Fluorescence lifetime measurements are becoming increasingly important in chemical and biological research. Time-domain lifetime measurements offer fluorescence multiplexing and improved handling of interferers compared with the frequency-domain technique. In this paper, an all solid-state, filterless, and highly portable light-emitting-diode based time-domain fluorimeter (LED TDF) is reported for the measurement of nanosecond fluorescence lifetimes. LED based excitation provides more wavelengths options compared to laser diode based excitation, but the excitation is less effective due to the uncollimated beam, less optical power, and longer latency in state transition. Pulse triggering and pre-bias techniques were implemented in our LED TDF to improve the peak optical power to over 100 mW. The proposed pulsing circuit achieved an excitation light fall time of less than 2 ns. Electrical resetting technique realized a time-gated photo-detector to remove the interference of the excitation light with fluorescence. These techniques allow the LED fluorimeter to accurately measure the fluorescence lifetime of fluorescein down to concentration of 0.5 μM. In addition, all filters required in traditional instruments are eliminated for the non-attenuated excitation/emission light power. These achievements make the reported device attractive to biochemical laboratories seeking for highly portable lifetime detection devices for developing sensors based on fluorescence lifetime changes. The device was initially validated by measuring the lifetimes of three commercial fluorophores and comparing them with reported lifetime data. It was subsequently used to characterize a ZnSe quantum dot based DNA sensor.

Fluorescence sensing is widely used in both chemistry and biology because it offers exceptional sensitivity, selectivity, and multiplexing capabilities. In high-performance liquid chromatography, fluorescence detectors generally offer an improvement of three order of magnitude in sensitivity compared to absorbance-based detectors.1 Flow cytometers also benefit from the increased sensitivity, but they also gain the added benefit of multiplicity, the ability to perform multiple measurements on the same sample. In one example, 14 independent measurement channels were combined to perform 100 assays in parallel.2 

Fluorescence sensing is most widely used in fluorescence microscopy. In this application, fluorophores are divided into endogenous fluorophores that preferentially accumulate in different cellular components such as the mitochondria.3 Other organic fluorophores such as ethidium bromide can be used to label nucleic acids.4 With the addition of quantum dots and fluorescent proteins, an entire “fluorescence toolbox” was created for simultaneously labeling a wide variety of cellular components.5 

Fluorescence microscopy is a powerful qualitative tool for determining where targets are located, but variations in probe delivery limit the use of the technique to quantify target concentration.6 In thicker samples, variations in sample absorption of both the excitation and emission light further confound quantification. Ratiometric fluorescence imaging solves these problems by taking the ratio of two fluorescence parameters. Since probe delivery and tissue absorption affect both parameters equally, the ratio removes these effects.7 Spectra ratiometric fluorescence sensing is performed using special fluorophores that shift their fluorescence peak in response to their environment. The fluorescence intensity is then measured at two different wavelengths.8 In order to detect targets for which these special fluorophores do not exist, dual-fluorophore probes have been developed that couple a constant-intensity fluorophore to a target-sensitive fluorophore for spectra ratiometric sensing.6 

Fluorescence lifetime sensing is an alternative method of ratiometric sensing. In this case, the sample is excited with a pulse of light. After the excitation pulse has ended, the light emitted from the sample quickly decays. The ratio of this decay is called the fluorescence lifetime. It can be measured by fitting a curve to the time-domain plot of fluorescence intensity, or by taking the ratio of the fluorescence integrated over two time-windows.9 These measurements offer all of the advantages of spectra ratiometric measurement as well as guaranteed matching of sample absorbance between the two signals since the same wavelengths are used for both time-points. As a result, fluorescence lifetime measurements have been widely used for vivo imaging10 and environmental sensing.11,12

The challenge of fluorescence lifetime sensing is that while a few fluorophores have fluorescence lifetime as long as several microseconds,13 most fluorophores have lifetimes on the order of nanoseconds. For example, green fluorescent protein has a fluorescence lifetime of 2.7 ns.14 Quantum dot fluorescent sensors have been built with a range of 11 ns to 19 ns for measuring distance to a gold nanoparticle quencher15 or 7 ns to 14 ns when measuring temperature.16 

Measuring fluorescence on a nanosecond time scale requires special instruments that can both modulate the excitation source and detect the emission fast enough. Instruments capable of measuring the fastest lifetimes for a given source and detector operate in the frequency domain. These systems measure the phase difference between the excitation signal and the detected signal to calculate the fluorescence lifetime.17 These systems are generally limited, however, to measuring samples with a single exponential decay. Time-Correlated Single Photon Counting (TCSPC) systems are better able to measure mixtures of fluorophores because they measure the whole exponential decay.18,19 Both frequency domain and TCSPC fluorescence lifetime instruments measure one point at a time, but they can be expanded to capture a whole image at a time using a time-gated optical intensifier as a sub-nanosecond shutter for a camera.20 Despite their differences, all three techniques are expensive, bulky, and fragile.

The development of nanosecond light sources based on light emitting diodes (LEDs) has led to the creation of a variety of portable fluorescence lifetime instruments. Early examples of LED based fluorescence lifetime instruments continued to use research instruments for the digitization and optical filters using the phase modulation method21,22 or TCSPC.23 More recently, compact systems have been built. Two self-contained fluorescence lifetime fluorimeter including a microprocessor were reported in 2008 and 2009. In 2008, our group reported a time-domain fluorimeter capable of distinguishing 1 ns, 2 ns, and 4 ns lifetime of fluorophores using a red laser diode for excitation.24 Excitation between 350 nm and 450 nm is necessary to excite some of the most commonly used fluorophores such as blue emitting fluorescent proteins25 and blue emitting quantum dots.26 The next year, a more sensitive system using a blue LED and the phase modulation technique to distinguish between fluorophores with lifetimes of 150 μs and 1.5 ns was reported.27 More recently, the phase modulation technique has been extended to accurately measure 4 ns fluorescence lifetimes with a portable fluorimeter.28 

This paper presents an extension of our earlier work on time-domain fluorimeter to use light emitting diodes to increase the options for excitation wavelengths. Unlike other LED based fluorimeters, our system uses no optical filters for either the excitation or the emission, so changing the excitation frequency is simply a matter of replacing the LED. In addition traditionally, time-domain methods for measuring fluorescence lifetime offer improved handing of interferers and multi-exponential decay but decreased sensitivity when compared to phase-modulation methods.29 Experimental result shows that LED TDF can detect fluorescein down to concentration of 0.5 μM.

The LED TDF reported in this paper operates with the same principle as the laser TDF we previously presented.24 As shown in Figure 1(a), the sample is repeatedly excited by a pulse of light, generating fluorescence decay after each excitation. A time-gated reset pulse creates time-interleaved windows to integrate the fluorescence decay in different sections. The reset pulse is produced with an incremental delay, Δt (normally 250 ps), so that the average fluorescence in each time bin can be calculated by subtracting the adjacent integration results. An avalanche photodiode (APD) is used to convert fluorescence into electrical signal during integration. Rapid lifetime determination9 (RLD) can be performed with 3 measurements, or the full decay curve can be constructed when more measurements are available. In this work 256 measurements were performed and the full decay data was used to extract the fluorescence lifetime by a least square fit.

FIG. 1.

This figure presents the timing of the fluorescence decay integration and system organization. (a) Two integration windows are interleaved by the time-gated reset pulse with a delay of 250 ps. In real application there are 256 different versions of integration windows produced by setting the reset pulse with different delay values. (b) The entire system consists of two printed circuit boards (PCBs), a testing sample (held by a 1.5 ml cuvette), and a personal computer (PC). The sample is secured on the PCBs by a sample holder. The PC and the PCBs are communicated through the Universal Serial Bus (USB).

FIG. 1.

This figure presents the timing of the fluorescence decay integration and system organization. (a) Two integration windows are interleaved by the time-gated reset pulse with a delay of 250 ps. In real application there are 256 different versions of integration windows produced by setting the reset pulse with different delay values. (b) The entire system consists of two printed circuit boards (PCBs), a testing sample (held by a 1.5 ml cuvette), and a personal computer (PC). The sample is secured on the PCBs by a sample holder. The PC and the PCBs are communicated through the Universal Serial Bus (USB).

Close modal

The system organization is depicted in Figure 1(b). LED TDF communicates with the personal computer (PC) through the Universal Serial Bus (USB). The entire system is divided into 4 blocks, including the pulse generation circuit, the sample excitation circuit, the fluorescence sensing circuit, and the Microcontroller Unit (MCU). Each block is described in detail in Secs. III B–III F.

Both the excitation pulse and the reset pulse originate from the three-stage ring oscillator circuit shown in Figure 2. The frequency is controlled by R1 (74 kΩ) and Cvar1 (Murata TZC3P300A110R00) tunable between 6.5 pF and 30 pF. They form a variable time constant, τRC, that ranges from 0.48 μs to 2.22 μs. The waveforms of V1, V2, and V3 are presented in Figure 3, corresponding to the case Cvar1 equals 30 pF. This capacitor creates a fast signal path from V1 to V3. When V1 toggles, V3 presents a step of ±ΔV. The amplitude of the step is set by the capacitor divider formed by Cvar1 and Cin, which is the input capacitor of the inverter U3 (NXP Semiconductor HEF4049BT, 652). After the fast signal path takes effect, the slow signal path constructed by the inverter U2 and R1 either charges or discharges V3 from its initial value to half the power supply, 2.5 V, the point where the inverter U3 toggles its state. The delay created from the second inverter stage, Td, is provided to both clock edges to determine the frequency of the oscillator (by assuming the delay in other stages can be ignored).

FIG. 2.

In the schematic of the pulsing circuit, the initial clock signal, CLK, is generated by a three-stage ring oscillator circuit with a tunable capacitor, Cvar1. CLK is then delayed, buffered, and applied to the tunable high-pass filters in generating excitation pulse and reset pulse. Two extra buffers between the output of the ring-oscillator and the delay line input are not depicted for simplicity.

FIG. 2.

In the schematic of the pulsing circuit, the initial clock signal, CLK, is generated by a three-stage ring oscillator circuit with a tunable capacitor, Cvar1. CLK is then delayed, buffered, and applied to the tunable high-pass filters in generating excitation pulse and reset pulse. Two extra buffers between the output of the ring-oscillator and the delay line input are not depicted for simplicity.

Close modal
FIG. 3.

This figure shows the waveforms of V1, V2, and V3 in ring oscillator circuit when Cvar1 is 30 pF. The step transition of V1 is coupled into V3 through Cvar1 by a step of ±ΔV. Then V3 is either charged or discharged to half the power supply where the inverter U3 toggles. V1 experiences the reverse of its state and the similar process takes place again. The overall frequency of the circuit is determined by the delay produced in the second inverter stage.

FIG. 3.

This figure shows the waveforms of V1, V2, and V3 in ring oscillator circuit when Cvar1 is 30 pF. The step transition of V1 is coupled into V3 through Cvar1 by a step of ±ΔV. Then V3 is either charged or discharged to half the power supply where the inverter U3 toggles. V1 experiences the reverse of its state and the similar process takes place again. The overall frequency of the circuit is determined by the delay produced in the second inverter stage.

Close modal

The value of Cin can be calculated by probing V3 and measuring the corresponding ΔV (3.6 V and 1.8 V when Cvar1 equals 6.5 pF and 30 pF). The calculated Cin was 3.6 pF, which is lower than the maximum value of 7.5 pF indicated in the device data-sheet. In calculation, the parasitic capacitance of the probe (Tektronix P6139A), 8 pF, was taken into account. Cin was then used to determine the real value of ΔV (4.5 V and 1.4 V) when probe was removed. The frequency of CLK, by ignoring the delay provided by other stages, follows

\begin{equation}f_{CLK} = \frac{1}{{2T_d }} = \frac{1}{{2(C_{{\mathop{\rm var}} 1} + C_{in})\ln \left(1 + \displaystyle\frac{{2\Delta V}}{{Vdd}}\right)}},\end{equation}
fCLK=12Td=12(C var 1+Cin)ln1+2ΔVVdd,
(1)

substituting Cvar1 (30 pF), Cin (3.6 pF), and the probe-free ΔV (4.5 V) into Eq. (1), the calculated frequency as Cvar1 equals 30 pF is 195.3 kHz, which is close to the measured 212.2 kHz. Similarly, by replacing Cvar1 with 6.5 pF and ΔV with 1.4 V, the calculated frequency when Cvar is 6.5 pF equals 1.504 MHz, and the measured value was 1.494 MHz. For both cases the error is below 8%.

CLK feeds into two 8-bit delay-line circuits with delays separately controlled by the MCU through a level translator. E1 and R1 are buffered by inverters to sharpen their edges and then high-pass filter R2-Cvar2 and R3-Cvar3 reshape the square wave CLK into a pulse. The Full-width at Half Max (FWHM) of the excitation pulse can be adjusted between 20 ns and 70 ns. The reset pulse range is between 30 ns and 100 ns because R3 is chosen larger than R2.

After the pulse is produced, 3 inverter stages decrease the rise and fall time. The last stage involves 4 discrete inverters in parallel for the sufficient driving capability required by the succeeding excitation circuit. The period of CLK was tuned to ∼1 μs, the width of the excitation pulse was set to 20 ns, and the reset pulse was set to 90 ns. This configuration provides a 910 ns integration length and a resolution of 250 ps. It also ensures that the APD is only enabled after the excitation pulse, so no excitation light can be detected during the excitation phase. This mechanism essentially creates an electrical filter to replace the emission filter used in traditional systems. Figure 4 depicts the waveforms of the three signals, CLK, excitation pulse, and reset pulse, used in practical application. The delay of the reset pulse with respect to the excitation pulse is controlled by the MCU. The delay value is updated by an 8-bit serial data sent from the MCU.

FIG. 4.

In the presented waveforms, the delay of the reset pulse with respect to the excitation pulse is 0 ns. It can be delayed incrementally with a resolution of integer number of 250 ps through the control of the MCU. Meanwhile the delay of the excitation pulse stays the same. The initial delay for both excitation pulse and reset pulse with respect to CLK can be set by software.

FIG. 4.

In the presented waveforms, the delay of the reset pulse with respect to the excitation pulse is 0 ns. It can be delayed incrementally with a resolution of integer number of 250 ps through the control of the MCU. Meanwhile the delay of the excitation pulse stays the same. The initial delay for both excitation pulse and reset pulse with respect to CLK can be set by software.

Close modal

A 405 nm LED was driven by the excitation circuit shown in Figure 5. Q1 is composed of 4 RF bipolar transistors (NXP Semiconductors /BFG21W, 115). Two other transistors were connected between the emitter of Q1 and the ground, acting as the pull down switch Q2. It is constantly biased by a resistor-divider circuit at ∼1 V. The cathode of the LED was either connected to the ground or to a −2 V bias point to extend the peak power of the LED during the excitation. The circuit was characterized by measuring the average current I1 and I2, as well as the average power of the LED. For the non-pre-offset case and −2 V pre-offset case, the measured average current, the calculated peak current, the measured average power, and the calculated peak power of the LED are listed in Table I. In the measurement a 0.5 Ω resistor was added between the power supply and node A to sense the voltage drop, while in actual use it is shorted. The calculated average current was converted into peak current, by assuming a rectangular pulse. The peak power of the LED was converted from its average value in the similar way. By applying the −2 V pre-offset voltage, the peak LED power was increased by a factor of 2.5 over the non-pre-offset case. The peak LED driving current in pre-offset case was increased by 112% over the maximum forwarding current (100 mA) stated by the device datasheet (10% duty cycle, <100 us pulse width). The enhanced peak LED power is comparable to the laser diode power demonstrated in our previous laser TDF.24 The pre-bias technique was used to measure low concentration samples by increasing the fluorescence intensity.

FIG. 5.

In the schematic of the excitation circuit, 4 RF bipolar transistors are connected in parallel to construct Q1. Two bipolar transistors form Q2. Q1 is actively triggered by the excitation pulse. When it is turned off, the LED is quickly shut off by Q2 biased at a constant voltage, 1 V. Resistor R1 is used to measure the average current consumed by the circuit, while it is shorted in actual use. The cathode of the LED can be grounded or pre-biased at a negative voltage, −2 V, to extend the peak optical power of the LED.

FIG. 5.

In the schematic of the excitation circuit, 4 RF bipolar transistors are connected in parallel to construct Q1. Two bipolar transistors form Q2. Q1 is actively triggered by the excitation pulse. When it is turned off, the LED is quickly shut off by Q2 biased at a constant voltage, 1 V. Resistor R1 is used to measure the average current consumed by the circuit, while it is shorted in actual use. The cathode of the LED can be grounded or pre-biased at a negative voltage, −2 V, to extend the peak optical power of the LED.

Close modal
Table I.

Excitation circuit characterization result under non pre-offset and −2 V pre-offset case.

  Non pre-offset−2 V pre-offset
  Average 
I1 (mA) Peak 150 300 
  Average 1.322 4.325 
I2 (mA) Peak 66.1 211.75 
  Average 0.832 2.12 
LED power (mW) Peak 41.6 106 
  Non pre-offset−2 V pre-offset
  Average 
I1 (mA) Peak 150 300 
  Average 1.322 4.325 
I2 (mA) Peak 66.1 211.75 
  Average 0.832 2.12 
LED power (mW) Peak 41.6 106 

The fluorescence emitted from the sample is detected by the APD and the photo current is conditioned by the sensing circuit before digitization. The schematic of the sensing circuit is depicted in Figure 6. The differential structure is used to improve the power supply noise rejection and match the parasitic effects. When reset pulse is high, Q1 is conducting and the voltage at node A is clamped at ∼3.3 V. As the reset pulse goes low, Q1 stops driving node A and the photon induced current is converted into voltage through the parasitic capacitor Cp1. The gating mechanism of the APD, in combination with the time-interleaved integration window specified by the delay-line, guaranteed that all segments of the fluorescence decay are integrated incrementally with the resolution of 250 ps.

FIG. 6.

In the schematic of the sensing circuit, fluorescence signal is first integrated through the APD, D1, at the sensing node A. This node is reset by transistor Q1 controlled by the reset pulse. The high frequency signal at node A is low-pass filtered by R3-Cp3 and buffered by U1. An identical signal path is used to construct the differential structure in matching the circuit and reducing the power supply noise. The differential signal is then amplified, inverted, and low-pass filtered before digitization. DC bias voltages for some circuit nodes are denoted in this figure when APD is shielded from the light. An optional APD, D2, can be added as the dummy device to better match the differential circuitry.

FIG. 6.

In the schematic of the sensing circuit, fluorescence signal is first integrated through the APD, D1, at the sensing node A. This node is reset by transistor Q1 controlled by the reset pulse. The high frequency signal at node A is low-pass filtered by R3-Cp3 and buffered by U1. An identical signal path is used to construct the differential structure in matching the circuit and reducing the power supply noise. The differential signal is then amplified, inverted, and low-pass filtered before digitization. DC bias voltages for some circuit nodes are denoted in this figure when APD is shielded from the light. An optional APD, D2, can be added as the dummy device to better match the differential circuitry.

Close modal

In the succeeding stage, the signal at node A is low-pass filtered by R3 and the input parasitic capacitor of U1, Cp3. The measured value of Cp3 was 4.44 pF, which provides a 39.8 kHz low-pass bandwidth before the signal is buffered. A subtractor made up of U3 and resistor R7 ∼ R10 subtracts the buffered signal from the charge injection effect to obtain the actual signal. It is amplified with a gain of 300 provided by the ratios among R7 ∼ R10. The polarity of the signal was then recovered by a unity gain inverter constructed by U4. Finally, the signal is low-pass filtered at the bandwidth of 1.6 Hz by the low-pass filter formed by C1 and R13.

With the APD shielded from light, the bias voltages were measured and their values are presented in Figure 6. The circuit operates at 5 V single supply, except for the bias voltage of APD and the reset voltage of Q1 and Q2. The emitters of Q1 and Q2 are biased at 3.3 V to prevent base-emitter breakdown when the reset pulse goes high. Resistor R1 is used to limit the peak base current below 15 mA. The maximum output current of U1 is 30 mA, which limits the output voltage of U1 below 3.9 V through resistor R7. However, the saturation of U1 is not an issue because it only occurs after the output of U3 reaches the power rails, at which point the circuit has reached its sensing limit. U3 is biased near the middle of power supply in order to have the maximum signal swing and the best linearity. The maximum output of the inverter made by U4 is ∼3.8 V. This value was designed less than 5 V because the succeeding analog-to-digital converter circuit in the MCU operates at 3.3 V.

The high voltage used to reverse bias the APD was produced by a commercial DC-DC converter (EMCO G05). As shown in Figure 7, 2 unity gain buffers with the maximum output current of 60 mA drives the DC-DC converter. The input of the buffer can be controlled either by a potentiometer or a digital-to-analog converter (DAC). The configuration of the DAC is performed by the MCU in software. The potentiometer was used to debug the biasing circuitry initially.

FIG. 7.

In the high voltage generation circuit, the commercial DC-DC converter (EMCO G05) generates up to 150 V bias voltage for APD. The ripple of the output is less than 26 mV. The DC-DC converter is driven by two parallelly connected unity gain buffers with the maximum output current of 60 mA. Either a potentiometer or a digital-to-analog (DAC) converter can be used to bias the buffer. The use of the potentiometer simplifies the debug process, while the use of DAC provides software tunability of the APD bias voltage in response to different fluorescence strengths. In Figure 7 the decoupling capacitors are not depicted for simplicity.

FIG. 7.

In the high voltage generation circuit, the commercial DC-DC converter (EMCO G05) generates up to 150 V bias voltage for APD. The ripple of the output is less than 26 mV. The DC-DC converter is driven by two parallelly connected unity gain buffers with the maximum output current of 60 mA. Either a potentiometer or a digital-to-analog (DAC) converter can be used to bias the buffer. The use of the potentiometer simplifies the debug process, while the use of DAC provides software tunability of the APD bias voltage in response to different fluorescence strengths. In Figure 7 the decoupling capacitors are not depicted for simplicity.

Close modal

The LED TDF is constructed by 2 printed circuit boards (PCBs) and a plastic spacer/cuvette holder. Two boards are stacked by 8 aligned through holes and 4 screws. As is shown in Figure 8, the overall size of the device is ∼3 stacked iPhone4s. The noisy pulsing circuit and the quiet sensing circuit were placed back to back to minimize the noise coupling. The back sides of two boards were poured with copper plate connected to the earth ground. This further reduces the coupling between two boards. The space between two boards is shielded from environmental light by sticking black paper slips to the side of the device. The surface of the spacer/cuvette holder was further filled with black color to prevent external light from interfering the measurement space.

FIG. 8.

The LED TDF is assembled by stacking two PCBs (a). The top board senses the fluorescence decay. The bottom board generates timing signals and excites the sample. They are aligned by 4 screws and electrically connected by a 5 cm long ribbon cable. The leads of the LED were bent with 90° in order to excite the sample without interfering with the APD during reset. The sample is secured in the LED TDF by the plastic spacer/cuvette holder (b). To prevent the external light from interfering with the measurement space, the LED TDF is further shielded from the outer space by black colored paper slips.

FIG. 8.

The LED TDF is assembled by stacking two PCBs (a). The top board senses the fluorescence decay. The bottom board generates timing signals and excites the sample. They are aligned by 4 screws and electrically connected by a 5 cm long ribbon cable. The leads of the LED were bent with 90° in order to excite the sample without interfering with the APD during reset. The sample is secured in the LED TDF by the plastic spacer/cuvette holder (b). To prevent the external light from interfering with the measurement space, the LED TDF is further shielded from the outer space by black colored paper slips.

Close modal

The LED leads are bent with a 90° angle. A 1.5 ml disposable cuvette (Fisherbrand 14-955-127) was placed between the LED and APD during the measurement. The angle ensures that most of the light emitted from the LED does not interfere with the APD during the reset. To secure the cuvette, a plastic spacer/cuvette holder was customized built from a 6.5 mm thick nylon plastic sheet. A 5 mm long conducting ribbon cable (Wurth 686610050001) and 2 flexible connectors (FCI 10018783-11010TLF) are used for board connection.

Figure 9 shows the details of the PCBs before they are assembled. The board named “pulsing board” consists of a PIC32 microcontroller circuit powered by PC through a USB cable. The rest of the board is populated with the pulsing circuit and excitation circuit, which use a different power supply over MCU to reduce the coupling induced by the digital pulse signals. The board referred as “sensing board” was used only for sensing the fluorescence and no digital circuit is distributed on it. The multi-stage sensing circuit was laid out in a symmetric manner to better match the parasitic effects in the differential signal path. Table II lists all the solid-state components used in the LED TDF.

FIG. 9.

These pictures show the front and back view of the printed circuit board used in the LED TDF. The pulsing board is on the left and it contains all the digital circuits. The fluorescence signal is detected by the sensing circuit placed on the right, which is made only with the analog signal chain and bias circuits. Ground plane is built on the back of both boards for reducing the noise coupling.

FIG. 9.

These pictures show the front and back view of the printed circuit board used in the LED TDF. The pulsing board is on the left and it contains all the digital circuits. The fluorescence signal is detected by the sensing circuit placed on the right, which is made only with the analog signal chain and bias circuits. Ground plane is built on the back of both boards for reducing the noise coupling.

Close modal
Table II.

Solid state components used in the system.

 ComponentManufacturer /Part numberQuality
PCB Advanced Circuits 
MCU Microchip Technology /PIC32MX795F512L-80I/PT 
Level translator Texas Instruments /CD40109BPWR 
Delay line Maxim Integrated /DS1124U-25+-ND 
Hex-inverter NXP semiconductors /HEF4049BT,653 
RF transistor NXP Semiconductors /BFG21W, 115 10 
405 nm LED Bivar Inc/UV5TZ-405-15 
365 nm LED Marktech Optoelectronics /MT3650N3-UV 
Voltage regulator 3.3 V Microchip Technology /TC1262-3.3VDBTR 
10 Voltage regulator 5 V Microchip Technology /TC1262-5.0VDB 
11 Potentiometer Bourns Inc/3296Y-1-473LF 
12 Flexible connector FCI/SFW10R-1STE1LF 
13 Ribbon jumper cable Wurth Electronics Inc /686610050001 
14 DAC Microchip Technology /MCP4921-E/MS 
15 DC-DC converter EMCO/G05 
16 Op amp Analog Devices Inc /AD8542ARMZ-REEL 
17 APD Hamamatsu/Si APD S2382 
18 SMD resistors   45 
19 SMD capacitors   80 
 ComponentManufacturer /Part numberQuality
PCB Advanced Circuits 
MCU Microchip Technology /PIC32MX795F512L-80I/PT 
Level translator Texas Instruments /CD40109BPWR 
Delay line Maxim Integrated /DS1124U-25+-ND 
Hex-inverter NXP semiconductors /HEF4049BT,653 
RF transistor NXP Semiconductors /BFG21W, 115 10 
405 nm LED Bivar Inc/UV5TZ-405-15 
365 nm LED Marktech Optoelectronics /MT3650N3-UV 
Voltage regulator 3.3 V Microchip Technology /TC1262-3.3VDBTR 
10 Voltage regulator 5 V Microchip Technology /TC1262-5.0VDB 
11 Potentiometer Bourns Inc/3296Y-1-473LF 
12 Flexible connector FCI/SFW10R-1STE1LF 
13 Ribbon jumper cable Wurth Electronics Inc /686610050001 
14 DAC Microchip Technology /MCP4921-E/MS 
15 DC-DC converter EMCO/G05 
16 Op amp Analog Devices Inc /AD8542ARMZ-REEL 
17 APD Hamamatsu/Si APD S2382 
18 SMD resistors   45 
19 SMD capacitors   80 

The MCU was programmed as the USB slave of the host PC. Figure 10 illustrates the overall control flow. Upon configured, the device begins to execute the measurement. While the free running excitation pulse produces fluorescence decay and the reset pulses specify the integration window, the DC output of the sensing circuit is periodically sampled and digitized by the embedded 10-bit ADC in the MCU. The result for each conversion is recorded in a 16 k memory in the MCU. Then the data is transmitted to PC through the USB. The MCU performs the configuration request once again before the next measurement. During the idle state, the MCU keeps polling through the USB module until the configuration word and the start command are detected. The configuration word is 4-byte binary data. The first byte is used for the MCU to recognize the configuration operation. The second byte sets the initial reset pulse delay. During the measurement this delay value is updated by a step of integer multiple of 250 ps. The third byte defines the excitation pulse delay value, which keeps constant for the entire measurement. The last byte sets the number of samples digitized by the ADC for a certain delay. The start command is a 1 byte binary data for the MCU to recognize that a new measurement is permitted to execute.

FIG. 10.

This figure shows the overall control flow of the MCU programmed as the USB slave of the PC. In PC a MATLAB script is created to configure the measurement, sends command to the MCU, and receives data from the MCU. When this script is executed in PC, the configuration word is first sent to the MCU. The MCU decodes the word by comparing it with the look-up table in its memory. After the configuration is completed, the MCU sends a configuration complete signal to PC. As the PC recognizes this signal, it sends the start command to the MCU to start the measurement. The configured measurement is then performed until the configured integration steps are completed. At the end of the measurement, digitized fluorescence data is sent to PC. The LED TDF stays in idle state until the next configuration word is received.

FIG. 10.

This figure shows the overall control flow of the MCU programmed as the USB slave of the PC. In PC a MATLAB script is created to configure the measurement, sends command to the MCU, and receives data from the MCU. When this script is executed in PC, the configuration word is first sent to the MCU. The MCU decodes the word by comparing it with the look-up table in its memory. After the configuration is completed, the MCU sends a configuration complete signal to PC. As the PC recognizes this signal, it sends the start command to the MCU to start the measurement. The configured measurement is then performed until the configured integration steps are completed. At the end of the measurement, digitized fluorescence data is sent to PC. The LED TDF stays in idle state until the next configuration word is received.

Close modal

The top-level signal control is performed by a customized MATLAB script. In this script all the configuration options are specified. Upon executed, the script creates a serial port on the MCU and sends out the configuration word. A start command is then sent to trigger the configured measurement. While the measurement is performed, the PC waits the data until a measurement complete signal is sent from the MCU. Data is transmitted through multiple packages and is finally saved in a .mat file in PC. A following fitting process is performed in a separate MATLAB script.

The reported LED TDF was first verified by testing and calibrating the signals generated from the pulsing board and sensing board. Then the entire system was used to test 3 commercial fluorophores, including Qdot 585 (Life technologies/Qtracker 585), fluorescein (Online science mall/CAD# 518-47-8), and Lucifer yellow (MP Biomedicals/M3415). Finally, the system was used to measure a ZnSe Qdot based DNA sensor.

A 5 GS/s, 1 GHz oscilloscope (Tektronix MSO 4104) was used to detect the electrical excitation pulse. The quality of the excitation light produced by the LED was detected by an APD reverse biased at 130 V. The output of the APD was probed by the oscilloscope operating with 50 Ω input impedance. Figure 11 shows the waveform of the electrical excitation pulse and the transient optical power detected by the APD. The excitation light exhibited a 6 ns falling edge, from 90% to 10%, according to this plot. However, the falling edge for the excitation pulse was less than 2 ns. This discrepancy could be due to the delay induced by APD package capacitance (3 pF) and the coaxial cable connecting the APD and the oscilloscope.

FIG. 11.

This figure shows the waveform of the electrical excitation pulse and the corresponding LED output detected by the APD reverse biased at 130 V. The fall time of the excitation pulse was less than 2 ns, while the fall time of the LED output is 6 ns, from 90% to 10%.

FIG. 11.

This figure shows the waveform of the electrical excitation pulse and the corresponding LED output detected by the APD reverse biased at 130 V. The fall time of the excitation pulse was less than 2 ns, while the fall time of the LED output is 6 ns, from 90% to 10%.

Close modal

To better characterize the excitation light especially its falling edge, the APD-oscilloscope was replaced with the sensing circuit. In this test-bench, the delay caused by the coaxial cable and the parasitic capacitor of the probe was eliminated, since the APD was integrated on the board and the length of traces connected with the integration node was minimized in design. The leads of the LED were 90° bended with respect to the APD. A foil paper covered with transparent tape was used to reflect the light into the APD. In measurement, the MCU controlled the delay of the reset pulse such that the excitation light was integrated in 128 time-interleaved windows. Figure 12 is the transient optical intensity recovered by the MATLAB script. The fall time is 1 ns. The topography of the recovered pulse is closer to the shape of the excitation pulse in terms of the initial overshoot. The variation presented during the pulsing period could be due to the non-linearity inherent in the delay-line circuit and the jitter presented along the clock chain. Above observation indicates that the excitation light can be quenched in less than 2 ns, so that the integration of the fluorescence decay can be performed just 2 ns after the excitation pulse is toggled to ground to maximize the detected fluorescence. The measurement also suggests that the speed of the sensing circuit is potentially faster than APD-oscilloscope system in capturing time-resolved periodical signals.

FIG. 12.

In the recovered transient optical pulse measured by the sensing circuit, the fall time is 1 ns, which indicates that the sensing circuit can more accurately sense the fast optical signal than the APD-oscilloscope system.

FIG. 12.

In the recovered transient optical pulse measured by the sensing circuit, the fall time is 1 ns, which indicates that the sensing circuit can more accurately sense the fast optical signal than the APD-oscilloscope system.

Close modal

The system was then constructed for fluorescence lifetime measurement. During the measurement, de-ionized water was used as a blank. Then the blank was replaced with the testing sample. This result was the sum of the blank signal and the fluorescence decay. Therefore, by subtracting the blank signal, the decay component was extracted. Figure 13 illustrates the blank response and the real sample induced responses for Lucifer yellow (10 μM). The subtraction result is also plotted to the right. Figure 13(b) was the integration of the fluorescence, so the recovery of the fluorescence decay required a further differentiation operation. However, since the time constant, τ, would not change when an exponential function is integrated, the subtraction result in Figure 13(b) was directly used for single-exponential data fitting.

FIG. 13.

The integration results for the blank and the real sample are presented in (a). By subtracting the blank from the real sample induced response, the fluorescence decay can be extracted, as is shown in (b).

FIG. 13.

The integration results for the blank and the real sample are presented in (a). By subtracting the blank from the real sample induced response, the fluorescence decay can be extracted, as is shown in (b).

Close modal

The extracted fluorescence decay from the measurement data was further used to estimate the lifetime of the material. This was achieved by applying the least square data fitting method in MATLAB. The fitting result corresponding to the previous example is depicted in Figure 14. In the fitting algorithm, one blank measurement result is subtracted from another to obtain the noise. This peak to peak value is used as the threshold to determine the valid data points to be used in the fitting process. Based on the same measurement protocol, two other commercial fluorescence materials were measured. The result of the testing is plotted in a semilog axis (Figure 15) for the comparison. The measured lifetime for Lucifer Yellow (10 μM) was 5.6 ns versus the reported value of 5.7 ns;30 the measured lifetime of Fluorecein (10 μM)) was 4.1 ns versus the reported value of 4.0 ns;30 Qdot 585 (0.2 μM) indicated a 18.8 ns lifetime in measurement and the reported value was 19.5±1.6 ns.15 

FIG. 14.

This figure shows the single-exponential data fitting result based on extracted fluorescence decay curve. In determining how many data points to be included in the data fitting process, the blank was measured twice, and one was subtracted from another to obtain the peak-to-peak noise of the system. This value was used to determine the threshold value, where all data larger than this value was used in the fitting process, while the rest of the data was abandoned.

FIG. 14.

This figure shows the single-exponential data fitting result based on extracted fluorescence decay curve. In determining how many data points to be included in the data fitting process, the blank was measured twice, and one was subtracted from another to obtain the peak-to-peak noise of the system. This value was used to determine the threshold value, where all data larger than this value was used in the fitting process, while the rest of the data was abandoned.

Close modal
FIG. 15.

The reconstructed fluorescence decay curves for 3 commercial fluorophores are presented in semilog plot. The dotted line is the measurement result. Data below the threshold value determined by the processing script has been removed. The solid lines are the fitting result based on the single exponential decay model. The decay rate and the initial amplitude for each material are denoted.

FIG. 15.

The reconstructed fluorescence decay curves for 3 commercial fluorophores are presented in semilog plot. The dotted line is the measurement result. Data below the threshold value determined by the processing script has been removed. The solid lines are the fitting result based on the single exponential decay model. The decay rate and the initial amplitude for each material are denoted.

Close modal

Multiple fluorescein water solutions were prepared with different concentrations (0.05 μM, 0.1 μM, 0.25 μM, 0.5 μM, 0.75 μM, 1 μM, 5 μM, and 25 μM) in order to determine the detection limit of the system. Samples were measured in 8 separate groups for each concentration. Then the average value and the standard deviation for each group were calculated. The distribution of the measurement result is plotted in Figure 16. As the sample was diluted less than 0.5 μM, the measured lifetime began to roll off with increased standard variation. This is due to the fact that low concentration sample presents fluorescence decay with lower initial amplitude, which makes the constructed full decay curve susceptible to system noise, especially the temperature variation. The data suggests that the concentration limit of detection for the system is 0.5 μM when fluorescein is used as the verification material. In frequency domain fluorimeter, the detection limit is resulted from the fact that the interference of the excitation light with the APD becomes increasingly significant when fluorescence signal keeps decreasing. In our system, however, there is no such effect because the time-gated reset pulse prevents the APD from sensing the excitation pulse without using any optical filters.

FIG. 16.

Fluorescein water solutions were prepared with 8 concentrations. Each concentration was measured in 8 groups with the lifetime separately evaluated. Then these data was used to calculate statistical properties including the average lifetime and the standard deviation. This plot indicates that the concentration limit of detection for the system is 0.5 μM. Because for samples with lower concentration the measured lifetime deviates from the reported value and the uncertainty grows greater than 0.2 ns. Whereas the higher concentrations consistently provide measured lifetime close to the reported value and the standard deviation is no greater than 0.2 ns.

FIG. 16.

Fluorescein water solutions were prepared with 8 concentrations. Each concentration was measured in 8 groups with the lifetime separately evaluated. Then these data was used to calculate statistical properties including the average lifetime and the standard deviation. This plot indicates that the concentration limit of detection for the system is 0.5 μM. Because for samples with lower concentration the measured lifetime deviates from the reported value and the uncertainty grows greater than 0.2 ns. Whereas the higher concentrations consistently provide measured lifetime close to the reported value and the standard deviation is no greater than 0.2 ns.

Close modal

ZnSe Qdots were synthesized using the method reported by Hines.31 Diethylzinc and selenium powder dispersed in trioctylphosphine (TOP) were injected into hot 1-hexadecylamine (HDA) at 300 °C and ZnSe Qdots were grown in that mixture at 275 °C. Aliquots were extracted from the reaction mixture 20 min after precursor injection. The Qdots were subsequently extracted from the reaction mixture, modified with 11-mercaptoundecanoic acid (MUA) and dispersed in PBS buffer using the protocol reported by Wang.32,33 Qdot-ssDNA conjugates were prepared by linking the carboxyl group of MUA to the amino group of amino-modified dC25 ssDNA. The Qdot-ssDNA conjugates were hybridized with complementary free dG25 ssDNA to obtain Qdot-dsDNA conjugates. The fluorescence emission spectra of ZnSe Qdots and Qdot-DNA conjugates (shown in Figure 4S34) were initially measured using a Flurolog-3 Spectrofluorometer (Horiba Jobin Yvon) with an excitation wavelength of 350 nm. The peak emission wavelength of the Qdots was 396 nm and their average size was estimated to be ∼3 nm using a reported correlation between Qdot diameter and fluorescence emission peak wavelength.35 All samples were subsequently analyzed using the portable time domain LED fluorimeter containing 2 μM of Qdots or Qdot-DNA conjugates in PBS buffer solution. The peak emission intensity of the Qdots increased by a factor of 5 after conjugation with ssDNA and by an additional factor of 1.5 after hybridization with complementary ssDNA (Figure 5S34).

Each sample containing ZnSe Qdots was analyzed by using the LED TDF five times and the average lifetime as well as the standard lifetime error was calculated. For these measurements the 405 nm LED was replaced with a 365 nm LED to match the absorption peak wavelength of the sample (Figure 6S34). Figure 17 shows the measured fluorescence lifetime and the corresponding error for each sample. The observed fluorescence lifetime of the Qdots increased by 4 ns after conjugation with ssDNA and by an additional 2 ns after hybridization with complementary ssDNA. These changes in fluorescence lifetime are consistent with the observed increase in peak emission intensity of the Qdots after conjugation to ssDNA and dsDNA (Figures 4S34 and 5S34).

FIG. 17.

This figure presents the lifetime change when ZnSe Qdots were conjugated with ssDNA and further hybridized with the complementary ssDNA (target). Lifetime change indicated from the figure agrees with the intensity change.

FIG. 17.

This figure presents the lifetime change when ZnSe Qdots were conjugated with ssDNA and further hybridized with the complementary ssDNA (target). Lifetime change indicated from the figure agrees with the intensity change.

Close modal

We developed a portable LED TDF by applying multi-step time-interleaved integration assisted by time-gated electrical resetting. The peak optical power produced from the pulsed LED reached over 100 mW to excite low concentration testing samples that are usually measured by laser based system. The resetting scheme eliminated the requirement of both excitation and emission filters in traditional instruments. Our system is an alternative to phase domain fluorimeter and we have demonstrated its high portability, high flexibility, and high sensitivity in measuring commercial fluorophores. The system was successfully used to characterize a ZnSe Qdot based biochemical sensor. This potentially makes it very attractive for the development of biochemical sensors and other instruments whose operation involves the measurement of fluorescence lifetime.

This work was supported by the National Science Foundation (NSF) (Award No. ECCS-1128558), by the Dev and Linda Gupta Endowment, and by the University of Massachusetts via the CVIP Technology Development Fund.

1.
L. R.
Snyder
,
J. J.
Kirkland
, and
J. L.
Glajch
,
Practical HPLC Method Development
(
John Wiley and Sons
,
2012
).
2.
B. S.
Edwards
,
T.
Oprea
,
E. R.
Prossnitz
, and
L. A.
Sklar
, “
Flow cytometry for high-throughput, high-content screening
,”
Curr. Opin. Chem. Biol.
8
,
392
398
(
2004
).
3.
J. W.
Lichtman
and
J.-A.
Conchello
, “
Fluorescence microscopy
,”
Nat. Methods
2
,
910
919
(
2005
).
4.
J. R.
Lakowicz
,
Principles of Fluorescence Spectroscopy
(
Springer
,
2007
).
5.
B. N. G.
Giepmans
,
S. R.
Adams
,
M. H.
Ellisman
, and
R. Y.
Tsien
, “
The fluorescent toolbox for assessing protein location and function
,”
Science
312
,
217
224
(
2006
).
6.
M. F.
Kircher
,
R.
Weissleder
, and
L.
Josephson
, “
A Dual fluorochrome probe for imaging proteases
,”
Bioconjug. Chem.
15
,
242
248
(
2004
).
7.
A. P.
Demchenko
, “
The concept of λ-ratiometry in fluorescence sensing and imaging
,”
J. Fluoresc.
20
,
1099
1128
(
2010
).
8.
B.
Valeur
and
I.
Leray
, “
Design principles of fluorescent molecular sensors for cation recognition
,”
Coord. Chem. Rev.
205
,
3
40
(
2000
).
9.
R. M.
Ballew
and
J. N.
Demas
, “
An error analysis of the rapid lifetime determination method for the evaluation of single exponential decays
,”
Anal. Chem.
61
,
30
33
(
1989
).
10.
M. Y.
Berezin
and
S.
Achilefu
, “
Fluorescence lifetime measurements and biological imaging
,”
Chem. Rev.
110
,
2641
2684
(
2010
).
11.
G.
Neurauter
,
I.
Klimant
, and
O. S.
Wolfbeis
, “
Microsecond lifetime-based optical carbon dioxide sensor using luminescence resonance energy transfer
,”
Anal. Chim. Acta
382
,
67
75
(
1999
).
12.
A.
Mills
, “
Oxygen indicators and intelligent inks for packaging food
,”
Chem. Soc. Rev.
34
,
1003
1011
(
2005
).
13.
W.
Zhong
,
P.
Urayama
, and
M.-A.
Mycek
, “
Imaging fluorescence lifetime modulation of a ruthenium-based dye in living cells: The potential for oxygen sensing
,”
J. Phys. D: Appl. Phys.
36
,
1689
(
2003
).
14.
K.
Suhling
,
D. M.
Davis
,
Z.
Petrasek
,
J.
Siegel
, and
D.
Phillips
, “
Influence of the refractive index on EGFP fluorescence lifetimes in mixtures of water and glycerol
,”
Biomark. Biol. Spectrom. Imaging
4259
,
92
101
(
2001
).
15.
S. H.
Ko
,
K.
Du
, and
J. A.
Liddle
, “
Quantum-dot fluorescence lifetime engineering with DNA origami constructs
,”
Angew. Chem., Int. Ed.
52
,
1193
1197
(
2013
).
16.
P.
Haro-González
,
L.
Martínez-Maestro
,
I. R.
Martín
,
J.
García-Solé
, and
D.
Jaque
, “
High-sensitivity fluorescence lifetime thermal sensing based on CdTe quantum dots
,”
Small
8
,
2652
2658
(
2012
).
17.
J.
Guo
and
S.
Sonkusale
, “
A CMOS imager with digital phase readout for fluorescence lifetime imaging
,” in
Proceedings of the ESSCIRC (ESSCIRC)
, pp.
115
118
(
2011
).
18.
W.
Becker
,
A.
Bergmann
,
K.
Koenig
, and
U.
Tirlapur
,
Picosecond Fluorescence Lifetime Microscopy by TCSPC Imaging
(
SPIE
,
2001
), Vol.
4262
, pp.
414
419
.
19.
W.
Becker
 et al,
High Resolution TCSPC Lifetime Imaging
(
SPIE
,
2003
), Vol.
4963
, pp.
175
184
.
20.
K.
Dowling
,
S. C. W.
Hyde
,
J. C.
Dainty
,
P. M. W.
French
, and
J. D.
Hares
, “
2D fluorescence lifetime imaging using a time-gated image intensifier
,”
Opt. Commun.
135
,
27
31
(
1997
).
21.
P.
Harms
,
J.
Sipior
,
N.
Ram
,
G. M.
Carter
, and
G.
Rao
, “
Low cost phase-modulation measurements of nanosecond fluorescence lifetimes using a lock-in amplifier
,”
Rev. Sci. Instrum.
70
,
1535
1539
(
1999
).
22.
T.
Iwata
,
T.
Kamada
, and
T.
Araki
, “
Phase-modulation fluorometer using an ultraviolet light-emitting diode
,”
Opt. Rev.
7
,
495
498
(
2000
).
23.
A. G.
Ryder
,
S.
Power
,
T. J.
Glynn
, and
J. J.
Morrison
, “
Time-domain measurement of fluorescence lifetime variation with pH
,”
Biomark. Biol. Spectrom. Imaging
4259
,
102
109
(
2001
).
24.
C. D.
Salthouse
,
R.
Weissleder
, and
U.
Mahmood
, “
Development of a time domain fluorimeter for fluorescent lifetime multiplexing analysis
,”
IEEE Trans. Biomed. Circuits Syst.
2
,
204
211
(
2008
).
25.
N. C.
Shaner
,
P. A.
Steinbach
, and
R. Y.
Tsien
, “
A guide to choosing fluorescent proteins
,”
Nat. Methods
2
,
905
909
(
2005
).
26.
D.
Gerion
 et al, “
Synthesis and properties of biocompatible water-soluble silica-coated CdSe/ZnS semiconductor quantum dots
,”
J. Phys. Chem. B
105
,
8861
8871
(
2001
).
27.
H.
Lam
,
Y.
Kostov
,
G.
Rao
and
L.
Tolosa
, “
A luminescence lifetime assisted ratiometric fluorimeter for biological applications
,”
Rev. Sci. Instrum.
80
,
124302
(
2009
).
28.
J.
Kissinger
and
D.
Wilson
, “
Portable fluorescence lifetime detection for chlorophyll analysis in marine environments
,”
IEEE Sens. J.
11
,
288
295
(
2011
).
29.
B. B.
Collier
and
M. J.
McShane
, “
Time-resolved measurements of luminescence
,”
J. Lumin.
144
,
180
190
(
2013
).
30.
See http://www.iss.com/resources/research/technical_notes/K2CH_FLT.html for Terpetschnig, ISS Technical Notes Fluorescence Lifetime, 2013.
31.
M. A.
Hines
and
P.
Guyot-Sionnest
, “
Synthesis and characterization of strongly luminescing ZnS-capped CdSe nanocrystals
,”
J. Phys. Chem.
100
,
468
471
(
1996
).
32.
J.
Wang
,
P.
Lei
,
S. T.
Andreadis
, and
T. J.
Mountziaris
, “
Detection of DNA hybridization via fluorescence intensity variations of ZnSe-DNA quantum dot biosensors
,”
Anal. Lett.
45
,
227
241
(
2012
).
33.
J.
Wang
and
T. J.
Mountziaris
, “
Homogeneous immunoassays based on fluorescence emission intensity variations of zinc selenide quantum dot sensors
,”
Biosens. Bioelectron.
41
,
143
149
(
2013
).
34.
See supplementary material at http://dx.doi.org/10.1063/1.4873330 for detailed analysis of pulsing circuit and optical properties of ZnSe Qdot based DNA sensors.
35.
B. C.
Mei
 et al, “
Dilution effects on the photoluminescence of ZnSe quantum-dot dispersions
,”
Appl. Phys. Lett.
93
,
083114
083114
(
2008
).

Supplementary Material