Article

Cite This: Anal. Chem. 2018, 90, 7569−7577 pubs.acs.org/ac

Automated 3D-Printed Microfluidic Array for Rapid Nanomaterial-

Enhanced Detection of Multiple Proteins
Karteek Kadimisetty,† Spundana Malla,† Ketki S. Bhalerao,† Islam M. Mosa,†,‡ Snehasis Bhakta,†
Norman H. Lee,# and James F. Rusling*,†,§,∥,⊥
†
Department of Chemistry and §Institute of Material Science, University of Connecticut, Storrs, Connecticut 06269, United States
‡
Department of Chemistry, Tanta University, Tanta 31527, Egypt
∥
Department of Surgery and Neag Cancer Center, UConn Health, Farmington, Connecticut 06032, United States
⊥
School of Chemistry, National University of Ireland, Galway H91 TK33, Ireland
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
Downloaded via INDIANA UNIV PURDUE UNIV AT IN on January 17, 2020 at 21:44:18 (UTC).

#
Department of Pharmacology & Physiology, George Washington University, Washington, D.C. 20037, United States
*
S Supporting Information

ABSTRACT: We report here the fabrication and validation of a novel 3D-printed, automated immunoarray to detect multiple
proteins with ultralow detection limits. This low cost, miniature immunoarray employs electrochemiluminescent (ECL)
detection measured with a CCD camera and employs touch-screen control of a micropump to facilitate automated use. The
miniaturized array features prefilled reservoirs to deliver sample and reagents to a paper-thin pyrolytic graphite microwell
detection chip to complete sandwich immunoassays. The detection chip achieves high sensitivity by using single-wall carbon
nanotube−antibody conjugates in the microwells and employing massively labeled antibody-decorated RuBPY−silica
nanoparticles to generate ECL. The total cost of an array is $0.65, and an eight-protein assay can be done in duplicate for
$0.14 per protein with limits of detection (LOD) as low as 78−110 fg mL−1 in diluted serum. The electronic control system costs
$210 in components. Utility of the automated immunoarray was demonstrated by detecting an eight-protein prostate cancer
biomarker panel in human serum samples in 25 min. The system is well suited to future clinical and point-of-care diagnostic
testing and could be used in resource-limited environments.

R eliable early detection is currently the best hope for

successful cancer patient outcomes.1−3 Measuring protein
biomarkers overexpressed into blood due to cancer has great
Thus, low cost, automated diagnostic platforms that can
rapidly detect multiple cancer biomarkers with high sensitivity
and specificity will be valuable tools for future personalized
early diagnostic potential1,4−8 that is still largely untapped in patient care of cancer and other diseases, as well as for research
the clinic.9 Enzyme-linked immunosorbent assays (ELISA) applications requiring measurement of low abundance
have long served as the method of choice for single protein proteins.9 In the present paper, we describe a new, low cost
diagnostic tests with limits of detection (LOD) typically in the 3D-printed array for multiple protein detection using
3−40 pg/mL range.10,11 More recently, bead-based optical and automated reagent delivery with a simple user interface for
electrochemiluminescent (ECL) methods have been marketed rapid assays.
for multiplexed protein assays but do not improve on ELISA Desktop 3D-printers can be used to rapidly design, optimize,
LODs.12−15 A single-protein counting technology known as and fabricate low cost, high performance microfluidic
Simoa features automation and LODs of 4−200 fg mL−1 for devices.22−27 3D-printing enables rapid prototyping of single-
multiplexed assays.16 However, reliable bioanalytical devices unit devices while avoiding expensive masters or masks
that offer automated, low cost, highly sensitive, multiplexed necessary for the alternative methods of microfluidic device
assays for clinical protein detection are still lacking.
Immunoarrays have been reported that measure small numbers Received: March 16, 2018
of proteins with accuracy, reliability, and automation but usually Accepted: May 20, 2018
lack one key attribute, usually low cost or high sensitivity.17−22 Published: May 21, 2018

© 2018 American Chemical Society 7569 DOI: 10.1021/acs.analchem.8b01198

Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

Figure 1. Representations of 3D-printed immunoarray: (A) 3D-printed microfluidic array with chambers to hold sample, wash buffers, detection
nanoparticles, and coreactant for ECL generation. Array is shown on left without detection chip and on right bonded to pyrolytic graphite sheet
(PGS) microwell detection chip with reagent and sample chambers filled with dye solutions for visualization; (B) representative disposable PGS chip
with heat transferred microwells printed using hydrophobic toner ink. Inset illustrates sandwich immunoassay on a single-wall carbon nanotube
forest (SWCNT) in one microwell.

fabrication such as photo lithography and soft lithography.28−30 deliver samples and reagents. This immunoarray measured
Device assembly tasks required when using precision cutting, eight prostate cancer biomarker proteins in human serum in 25
molding, and machining for fabrication are minimized by 3D- min with LODs as low as 85−110 fg mL−1. We verified
printing to produce nearly complete microfluidic devices. accuracy for the eight proteins in serum and successfully
Plans for 3D-printed objects are developed using computer- demonstrated analysis of human serum samples.

■
aided design (CAD) software, and the CAD file is processed to
generate print instructions that are uploaded to the printer.23,24 EXPERIMENTAL SECTION
Optimization is achieved at a fraction of cost and time of
Chemicals and Materials. Poly(diallyldimethylammonium
lithography, and the final optimized prototype becomes the
chloride) (PDDA), poly(acrylic acid) (PAA), bovine serum
usable device. While lithography can presently achieve better
albumin (BSA), 1-(3-(dimethylamino)propyl)-3-ethylcarbodi-
resolution than 3D-printing, ongoing advances in 3D print
imidehydrochloride (EDC), and N-hydroxysulfosuccinimide
resolution and speed are underway.31 However, stereolitho-
(NHSS) were from Sigma. Human prostate cancer patient
graphic (SLA) 3D-printers can achieve channel widths of 150 serum samples were from George Washington University
μm and structural features of 95 μm32 and are well suited for (GWU) Hospital under IRB ethical approval. Calf serum as a
many bioanalytical microfluidic devices. Recent analytical surrogate for human serum was used to dissolve standard
applications include smartphone-controlled biosensors,33−37 proteins for calibrations.
electrochemical sensors,38,39 optics for SPR,40 and other Ru(bpy)32+-silica nanoparticles (RuBPY−SiNP) were pre-
biomedical sensors.41 pared by a previously described method,50 had a diameter of
Our research team has 3D-printed microfluidic devices that 110 ± 13 nm from TEM (Figure S1A, SI) and are stable for
include an electrochemical hydrogen peroxide sensor, electro- one month or more at 4 °C. RuBPY−SiNPs of about 100 nm
chemiluminescent DNA sensor,42 a genotoxicity chemistry were found to provide a good balance between mobility in a
reactor with DNA damage end point,43 and prototype microfluidic system while retaining good detection sensitivities.
immunoarrays for proteins.22,44,45 Earlier, we used polymer RuBPY−SiNP was coated successively with PDDA and PAA
molding to develop ultrasensitive microfluidic immunoarrays followed by attachment of detection antibodies (Ab2) using
that detect multiple proteins by combining massively labeled EDC−NHSS to link amine groups of antibodies to COOH
antibody-coated magnetic beads with nanostructured electro- groups of PAA on the RuBPY−SiNP surface.50 Optimized
chemical sensors.46−48 We used precision cutting and concentrations of Ab2 used for attachment to RuBPY−SiNP
machining to develop microfluidic ECL arrays combining were 8 μg mL−1 for PSA, PSMA, IGFBP-3, VEGF-D, and PF-4,
antibody-coated silica nanoparticles containing 0.5 million 5 μg mL−1 for IGF-1 and CD-14, and 3 μg mL−1 for GOLM-1.
Ru(bpy)32+ (RuBPY) ions with single-walled carbon nanotube Once Ab2 was covalently attached, bicinchoninic acid (BCA)
(SWCNT) forest sensors.49 Both methodologies detect up to total protein assays51 were used to determine the number of
four proteins with ultralow detection limits (LOD) of 5−100 fg antibodies per particle, which gave Ab2:RuBPY−SiNP ratios of
mL−1 in less than 1 h. ∼32:1 in all cases. An optical absorbance assay measured
In the current paper, we describe a low cost, miniature 3D- 400 000 RuBPYs per nanoparticle. Ru(bpy)32+ concentration in
printed immunoarray to detect multiple proteins using ECL the silica nanoparticle dispersions was estimated using a
with CCD camera measurement. We integrated an automated calibration curve for dissolved Ru(bpy)32+ absorbance at 457
micropump controller with the unibody 3D-printed device to nm. The number of Ru(bpy)32+ ions per particle was obtained
7570 DOI: 10.1021/acs.analchem.8b01198
Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

by dividing the Ru(bpy)32+ concentration found by the total P113689 ND) to desired sizes to fit array dimensions. The
number of Ru(bpy)32+ silica nanoparticles.21,50 The resulting pyrolytic graphite chip was patterned by printing an inkjet
Ab2−RuBPY−SiNP particles were incubated in 2 mg mL−1 toner microwell template onto glossy paper53 and then heat
BSA to minimize nonspecific binding, washed, and stored at 4 transferring onto the pyrolytic graphite to make 16 microwells
°C. In the assay, ECL was generated from RuBPY−SiNP by that are about 10−15 nm deep and can hold volumes of ∼1 μL.
applying 1.0 V vs Ag/AgCl from an on-board three-electrode Densely packed single-walled carbon nanotube (SWCNT)
potentiostat using 500 mM tripropylamine (TPrA) coreactant forests were assembled in the microwells using previously
with 0.05% Tween-20 (T20) and 0.05% Triton-X in 0.2 M described methods54,55 to increase surface area and enhance
phosphate buffer (See Scheme S1, ECL pathway). amounts of capture antibodies (Ab1) attached. Figure S1B,C
3D-Printed Microfluidic Device. A Form 2 (Formlabs) shows SEM images of the surface of the PGS sheets with
stereolithographic (SLA) desktop 3D-printer was used to microwells prepared from hydrophobic printer toner. Figure
achieve high resolution (∼25 μm) with low surface rough- S1C shows the surface of PGS inside a microwell region.
ness.32 CAD files of the arrays were prepared using 123D Tapping mode atomic force microscopy (AFM) showed
software (Autodesk), converted to the appropriate format, and increased roughness from the bare PGS surface (35 nm) to
then uploaded to the printer to fabricate the desired object. the SWCNT forest surface from vertically aligned SWCNTs
Clear photo curable resin (GPCL02 Formlabs clear resin) was (47 nm), (Figure S1D,E). The terminal carboxyl groups of
used to produce compact plastic arrays with internal chambers SWCNTs were activated using EDC−NHSS to attach capture
to hold assay reagents, sample, and detection chip. Freshly antibodies (Ab1) by amidization, with a resulting decrease in
printed arrays were cleaned by flushing, bathing, and sonicating the AFM roughness factor to 38 nm. (Figure S1F). Chips with
in isopropanol for 10−15 min to remove uncured resin, Ab1 attached are stable for at least 1 week at 4 °C.
followed by air drying. Before use, arrays were coated with The eight proteins in the prostate cancer biomarker panel are
acrylic spray (Krylon) to improve optical clarity from slightly listed in Table S1. Capture antibodies for each protein were
opaque to 90% transmittance to visible light.42 The reference attached to SWCNT wells at 100 μg mL−1 concentration.
electrode, Ag/AgCl, and counter electrode platinum wires were Antibody-coated arrays were incubated with 2% BSA for 1 h at
then inserted into grooves printed so that they will lie opposite room temperature to minimize NSB.
along the microwell row in the detection array in the fully The 3D-printed array features nine chambers, one inlet port,
assembled array and one outlet port leading to the interconnected detection
The complete printed array has dimensions (L) length = 40 chamber. Chambers holding different solutions are in series
mm, (W) width = 35mm, and (H) height = 3.5 mm (Figure with an air-filled chamber to avoid intermixing of reagents. The
1A). Internal reagent and sample chambers have dimensions of first reagent chamber next to the pump inlet port is filled with
25 × 2 × 1 mm (L x W x H) and volumes of 48 ± 2 μL. The tripropylamine (TrPA) ECL coreactant, and the second and
detection channel has an open bottom that is converted into a third chambers are filled with wash buffer: 10 mM PBS + 0.05%
closed detection chamber by bonding it with adhesive onto a Tween 20, pH 7.4. These last two chambers are not separated
paper-thin pyrolytic graphite microwell detection chip cut from by an air gap, since the two solutions are identical and
pyrolytic graphite sheet (PGS, Panasonic).52 The detection intermixing is not an issue. The fourth chamber holds air
chamber is 30 × 2 × 0.6 mm (L × W × H) to accommodate followed by the fifth chamber containing a dispersion of the
the sample and reagents volumes to be delivered from ECL detection nanoparticles, Ab2−RuBPY−SiNP. This sixth
upstream. Chamber sizes were optimized to efficiently contain chamber is another air gap, followed by a wash buffer chamber.
reagent and sample with low dead volumes. The volume of the The eighth chamber is air-filled followed by the final chamber
detection chamber is slightly less than 50 μL and containing sample or standard solution directly before the
accommodates a 16-microwell detection chip, with the detection chamber (Figure 1A). The inlet port is connected to
microwells separated in space and preassigned for duplicate the programmed micropump, which facilitates sequential
measurements of each analyte protein (Figure 1B). Sample and delivery of reagents on an optimized time schedule.
reagents were preloaded into the array using a micropipette Automation and User Interface. A touch-screen operated
prior to the immunoassay and stored at 4 °C until use. No controller system employs simple user commands to start the
reagent degradation was found for 1 or 2 day storage. assay. The micropump user interface uses a voltage-controlled
To minimize nonspecific binding (NSB) of sample oscillator (VCO) latch and a digital-to-analog convertor (DAC)
constituents to the immunoarray, we incubated arrays with 2 latch. The VCO controls on and off cycles of the micropump
mg mL−1 bovine serum albumin (BSA) for 1 h prior to use. and times assay events. The DAC controls amplitude and
Reagents and samples were added into their chambers via small frequency of the micropump, which controls flow rate. Three
loading ports present on the top of the array. By placing a pipet DACs are connected to three submicrocontrollers to control
tip into these ports, assay reagents were loaded with good three micropumps to enable three simultaneous assays. Another
reproducibility. Vent holes were strategically placed at the end DAC was used to equip a three-electrode potentiostat to apply
of each chamber so that solutions fill uniformly and efficiently, 1.0 V vs Ag/AgCl to drive the ECL detection reaction (see SI).
while air is expelled. Vent holes and loading ports were closed Components are integrated with an Arduino microcontroller
with single-sided transparent tape to form an airtight and the touch LED screen to input user commands. A small 4.5
continuous channel to sequentially deliver reagents down- V rechargeable lithium ion battery powers the control system
stream to the detection chamber. and drives ECL. Once the controller is turned on using an on/
The detection chamber is completed by attaching the off toggle switch, the user can start the immunoassay with a
nanostructured pyrolytic graphite sheet (PGS) microwell preset program or press a program button to enter new settings
chips via double-sided adhesive under the 3D-printed detection to control flow rates and timing of immunoassay steps. The
channel. These detection chips were prepared by cutting 70 μm start button initiates the immunoassay; the “IDLE” button
thick PGS (Panasonic Industrial Devices and Solutions, turns into “RUN”, which is an indication of pump initiation. A
7571 DOI: 10.1021/acs.analchem.8b01198
Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

Figure 2. Immunoassay controller with touch-screen user interface for immunoarray. A microfluidic array connected to a micropump is shown with
dye-filled reagent chambers and graphite detection chip. Inset figures show multiple immunoassay steps along with messages to inform the user.

digital timer indicating step number is displayed so that the user

can monitor assay progress (Figure 2). When the immunoassay
■ RESULTS
Development of an assay for an eight-protein prostate cancer
is complete, the screen shows “Done” and “Measure OFF” is
biomarker panel in serum was chosen to evaluate the new 3D-
displayed. At this point, no voltage is applied, but once the user
printed immunoarrray. About 160 000 prostate cancers are
touches the “Measure OFF” button, the screen turns to a green
predicted for men in the U.S. in 2017 with 27 000 deaths.56
“Measure ON” sign that applies voltage for the ECL reaction
Current diagnosis of prostate cancer using the prostate specific
with CCD camera data collection. While the ECL potential is
antigen (PSA) biomarker test with a decision threshold of 4 ng
being applied, “Measure ON” is displayed for 180 s to indicate
that detection is in progress (Figure 2). In developing assays for mL−1 yields a clinical specificity of 63% and sensitivity of 35%
the immunoarrays, we optimized conditions to attain the best for disease detection57,58 and can result in overdiagnoses
analytical performance characteristics including high sensitivity, leading to unnecessary surgery. Also, the serum PSA test is
wide dynamic range, ultralow detection limits, and good unable to identify life-threatening aggressive forms of the
reproducibility. disease.59 Our long-term goal is to establish a panel of prostate
Assay Procedure. The arrays were filled with immunore- cancer serum biomarkers to facilitate distinguishing aggressive
agents and sample and were then connected to the automated from indolent forms of the disease.60−62
interface module. Pumps and pumping cycles are initiated by Members of the eight-protein biomarker test panel (Table
pressing “start”. Step 1 flows sample to detection chamber; the S1) were chosen from prior literature. A subset of this panel
stopped-flow incubation occurs and then washing. Step 2 was shown to facilitate surgical decisions for prostate cancer
delivers ECL RuBPY−SiNP dispersion, incubates, washes, and patients. Vascular endothelial growth factor-D (VEGF-D) is
then delivers coreactant TrPA. For detection, 1.0 V vs Ag/AgCl implicated in prostate cancer angiogenesis.63,64 Insulin-like
was applied from the onboard potentiostat with the array under growth factor IGF-1 and IGF-binding protein IGFBP-3 are
a CCD camera in a dark box for ECL generation and implicated in cell survival.65 Serum monocyte differentiation
measurement. Captured ECL light is analyzed with appropriate antigen CD14 (CD14) is an inflammation marker.66 Golgi
software to relate light intensities to concentration of membrane protein 1 (GOLM-1) is a prostate cancer gene
analyte.43,44 Micropumps were optimized to deliver reagents fusion protein.67,68 We also included general prostate cancer
reproducibly at designated timed intervals at 130 ± 5 μL min−1 biomarkers PSA, platelet factor-4 (PF-4), and prostate specific
flow rates. When more than one assay is desired, synchroniza- membrane antigen (PSMA), for which we had previously
tion of three micropumps in ensured by adjusting frequency developed multiplexed immunoassays in serum.69
and amplitude via the user interface. Optimized incubation Initially, two-protein assays were used to optimize reprodu-
times for sample and label in the multiplexed detection cibility and test compatibility of antibody pairs selected. A
chamber were 10 min. For the duplex protein detection, the single two-antibody RuBPY−SiNP particle was used for
incubation time for the sample is 14 min and for the RuBPY detection of each protein pair. Duplex assays coupled PSA
label is 12 min, enabling a slightly better LOD. ECL was and PSMA, VEGF-D and PF-4, IGF-1 and CD-14, and IGFBP-
generated at 1.0 V vs Ag/AgCl for 180 s camera accumulation 3 and GOLM-1 (Table S1) as simultaneously detected pairs.
times using 500 mM tripropylamine in 200 mM PBS + 0.05% Relative ECL responses from the 3D-printed immunoarray with
Tween 20 + 0.05% Triton-X at pH 7.5. Relative ECL intensities three trials of controls and 1 pg mL−1 PSA and PSMA in
were obtained by analyzing the light intensities of sample undiluted calf serum gave an average relative standard deviation
concentrations and dividing by the signal for protein-free (RSD) of ±5% for well-to-well (n = 8) and ±8% for array-to-
controls. array (n = 3). We constructed calibration curves for all eight
7572 DOI: 10.1021/acs.analchem.8b01198
Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

Figure 3. Recolorized CCD images for five arrays showing increase in ECL light with increase in concentration for all eight proteins on a single array
with an acquisition time of 180 s at 1.0 V Ag/AgCl in the presence of 500 mM TrPA.

Figure 4. Calibration data for multiplexed detection in undiluted calf serum from ECL responses at 1.0 V vs Ag/AgCl. (A−H) are multiplex assay
calibration curves for IGF-1, PSA, PF-4, CD-14, VEGF-D, GOLM-1, PSMA, and IGFBP-3. Standard deviations for n = 4.

proteins in duplex assays and found reproducible ECL signals we sacrificed detection limits slightly by shortening incubation
with RSDs ranging from ±7−13%. ECL signal intensities were periods to achieve a shorter assay time of 25 min.
divided by ECL intensity from protein-free controls and In the eight-protein assay, the pump delivers sample or a
expressed as relative ECL intensities, which were plotted mixture of standard proteins from the sample chamber to the
against concentration for calibration (Figure S2, see Figure 3 detection microwells, followed by stopped-flow incubation for
for typical raw data). Dynamic ranges were from 0.1 to 1000 pg 10 min, then washing with PBS at pH 7.4. The pump then
mL−1 for PSA, PSMA, VEGF-D, IGF-1, CD-14, and IGFBP-3 delivers RuBPY−SiNP dispersion to the detection microwells,
and 0.1 to 10 000 pg mL−1 for GOLM-1 and PF-4. Limits of stop-flow incubates for 10 min, then washes with 10 mM PBS
detection of 78−100 fg mL−1 were obtained for all eight analyte and PBS Tween 20 (pH 7.4). TPrA solution is then delivered
proteins in 35 min assays. and flow stopped, and ECL responses are measured by the
Multiplexed Detection. Once we optimized and charac- CCD camera in a dark box while applying 1.0 V vs Ag/AgCl to
terized performance with duplex assays, we progressed to the detection chip. A representative recolorized ECL image of
detecting all eight proteins simultaneously in duplicate. The calibration results (Figure 3) shows increased ECL light with
eight different capture antibodies were immobilized in two increase in analyte protein concentrations. Figure 4 shows the
individual SWCNT sensor microwells each in the order of IGF- resulting calibration graphs for all eight proteins, six of which
1, PSA, PF-4, CD-14, VEGF-D, GOLM-1, PSMA, and IGFBP-3 were linear, and the remaining two were slightly curved but still
(Figure 3). A detection label dispersion was prepared by mixing fully acceptable for reliable sample concentration determina-
the duplex Ab2−RuBPY−SiNP assay labels, label 1 for PSA and tions. Acceptable dynamic ranges from 0.5 pg mL−1 to 10 ng
PSMA, label 2 for VEGF-D and PF-4, label 3 for CD-14 and mL−1 were obtained for all proteins with LODs from 110 to
IGF-1, and label 4 for GOLM-1 and IGFBP-3. These four 500 fg mL−1 for this multiplexed protocol; the difference in
nanoparticle types were mixed in equal proportions and ECL signals for each biomarker was represented as their
delivered to the detection chamber after analyte protein sensitivities (Table S5). Among the eight proteins under study,
binding to complete the sandwich immunoassay. As the levels calibration curves for CD-14 and VEGF-D showed a nonlinear
of these proteins in human serum samples allowed (Table S1), curve fitting; these nonlinear curve fittings are common and
7573 DOI: 10.1021/acs.analchem.8b01198
Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

Figure 5. Statistical summary of patient sample results using clustered multiple variable graphs for each protein. (A,B) Box-and-whisker plots show
data point distribution classified as cancer-free and prostate cancer samples for each biomarker. The plots present lower, upper quartile, and median
values along with minimum and maximum ranges.

acceptable in ligand binding assays.70 Accuracy of the cancer and cancer-free patient samples. Receiver operating
measurements was evaluated using spiked serum and by characteristic (ROC) plots can provide estimates of accuracy
measuring recovery. Known concentrations of all target for clinical diagnostic predictions.74 Preliminary ROC analyses
proteins from 750 fg mL−1 to 2.5 ng mL−1 were spiked into were done for the 12 sample data set (see SI), and results
diluted human serum, and percent recovery was determined suggest the possibility that this panel may be useful in
from the measured concentration. Percent recovery values for distinguishing between prostate cancer and cancer-free patients
individual proteins (Table S2) were in the analytically and perhaps even between aggressive and indolent cancer
acceptable range71 of 100 ± 14%, demonstrating reasonably patients. These results are encouraging but still very
good accuracy, and the absence of significant antibody cross preliminary, and hundreds of samples will need to be analyzed
reactivity. before definitive conclusions can be drawn.

■
Single-protein ELISA assays of all eight proteins were not
possible due to the lack of sufficient sample volumes. However,
we compared PSA concentrations in serum samples using the DISCUSSION
multiplex immunoarray with ELISA data supplied by George The results above successfully demonstrate a miniature 3D-
Washington University Hospital and found good correlation printed ECL immunoarray operated automatically by a touch-
between the two methods (Figure S3). screen interface for simultaneous multiplexed detection of
Human Serum Samples. We used the multiplexed proteins in aqueous samples. Results for an eight-protein
immunoarray to measure the prostate cancer protein panel in prostate cancer biomarker panel demonstrate that the
12 human serum samples. Gleason scores (GS) are used to immunoarray can achieve sensitive multiplexed protein
grade prostate cancer tumors,72 and 6 or less means that the detection with LODs of ∼110 fg mL−1 and dynamic ranges
tumor tissue is well differentiated and likely to grow slowly, of 0.5 pg mL−1 to 10 ng mL−1 (Figures 3 and 4). Spike and
suggesting a less aggressive form of cancer. A Gleason score of recovery tests confirmed accurate measurement of target
7 is intermediate, and 8 or more is categorized as the most proteins in diluted human serum (Table S2). Using sample
aggressive. The patient samples included four cancer-free dilutions to bring protein concentrations into the dynamic
individuals, four samples with nonaggressive (indolent) forms ranges, as little as 1 μL of serum was used for an eight-protein,
of prostate cancer with GS ≤ 6, and four aggressive forms of 25 min assay. Lower detection limits of 78−100 fg mL−1 and
prostate cancer with GS ≥ 8. We label cancer-free (CF) wider dynamic ranges up to 0.1 pg mL−1 to 10 ng mL−1 can be
samples as CF1 to CF4, indolent cancer samples as I1 to I4,
obtained by using longer incubations resulting in 35 min assay
and aggressive samples as A1 to A4. Guided by normal serum
times, demonstrated in duplex assays for the eight biomarkers.
levels of the biomarkers (Table S1), we diluted serum samples
Here, better performance in the lower protein concentration
up to 750-fold in PBS prior to analysis to bring ECL responses
into the linear or sensitive curvilinear ranges of the calibration range is most likely due to allowing more time for antibodies on
curves. Concentrations of all eight biomarkers in the samples sensors and detection particles to incubate and bind with the
are shown in Table S2. Clustered multiple variable (CMV) analyte proteins.75
graphs (Figure 5) show visual comparisons of multiple The immunoarray requires only a few touch-screen
biomarker concentrations obtained from the multiplexed assays, commands to start and complete the immunoassay automati-
which clearly segregate into the subgroups of cancer-free vs cally, at minimum “RUN” and “Measure ON” commands
cancer patient.73 The central boxes, in blue for cancer-free and (Figure 2). Assay parameters for pump on−off cycles that
black for cancer patients, indicate values that lie in lower to regulate incubation times can be changed with touch-screen
upper quartile (25th to 75th percentile). The horizontal line in commands if required. The automated immunoarray is robust
the each box represents median values, while the vertical lines and can in principle be adapted to sandwich assays for any
and the box ends represent the range from minimum to analyte macromolecule, for which two selective binding agents
maximum concentration. with two distinct epitopes are available, e.g., antibodies or
These results on a limited data set suggest that VEGF-D, aptamers. Good reproducibility was achieved with RSD ranging
CD-14, PF-4, and PSMA differentiate well between prostate from ±1 to ±13% for the eight proteins, with most RSDs less
7574 DOI: 10.1021/acs.analchem.8b01198
Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

than ±8% for well-to-well multiplexed detection and ±8% for In summary, results above demonstrate the fabrication and
array-to-array (n = 3). utility of a novel 3D-printed ECL immunoarray to detect
The cost of fabricating the 3D-printed arrays was about $0.65 multiple proteins simultaneously with ultralow detection limits.
in materials, and the total assay cost for a single multiplexed The new low cost immunoarray with touch-screen control
assay including all the immunoreagents to detect eight proteins enables simple automated operation with low sample volumes
in duplicate was ∼$1.10, or $0.14 per protein. Low volume and fast assays. This system should be well suited to future
reagent chambers (50 μL) minimize the overall cost per assay. clinical and point-of-care diagnostic testing as well as in
Microwells on these detection chips are important to facilitate resource-limited environments, At present, efforts are underway
SWCNT assembly and antibody attachment without cross to analyze a much larger number of patient samples to validate
contamination. The microwells are rapidly fabricated by our the biomarker panel used here for advanced prostate cancer
computer print−heat transfer method.53 (Figure 1B). diagnostics.
3D-printing enabled fast development and optimization of
unique design features like strategically arranged air gap
chambers, vent holes for reproducible liquid loading, and
■
*
ASSOCIATED CONTENT
S Supporting Information
interconnected inlet and outlet ports leading to a detection The Supporting Information is available free of charge on the
channel with low dead volume that streamlines multitask ACS Publications website at DOI: 10.1021/acs.anal-
immunoassay operations. 3D-printing also facilitates reprodu- chem.8b01198.
cible mass production of disposable microfluidic devices. For Includes TEM, SEM, and AFM images for particle size
example, we printed 22 fully functional microfluidic devices at a and sensor characterization; ECL pathway scheme, list of
time on an SLA desktop printer and post-treated them all prostate cancer biomarkers, calibration curves for two
within 6.5 h (∼30 min/array). Thus, when working with protein assays, percentage recovery data from spike and
human samples, diagnostic platforms can be disposable and recovery tests, comparative data for PSA detection from
destroyed after the assay to avoid pathogen contamination ECL array vs ELISA, preliminary statistical analysis of
issues. A new array is readily plugged into the automated sample data, and automation system details (PDF)

■
control system to analyze the next sample. The arrays can also
be reused if desired following an extensive cleaning (see SI).
The automation module features simple electronics with AUTHOR INFORMATION
touch-screen user input that is portable and user-friendly and Corresponding Author
can be configured for several simultaneous assays. Micropumps *E-mail: [email protected].
were programmed with open source Arduino mega and digital ORCID
commands to achieve complete automation without complex James F. Rusling: 0000-0002-6117-3306
software. Power is supplied by a low cost rechargeable lithium
Notes
ion battery allowing up to nine assays with a single recharge. All
The authors declare no competing financial interest.

■
components are commercially available with a total cost of
$210 for a single-micropump system and $375 for a three-
micropump system. An on-board three-electrode potentiostat ACKNOWLEDGMENTS
allows application of precise voltage for ECL generation This work was supported financially by an Academic Plan
powered by the same rechargeable battery. Grant from The University of Connecticut and in part by Grant
Thin, semiflexible pyrolytic graphite sheets (PGS) facilitated no. EB016707 from the National Institute of Biomedical
inclusion of a low cost detection chip. PGS is a unique Imaging and Bioengineering (NIBIB), NIH. The authors thank
alternative to bulk pyrolytic graphite, as they are paper thin, John M. Fikiet from UCONN, School of Engineering
highly conductive, cheap, and withstand organic solvents. After electronics shop for assistance with integration of the touch-
microwells were printed on small cut pieces of PGS, they are screen user interface.
fully compatible with DMF-based assembly of SWCNT forests
that provide nanostructured surfaces for enhanced capture
antibody coverage, which enhances sensitivity.47,54 Detection
■ REFERENCES
(1) Hanash, S. M.; Baik, C. S.; Kallioniemi, O. Nat. Rev. Clin. Oncol.
antibody-decorated RuBPY−SiNP as detection probes provide 2011, 8, 142−150.
0.4 million Ru(bpy)32+ labels per bound protein analyte to (2) National Cancer Institute. Cancer Screening and Early Detection
greatly enhance the ECL response.20,49 Research. https://www.cancer.gov/research/areas/screening (accessed
February 15, 2018).
Accurate detection of the eight low abundance biomarker (3) Cohen, J. D.; Li, L.; Wang, Q.; Thoburn, C.; Afsari, B.; Danilova,
proteins in the presence of thousands of other proteins in L.; Douville, C.; Javed, A. A.; Wong, F.; Mattox, A.; et al. Science 2018,
human serum at concentrations up to mg mL−176 support the 359, 926.
high specificity and selectivity of the 3D-printed immunoarray. (4) Giljohann, D. A.; Mirkin, C. A. Nature 2009, 462, 461−464.
For clinical samples, sample volume is often a big concern but (5) Kulasingam, V.; Diamandis, E. P. Nat. Clin. Pract. Oncol. 2008, 5,
can be as low as 1 μL in our assays. 588−599.
Results for the eight proteins detected in spiked serum and (6) Kingsmore, S. F. Nat. Rev. Drug Discovery 2006, 5, 310−320.
prostate cancer samples show proof of concept for the practical (7) Rusling, J. F.; Kumar, C. V.; Gutkind, J. S.; Patel, V. Analyst 2010,
135, 2496−2511.
clinical utility of the immunoarray. While the number of (8) Dixit, C. K.; Kadimisetty, K.; Otieno, B. A.; Tang, C.; Malla, S.;
samples analyzed is much too small to draw definitive Krause, C. E.; Rusling, J. F. Analyst 2016, 141, 536−547.
conclusions, these early results (Figures 5, S4,S5, and Table (9) Rusling, J. F. Anal. Chem. 2013, 85, 5304−5310.
S3) also suggest that there may be a significant future role in (10) Ward, A. M.; Catto, J.; Hamdy, F. Ann. Clin. Biochem. 2001, 38,
prostate cancer diagnostics for this biomarker panel. 633−651.

7575 DOI: 10.1021/acs.analchem.8b01198

Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

(11) Srinivas, P. R.; Kramer, B. S.; Srivastava, S. Lancet Oncol. 2001, (41) Xu, Y.; Wu, X.; Guo, X.; Kong, B.; Zhang, M.; Qian, X.; Mi, S.;
2, 698−704. Sun, W. Sensors 2017, 17, 1166−2003.
(12) Beveridge, J. S.; Stephens, J. R.; Williams, M. E. Annu. Rev. Anal. (42) Bishop, G. W.; Satterwhite-Warden, J. E.; Bist, I.; Chen, E.;
Chem. 2011, 4, 251−273. Rusling, J. F. ACS sensors 2016, 1, 197−202.
(13) Roche Diagnostics. Roche Diagnostics USA Home Page. (43) Kadimisetty, K.; Malla, S.; Rusling, J. F. ACS Sensors 2017, 2,
https://usdiagnostics.roche.com/en/core_laboratory/reagents/full- 670−678.
menu.html (accessed March 15, 2018). (44) Kadimisetty, K.; Mosa, I. M.; Malla, S.; Satterwhite-Warden, J.
(14) Meso Scale Diagnostics. Meso Scale Diagnostics Home Page. E.; Kuhns, T. M.; Faria, R. C.; Lee, N. H.; Rusling, J. F. Biosens.
http://www.mesoscale.com (accessed March 15, 2018). Bioelectron. 2016, 77, 188−193.
(15) Perkin Elmer. Protein Detection. http://www.perkinelmer.com/ (45) Rusling, J. F.; Kadimisetty, K.; Malla, S.; Bishop, G. W.;
category/protein-detection (accessed March 15, 2018). Satterwhite-Warden, J. E. Low Cost 3D-Printed Biosensor Arrays for
(16) Meissner, E. G.; Decalf, J.; Casrouge, A.; Masur, H.; Kottilil, S.; Protein-based Cancer Diagnostics based on Electrochemilumines-
Albert, M. L.; Duffy, D. PLoS One 2015, 10 (7), e0133236. cence. Proceedings of 9th International Joint Conference on
(17) Kelley, S. O.; Mirkin, C. A.; Walt, D. R.; Ismagilov, R. F.; Toner, Biomedical Engineering Systems and Technologies, Rome, Italy,
M.; Sargent, E. H. Nat. Nanotechnol. 2014, 9, 969−980. February 21−23, 2016; pp 17−22; https://dl.acm.org/citation.
(18) Rusling, J. F.; Bishop, G. W.; Doan, N.; Papadimitrakopoulos, F. cfm?id=3093393.3093394.
J. Mater. Chem. B 2014, 2, 12−30. (46) Chikkaveeraiah, B. V.; Mani, V.; Patel, V.; Gutkind, J. S.;
(19) Tak For Yu, Z.; Guan, H.; Ki Cheung, M.; McHugh, W. M.; Rusling, J. F. Biosens. Bioelectron. 2011, 26, 4477−4483.
Cornell, T. T.; Shanley, T. P.; Kurabayashi, K.; Fu, J. Sci. Rep. 2015, 5, (47) Malhotra, R.; Patel, V.; Chikkaveeraiah, B. V.; Munge, B. S.;
11339. Cheong, S. C.; Zain, R. B.; Abraham, M. T.; Dey, D. K.; Gutkind, J. S.;
(20) Kadimisetty, K.; Malla, S.; Sardesai, N. P.; Joshi, A. A.; Faria, R. Rusling, J. F. Anal. Chem. 2012, 84, 6249−6255.
C.; Lee, N. H.; Rusling, J. F. Anal. Chem. 2015, 87, 4472−4478. (48) Krause, C. E.; Otieno, B. A.; Bishop, G. W.; Phadke, G.;
(21) Fragoso, A.; Latta, D.; Laboria, N.; von Germar, F.; Hansen- Choquette, L.; Lalla, R. V.; Peterson, D. E.; Rusling, J. F. Anal. Bioanal.
Hagge, T. E.; Kemmner, W.; Gartner, C.; Klemm, R.; Drese, K. S.; Chem. 2015, 407, 7239−7243.
O’Sullivan, C. K. Lab Chip 2011, 11, 625−631. (49) Kadimisetty, K.; Malla, S.; Sardesai, N. P.; Joshi, A. A.; Faria, R.
(22) Tang, C. K.; Vaze, A.; Rusling, J. F. Lab Chip 2017, 17, 484− C.; Lee, N. H.; Rusling, J. F. Anal. Chem. 2015, 87, 4472−4478.
489. (50) Sardesai, N.; Pan, S.; Rusling, J. Chem. Commun. 2009, 4968−
(23) Gross, B. C.; Erkal, J. L.; Lockwood, S. Y.; Chen, C.; Spence, D. 4970.
M. Anal. Chem. 2014, 86, 3240−3253. (51) Walker, J. M. Basic Protein Peptide Protocols; Methods in
(24) Bishop, G. W.; Satterwhite-Warden, J. E.; Kadimisetty, K.; Molecular Biology; Springer, 1994; pp 5−8.
Rusling, J. F. Nanotechnology 2016, 27 (8pp), 284002. (52) Basic Properties and Application Examples of PGS Graphite
(25) Ho, C. M. B.; Ng, S. H.; Li, K. H. H.; Yoon, Y.-J. Lab Chip 2015, Sheet. http://www.mouser.com/pdfdocs/thermalgraphitesheets.pdf
15, 3627−3637. (accessed March 15, 2018.)
(26) O’Neill, P. F.; Ben Azouz, A.; Vázquez, M.; Liu, J.; Marczak, S.; (53) Tang, C. K.; Vaze, A.; Rusling, J. F. Lab Chip 2012, 12, 281−
Slouka, Z.; Chang, H. C.; Diamond, D.; Brabazon, D. Biomicrofluidics 286.
2014, 8, 052112. (54) Malhotra, R.; Papadimitrakopoulos, F.; Rusling, J. F. Langmuir
(27) He, Y.; Wu, Y.; Fu, J.; Gao, Q.; Qiu, J. Electroanalysis 2016, 28, 2010, 26, 15050−15056.
1658−1678. (55) Sardesai, N. P.; Kadimisetty, K.; Faria, R.; Rusling, J. F. Anal.
(28) Whitesides, G. M.; Ostuni, E.; Takayama, S.; Jiang, X.; Ingber, Bioanal. Chem. 2013, 405, 3831−3838.
D. E. Annu. Rev. Biomed. Eng. 2001, 3, 335−373. (56) American Cancer Society. Key Statistics for Prostate Cancer.
(29) Duffy, D. C.; McDonald, J. C.; Schueller, O. J.; Whitesides, G. https://www.cancer.org/cancer/prostate-cancer/about/key-statistics.
M. Anal. Chem. 1998, 70, 4974−4984. html (accessed March 15, 2018).
(30) Qin, D.; Xia, Y.; Whitesides, G. M. Nat. Protoc. 2010, 5, 491− (57) Crawford, E. D.; Leewansangtong, S.; Goktas, S.; Holthaus, K.;
502. Baier, M. Prostate 1999, 38, 296−302.
(31) Tumbleston, J. R.; Shirvanyants, D.; Ermoshkin, N.; (58) Hernández, J.; Thompson, I. M. Cancer 2004, 101, 894−904.
Janusziewicz, R.; Johnson, A. R.; Kelly, D.; Chen, K.; Pinschmidt, (59) Kumar-Sinha, C.; Tomlins, S. A.; Chinnaiyan, A. M. Nat. Rev.
R.; Rolland, J. P.; Ermoshkin, A.; Samulski, E. T.; DeSimone, J. M. Cancer 2008, 8, 497−511.
Science 2015, 347, 1349−1352. (60) Oon, S. F.; Pennington, S. R.; Fitzpatrick, J. M.; Watson, R. W.
(32) Macdonald, N. P.; Cabot, J. M.; Smejkal, P.; Guijt, R. M.; Paull, Nat. Rev. Urol. 2011, 8, 131−138.
B.; Breadmore, M. C. Anal. Chem. 2017, 89, 3858−3866. (61) Kumar-Sinha, C.; Tomlins, S. A.; Chinnaiyan, A. M. Nat. Rev.
(33) Coskun, A. F.; Wong, J.; Khodadadi, D.; Nagi, R.; Tey, A.; Cancer 2008, 8, 497−511.
Ozcan, A. Lab Chip 2013, 13, 636−640. (62) Oon, S. F.; Fanning, D. M.; Fan, Y.; Boyce, S.; Murphy, T. B.;
(34) Coskun, A. F.; Nagi, R.; Sadeghi, K.; Phillips, S.; Ozcan, A. Lab Fitzpatrick, J. M.; Watson, R. W. Prostate 2012, 72, 1523−1531.
Chip 2013, 13, 4231−4238. (63) Dawson, D. W.; Volpert, O. V.; Gillis, P.; Crawford, S. E.; Xu,
(35) Wei, Q.; Nagi, R.; Sadeghi, K.; Feng, S.; Yan, E.; Ki, S. J.; Caire, H.-J.; Benedict, W.; Bouck, N. P. Science 1999, 285, 245−248.
R.; Tseng, D.; Ozcan, A. ACS Nano 2014, 8, 1121−1129. (64) Kaushal, V.; Mukunyadzi, P.; Dennis, R. A.; Siegel, E. R.;
(36) Wei, Q.; Luo, W.; Chiang, S.; Kappel, T.; Mejia, C.; Tseng, D.; Johnson, D. E.; Kohli, M. Clin. Cancer Res. 2005, 11, 584−593.
Chan, R. Y. L.; Yan, E.; Qi, H.; Shabbir, F.; Ozkan, H.; Feng, S.; (65) Chan, J. M.; Stampfer, M. J.; Giovannucci, E.; Ma, J.; Pollak, M.
Ozcan, A. ACS Nano 2014, 8, 12725−12733. Growth Horm. IGF Res. 2000, 10 (Suppl 1), S32−S33.
(37) Roda, A.; Michelini, E.; Cevenini, L.; Calabria, D.; Calabretta, (66) Butkus de Aguiar, B.; Girardi, I.; D'Avila Paskulin, D. D.; de
M. M.; Simoni, P. Anal. Chem. 2014, 86, 7299−7304. Franca, E.; Dornelles, C.; Suparregui Dias, F.; Bonorino, C.; Sampaio
(38) Erkal, J. L.; Selimovic, A.; Gross, B. C.; Lockwood, S. Y.; Alho, C. Immunol. Invest. 2008, 37, 752−769.
Walton, E. L.; McNamara, S.; Martin, R. S.; Spence, D. M. Lab Chip (67) Xu, J.; Zheng, S. L.; Isaacs, S. D.; et al. Proc. Natl. Acad. Sci. U. S.
2014, 14, 2023−2032. A. 2010, 107, 2136−2140.
(39) Bishop, G. W.; Satterwhite, J. E.; Bhakta, S.; Kadimisetty, K.; (68) Varambally, S.; Laxman, B.; Mehra, R.; Cao, Q.; Dhanaskearan,
Gillette, K. M.; Chen, E.; Rusling, J. F. Anal. Chem. 2015, 87, 5437− S. M.; Tomlins, S. A.; Granger, J.; Vellaichamy, A.; Sreekumar, A.; Yu,
5443. J.; et al. Neoplasia 2008, 10, 1285−1294.
(40) Hinman, S. S.; McKeating, K. S.; Cheng, Q. Anal. Chem. 2017, (69) Chikkaveeraiah, B. V.; Bhirde, A.; Malhotra, R.; Patel, V.;
89, 12626−12630. Gutkind, J. S.; Rusling, J. F. Anal. Chem. 2009, 81, 9129−9134.

7576 DOI: 10.1021/acs.analchem.8b01198

Anal. Chem. 2018, 90, 7569−7577
Analytical Chemistry Article

(70) Findlay, J. W.; Dillard, R. F. AAPS J. 2007, 9, E260−E267.

(71) Shah, V. P.; Midha, K. K.; Findlay, J. W.; Hill, H. M.; Hulse, J.
D.; McGilveray, I. J.; McKay, G.; Miller, K. J.; Patnaik, R. N.; Powell,
M. L.; et al. Pharm. Res. 2000, 17, 1551−1557.
(72) Lilja, H.; Ulmert, D.; Vickers, A. J. Nat. Rev. Cancer 2008, 8,
268−278.
(73) McGill, R.; Tukey, J. W.; Larsen, W. A. Am. Stat. 1978, 32, 12−
16.
(74) Hajian-Tilaki, K. Casp. J. Intern. Med. 2013, 4, 627−635.
(75) Krause, C. E.; Otieno, B. A.; Latus, A.; Faria, R. C.; Patel, V.;
Gutkind, J. S.; Rusling, J. F. ChemistryOpen 2013, 2, 141−145.
(76) Pieper, R.; Gatlin, C. L.; Makusky, A. J.; Russo, P. S.; Schatz, C.
R.; Miller, S. S.; Su, Q.; McGrath, A. M.; Estock, M. A.; Parmar, P. P.;
et al. Proteomics 2003, 3, 1345−1364.

7577 DOI: 10.1021/acs.analchem.8b01198

Anal. Chem. 2018, 90, 7569−7577