|Home | About | Journals | Submit | Contact Us | Français|
A novel system that has enabled the measurement of single-cell oxygen consumption rates is presented. The experimental apparatus includes a temperature controlled environmental chamber, an array of microwells etched in glass, and a lid actuator used to seal cells in the microwells. Each microwell contains an oxygen sensitive platinum phosphor sensor used to monitor the cellular metabolic rates. Custom automation software controls the digital image data collection for oxygen sensor measurements, which are analyzed using an image-processing program to yield the oxygen concentration within each microwell versus time. Two proof-of-concept experiments produced oxygen consumption rate measurements for A549 human epithelial lung cancer cells of 5.39 and 5.27 fmol/min/cell, closely matching published oxygen consumption rates for bulk A549 populations.
The development of a single-cell metabolic rate monitoring system with fmol/minute resolution is essential to the mission of the Microscale Life Sciences Center (MLSC) at the University of Washington, a National Institutes of Health (NIH) National Human Genome Research Institute (NHGRI) Center of Excellence in Genomic Science (CEGS), www.life-on-a-chip.org . This mission includes the development of a high-throughput multiparameter integrated system to monitor in parallel cellular parameters such as respiration rates, gene expression, protein expression, cytokine expression, membrane integrity, and ion gradients.
The evolving technology, when complete, will permit real-time investigations of cellular heterogeneity using experiments that examine cell damage and death pathways including neoplastic progression (cell proliferation) as seen in cancer or pyroptosis (pro-inflammatory cell death), as seen in heart disease and stroke. The premise driving fmol/min measurement resolution is that cell behavior may not be distributed simply about the mean of a large population; outliers may, in fact, tip the scale to either cell proliferation or cell death.
Advancements in single-cell analysis systems are emerging –; however, real-time high-throughput multiparameter single-cell measurements have remained elusive. Such a system would logically incorporate microfluidics, cell trapping, environmental control, automation software, fluorescence microscopy, and data analysis. Oxygen was selected as the initial parameter of interest due to the inherent difficulty associated with this measurement at the required detection resolutions. The technological development needed to enable these advances lays a strong foundation upon which to incorporate subsequent parameters.
Oxygen sensor systems based on the quenching-dependent emission from phosphor embedded polymers are becoming increasingly popular –. Absorption of visible or UV radiation from a laser, for example, excites the phosphor into a short-lived singlet state whereupon it nonradiatively relaxes to a lower level triplet state. Light is often emitted as the electron relaxes from the triplet state back to the ground state. The ensemble emission of the phosphor can be empirically represented by the following exponential decay function:
where I(t) is the intensity of the phosphorescence as a function of time, I0 is the maximum intensity at time, t = 0, and the lifetime τ is the average time the phosphor stays in the excited state following excitation. The lifetime of the decay of phosphorescence is typically on the order of microseconds to seconds compared with typical fluorescence lifetimes of nanoseconds .
Oxygen sensing is achieved because the triplet state of the phosphor is at a similar energy level as the triplet ground state of molecular oxygen. The presence of oxygen provides a molecular orbital overlap that allows for a nonradiative relaxation pathway for excited electrons within the phosphor molecule. This additional decay path manifests itself as a shortening of the observed lifetime. While other oxygen detection experiments have measured the exponential decay curve directly , it is also possible for the sensor to be excited with sinusoidal intensity modulated light and to measure the phase shift of the emission. Emission from a sinusoidally driven sensor follows with the same frequency, but is delayed in phase and reduced in amplitude relative to the excitation. Measuring the phase shift rather than the exponential decay has the important benefits of improved signal-to-noise ratio, greater sensor output dynamic range, shorter excitation exposure times, and allowing for experimental optimization techniques.
In order to assign a value to the observed decrease in lifetime or phase, it is necessary to model the decay. A first approach might be to model the phosphorescence with only a single-component lifetime or phase as described by a linear Stern–Volmer equation referred to here as Model 1
where τ0 and Φ0 are the unquenched emission lifetime and phase, respectively, while τ and Φ are the same quantities in the presence of different quencher concentrations, KSV is the Stern–Volmer constant, and [q] is the quencher concentration . Since the main quenching agent present in the microwells is oxygen, [q] can be replaced by the aqueous concentration of oxygen [O2].
In many cases, the phosphorescent quantum yield is not modeled well by the linear Model 1 Stern–Volmer relationship. A downward deviation from the linear model is observed, and as a result of the nonlinearity, it is necessary to calibrate following a different model. A commonly used power law quenching model, referred to here as Model 2 is
where K and m are the fitting parameters . Model 2, while purely empirical, improves calibration, but has no physical or chemical significance beyond provision for improved calibration fits.
A third Stern–Volmer equation models two distinct quenching sites in the sensor. Both sites are quenchable, but they have different quantum yields and will be referred to here as Model 3
where f01 and f02 are the fraction of the total emission from each component under unquenched conditions and KSV1 and KSV2 are the associated Stern–Volmer quenching constants for each component .
In order to practically utilize a Stern–Volmer calibration model, it is necessary to experimentally calibrate at known oxygen concentrations. The abundance of oxygen in an aqueous solution is commonly expressed as a molar concentration [O2(aq)] by
where KH is Henry’s constant and PO2 is the partial pressure of oxygen. Equation (5) states that the concentration of oxygen dissolved in water is directly proportional to Henry’s constant multiplied by the partial pressure of oxygen in the gas in contact with the water.
According to Benson and Krause, Jr., who empirically fit experimental data for temperature ranging from 0 °C to 45 °C, Henry’s constant for oxygen in water as a function of temperature is approximately
where KH is Henry’s constant in M/atm and T is temperature in Kelvin . An equation is also presented in  to include the effects of salinity, but for this study the salinity has been considered to be negligible. After a number of sensor lifetime or phase readouts at known temperature and partial pressure are collected and fit to an accurate Stern–Volmer model, any sensor readout can be converted to aqueous concentration of oxygen.
Several 1 × 1 cm chips containing an array of microwells are etched on 3-inch borosilicate glass wafers using standard photolithography and etching techniques (Fig. 1). The microwells are 150 μm in diameter, 50 μm deep, and spaced 200 μm apart center-to-center. A 3 μm high lip encircles the top of each microwell. The sensors at the bottom circumference of each microwell consist of a fused matrix of polystyrene beads embedded with a platinum phosphor . The wafer level deposition allows for efficient creation of numerous wells with sensors in parallel.
Sealing each microwell is essential for metabolic rate experiments. Each well, which contains one or more cells, a sensor, and cell media needs to be diffusionally isolated from the external surroundings to measure oxygen metabolism. The sensor monitors the decreasing aqueous oxygen concentration of the cell media as cells consume oxygen. If the microwell is not sealed, exterior oxygen could refresh the microwell oxygen supply, and the respiration rate of the cell could not be accurately determined. A gold oxygen barrier is pushed onto the seal ridge of the microwell by a piston to make a seal, while the compliance layer compensates for any off-axis tilt between the contacting surfaces (Fig. 2).
The lid actuator is designed to press an oxygen barrier down onto the microwells (Fig. 3). A 24-well tissue culture plate is modified to accommodate two securely attached anchors on either side of a well containing the microwell array and cell media. The spanning bar transfers the force generated by the wing nuts through the piston down to the oxygen barrier. The piston has a flat polished tip to which the compliance layer and oxygen barrier are attached.
An environmental chamber is needed to maintain a constant and controllable temperature for the cells under investigation, which is of critical importance when following biological protocols. Temperatures adequate for cell assays range approximately from 35 °C to 37 °C. A commercially available device (PeCon GmbH, Erbach, Germany; TempContol 37-2 digital, heating unit, humidifier system, and microscope stage chamber) that mounts onto the stage of an inverted microscope (Zeiss, Thornwood, NY; Axiovert 200M) accommodates the placement of the lid actuator apparatus inside. The environmental chamber and temperature control system help maintain the required thermal environment for biological experiments.
A frequency-domain technique for experimentally determining the phosphorescence phase has been developed called the on–off frequency-domain (OOFD) method , which is a modification of a method described by Shonat et al. . In the OOFD method, a series of images of the sensors is acquired. The sensor is pulsed on and off with laser light at a fixed pulse width, W and the timing of each camera exposure is delayed by ΦD with respect to the laser (Fig. 4). At an initial phase delay of zero, a Q × Q pixel digital image of the sensors is acquired representing the first of F frames in a single kinetic series. Subsequent phase-delayed images are similarly obtained and concatenated to the end of the growing kinetic series until the Fth frame is obtained. Each frame contains images of the same N sensors that differ in phase-dependant intensity. Each of the K kinetic series yields one oxygen concentration point in time for each sensor. Image processing follows (described in the next section), which extracts the individual sensor areas for each kinetic series and plots the integrated intensities of each sensor as a function of phase. A fourth-order polynomial is fit to the data, whereupon the phase Φ is determined by taking the derivative of the fit polynomial and finding the root that corresponds to the phase. This phase, of course, changes with different oxygen quenching levels.
The data collection hardware enabling the oxygen sensors is shown in Fig. 5. The function generator (Tektronix, Richardson, TX; AFG 310 Arbitrary Function Generator) is set to produce a square wave (5 Hz, 50% duty cycle). The delay generator (Berkeley Nucleonics Corporation, San Rafael, CA; Model 555 Pulse/Delay Generator) produces two in-phase square pulses (50 μs pulse width) when triggered by the function generator. One pulse is sent to the laser (Power Technology Inc., Little Rock, AR; Diode Laser PPMT LD1539 405 nm), while the other pulse is sent to the camera (Andor, South Windsor, CT; iStar DH734-18F-73 Imaging ICCD). The pulses are continually sent and serve to synchronize the laser and the camera.
The personal computer (Dell, Round Rock, TX; Precision 370-P4 2.80 GHz, 1.00 GB RAM) contains the hard drive and the Andor camera Peripheral Component Interconnect (PCI). The camera controller PCI connects to the camera via a shielded bundle cord. There is also a 15 pin D-Sub auxiliary port on the PCI used to trigger the external shutter controller (Uniblitz, Rochester, NY; Shutter Driver/Timer Model VMM-T1) via the automation program. The normally-closed external shutter opens with Transistor-Transistor Logic (TTL) high. The microwell array sitting in the lid actuator device is placed on the inverted microscope stage (Zeiss, Thornwood, NY; Axiovert 200M).
The laser light used to excite the platinum phosphor sensors passes through a 5× beam expander and an optical diffuser used to reduce the optical density and spread out the Gaussian profile. It reflects off a dichroic and illuminates the sensors after traveling through a 10× objective (Zeiss, Thornwood, NY; A-Plan 10X/0.25 Ph1Var1 1020-863). No excitation filter is necessary since the laser is fixed at the desired 405 nm excitation wavelength. The phosphorescence emission of the sensors travels back through the objective, passes through the dichroic, which blocks any laser light, through a 585 nm long pass filter (Omega, Brattleboro, VT; XF3090 585 ALP) and enters the camera. While it is true that it is possible to damage cells with near-UV laser light, 405 nm or lower excitation wavelengths for live-cell probes and sensors are commonplace –, and there are many commercially available cellular stains that are excited in the UV to near-UV range. By reducing exposure times and using diffuse excitation light, possible side effects to cell heath are minimized.
The automation of the sensor data acquisition is orchestrated by software written in a procedural programming language supplied with the camera called AndorBasic. Referring to Fig. 6, the automation steps are explained. First, the relevant acquisition parameters outlined in Fig. 4 are defined (Step 1) including: intensifier pulse width W, phase delay between subsequent frames ΦD, number of frames in kinetic series F, and number of kinetic series in the kinetic series batch K.
Next, several camera settings are defined (Step 2). The trigger mode is set to external. The data type is set to “background corrected” to subtract a background image including any dark noise or fixed pattern noise that is present from each image in the kinetic series. The number of accumulations is set and specifies how many multiple scans of each frame in the kinetic series are added together in computer memory. The gain is set to define how much the emission signal is amplified within the camera. The binning is set to specify how many neighboring pixels on the CCD should be combined to make the resulting Q × Q image.
The program then enters a loop that collects K kinetic series. Since the frames of the kinetic series are background corrected, a background image must be collected (Step 3) before the frames of the kinetic series are acquired. The external shutter is then opened (Step 4) to let the pulsing laser light through to the sensors, and the kinetic series is acquired (Step 5). The external shutter is immediately closed (Step 6) to prevent the laser light from reaching the sensors and cells. The kinetic series is then converted to ASCII format and saved on the hard drive with a number appended to the end of the file name that matches the loop index (Step 7). All kinetic series needed for the experiment are collected in this manner by going through the loop K times.
A human epithelial lung cancer A549 cell line was obtained from the American Type Culture Collection (Manassas, VA). Cells were cultured as attached cells in Dulbecco’s Modified Eagle’s Media (DMEM) with 4.5 g/L, 2 mM glutamine, and Penicillin-Streptomycin (Cambrex Bioproducts, Walkersville, MD; catalog # 12-614, 17-602, 17-602). The media was supplemented with 10% Fetal Clone III (Hyclone, Logan, UT; catalog # SH30109.03) in 5% CO2 at 37 °C. Cells were harvested using 0.25% trypsin in versene (Cambrex Bioproducts, Walkersville, MD; catalog # 17-160, 17-711).
Microwell array chips were placed in individual wells of a 24-well tissue culture plate and incubated at room temperature in 70% ethanol for 10 min. Ethanol was decanted and each array was rinsed with sterile Dulbecco’s phosphate buffered saline (Invitrogen, Grand Island, NY; catalog # 14200-075), diluted to 1×, adjusted to pH 7, and rinsed a second time in the modified growth media that was used for cell culture. 0.5 mL of the same modified growth media was added to each well of the 24-well tissue culture plate containing a microwell array chip. The microwells were seeded by adding 100 μL of cell suspension (approximately 500–10 000 cells) to each well containing a chip. The cells were incubated 16–20 h at 37 °C allowing time for them to settle and attach to the microwell array surface and the bottom of the microwells depending on where they settled.
Before the data collection, the cells were stained with a cell permeable DNA stain, Hoechst 33342, (Molecular Probes, Inc, Carlsbad, CA; catalog # H1399) by adding 6.0 μL to the well from a 1 mM stock solution in distilled water. Adding the Hoechst dye before the experiment allowed enough time for the dye to permeate the cells for imaging at the end of the experiment. After imaging the cells with the Hoechst dye, 3 μL from a 1 mg/mL stock solution of propidium iodide (PI) (Sigma-Aldrich, St Louis, MO; catalog # P4170) was added to the well. Images of the PI stained cells were acquired after 15 min.
The image processing program [written in MATLAB and implemented in MATLAB Version 126.96.36.19920 (R14)] processes the multiple kinetic series collected during an experiment and produces a plot of oxygen concentration versus time for each sensor. It consists of six major components: A) Locate Sensors, which determines the sensor locations in a kinetic series; B) Integrate Intensity, which integrates the pixel intensity values of each sensor in each frame of the kinetic series; C) Estimate Phase, which determines the phase for each sensor by fitting a curve to the integrated intensity data and locating the peak; D) Process All Kinetic Series, which processes every kinetic series collected during an experiment and stores the phase data in a matrix; E) Convert Phase, which converts all phase values to aqueous oxygen concentration with a calibration curve; and F) Plot, which produces the final plot of oxygen concentration versus time for each sensor. Each of these components are now discussed in detail.
First, the ring-shaped sensors imaged in a kinetic series are located. Let represent a kinetic series, where f = 1, … , F is the frame index and Q is the number of pixel rows and columns in each frame. Each frame is a phase dependent intensity map of the sensors, which remain in the same spatial position. Let ζQ×Q be a binary map containing a ring located in the image center
where X = (x, y) are the pixel locations within the image domain and α and β are inner and outer radius of the ring in pixels, respectively. The N sensors are located by convolving the first frame of a kinetic series with sensor filter ζQ×Q and searching for the N peaks in the resulting image. Let ΓQ×Q be the resulting convolution image
Let be a vector of length N with components l1, l2, … , lN containing the sensor location coordinates
where n = 1, … , N is the sensor index and Xn is the pixel location in ΓQ×Q corresponding to the nth local maxima. Once the sensor coordinates are determined, a filter set is made, which contains n binary images of a ring located at each sensor position. Let be a filter set, where n = 1, … , N is the frame index. Each frame is a binary map of a ring centered at the nth location in
where X = (x, y) are the pixel locations within the image domain and α and β are inner and outer radius of the ring in pixels, respectively.
Next, the integrated intensity for each sensor in each frame of the kinetic series is determined. Let IN×F be an N row by F column matrix containing the integrated intensity for each sensor in each frame of the kinetic series. Each location in IN×F equals the integration of the Hadamard product of the nth frame of the filter set with the fth frame of the kinetic series
where n = 1, … , N is the sensor index, f = 1, … , F is the kinetic series frame index, and Ω is the image domain.
Then, the phase for each sensor in the kinetic series is estimated. Let be a phase delay vector of length F with components d1, d2, … , dF
where f = 1, … , F is the kinetic series frame index and ΦD is the phase delay between consecutive frames in a kinetic series. The program performs a fourth-order polynomial least-squares error fit of the integrated intensity in the nth row in IN×F to the phase delay vector . That is, it finds parameters α1, α2, α3, α4, and α5 of the equation
that minimizes the total square error between the data in each row of IN×F and , where n = 1, … , N is the sensor index and f = 1, … , F is the kinetic series frame index. A continuous fourth-order polynomial function denoted gn(Φ) with parameters α1, α2, α3, α4, and α5 is created for each sensor, where n = 1, … , N is the sensor index. Let be a vector of length N with components v1, v2, … , vN containing the estimated phase Φ for each sensor. Each phase value is estimated by taking the derivative of g and finding the appropriate root in g’, which is between 0° and 90°
where n = 1, … , N is the sensor index.
The program now repeatedly processes and stores each K kinetic series collected during an experiment. Let JK×N be a K row by N column matrix containing the estimated phase for each sensor in all kinetic series
where k = 1, … , K is the kinetic series index and n = 1, … , N is the sensor index.
Each phase value in JK×N must be converted to an aqueous oxygen concentration. In this example the conversion is accomplished after fitting calibration data to a three-point Model 3 Stern–Volmer calibration curve (4).
Let be a vector of length 3 with components P1, P2, and P3 containing three partial pressure calibration points 0%, 10%, and 20%, respectively. Let be a vector of length 3 with components c1, c2, and c3 containing three aqueous oxygen concentration values (ppm) corresponding to the three partial pressure calibration points in calculated from (5) and (6) combined
where a = 1, … , 3, k is the molecular weight of O2 (mg/mol), ψ is the atmospheric pressure (atm), and η is the media temperature (K). Let be a vector of length 3 with components q1, q2, and q3 containing three average phase calibration points experimentally determined at each partial pressure in . Let be a vector of length 3 with components n1, n2 and n3 containing , the y axis of the Stern–Volmer plot
where a = 1, … , 3. Next, the program performs a model 3 Stern–Volmer (4) least-squares error fit of the calibration phase data vector to calibration concentration vector . That is, it finds parameters f01, KSV1, f02, KSV2 of the equation
that minimizes the total square error between data vectors and fit to (4). All phase values in matrix JK×N are then converted to concentration values and stored in matrix OK×N by solving the following equation for O:
where k = 1, … , K is the kinetic series index and n = 1, … , N is the sensor index.
Finally, a plot of oxygen concentration versus time for each sensor is made. Let be a time axis vector of length K with components t1, t2, … , tK
where k = 1, … , K is the kinetic series index and T is the time between consecutive kinetic series acquisitions during an experiment. Plotting each column in OK×N versus yields a graph showing the aqueous concentration of oxygen in each microwell as a function of time.
The lifetime and phase for partial pressure O2 levels of 0%, 5%, 10%, 15%, and 20% bubbled into 3 mL of water in contact with the microwell array were experimentally determined using the OOFD method, as well as a PMT and MLE method. The PMT and MLE methods have been included in this study to aid in validating our new methodologies and are generally described in  and , respectively. All data were taken from a single sensor from the microwell array at 31.7 ± 0.3 °C and 1.0 atm. Each calibration point is the mean value of 30 consecutive lifetime or phase estimates. The three Stern–Volmer models (2–4) were then fit to the five calibration points for each of the three acquisition methods.
After a visual comparison between the three different Stern–Volmer models fit to the calibration data, Model 3 appeared to be the best fitting for all three acquisition methods. This qualitative finding is in agreement with a more rigorous calibration study on transition-metal complex oxygen sensors . In addition, a three-point calibration (0%, 10%, and 20% PO2) gives an almost identical fit as the full five-point calibration for all three acquisition methods and suffices for an accurate calibration (Fig. 7). Further reference of Model 3 calibration implies a three-point calibration.
Several suspected reasons for the observed downward curvature of the data, as shown in Fig. 7, have been reported , , –: multiple lifetimes of the phosphor, heterogeneity of the sensor matrix (site heterogeneity), systematic errors in quantitative emission intensity using gated emission and excitation, and/or oxygen permeability of the sensor polymer. The two quenching sites modeled by the Model 3 Stern–Volmer equation are most likely the phosphor on the surface of the beads and the phosphor in the bulk of the beads . The sites on the surface are more readily accessible to the free oxygen compared with the sites buried in and bonded to the polystyrene.
The results presented in this section provide fundamental validation for the single-cell oxygen consumption system by high-lighting two proof-of-concept biological experiments. Several microwell arrays were randomly seeded with a dilute solution of human epithelial lung cancer cells as described above. Two were chosen that appeared to have a minimum number of cells (0–6) trapped in each microwell.
Each experiment began with adding Hoechst 33342 dye to the well containing the microwell array chip. The lid with the gold oxygen barrier was then pressed onto the top of the microwells with a force estimated to be in the range of five to ten pounds. Shortly thereafter, the data collection process was initiated. The data were processed by the image processing program to produce a plot of the oxygen concentration, as a function of time for each microwell. The cells stained with the Hoechst dye were then imaged 30 min after the dye was initially added. PI stain was added and the cells were again imaged 15 min later.
Hoechst 33342 freely enters living or dead cells and binds to the DNA whereupon it fluoresces . A cell undergoing apoptosis shows signs of DNA fragmentation, which can be observed in images of the Hoechst stained cells ,  (Fig. 8). An image of propidium iodide (PI) stained cells was also collected (not shown). PI readily permeates through compromised cellular membranes of nonviable cells and binds to the DNA. The cell membrane of cells undergoing apoptosis remains intact and functions normally .
When using PI and Hoechst together, it is possible to classify cells as either necrotic (accidental cell death caused by injury, inflammation, etc.), apoptotic (programmed cell death), or live. Necrosis in a cell can be identified by no nuclear fragmentation in combination with damage to the cell membrane. Apoptosis, on the other hand, can be identified by the presence of nuclear fragmentation in combination with cell membrane integrity. A cell exhibiting both nuclear fragmentation and membrane damage is classified as late apoptotic . If an image of a cell stained with both Hoechst and PI does not indicate apoptosis or necrosis, the cell is fully intact and is most likely healthy and living .
The microwell array used in the first experiment appears to contain a fragmented nucleus (fuzzy image of cellular DNA) in microwell #5, a cluster of three intact nuclei (sharp image of cellular DNA) and two fragmented nuclei in well #6, and one intact nucleus and one fragmented nucleus in microwell #7. No PI penetrated the cell membranes indicating a mixture of apoptotic and live cells, as identified in Fig. 8.
Fig. 9 shows the oxygen consumption results of the first experiment. The oxygen concentration of the media in microwell #6 decreased in oxygen concentration from 6.2 to 0.0 ppm in approximately 9 min resulting in a consumption rate of 0.69 ppm/min for three live cells or an average of 0.23 ppm/min/cell. Each microwell is 50 μm deep and 150 μm in diameter resulting in a volume of 0.88 nL assuming a perfect cylinder. Optical profile images have shown that the bottom edges of the microwells are not square but instead curve due to the etching process reducing the volume by approximately 15% to 0.75 nL. Using this volume and the molecular weight of oxygen of 32 g/mol to convert from parts per million, the resulting average consumption rate for the three live cells in microwell #6 is 5.39 fmol/min/cell.
The remaining microwells showed a minimal decrease in oxygen concentration over time. The slight negative slope probably indicates a small leak in all microwell seals. One might expect to see well #7 have a decrease in oxygen, but even though that particular cell seems to be living as indicated by the dyes, it is possible that it has entered an early stage of apoptosis before DNA fragmentation started but after oxygen consumption has ceased. The high variance of sensor #9 is attributed to low emission intensity (low signal-to-noise ratio) due to poor sensor deposition in that particular well.
The microwell array used for the second experiment appeared to contain one intact nucleus in microwell #1, one intact nucleus in microwell #7, and zero cells in the remaining microwells. Again, no PI penetrated the cell membranes indicating that both cells were most likely live cells. The oxygen concentration of the media in microwell #7 decreased from 7.5 to 3.0 ppm over 20 min resulting in a consumption rate of 0.225 ppm/min/cell or 5.27 fmol/min/cell. Microwell #1 reported no significant decrease in oxygen besides the negligible negative slope comparable to the empty wells.
The two proof-of-concept experimental results are now compared with published bulk oxygen consumption rates of the same A549 human epithelial lung cancer cells. Ahmad et al.  reports a bulk consumption rate of 32 nmol/min/10 million cells, which gives an average respiration rate of 3.2 fmol/min/cell. Gardner et al.  reports a bulk consumption rate of 48–57 nmol/min/10 million cells giving an average respiration rate of 4.8–5.7 fmol/min/cell.
Table I summarizes the experimental results from Ahmad, Gardner, and the MLSC. The closely matching estimation of the oxygen consumption rate of a single A549 cell compared with published rates gives us confidence that we have reached an important developmental milestone on our path towards robust single-cell investigations.
The proof-of-concept experiments presented in this paper highlight a successful architecture for measurement of oxygen metabolism at rates in the fmol/min range. While the automation, image processing, and calibration components of this system are robust, some mechanical issues need to be improved upon to enable the technical reproducibility necessary for vigorous biological experiments.
The three main mechanical bottlenecks, which are close to being solved, are the flexibility of the plastic well that the chip rests on, the lid actuator device, and the material used for the oxygen barrier. The first experiment was forced to end after 14 min because of a broken chip, as seen in Fig. 8(a). The flexibility of the bottom of the wells in the plastic 24-well plate permits the chip to bend and break. A new single-well plate is being developed with a rigid quartz window for the chip to rest on to prevent this from happening.
A new lid actuator is being designed to press the oxygen barrier down onto the microwell array with more control and perpendicularity. A load cell placed in the force axis will allow for real-time measurements of the force being applied for accurate force control and repeatability. In addition, several other oxygen barrier materials are currently being tested to identify a more reliable lid needed to further increase experiment efficiency.
The work of D. Meldrum and M. Lidstrom was supported in part by the National Institutes of Health NHGRI Centers of Excellence in Genomic Science under Grant 5 P50 HG002360.
Timothy W. Molter was born in Port Washington, WI, in 1979. He received the B.S. degree in physics from the University of Wisconsin, La Crosse, in 2001, the B.S.E.E. degree from the University of Minnesota, Minneapolis, in 2004, and the M.S. degree in electrical engineering from the University of Washington, Seattle, in 2006.
During graduate school, he worked as a Teaching Assistant in the Department of Electrical Engineering and as a Research Assistant at the Microscale Life Sciences Center (MLSC), University of Washington. He is currently a Research Engineer at the MLSC specializing in image processing, development of new automated research platforms, and optimization of a variety of projects.
Mark R. Holl (M’01) was born in Champaign, IL. He received the B.S. degree in mechanical engineering from Washington State University, Pullman, in 1986, the M.S. degree in mechanical engineering from the University of Washington, Seattle, in 1990, and the Ph.D. degree in mechanical engineering from the University of Washington in 1995. Graduate studies were performed with Prof. Garbini and Proc. Kumar at the University of Washington in the field of systems modeling and manufacturing automation.
During his undergraduate student years he worked as an Instrument Maker in the engineering machine shops at Washington State University. Prior to graduate research, he worked for two years at Boeing Commercial Aircraft in Manufacturing, Research, and Development in factory support and integrated automated system functional test development. Upon completion of graduate studies he engaged a five-year relationship with Prof. Yager in Bioengineering to assist in the development and commercialization of micro total analysis system technologies. He was a principal inventor with Yager of a laminate-based microfluidic technology and a founding member of a startup company born in part of this work, Micronics, Inc. He guided initial integrated system and microfluidic cassette development for Micronics thru 1998, and then returned to academic research and development work with Yager from 1998 to 1999. In 2000, he joined in a six-year collaborative effort with Prof. D. Meldrum to develop microfluidic technologies for biomedical and genome science applications. During this time, he maintained research development engineering, laboratory management, and Research Assistant Professor responsibilities in support of laboratory objectives. In 2007, he moved with the laboratory of Prof. Meldrum to contribute to the launch effort of a new Center for Ecogenomics in the Biodesign Institute, Arizona State University, Tempe campus. His research interests include microfabrication technologies, systems integration for total analysis systems, integration for micro and nanotechnology components in robust and simple to use formats, microscale systems for biological applications, bioprocess automation, process sensors, and process characterization and control with emphasis on biomedical, genomic, and proteomic science applications.
Dr. Holl is a member of American Association for the Advancement of Science (AAAS), ASME, The International Society for Optical Engineers (SPIE), and an Investigator with the NIH Center of Excellence in Genomic Sciences, and the Microscale Life Sciences Center (MLSC), Arizona State University.
Joseph M. Dragavon was born in Monroe, WA, in 1979. He received the B.S. degree in chemistry from the University of Puget Sound, Tacoma, WA, in 2001 and the Ph.D. degree in chemistry from the University of Washington, Seattle, in 2006, for his dissertation titled “Development of a cellular isolation system for real-time single cell oxygen consumption monitoring.”
He is currently a Postdoctoral Research Assistant at the University of Hull, Kingston upon Hull, U.K.
Sarah C. McQuaide was born in Ventura, CA, in 1976. She received the B.S. degree in mechanical engineering from California Polytechnic State University, San Luis Obispo, in 1999 and the M.S. degree in mechanical engineering from the University of Washington, Seattle, in 2002, where she worked on deformable micromirrors for 3-D retinal scanning display systems in the Human Interface Technology (HIT) Laboratory.
In 2001, she received a grant from Boeing and Lufthansa to study MEMS fabrication at the Technische Universitaet,Berlin, Germany. Currently, she is a Research Engineer at the Microscale Life Sciences Center working on the design and fabrication of systems for single-cell analysis.
Judith B. Anderson was born in Frankfort, KY. She received the B.S. degree in social sciences and microbiology from Michigan State University, East Lansing, in 1974.
Currently, she is a Research Scientist with the Department of Pathology, University of Washington, directing the cell culture facilities and studying diseases related to aging.
A. Cody Young received the B.A. degree in physics from Colorado College, Colorado Springs, and the M.S. degree in physics from the University of Washington, Seattle, in 1999. Currently, he is working towards the Ph.D. degree at the Department of Physics, University of Washington. His dissertation is on the development and characterization of ultrafast laser generated optical devices
Before returning to graduate school in 2002, he was employed at Los Alamos National Laboratory on a variety of astrophysical and computational problems. His research assistantships have included studying starquakes at Los Alamos National Laboratory, cellular automata at the Santa Fe Institute, and inorganic superlattices at Arizona State University.
Lloyd W. Burgess was born in Syracuse, NY. He received the B.S. and M.S. degrees in chemistry from Syracuse University, Syracuse, and the Ph.D. degree in analytical chemistry from the Virginia Polytechnic Institute, Blacksburg, in 1985.
Currently, he is a Research Professor with the Chemistry Department, University of Washington. He directs a sensors and instrumentation group investigating new concepts in the development of integrated multicomponent analysis systems for real-time industrial, biomedical, and environmental monitoring applications.
Mary E. Lidstrom was born in Oregon. She received the B.S. degree in microbiology from Oregon State University, Corvallis, and the M.S. and Ph.D. degree in bacteriology from the University of Wisconsin, Madison. She conducted postdoctoral work as a Leverhulme Fellow in Microbiology at the University of Sheffield, and has held academic appointments in the Center for Great Lakes Studies in Milwaukee, WI, and in Environmental Engineering Science at the California Institute of Technology. Currently, she is a Professor of Chemical Engineering and Microbiology and holds the Frank Jungers Chair of Engineering at the University of Washington, where she is the Vice Provost of Research.
Dr. Lidstrom is a member of the American Society for Microbiology and is a Fellow of the American Academy of Microbiology and the American Association for the Advancement of Science.
Deirdre R. Meldrum (M’93–SM’00–F’04) received the B.S. degree in civil engineering from the University of Washington, Seattle, in 1983, the M.S. degree in electrical engineering from the Rensselaer Polytechnic Institute, Troy, NY, in 1985, and the Ph.D. degree in electrical engineering from Stanford University, Stanford, CA, in 1993.
As an Engineering Co-op student at the NASA Johnson Space Center in 1980 and 1981, she was an Instructor for the astronauts on the Shuttle Mission Simulator. From 1985 to 1987, she was a member of the Technical Staff at the Jet Propulsion Laboratory working on the Galileo spacecraft, large flexible space structures, and robotics. From 1992 to 2006, she was a Professor of Electrical Engineering and Director of the Genomation Laboratory, University of Washington. Currently, she is Dean of the Ira A. Fulton School of Engineering, Professor of Electrical Engineering, and Director of the Center for EcoGenomics, Biodesign Institute, Arizona State University. Her research interests include genome automation, microscale systems for biological applications, ecogenomics, robotics, and control systems.
Dr. Meldrum is a member of American Association for the Advancement of Science (AAAS), ACS, AWIS, HUGO, Sigma Xi, and SWE. Her honors include an NIH Special Emphasis Research Career Award (SERCA) in 1993, a Presidential Early Career Award for Scientists and Engineers in 1996 for advancing DNA sequencing technology, Fellow of AAAS in 2003, Director of an NIH Center of Excellence in Genomic Sciences called the Microscale Life Sciences Center from 2001 to 2011, and Senior Editor for the IEEE Transactions on Automation Science and Engineering (2003–present).
Note to Practitioners—In the field of biology, the majority of cellular response studies have been performed on large populations of cells, providing only the average cellular response to a stimulus. Recent advances have revealed cell-to-cell heterogeneity among previously presumed homogeneous populations and have suggested that this heterogeneity is an important factor for understanding many diseases such as cancer, heart disease, and stroke. It may be necessary to monitor the response of individual cells within a population to a given stimulus. Identifying and tracking the progression of detrimental variants that have slipped by the safety mechanisms of the body is key to developing the next generation of patient treatments. These outliers can be identified by several measurable parameters including oxygen consumption rates, which is the focus here.