.2 μm and pore size of 2.0 ± 0.4 μm (scale:5 μm).

2.2. ARROW biosensor device and experimental setup

Antibiotic resistance plasmids extracted using the sample processing devices were detected simultaneously with single molecule sensitivity using the ARROW optofluidic biosensor chip (S.4). The optofluidic sensor has a planar solid core (SC) waveguide orthogonally intersecting LC waveguides ^42,43. For multiplexed target detection using spatial multiplexing, the SC waveguide is designed to expand into a wide MMI waveguide section³⁴. This chip features (Fig. 3a) a single MMI waveguide that intersects three separate LC waveguides (Channel 1 [Chi], Channel 2 [Ch2] & Channel 3 [Ch3]). The inlets of the three LC waveguides start at separate positions and merge together to a common outlet after intersecting the MMI waveguide. The MMI waveguide is wide enough to support several optical modes with different propagation constants. The modes constructively interfere with each other and produce well-defined spot patterns at specific lengths along the waveguide for a certain wavelength (Fig. 3a). The positions of the LC waveguides are designed to intersect the MMI waveguide at these specific lengths and thus get excited by the corresponding spot patterns generated by the MMI waveguide. Here (Fig. 3b), the MMI waveguide of refractive index n (1.46) and width W (75 μm) is designed to intersect the LC waveguides at positions L_j (L₁ = 1676 μm, L₂ = 2243 μm & L₃ = 3350 μm). The dimensions are chosen such that the MMI

$$N_j\cdot L_j=\frac{n\cdot W^2}{\lambda }$$ Eq.1

waveguide generates N_j number of spots (11, 8 & 5) in the LC waveguides (Ch1, Ch2 & Ch3 respectively) when excited by light with wavelength λ (556 nm) (Eq.1).

Figure 3.

(a) Cartoon of ARROW optofluidic chip system for 3X multiplexed detection. Fluorescently labelled targets are introduced in the three LC waveguides. The signals generated by targets from each channel upon excitation by the MMI waveguide are collected simultaneously by a 3x1 Y coupler. (b) Top down image of an actual chip (Scale: 1 mm). (c) Fluorescent image of the MMI patterns in the dye filled LC channels (Ch1 to Ch3: bottom to top) when excited by the MMI waveguide with a 556 nm laser.

Fig. 3c shows the spot patterns produced in the fluorescent dye filled LC waveguides when the MMI waveguide is excited. About 4 μL of fluorescently tagged targets are introduced at the inlets of the LC channels and negative pressure is applied at the outlet to flow them across the MMI waveguide intersection. The micron scale waveguide dimensions of the chip results in femtoliter scale excitation volume and thus allows for rapid detection of single particles using very low sample volumes. The MMI waveguide is coupled to a 556 nm solid state laser via a butt-coupled fiber. The targets from each LC waveguide get excited by the MMI waveguide and generate fluorescent signals. The signals from each LC waveguide are collected by three separate collection SC waveguides which have an optimized 3X1 Y-coupler geometry to send the signals to an avalanche photo detector through a single output waveguide. Scattered excitation light is filtered out off-chip using a band pass filter. Based on the channel from which the target signal is produced, the specific MMI excitation patterns encode the spatial information in the time domain fluorescence signal. The target information is fully encoded in this fluorescence signal which is deciphered using a signal processing algorithm. This enables simultaneous detection of three different targets when introduced in the three LC channels of the biosensor chip. The total time to analyze a flow through sample volume was eight minutes for both singleplex and multiplex detection.

3.0. Results and discussion

3.1. Single plasmid detection using MMI waveguide

The ability of the ARROW bio-sensor device (Section 2.2) to accurately detect antibiotic resistance plasmids with single molecule sensitivity is demonstrated first. Plasmids that were target-specifically captured and eluted (Section 2.1.ii) were diluted ≈15X and tagged with 1 μM of green fluorescent nucleic acid staining dye (POPO-3). The staining dye has high affinity for nucleic acids and upon binding to the plasmids, becomes fluorescent. Each plasmid is size-based tagged with a multitude of such fluorophores and thus emits bright fluorescent signals when excited. Each type of antibiotic resistant plasmid is assigned a specific LC channel in the optofluidic chip (pUC19-KPC--Ch1, pUC19-NDM--Ch2, pUC19-VIM--Ch3 [Figure.4]).

Figure 4.

(a.i) Fluorescence trace for single plex detection of individual KPC. Fluorescently labelled KPC plasmids are flowed in Ch1 and buffer with dye in Ch2 and Ch3. The MMI waveguide is excited by 556 nm laser ((b) and (c) are signals of NDM and VIM plasmids from similar experiments). (a.ii) Zoomed in signal of a KPC plasmid shows an event with 11 peaks corresponding to the MMI pattern in Ch1. (a.iii) Histogram map of normalized S(t)j of a correctly identified KPC signal.

4 μL of fluorescently labelled pUC19-KPC plasmids were introduced in Chi, while just buffer with nucleic acid staining dye was flowed in Ch2 and Ch3. The MMI waveguide was excited and generates a pattern with 11 spots in Ch1. Fig. 4a(i) shows the fluorescence trace with signals from individual plasmids, defined as photon counts above a set threshold (black dotted line: highest signal from experiments done with negative controls, S5: Fig. 2). Zooming into a single event shows signal from a KPC plasmid with 11 peaks corresponding to the excitation pattern in Ch1 (Fig. 4a(ii). A signal processing algorithm (Eq.2) is used to identify all the signals in the trace ³⁶. F(t) is the measured time domain fluorescence signal and N_j is the number of spots in the excitation pattern generated by the MMI in each LC channel. The product operation in Eq.2 corresponds to shifting and multiplying F(t) several times by the characteristic time δt_j for a given spot number. If a particular spot number matches with that of the pattern in the signal, it gives rise to an enhanced S(t) value. To account for undesired enhancement due to background noise, the signals are normalized with the average background fluctuations (σ). δt_j is the characteristic time the target takes to travel between

$$S{\left(t,\delta t_j\right)}_j=\left\{{\Pi }_{i=0}^{N_j-1}\frac{F(t-i\cdot \delta t_j)}{\sigma }\right\}$$ Eq.2

adjacent excitation spots, calculated using the total duration individual plasmids take to pass the MMI waveguide intersection. σ is the average background noise (1.53 counts/10μs) plus standard deviation. The fluorescence signals are correctly identified based on which channel S(t) value dominates. Fig. 4a(iii) shows the S(t) values of a correctly identified KPC signal (from Fig. 4a(ii)). The signal has a large S(t) for channel one with N₁ = 11 and is low in channels two and three.

S6: Fig. 3a shows the distribution of the S(t) values for KPC plasmids. The distribution clearly shows most of these events to have very high S(t) in Ch1 and very low values in the other two channels, thus enabling correct target identification. Similar experiments were done for single-plex detection of pUC19-NDM and pUC19-VIM plasmids (Figs. 4b and 4c) and were correctly identified with an accuracy of 87%. NDM plasmids gave signals with 8 peaks and VIM plasmids generated signals with 5 peaks corresponding to the MMI pattern generated in the LC channel to which they were assigned (Figs. 4b(ii) and 4c(ii)). Furthermore, using the dimensions of the excitation volume and the time duration of the signals, an average concentration of 3x10⁶copies/mL of KPC and VIM plasmids and 8.5x10⁵copies/mL of NDM plasmids were estimated. The geometry of the SC collection Y coupler waveguides was carefully designed ³⁶ for efficient and unbiased collection of signals from all three channels. This is evident from the similar SNR (an average of 12.2) of the signals from all three above experiments.

3.2. 3X multiplexed detection of single plasmids

For amplification-free multiplex detection, fluorescently tagged KPC, NDM and VIM plasmids (Section 3.1) were introduced (Ch1, Ch2 and Ch3 respectively) in the chip simultaneously. The three different plasmids flowing in the separate LC channels of the chip were excited simultaneously by the MMI waveguide. Upon excitation by the MMI waveguide, the plasmids generated signals with distinct patterns (signals with 11, 8 or 5 peaks) which has the spatial information encoded in them. The signals from the three channels were collected together and sent to a single detector using the 3x1 SC collection Y coupler waveguide. Fig. 5a shows the fluorescence trace with signal from all three antibiotic resistance targets detected and decoded simultaneously. The signal processing algorithm S(t) described in section 3.1 was applied to the signals collected by the detector and was used to identify the events, eliminating the need for any off-chip signal decomposition elements for correct target identification. Fig. 5b shows the analysis for the S(t) of an event that clearly identifies that signal as coming from channel 1. The distribution of the S(t) values of all the events detected above the background threshold is shown in S6: fig. 3b. Events with dominant S(t)₁ are signals from individual KPC plasmids, and those of NDM and VIM plasmids are determined similarly. Plasmids were detected at a rate of 17±2 events/min with an average signal strength of 20±7 photon counts/10μs (measurements from three multiplexed experiments done using a sample set from section 2.1.ii). Fluctuations in the fluorescent signals arise due to distribution of the velocity of the particles and their cross-sectional position with respect to the MMI waveguide. Nevertheless, signals were detected with an average SNR of 13±4.5 (maximum of 37) by the chip. This demonstrates the rapid sensing capability of the optofluidic chip and shows the chip to have very good sensitivity to detect single plasmids. Furthermore, using the dimensions of the excitation volume and the time duration of the signals, an average of 2.0±0.6x10⁶ copies/mL of KPC and VIM plasmids and 9.0±2.0x10⁵ copies/mL of NDM plasmids were detected. Since the high sensitivity of the biosensor enables direct counting of individual plasmids, the concentrations of the plasmids estimated form the fluorescence trace can be used for quantitative analysis of the targets as was demonstrated before for singleplex detection of Ebola RNA over six logs, reaching a detection limit of a few pfu/mL³¹.

Figure 5.

(a) Fluorescence trace with all three antibiotic resistant plasmids detected simultaneously when the MMI is excited. KPC, NDM and VIM plasmids were flowed in the chip simultaneously in Ch1, Ch2 and Ch3 respectively. Targets were correctly identified by decoding the signals using eq 2. (b) Example of an event analyzed with S(t) clearly identifying the signal to be from channel 1.

4.0. Conclusions

In conclusion, we demonstrated for the first time, amplification-free, on-chip multiplexed detection of three antibiotic resistance plasmids with single molecule sensitivity. To mimic clinical scenarios, plasmids were extracted in a target-specific manner from template E. coli cells spiked in whole human blood, using disk-based separation of cells and microfluidic chips with functionalized polymer monoliths. The total time necessary to go from blood sample to analysis for a target was 60 minutes, including 40 minutes for cell separation, lysis, and extraction, 12 minutes for monolith capture and labeling, and 8 minutes for optofluidic analysis. While the short optical readout time already applies to the multiplexing case, the sample preparation steps can be carried out in parallel for multiple targets, ensuring compatibility of the multiplex analysis process with a one hour sample-to-answer time. Apart from detecting the presence of antibiotic resistance genes in bacteria, it is also important to simultaneously determine the bacterial species which is causing the disease in a patient. This requires scaling up multiplexing capabilities, which can be done on the ARROW optofluidic chip platform by combining spatial, spectral, and velocity-based multiplexing ^36,44.