Optics-Integrated Microfluidic Platforms for Biomolecular Analyses

Abstract

Compared with conventional optical methods, optics implemented on microfluidic chips provide small, and often much cheaper ways to interrogate biological systems from the level of single molecules up to small model organisms. The optical probing of single molecules has been used to investigate the mechanical properties of individual biological molecules; however, multiplexing of these measurements through microfluidics and nanofluidics confers many analytical advantages. Optics-integrated microfluidic systems can significantly simplify sample processing and allow a more user-friendly experience; alignments of on-chip optical components are predetermined during fabrication and many purely optical techniques are passively controlled. Furthermore, sample loss from complicated preparation and fluid transfer steps can be virtually eliminated, a particularly important attribute for biological molecules at very low concentrations. Excellent fluid handling and high surface area/volume ratios also contribute to faster detection times for low abundance molecules in small sample volumes. Although integration of optical systems with classical microfluidic analysis techniques has been limited, microfluidics offers a ready platform for interrogation of biophysical properties. By exploiting the ease with which fluids and particles can be precisely and dynamically controlled in microfluidic devices, optical sensors capable of unique imaging modes, single molecule manipulation, and detection of minute changes in concentration of an analyte are possible.

Introduction

Despite the burgeoning activity in the field of optofluidics, there are many open opportunities to integrate optical tools with biological microfluidic handling systems to perform previously impossible analyses. Many optics-integrated microfluidic systems demonstrate comparable sensitivity to nonmicrofluidic systems, but most do not take full advantage of the potential to integrate sample handling and parallelize assays for ease of biological analysis. Assay systems that simplify and multiplex measurements and diagnostics could expand the reach and impact of optofluidic systems by making some types of high-throughput measurements possible and by making technology available to more diverse settings.

Biological samples are diverse yet sparse solutions of cells, proteins, and small molecules, and are typically acquired in limited fluid volumes (milliliters to microliters), making them especially difficult to process. The issues of sample complexity and low molecular concentration, however, can be significantly decoupled from the problem of detection and quantification of analytes by incorporating microfluidic methods. Myriad developments in microfluidics over the past two decades have demonstrated that biological samples can be successfully manipulated and separated from undesired components on-chip. The inherent heterogeneity in biological systems is a complicating factor that often requires the same assay be performed many times, a job that microfluidics are well suited for. In addition, many optical sensing systems can detect low concentrations of analyte, often with limits of detection in the pico- to attomolar range. In this review, we will attempt to provide the reader with an introduction to biological sample preparation in microfluidics as well as available cellular and subcellular techniques for optical manipulation and sensing in the hope to inspire new and interesting marriages between these fields. To this end, we will discuss manipulations and separations at molecular and cellular levels using optical techniques, hydrodynamics, and other force fields, in addition to optical tools commonly used in sensing systems, and optical sensing systems themselves.

Materials and Methods

Physical separations

Separations are often required when working with biological samples to isolate a desired entity, sort a mixed population, or concentrate highly dilute mixtures. Because of the large variety of biological molecules that exist in cellular environments and the rarity of many of those molecules, separation is frequently the first step in biological analysis. By integrating these separation steps into a microfluidic system, it is possible to perform efficient and highly precise separations over a large range of particle sizes, from individual molecules and proteins (on the order of angstroms) to cells (on the order of micrometers) to small model organisms such as Caenorhabditis elegans (on the order of millimeters), in small volumes of liquid (on the order of nanoliters). Optical separations by themselves are limited by the analytical meaningfulness of optical differences. For example, the refractive index of individual cells within isogenic cell populations is often highly heterogeneous, depending on size and shape, making separation difficult. Consequently, although high-throughput and single-cell optical separation has proven very useful, there remain opportunities to improve the large-scale handling of particles. Before delving into more specialized optical separations, we will briefly mention some relevant nonoptical separations that supplement optical separations and could well be used to complement other optical sensing techniques. For a more thorough review of individual techniques, we point the reader to references cited in this section.

Nonoptical separations

Within nonoptical separations, there are primarily two types of separation: those that occur passively due to hydrodynamic forces (pressure fields) within a microfluidic device, and those that occur due to imposed electrical or magnetic fields. As hydrodynamic forces are intrinsically required for microfluidic flow, they are particularly easy to integrate as a means of separation. The laminar nature of microfluidic flow results in a replicable system whose flow field can be modeled relatively simply, dramatically streamlining device design. Cells or other particles with different size, shape, or stiffness passing through a specifically designed flow field experience differential forces that direct them to different regions of the flow, allowing them to be separated passively. Applications are typically limited to populations where physical properties contrast significantly, but recent developments in inertial microfluidics have shown that micro- and nanoparticles can be separated in a high-throughput manner (600 μl/min) with high purity (90–99%), demonstrating potential for cell throughput close to the order of magnitude of flow cytometers (1, 2). Although inertial microfluidics operate at higher Reynolds numbers, they remain laminar. Despite exposure of biologicals to higher shear stress compared with noninertial microfluidics, shear forces are much lower than what samples experience when they are pipetted, and separation in this manner may be gentler than centrifugation purification or physical filtration. Operation windows of inertial devices, however, are quite small and careful design is required (3). Microfluidic chromatography is another common method for which implementation is straightforward. High-resolution size separation over a large size range can be performed using deterministic lateral displacement, where regularly spaced features within the microfluidic channel bump different sized particles into different streamlines (4, 5). For biological entities with a defined immunophenotype, microfluidic channel surfaces may be coated with antibodies before sample addition to perform affinity chromatography (6). The concept is simple, however, the requirement for a sample with a defined immunophenotype may be limiting, as cells must come into sufficient contact with channel walls to be captured, and antibodies are not only costly, but may also exhibit nonspecific binding.

Externally applied fields may also be applied to microfluidic channels using principles of field flow fractionation (7). Separation by magnetic field is possible, but requires the use of magnetically responsive particles and antibodies in addition to an external magnetic field (6). Often the most useful technique in this category is dielectrophoresis (DEP) (8). Differences in dielectric properties of sample constituents allow continuous separation using an inhomogeneous electric field in DEP. The primary challenge of DEP is that many characteristics can impact the dielectric properties of the cell, including, but not limited to, cell size, and cell membrane and cytosolic content (which impacts permeability, capacitance, and conductivity). However, DEP has the significant advantage of being label-free and a reasonably sensitive method for separation despite its nonspecific nature (9, 10, 11, 12, 13, 14). The applicability of DEP must be weighed carefully, as more complicated design considerations are involved, and exposure to an electrical field may damage or alter biological samples.

Optical property-based separations

Label-free methods

In general, optical manipulations operate on objects of approximately micron scale and have high spatial resolutions in addition to being label-free and operating in a noncontact mode (15, 16). Techniques that require labels have certain disadvantages; interference with molecular structures whose physical conformation is desired and interaction with biological processes of interest are both potential sources of error in labeled techniques, not to mention the added expense of additional reagents. The use of optical manipulation techniques on-chip often still requires bulky external equipment and patient optical alignment, however, external optics also allow optics to be decoupled from the microfluidics, enabling on-the-spot adjustment of key parameters rather than requiring further fabrication and redesign. Similar to DEP, optical manipulation can be very precise, but may require more complicated fabrication or alter cell physiology when high-powered lasers are involved.

Investigating physical properties on the single-cell and single-protein level is an ideal application point for integrating optics and microfluidics. The use of optical tweezers (OT) to apply force, torque, and other fine-scale manipulations in a massively parallel fashion provides a window into the world of heterogeneity in both mechanical and optical senses. OTs use spatial light gradients to focus micron-sized particles to the center of a laser beam due to differences in radiation pressure. Classic examples are the measurement of the movement of the biological motor kinesin, and the characterization of the elasticity of DNA through attachment of these nanoscale objects to micrometer-sized spheres (17, 18, 19).

Typical OT set-ups require high-power infrared lasers to focus particles, necessitating substantial modification to commercial microscopes and limiting the number of manipulable entities. The high-powered lasers needed could also lead to photodamage of biological molecules. Recently developed microfluidic alternatives for OTs have used plasmonic nanostructures to create evanescent waves at the interface of thin metal surfaces, generating a very strong near-field effect capable of trapping dielectric particles (20, 21). Due to the relative system-level optical simplicity of optics on microfluidic systems in comparison to OTs, the localized nature of near-field effects, and the overall compactness of systems relying on surface plasmon resonance (SPR), plasmonics-based trapping lends itself to integration with microfluidic devices. SPR techniques on-chip is very promising for precise particle manipulations, including subdiffraction-limit trapping (22).

Efficient assay parallelization leads to greater ease of sampling larger subsets of a population, thereby informing our perspective on heterogeneity. A current limitation for optical manipulations is the inefficiency of scaling up to many optical traps on a single microfluidic device. Although technologies like holographic optical tweezers and opto-electronic tweezers allow large-scale manipulation, these systems are expensive and optical contrast between different types of cells is often low, making manipulation difficult and again leading to the use of high power lasers (23, 24). In a similar vein, projection of optical images onto microfluidic devices have been used to passively separate cells based on their optical contrast and polarizability (which increases as size increases), however, this relies on changes in optical contrast and polarizability being useful and meaningful ways to separate cells, which is not always the case (25).

A technique that may enable high-throughput and precise manipulation on-chip has recently been described that employs lower power lasers than those typically required for multiplexing optical traps (26). In this case, multiple traps are created at antinodes of a standing wave from a single laser and traps are repositioned by using thermal energy to change the phase of light waves. Rudimentary sorting has been performed by holding bead-conjugated biological samples in one set of traps while applying small lateral pressure fields to push untrapped particles out of the device, or alternatively, by flow switching based on Raman spectra obtained from single cells (16, 26). This application relies heavily on the creation of highly stable traps that would experience significant drift outside of a well-defined microfluidic environment, and the ability to switch fluid flow quickly, a clear strength of microfluidic systems. Continued effort to create large arrays for simultaneous manipulation of individual particles in an efficient manner will likely remain a significant goal because of its potential impact on easing the investigation of heterogeneity and other fields reliant on big data for answers, for example genomic studies.

Labeled separations

Images are one of the highest content data formats; a typical photosensor has several megapixels worth of data points. The conceptually simple idea of separation based on image properties and features is an integral part of many biological assays that can be supplemented through the use of microfluidic devices. By using microfluidic chips, micrometer-scale specimens can be imaged either in series or in parallel, both of which can enable high-throughput modalities (27). This is a common theme over a broad range of phenotyping applications, including both small organisms such as C. elegans as well as cells (28, 29, 30). The ability to restrict the movement of cells and model organisms while orienting asymmetrical objects reproducibly via hydrodynamic forces within microfluidic devices eases the time burden of physical manipulation as well as simplifies quantitative comparison of specimens. For living animals, various schemes are available for immobilization within devices, including cooling and compression, which are often much gentler compared with the usual glue or anesthetic methods (29, 31). Because these chips are made of the flexible polymer polydimethylsiloxane (PDMS), chips are inexpensive to make, optically transparent, and compatible with biological specimens. The material properties of PDMS also enable on-chip membrane valves to conveniently control fluid flow. Although the use of microfluidics in this manner requires additional pressure sources and controls, time involved in operation and handling is orders of magnitude less compared with individually mounting and imaging animals or cells. Thus, for high-throughput experiments or experiments requiring large numbers of samples, as in genetic screens and phenotyping experiments, more comprehensive arrays of experiments come within reason for a single researcher to perform.

Many separations based on image properties are facilitated by fluorescent labeling, but separation through image-based phenotyping is not exclusively confined to labeled reagents. Given that an object in a population to be sorted has sufficient regional variation that can be identified via brightfield, darkfield, or other modes of microscopy, these types of images can be used as the basis for separation. However, the use of transgenic cells and species that encode fusion proteins for reporter systems are common because of the ability to label proteins in subpopulations of cells in vivo. Subcellular localizations of proteins can often be determined as well (although how significantly fusion proteins affect localization is a matter of debate), and autofluorescence is typically low enough in most cells and organisms that differentiating background from foreground is relatively simple. Compared to a label-free system, a subcellularly localized reporter system can give much more detailed information about whole-system function because tagged proteins are actively involved in cellular processes. In addition, the number of reporters is theoretically only limited by the number of optically orthogonal reporters possible within the color spectrum, although in practice excitation and emission spectra often overlap significantly, making it impractical to use more than three or four reporters. This has recently driven the development of reporters that emit in near-infrared range. Despite the usefulness of fluorescent reporters, the effort required to transform organisms and transfect cells can be significant, and damage due to photobleaching, as well as differences in the rate of photobleaching at different locations within the cell makes long-term monitoring more challenging for parallelized methods (32). If intracellular or intraorganismal protein distribution is not required, this additional effort may not be warranted and label-free sensing may be more appropriate.

On microfluidic chips, active separation of samples may involve acquisition of an image of the specimen, followed by actuating a valve on device based on optical feedback (i.e., based on specific image features), limiting the specimen to following a single path of least fluidic resistance. Recovery of specific specimens out of a large microfluidic array is sometimes only possible through the use of optical methods like OT that can provide control over individual specimens, however, this increases cost and power input significantly while requiring more user intervention (33). In contrast to most optical separation methods, using image features as criteria for separation is extremely versatile because separation need not be based upon a single feature, such as refractive index or physical size, but upon combinations of many features predefined by the user. Machine learning algorithms may be used to produce these combinations, limiting user-introduced bias toward phenotypes most evident to the human eye. However, even simple image processing followed by the application of heuristics relevant to the application of interest can be of use in situations where a desired phenotype is predefined (33). Although computer vision tools vary in terms of ease of use, there are many machine learning techniques that can be used out-of-the-box or with little modification. The reader may consult Elicieri et al. (34) for more insight into image analysis packages and image processing and computer vision examples. In addition to consumer software available for computer vision, continuing innovations in computer vision algorithms have resulted in consistent improvement in terms of accuracy, speed, and bias reduction in the ability to detect image features in an unsupervised manner.

Results

Optical tools for guiding light on-chip

In many optical sensing situations, the integration of multiple optical tools, such as lenses, waveguides, and lasers is critical to create a mechanism for sensing. Traditional optical systems for use with biomicrofluidics rely heavily on these components in the form of off-chip microscopy. In comparison, although most of the on-chip techniques described here are still dependent on external sensors (e.g., charge-coupled device or complementary metal oxide semiconductor sensors) to collect data, on-chip and partially on-chip optics confer advantages in terms of footprint, cost, and timescale while maintaining sensitivity expected from similar techniques. As many high-sensitivity optical techniques already require micro- or nanoscale structures to confine and propagate light, some type of microfabrication is already necessary in most cases. The addition of microfluidics, then, follows logically as a convenient way to control sample volume and handling without adding significantly more fabrication steps.

On-chip lens systems

Lenses are integral parts of most optical microscopy systems, but traditionally have little flexibility in terms of dynamic changes to focal length, necessitating precise translations in the z-direction to collect in-focus light from multiple image planes. In contrast to the rigidity of traditional lenses, cheaply produced on-chip lenses provide flexibility both in terms of chip design and material properties, and can be used to facilitate on-chip imaging in a variety of ways. Practically, lenses are some of the easiest elements to integrate, as many common materials in microfabrication can be manipulated into cheap and robust lens arrays using simple techniques (35, 36). Particularly in photometry applications, where wavefront fidelity is less important, simple converging and diverging lenses can be of use. Although variable-focus lenses were initially developed at the macroscale for use in spectacles, the fine scale at which they can be fabricated in microfluidics has yielded a small host of applications at the micron scale (37, 38, 39). Adjustable focal length microfluidic lenses are tunable by variation of refractive indices or lens shape. Refractive indices may be controlled through lens composition, or via thermal or electric fields, can be tuned pneumatically, through environmentally responsive materials such as hydrogels, or with electromagnetic fields (38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48).

A primary challenge in fabricating variable focus lenses that function by lens deformation is in creating lenses with large dynamic ranges and short response times; for example, pneumatic control of fluid-filled reservoirs behind thin, flexible membranes (composed of, for example, PDMS) is a common microlens approach that is limited by slow response times despite large tuning ranges and easy device integration (Fig. 1 A) (49, 50, 51). Although PDMS is optically transparent and has low losses in the visible range, light scattering through the polymer is not always negligible due to nanoparticles of silica within commercially available products (52). Alternatively, nanoliter-size droplets can be easily, robustly, and precisely formed in microfluidic devices whose focal length can be tuned by altering the surface tension at the droplet interface or electrical field within the microfluidic device (43, 49). Environmentally responsive lenses, such as hydrogels, have also been used to adjust focus in a passive manner (see Fig. 1 D). In one interesting example, red blood cells were used as microlenses, switching their focal length from negative to positive values by changing the chemistry of the buffer solution and using their focusing properties as a basis for phenotyping (53). Using electrowetting to induce changes in lens geometry by altering the liquid contact angle has distinct advantages in consistency of fluid deformation, response time, and the need for only small driving voltages as compared with large pressures (Fig. 1 B) (45, 46, 54). The creation of arrays of these lenses on-chip is simplified by the ability to address individual droplets to spatial locations, either hydrodynamically or through the use of electrophoresis or electrowetting on dielectrics (EWOD). Temporal modulation of an electric field has recently been used to both create two-dimensional (2D) monodisperse droplet arrays for use as microlenses and to achieve large deformation of microlenses for variable-focus tunability over millisecond timescales, a speed similar to that of the inertial response limit that can be affected by a common piezoelectric z-stepper (49, 55).

Figure 1.

Physical configuration of variable focus lenses. (A) Pneumatically tuned lens. (B) Lens controlled by electrowetting. (C) Hydrodynamic lens. (D) Environmentally responsive lens. Dashed lines indicate approximate limits of physical tunability. ITO, indium tin oxide. To see this figure in color, go online.

In addition to these vertical tunable lenses, tunable lenses composed of core and cladding fluids within horizontal microfluidic channels also enable dynamic modulation of lens shape through hydrodynamics or electrokinetics (Fig. 1 C) (47, 56, 57, 58). By exchanging fluids used for core and cladding, an additional parameter can be used for focal plane adjustment, further broadening the capabilities of this type of variable focus lens (59). Although these lenses are often easier to initially form, they can also be perturbed by small particles or bubbles within the flowing fluid, leading to significant chromatic and geometric aberrations and making them more difficult to maintain over long periods. Table 1 summarizes advantages and disadvantages of the primary types of dynamic lenses and example applications. Although acoustic tuning of microfluidic lenses has not been explored to our knowledge, at larger scales acoustic waves have been used to successfully tune lenses to allow the collection of three-dimensional (3D) images on submillisecond timescales (55).

Table 1.

Various Mechanisms for Employing Variable-Focus Lenses in Microfluidics and Their Advantages and Disadvantages

Tuning Mechanism	Advantages	Disadvantages	Application References
Pneumatic	easy to integrate with existing microscopy systems	some arrangements require manual alignment of several device layers	adaptive focal length with large viewing angle (39)
large focal length tuning range	require high pressure source (∼40 psi)	air-liquid interface manipulation used for lens tuning (48)
can incorporate as in-plane (horizontal) or out-of-plane (vertical) configurations	long-term deformation and mechanical fatigue
reasonable response time (100 ms)	require external hardware and control
Electrowetting and dielectric forces	high lens-shape repeatability	electrode incorporation	adaptive lens on flexible surface for improved POV (47)
fast response times (50 ms)	open-air systems nonsterile, prone to evaporation	improved stability of dielectrically adjustable lenses through liquid tuning (45)
require high voltages (∼100 V), potential for hydrolysis	focal point adjustment via electrowetting followed by UV polymerization (46)
only for planoconvex or planoconcave configurations
significant external hardware and control required
electrowetting: Joule heating and microbubbles
Hyrodynamic	molecularly smooth interfaces	slow response time (s)	liquid core/liquid cladding lens (56)
many lens shapes possible	precise pressure control necessary (<1 psi resolution)	fluid diffusion-based lens (59)
easily disrupted by liquid impurities
Responsive material	ability to alter lens shapes based on meaningful physiological change	slow response time (s)	chemosensitive hydrogels (42)
variety of lens shapes possible	red blood cell as tunable lens (53)
passive

Independent of advantages related to tunability, lenses integrated in microfluidics have been used to improve spatial resolution of fluorescence imaging as well as improve fluorescence intensity signals by capturing supercritical angle fluorescence (35, 60). Other devices have incorporated large matrices of inkjet-printed micron-sized lenses composed of the photoresist SU-8 as a simple way to improve sample focus and magnification (61, 62). These could be useful for multiplexed droplet measurements at reduced cost and timescales, a particularly important quality for combinatorial problems.

Despite the usefulness of lenses to improve spatial resolution, the low efficiency of light collection through lens materials and the inability to decouple field of view from resolution has resulted in some microfluidic systems adopting lens-free configurations in which microfluidic chips are directly coupled to charge-coupled device or complementary metal oxide semiconductor sensors (63, 64). The use of microfluidics in these systems has enabled large-scale multiplexing and analysis of a great variety of biomarkers, as well as DNA, concentrations of therapeutic agents, and even small model organisms on platforms for rapid clinical diagnostics that require only microliter samples (65, 66, 67). One flavor of lens-free imaging collects holograms (interference patterns) to reconstruct 2D and 3D images (63, 67). A second type of lens-free imaging relies on sampling of transmitted light through a specimen in contact with a sensor array, capturing its shadow (63). In addition to multiplexing clinical diagnostics, lens-free optofluidics has also enabled pixel superresolution on-chip by several different methods: the highly controllable fluid flow in microfluidics can be used to effectively raster-scan cells and small multicellular organisms by imaging through apertures smaller than individual pixel dimensions (68), and by capturing multiple holograms at different specimen illumination angles (69, 70). Similarly, submicron apertures in a microfluidic device with a tilted channel can be used to acquire a pseudo-z-stack by capturing images as a sample travels through the channel (71). These conceptually simple lens systems have been some of the most influential optofluidic developments because of the large acceptance of similar techniques at the macroscale and prevalence in conventional imaging. They are also mechanically quite simple to reliably integrate compared with many other optical sensing tools, with a main barrier to widespread use being the off-line computational nature of the required algorithms.

Leveraging reflection

The optical properties of interfaces can give rise to on-chip mirrors. Through the use of thin metal coatings, liquid-air interfaces created by bubbles, inkjet technology, liquid-core/liquid-cladding systems, and liquid metals, light reflection has been achieved on device (35, 72). These methods can all be used on-chip to create optical chains that direct light to sensors. The use of reflective microprisms for 3D imaging (73) and improved detection of fluorescence signals through increased optical resonance without the sacrifice of a smaller field of view (35) are two interesting applications that leverage reflection to increase throughput and speed of imaging while also expanding the volume of data. Recently, stable liquid-liquid mirrors have been implemented for use as parallelized optical switches in an on-chip light router, realized by using immiscible liquids on a device where wettability can be controlled via application of an electric field (EWOD, electrowetting-on-dielectrics) (74). If multiple sensing regions are required on-chip, optical switches created by stable liquid-liquid mirrors could be used to direct light from different samples to a single, shared photosensor.

Traditional optics use solid-core waveguide components whose refractive index is higher than the surrounding materials to guide light. However, most liquids have refractive indices lower than those of typical cladding materials, which led to the development of specialized microfluidic tools for waveguides designed to couple light into fluidic channels (75). Many solutions have been proposed to enable efficient waveguides with minimal losses, but in particular the use of liquid-core/liquid-cladding (L²), antiresonant reflecting optical waveguides (ARROW), and nanophotonic structures have taken foothold (75, 76, 77).

ARROW and liquid core systems

ARROWs are constructed on-chip by layering dielectric materials (often SiO₂ and SiN; layer thickness depends on desired wavelength) in a microfluidic channel (78). These layers act as Fabry-Perot reflectors, which create constructive interference for in-phase light and destructive interference for out-of-phase light passing through the liquid core. ARROWs have successfully been used to measure fluorescence down to picomolar levels, are subject to relatively low losses, even in millimeter-sized channels, and can be implemented as single-mode waveguides (75, 79). ARROW does not require the large refractive index ratio that other liquid-core solutions need to achieve total internal reflection at dielectric surfaces. Compared to the coating of microfluidic channels with solid metals or fluorinated polymers, liquid/liquid waveguides can take advantage of laminar flow regimes in microfluidic channels to limit mixing of the two liquids, providing a much smoother optical surface along which light is guided, even for rough channels, decreasing overall losses. In comparison, this can be a limitation with thin metal layers, as deposition is nonuniform, particularly in rectangular microchannels (75). Liquid/liquid systems also allow modularity via the ability to switch between different liquid-core/liquid-cladding pairs at the expense of somewhat more complicated systems. However, when handling live biological specimens, in which water is often a primary component, liquid/liquid systems often prove difficult because of water’s relatively low refractive index (1.33) (80). Nanoporous materials and fluorinated polymers have also been proposed to enable liquid-core waveguides, however, these have found limited applications due to their involved fabrication processes.

Single influenza virus particle detection has recently been demonstrated in an ARROW-based device, using nanopores to simultaneously achieve small volume excitation of labeled particles and obtain a molecularly distinctive current signature resulting from the passage of a molecule through a nanopore in an electric field (Fig. 2) (81). Integration of both electrical and optical sensing within the same platform provides a further opportunity to improve upon the ability to differentiate between multiple molecular species, reaching fidelity of up to 100% (81). Despite complicating overall design and fabrication, multimodal detection schemes provide a specificity and fidelity that some purely optofluidic systems (e.g., phenotyping based on refractive index) lack due to the often lower specificity of label-free techniques (82).

Figure 2.

Single influenza virus particle detection. (A) The incorporation of both solid-core (orange) and liquid-core (blue) ARROW waveguides allow small volume optical excitation of fluorescent virus particles, whereas virus particle transversal of a nanopore results in transient drops in current. (B) Particles pass first through the nanopore, over which current changes are measured (top, black) and then through optical detection volume, where fluorescence signals from viruses (red) and nanoparticles (blue) are measured, resulting in a known dwell time. Blockade depths from both types of particles are very similar, but variable dwell times allow them to be distinguished. The incorporation of both optical and electrical detectors enables the separation of single viruses from a mixture of nanobeads and viruses (81). This is an unofficial adaptation of an article that appeared in an ACS publication. ACS has not endorsed the content of this adaptation or the context of its use. http://pubs.acs.org/doi/full/10.1021/nl502400x. To see this figure in color, go online.

Photonic crystals and optical cavities as sensors

The unique properties of photonic crystals (PhCS) allow the very accurate confinement of light within small volumes. PhCs are periodic dielectric micro- or nanostructures that act as band-pass filters, only propagating certain wavelengths of light depending on the periodicity of the structures and the refractive index of the medium. Because the medium is subject to dynamic control within microfluidic devices, the propagated wavelength can be altered simply. PhCs have led to very low-loss waveguides whose small structures are realizable using microelectromechanical systems techniques (83). Bragg gratings, a type of one-dimensional photonic crystal structure, have been used to detect unamplified and unlabeled genomic soy DNA as well as discriminate cells based on intracellular density (detection limit 2 × 10⁻⁵ RIU) by measuring wavelength shifts in infrared light in microfluidic structures (84, 85). Another early, but cogent example of the use of PhC waveguides in a dye laser used PDMS posts within a microfluidic channel as Bragg gratings in a very simple fabrication process to create a distributed feedback dye laser (86). The wide detection range of PhC-based sensors and familiar fabrication techniques used in creating PhC structures are promising for label-free sensing.

Recent innovations in fabrication processes using pico- and femtosecond lasers have also made precise microfabrication of internal structures in transparent materials possible, enhancing ease of integration of more intricate PhCs on device (87). Apart from use as a means for fabrication, a newly developed microfluidics-enabled method for delivering cargo such as live bacteria and protein into cells uses rapidly pulsed lasers to create cavitation at a SiO₂-coated porous membrane, transiently opening pores in cells seeded on the membrane with a delivery efficiency of nearly 100% (88).There is also significant potential for colloidal photonic crystals in waveguides and light filters on device—although assembly of colloidal photonic crystals has traditionally required a timescale on the order of hours, a demonstration of much shorter timescales for spatially selective assembly of colloidal crystals using optical patterns has been published, which could have interesting implications for the use of photonic colloids in microfluidics (89).

Optical cavities (also called optical resonators) are optical paths that enable the formation of a standing wave. For most applications, single-mode light is preferred because beam quality is better preserved, which is essential for ensuring sensitivity of many optical tools. In addition to profiling protein adsorption, DNA, virus, cell membrane adsorption (via membrane proteins), and cell adsorption are all candidates for quantification using optical cavities (90). One of the most important metrics for optical cavities is their Q-factor (Quality factor), which characterizes the sensitivity of the resonator to shifts in wavelength caused by adsorption of molecules to the surface of the cavity (90). Optofluidic resonators are sometimes implemented as ring-resonators, which can support standing wave patterns called whispering gallery modes (WGM) with extremely high Q-factors up to 10⁸ (91). High-Q WGM sensors have lower losses than low-Q sensors and can have sensitivity down to the detection of single molecules because of their ability to recirculate light around the optical cavity, effectively enhancing the small shifts in wavelength that indicate molecules have absorbed (Fig. 3) (91). In this way, WGM sensors can surpass the sensitivity of many enzyme-linked immunosorbent assays (ELISA) or SPR detection methods (90). The label-free nature of this technique is also advantageous, although for the detection of specific proteins, DNA, etc., a corresponding and specific antibody, aptamer or other binding agent must be adsorbed to the surface of the resonator. Multiplexing detection of multiple biomolecules can be easily achieved by exciting multiple WGMs in the same sensor (92). The structure sizes required for the detection of biomolecule presence are also on the micron scale, making them a natural choice for integration into microfluidics.

Figure 3.

Principle of WGM sensors. Light from a laser (red) is evanescently coupled into the dielectric toroid, sphere, cylinder, ring, or disk via a light guide, such as a tapered optical fiber. The wavelength of the laser is tuned so that the light traveling around the surface of the WGM remains in phase when it returns back to the point of coupling. As the resonant wavelength is approached, power is extracted from the fiber, decreasing the amount of light transmitted to the detector. When analytes (orange) are bound to the sensing surface through antibodies (green), the resonant wavelength of the system shifts due to local changes in the index of refraction. Resonance shifts down to 6 fm can be detected with appropriate detectors (90). To see this figure in color, go online.

Most fabrication methods for microfluidic WGM use toroidal glass capillaries and microspheres embedded in PDMS devices, which is labor intensive and requires specialized instrumentation for precise placement (93, 94). Because of this and sensitivity to thermal and mechanical drift, microfluidic WGM sensors are currently limited, despite their attractiveness as potential high-sensitivity detectors for resource poor regions (95, 96). Despite this, WGM remain an attractive option and have been used to detect a variety of biomolecules, including DNA and protein (97), detect cell adsorption (98), measure the anisotropy of coated surfaces (99), characterize the kinetics of membrane receptor adhesion and signaling (100, 101), or to provide the feedback mechanism for lasers (discussed below). In addition to this versatility, recent reports have shown improved sensitivity (on the order of femtomolars) as well as shorter sensing times (102). Potential for high-throughput drug screening applications based on combinations of this technology with microfluidics have been proposed that could help lessen the burden associated with the advent of personalized medicine, however, there are few current examples that use label-free, multiplexed assays to their fullest extent (103, 104). Thus, although there is certainly opportunity for improvement in fabrication methodology, microfluidic WGM sensors can be helpful for quantification of a number of aspects of biological importance. Another simpler technique that may be used for observing kinetics of enzymes and biological molecules in a highly parallel platform is to use zero mode waveguides, consisting of subdiffraction limit apertures, to reduce the volume of detection to femtoliters (105). However, in this type of waveguide, noise can be hard to eliminate and can obscure single molecule events, making it necessary to dilute the sample sufficiently (to approximately nanomolars) to maximize fluorescence fluctuation due to a single molecule’s presence (106).

Metasurfaces, planar structures that locally control amplitude, phase, and polarity of light, are also a significant area of interest for building small scale and highly controlled systems (107). Metasurfaces can be fabricated using photolithography, as opposed to the complex steps to fabrication that WGM require. Furthermore, at the nanoscale, many optical properties can be dynamically changed by tuning electromagnetic forces, through thermal heating, or optical stimulation, leading to potential for switchable optical materials (108). Development of metamaterials alongside microfluidic systems could provide a broader set of optical capabilities with familiar fabrication steps.

Interferometric and spectroscopic sensors

Interferometers in optical microfluidics are another common method used to measure changes in refractive index. A single source light wave is split into multiple light waves via a beam splitter. These two beams travel different paths, with the change in phase of the two beams creating interference patterns which depend on the change in refractive index of the medium. These phase shifts can be used to measure changes in fluid composition (i.e., presence of analytes), as well as pressure and flow rate on microfluidic device (109). Simple interferometric microfluidic refractive index sensors can have sensitivities of 10⁻⁵ RIU (refractive index units), the equivalent of ∼0.05% (w/w) change in solute concentration, and a limit of detection around 10⁻⁷ RIU (110, 111). The convective flow through microfluidic interferometers also permit diminished timescales compared with bulky cuvette interferometers, because many molecules of interest are physiologically at low concentrations and these rare molecules must be transported to a surface. By sandwiching a functionalized nanoporous layer between two flow layers, timescales can be decreased further, because the entire volume of fluid must pass through an array of nanopores where analytes may bind, providing a larger sensing surface (112). Another useful volume confinement technique that microfluidics enables is confinement of individual sample packets within nanoliter droplets made on-chip (113). Through droplet confinement, smaller reaction and photosensing volumes can be achieved and reactions can be performed in massively parallel fashion while contamination is prevented.

A recent example of interferometry in ecological diagnostics quantified bacteriophage concentrations in water samples by incubating phage with their bacterial hosts inside droplets (114). By performing this incubation within droplets on device, simultaneous large-scale analysis of multiple samples is possible. After incubation, each droplet is individually interrogated using interferometry to determine bacteriophage concentration. Some popular types of on-chip interferometers that are simple to fabricate include Fabry-Perot and Mach-Zehnder configurations. Although it has not yet been implemented in microfluidics, in one recent report, interferometric detection of Rayleigh (elastic) scattering with a limit of detection ∼10 pM was used to determine kinetics of binding and spatial position of individual protein molecules of <60 kDa (fibrinogen, BSA, IgG) (115). This platform could benefit from the tightly defined flow within microfluidic channels as well as the associated reduction in size and cost and potential increase in throughput.

Despite the success of Raman spectroscopy as an analytical technique, Raman scattering is very weak in comparison to elastic scattering, requiring minutes-long light collection (116), leading to the development of surface-enhanced Raman spectroscopy (SERS), where molecules are confined near a roughened surface to improve scattering characteristics. Microfluidics offer a convenient and simple way to deliver small amounts of analyte to the sensing location in a precise and controlled manner, the ability to separate heterogeneous biological samples, and the rapid detection of biomolecules. For example, microfluidic SERS have been used to detect drugs of abuse (117), dopamine (118, 119), and various other biomolecules (120, 121). Although levels of individual proteins cannot be detected in living cells due to the complexity of cells, a cellular fingerprint can be useful in comparing cells. Dochow et al. (16) used a Raman spectra fingerprint to train and automate the separation of circulating tumor cells from blood. A significantly simplified fabrication method that nanostructures a silver layer by oxygen plasma treatment may in the future make this technology much more accessible for those without specialized facilities (118). However, most microfluidic SERS platforms have a low sensitivity, making them useful for binary detection, but not for minute changes in concentration (117, 121).

Lasers as detectors

Lasers have been implemented with great success on microfluidic chips, not only for molecule or cell manipulation, but for sensing purposes as well (122). A laser is composed of an optical oscillator (a cavity), a gain medium (for the stimulated emission), and an external energy source to stimulate the gain medium (usually a second laser, but can be other sources) (122). In particular, Fabry-Perot cavities are simple resonators that are amenable to adaptation in microfluidics, and have found widespread use in many photonic applications (123, 124, 125). Benefits of using on-chip lasers include ease of use compared with typical bulky laser systems and tunability of the lasing wavelength through different solvent mixture concentrations (126).

One popular subset of optofluidic lasers uses organic dyes in liquid as the gain medium. Due to the tunability of liquid gain media, it is also possible to have multiple color lasing through the use of different dyes. Other optofluidic lasers have used photonic crystals or fluorescent molecules as the gain medium. The ELISA technique, used most often for detecting the presence of antibodies in biological systems, was recently redesigned as an optofluidic laser-based assay with a limit of detection at 38 aM by using fluorescent products of the enzymatic reaction as the gain medium (123). In this work, the laser turn-on time is the signal, allowing better isolation of the ELISA signal from other noise, such as excitation light leakage, leading to a six order of magnitude detection range. The use of biological gain mediums, which often use fluorescence resonance energy transfer between a fluorescent molecule donor and an acceptor such as Rhodamine to create the gain medium, are a particularly interesting new direction considering the number of biological assays that already use fluorescence as a reporter (123, 127, 128, 129).

Random lasers, for which, as the name suggests, the pathways over which light is amplified appears random, have also been demonstrated. Because random lasers do not require reflection via mirrors, the fabrication process of on-chip lasers can be further simplified (130). In fact, the seeming randomness of random lasers is caused by miniscule defects in the polymers that constitute the lasing medium, and by controlling pumping, random laser emission spectrum and directionality may be controlled (131). Although random lasing may seem undesirable, when one considers that many superresolution microscopy techniques use the stochastic excitation of fluorescent molecules to achieve subdiffraction limit resolution, it may in the future find applications in imaging.

Surface plasmon resonance on-chip

SPR arises when light incident upon noble metal structures has a frequency that matches the natural electron oscillation frequency of the metal, causing the resonant oscillation of electrons on the metal surface. This creates surface plasmon polaritons that propagate along the surface of the metallic feature and that will be disturbed by even small irregularities, such as protein binding. SPR is a standard tool used with refraction sensors to measure binding kinetics and affinities, however, macroscale measurement systems are bulky and expensive, whereas their microfluidic counterparts, are cheap, easy to fabricate in bulk (especially compared with more conventional prism-based SPR), simple to operate, and easy to integrate with existing systems (132, 133).

When SPR is created at surfaces of metallic nanofeatures, called localized surface plasmon resonance, or LSPR, the near field wave amplitude, and thus the refractive index, is enhanced at these features and rapidly decays with distance, enabling spatial resolution limited only by the size of the nanostructure used. LSPR systems require equivalent fabrication steps or fewer than SPR by using gold nanoparticles as nanoscale features. One common drawback of refractive index measurements off-chip is the long timescale required to detecting low concentrations of analyte; diffusion of analyte to the sensing surface is the rate-limiting step that makes high-throughput difficult. Hence, a large surface area/volume ratio is desirable, which microfluidic environments can supply through the use of functionalized nanopores. Other methods are often used in conjunction to further amplify a signal obtained via SPR, for example agglomeration of particles, dielectrophoresis, and the use of antibodies and aptamers to confine molecules to a surface (134, 135, 136). Microfluidic LSPR has been demonstrated to achieve a figure of merit approaching that of the theoretical limit of propagating SPR sensors, with reported limits of detection in the attomolar range (136, 137). Using LSPR on-chip, interactions between biological molecules, including immunoassays and protein surface binding kinetics, can be performed with high sensitivity and simplify the construction of necessary fluid-flow systems (138, 139). In a similar vein, the dynamics of cell secretion have been monitored by trapping cells on microfluidic chips with both LSPR and SPR modes to measure refractive index changes with excellent sensitivity and lateral resolution on the order of microns (Fig. 4) (140, 141). By trapping and culturing cells on-chip, secreted factors accumulate quickly enough to provide more than simple endpoint information about secretion rate. Furthermore, LSPR has been used for superresolution optical trapping of dielectric particles with a reduced power and laser intensity compared with conventional far-field optical trapping, a potential benefit for microfluidic devices that require parallel handling of objects (22). Localized surface plasmons are not only capable of enhanced light scattering, they are also sources of heat (several nW/nm³) when illuminated, allowing very fine temperature control over the plasmonic surface that could be useful in calorimetric applications (142).

Figure 4.

Optofluidic human blood immunoanalysis. (A) Schematic of optofluidic device. (B) System functional states during optofluidic ELISA. (C) Scanning electron microscopy image of cells conjugated to microbeads trapped by micropillar array in the microfluidic chamber. (D) Representative readout from LSPR experiment (141). This is an unofficial adaptation of an article that appeared in an ACS publication. ACS has not endorsed the content of this adaptation or the context of its use. http://pubs.acs.org/doi/full/10.1021/nn406370u. To see this figure in color, go online.

Discussion

The prospects for optics-integrated microfluidics are bright—the wide acceptance of the microfluidics field for a broad variety of applications has familiarized many researchers with microelectromechanical systems fabrication, resulting in a broad group that could benefit from having extended and uniquely on-chip optical sensing. The applicability of these techniques to biological and chemical analysis has been shown using a variety of inexpensive optical tools over a wide range of analyte size scales, from millimeter-long nematodes to individual molecules. Optical measurement methods are noncontact, with sensitivity and limits of detection rivaling those of comparable techniques at a lower cost—for commercial SPR sensors, typical sensitivities lie around 2 × 10⁻⁷ RIU (143), similar to high-end sensitivities produced in microfluidic devices (110, 111). Optical manipulation is extremely precise and has a multitude of uses that, in combination with separation and handling techniques in microfluidics, can be highly precise. In addition to cost and sample handling, microfluidics provide a timescale advantage: small scales and high surface area to volume ratios allow optical detection of small molecules in a greatly reduced timespan compared with conventional methods that operate in milliliters of bulk solution. Furthermore, many of the most sensitive optical techniques require nanoscale structures, which share some fabrication methods with microfluidic structures, making the combination of microfluidics and optics an obvious route toward more sophisticated sensing platforms. The remaining hurdles to broader usage are primarily related to user-friendliness of handling off-chip optical components, and not necessarily sensitivity. As more tools can be moved on-chip and operation simplicity increased, optofluidics will increasingly reach a broader audience. In its current state, biological techniques using optics-integrated devices are becoming appropriate for adoption by a more diverse set of researchers.

One of the greatest advantages of using optics-integrated microfluidics for sensing and detection is the potential to forgo labeling molecules, proteins, or cells of interest to identify them. Labels such as fluorescent proteins or dyes have the potential to interfere with biophysical properties of interest, the assay itself, and require greater time and expense for sample preparation. Many of the techniques reviewed here are label-free, although many still require binding of analytes to functionalized surfaces that must specifically bind only the desired analyte—in a complex biological sample, this could pose a potential issue. However, handling small biological samples and performing necessary preparatory steps is a strength of microfluidic systems that could be exploited to a greater extent to achieve the lab-on-a-chip ideal of removing the chip from the lab. The path toward this goal for optofluidic devices has been obscured in part by the often complicated off-chip readout systems used with optofluidic chips. Examples of stand-alone photonic microfluidic devices exist, but still number among the minority, despite the push for chips that could be used in resource-poor areas (111). Some excellent recent examples of lab-independent microfluidics that use smartphones (despite their relatively poor photosensors) to read out optical signals include a triplex diagnostic chip that combines preparatory steps for whole blood handling with a simple optical density measurement, as well as an implantable microfluidic intraocular pressure monitor (144, 145). With the improvement in ease of fabrication of chips incorporating the more sensitive techniques described here, lab-independent point-of-care chips for a greater variety of biomarkers could be readily developed at reasonable costs.

For online image- and data processing applications, which are useful in active sample separation, the limitation on throughput is set by computational power, which may be the next frontier. For instance, digital holography and similar, more computationally involved methods have seen relatively little usage compared with epifluorescent and confocal imaging. However, as miniaturized computers continue to reduce in price, and smartphones, already capable of image processing, become faster and even more ubiquitous, these computationally involved techniques may see heavier usage. In parallel, using cloud computing services to efficiently process sensor information may also see many future applications.

Perhaps the most significant advantage of optofluidic systems in relation to biophysical measurement is the uniquely large data sets that a well-integrated optofluidic system promises. In addition, even greater label-free sensitivity can be achieved through multimodal chips using integrated optics and electronics along with fluid control; this can provide comprehensive and multidimensional information about the analytes, which can be useful when profiling and attempting to distinguish between analyte species. Quantitative analysis of highly multidimensional data sets will only become more important as the community continues to strive for precision medicine platforms, where big data scenarios are inevitable. Thus far, most optofluidic applications have demonstrated equivalency with assays that can be performed with conventional equipment. However, investigations of biological heterogeneity at various scales using optofluidic tools has the potential for improving the ease of collecting data on protein binding and enzyme kinetics as well as cellular heterogeneity.

Resolution limitations can be partially resolved through computational techniques, but subdiffraction limit microscopy methods such as fluorescence photoactivation localization microscopy and stochastic optical reconstruction microscopy are not yet in routine usage with microfluidics. Although these techniques are extremely useful for applications that require precise localization of proteins and cellular structures, label-free sensors can be an excellent high-sensitivity alternative for purposes of detection and quantification involving less financial commitment. In addition, subpixel resolution methods and other signal-enhancing techniques have been described on-chip that presents another interesting avenue of research.

Looking to the future, the integration of optics with microfluidics has produced countless possibilities for combination and exploitation, particularly for biochemical applications that benefit from parallelization. The creative use of micro- and nanostructures is inherent to both many optical techniques as well as the foundations of microfluidics, making them a suitable pair capable of building upon each other. The further development of techniques for improving the integration of sensitive optical measurements with microfluidics will no doubt yield a breadth of tools that would be unattainable without one or the other.

Editor: Brian Salzberg.