Single-cell proteomic chip for profiling intracellular signaling pathways in single tumor cells.

Abstract

We describe a microchip designed to quantify the levels of a dozen cytoplasmic and membrane proteins from single cells. We use the platform to assess protein-protein interactions associated with the EGF-receptor-mediated PI3K signaling pathway. Single-cell sensitivity is achieved by isolating a defined number of cells (n = 0-5) in 2 nL volume chambers, each of which is patterned with two copies of a miniature antibody array. The cells are lysed on-chip, and the levels of released proteins are assayed using the antibody arrays. We investigate three isogenic cell lines representing the cancer glioblastoma multiforme, at the basal level, under EGF stimulation, and under erlotinib inhibition plus EGF stimulation. The measured protein abundances are consistent with previous work, and single-cell analysis uniquely reveals single-cell heterogeneity, and different types and strengths of protein-protein interactions. This platform helps provide a comprehensive picture of altered signal transduction networks in tumor cells and provides insight into the effect of targeted therapies on protein signaling networks.

Full text

Page 1

Single-cell proteomic chip for profiling intracellular
signaling pathways in single tumor cells
Qihui Shia,b, Lidong Qina,b, Wei Weia,c, Feng Gengd, Rong Fane, Young Shik Shina,b, Deliang Guod, Leroy Hooda,f,
Paul S. Mischela,g,h, and James R. Heatha,b,h,1
aNanosystems Biology Cancer Center,
Pasadena, CA 91125;
Columbus, OH 43210;
and
bDivision of Chemistry and Chemical Engineering, and
dDepartment of Radiation Oncology, Arthur G. James Comprehensive Cancer Center, Ohio State University Medical School,
eDepartment of Biomedical Engineering, Yale University, New Haven, CT 06520;
gDepartments of Pathology and Laboratory Medicine, and
cMaterials Science, California Institute of Technology,
fInstitute for Systems Biology, Seattle, WA 98109;
hMolecular and Medical Pharmacology, University of California, Los Angeles, CA 90095
Edited by Chad A. Mirkin, Northwestern University, Evanston, IL, and approved November 17, 2011 (received for review July 6, 2011)
We describe a microchip designed to quantify the levels of a dozen
cytoplasmic and membrane proteins from single cells. We use the
platform to assess protein–protein interactions associated with the
EGF-receptor-mediated PI3K signaling pathway. Single-cellsensitiv-
ity is achieved by isolating a defined number of cells (n ? 0–5) in
2 nL volume chambers, each of which is patterned with two copies
of a miniature antibody array. The cells are lysed on-chip, and the
levels of released proteins are assayed using the antibody arrays.
We investigate three isogenic cell lines representing the cancer
glioblastoma multiforme, at the basal level, under EGF stimulation,
and under erlotinib inhibition plus EGF stimulation. The measured
protein abundances are consistent with previous work, and single-
cell analysis uniquely reveals single-cell heterogeneity, and differ-
ent types and strengths of protein–protein interactions. This plat-
form helps provide a comprehensive picture of altered signal
transduction networks in tumor cells and provides insight into
the effect of targeted therapies on protein signaling networks.
A
from oncogenes is often distributed through multiple pathways
(2), potentially underlying the failure of most such inhibitors
(3). Measuring signal flux through multiple pathways, in response
to signal transduction inhibitors, may help uncover network inter-
actions that contribute to therapeutic resistance and that are not
predicted by analyzing pathways in isolation (4). The cellular and
molecular complexity of a solid tumor microenvironment (5) sug-
gests the need to study signaling in individual cancer cells.
Protein–protein interactions within signaling pathways are of-
ten elucidated by assessing the levels of relevant pathway proteins
in model and tumor-derived cell lines and with various genetic
and molecular perturbations. Such interactions, and the implied
signaling networks, may also be elucidated via quantitative mea-
surements of multiple pathway-related proteins within single cells
(6). At the single-cell level, inhibitory and activating protein–pro-
tein relationships, as well as stochastic (single-cell) fluctuations,
are revealed. However, most techniques for profiling signaling
pathways (7, 8) require large numbers of cells. Single-cell immu-
nostaining (9) is promising, and some flow cytometry (6) techni-
ques are relevant, as discussed below.
We describe quantitative, multiplex assays of intracellular sig-
naling proteins from single cancer cells using a platform called
the single-cell barcode chip (SCBC). The SCBC is simple in con-
cept: A single or defined number of cells is isolated within an
approximately 2 nL volume microchamber that contains an
antibody array (10) for the capture and detection of a panel of
proteins. The SCBC design (11) permits lysis of each individual
trapped cell.
Intracellular staining flow cytometry can assay up to 11 phos-
phoproteins from single cells (6). Our SCBC can profile a similar
size panel, but only for approximately 100 single cells per chip.
Each protein is assayed twice, yielding some statistical assessment
lthough signal transduction inhibitors occasionally offer
clinical benefit for cancer patients (1), signal flux emanating
for each experiment. The SCBC is a relatively simple platform
and only requires a few hundred cells per assay.
We used the SCBC to study signal transduction in glioblastoma
multiforme (GBM), a primary malignant brain tumor (12). GBM
has been genetically characterized, yet the nature of signaling
pathways downstream of key oncogenic mutations, such as epi-
dermal growth factor receptor activating mutation (EGFRvIII)
and phosphatase and tensin homolog (PTEN) tumor suppressor
gene loss associated with receptor tyrosine kinase (RTK)/PI3K
signaling, are incompletely understood (13–15). Single-cell ex-
periments may also help resolve the characteristic heterogeneity
of GBM.
We interrogated 11 proteins directly or potentially associated
with PI3K signaling (see SI Appendix, Methods I) through three
isogenic GBM cell lines: U87 (expressing wild-type p53, mutant
PTEN, andlow levels ofwild-type EGFR, no EGFRvIII) (16,17),
U87 EGFRvIII (U87 cells stably expressing EGFRvIII deletion
mutant), and U87 EGFRvIII PTEN (U87 cells coexpressing
EGFRvIII and PTEN) (18). Fig. 1 diagrams this biology. Each
cell line was investigated under conditions of standard cell
culture, in response to EGF stimulation, and after erlotinib treat-
ment followed by EGF stimulation. The proteins assayed repre-
sented RTKs and proteins signifying activation of PI3K and
MAPK signaling. They were (p- denotes phosphorylation) p-Src,
p-mammalian target of rapamycin (p-mTOR), p-p70 ribosomal
protein S6 kinase (p-p70S6K), p-glycogen synthase kinase-3
(p-GSK-3α/β), p-p38 mitogen activated protein kinase (p-p38α),
p-extracellular regulated kinase (p-ERK), p-c-Jun N-terminal
kinase (p-JNK2), p-platelet derived growth factor receptor β
(p-PDGFRβ), p-vascular endothelial growth factor receptor 2
(p-VEGFR2), tumor protein 53 (P53), and total EGFR.
Results
The SCBC for Quantitative, Multiplex Measurement of Intracellular
Signaling Proteins. The SCBC is comprised of a two-layer micro-
fluidic network (11) (Fig. 2A and SI Appendix, Methods II). Valves
isolate the chip into 120 microchambers for cell compartmenta-
lization, cell lysis, and protein assays (Fig. 2 B and C). Upon cell
loading, each microchamber contains zero to a few cells, which
are counted through the transparent chip. Cells are lysed via dif-
fusion of lysis buffer from the neighboring chamber (Fig. 2C and
SI Appendix, Methods II). Capturing the (transient) levels of phos-
phorylated proteins is a key objective. After testing literature re-
cipes (6–9) (see SI Appendix, Fig. S1), a protocol was developed.
Author contributions: Q.S., L.Q., and J.R.H. designed research; Q.S., L.Q., F.G., R.F., Y.S.S.,
and D.G. performed research; Q.S., W.W., L.H., P.S.M., and J.R.H. analyzed data; and Q.S.,
W.W., L.H., P.S.M., and J.R.H. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
1To whom correspondence should be addressed. E-mail: heath@caltech.edu.
This article contains supporting information online at www.pnas.org/lookup/suppl/
doi:10.1073/pnas.1110865109/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1110865109 PNAS ∣ January 10, 2012 ∣ vol. 109 ∣ no. 2 ∣ 419–424
ENGINEERING
BIOPHYSICS AND
COMPUTATIONAL BIOLOGY

Page 2

Following cell stimulation, the cells were trypsinized for 5 min
and all subsequent steps were done at or near ice temperatures.
The cells were centrifuged, resuspended in PBS, and introduced
into a prechilled SCBC within 15 min of stimulation. Within
30 min of stimulation, the cells had been lysed. The lysis buffer
contains phosphatase and protease inhibitors.
The SCBC antibody arrays begin as 20-μm-wide DNA bar-
codes (10). After SCBC assembly, they are converted into anti-
body barcodes using DNA-encoded antibody libraries (Fig. 2C
and SI Appendix, Methods I). Captured proteins are developed
by applying biotinylated detection antibodies and fluorophore-
labeled streptavidin. A barcode contains 11 antibody stripes,
an alignment stripe, and a control stripe. All ssDNA and antibody
reagents are provided in SI Appendix, Table S1. The uniformity of
the DNA barcodes was evaluated through the use of fluorophore-
labeled complimentary DNAs. The barcodes exhibited uniform
DNA loading, with coefficients of variation ≤11% (SI Appendix,
Methods III).
The antibody pairs were selected to detect only phosphory-
lated proteins (excepting p53 and EGFR). The barcode protein
assays exhibited sensitivities and dynamic ranges comparable to
the commercial ELISAs. See SI Appendix, Methods III for calibra-
tion and cross-reactivity data, as well as coefficients of variation
for all antibodies in the barcodes. A 2-h incubation used here
reaches >95% of maximal intensity for all assays (SI Appendix,
Methods III).
For the single-cell assays, experimental variation can also arise
from the location of cells in the microchambers prior to cell lysis,
because of the competition between the antibody/antigen binding
kinetics relative to protein diffusion times. The duplicate barcode
assays in each chamber, coupled with a Monte Carlo simulation,
allowed for estimation of this experimental variation to be
<15% (see SI Appendix, Methods III). The uncertainties are small
compared to the measured variation in protein levels, as de-
scribed below. The biologic variation can be extracted from
CVassay¼
Quantitative protein abundances (copy numbers for a given
protein) are utilized herein. Some standard proteins (used for ca-
librations) were not commercially available (e.g., p-VEGFR2 and
EGFRvIII), and so relative fluorescence intensities are used.
Furthermore, standard proteins may differ from the cellular pro-
teins (e.g., glycosylation levels may vary). Calibrations utilized
standard protein spiked into buffer, whereas SCBC protein levels
are measured in a more complex environment. Background levels
from microchambers containing zero cells, correlations of protein
signal strengths with numbers of cells (SI Appendix, Methods III),
and comparison of SCBC assays with traditional (Western blot-
ting) assays, as well as with flow cytometry (19) have all been mea-
sured to further validate the SCBC assays.
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
CVsystem2þ CVbiological
2
q
.
PI3K Pathway-Associated Protein Assays from Single GBM Cells and
Bulk Cell Lysates. Here we give protocols for stating whether a
protein was detected, and we provide comparisons of the SCBC
protein assays to assays from bulk cell lysate, including literature
results.
Fig. 3 shows heat map data from SCBC experiments on U87
EGFRvIII PTEN cells and from measurements on bulk popula-
tions of those cells. Individual microchamber data are shown in
SI Appendix, Fig. S2. The detection threshold for a given protein
and set of conditions (numbers of cells, cell line, stimulation
conditions) was defined by a signal/noise ðS∕NÞ ≥ 2. The signal
was the average reading from the duplicate barcode assays from
each microchamber, averaged over all experiments for a given set
of conditions. The noise was estimated from the negative control
DNA stripe within each barcode. See SI Appendix, Table S2 for
the calculated average S∕N. For example, for single U87 EGFR-
vIII PTEN cells stimulated with EGF, we observed nine proteins
(S∕N levels are included after each protein name): EGFR (130),
p53 (13), p-VEGFR2 (14), p-ERK (12), p-p38α (11), p-GSK3α/β
(12), p-p70S6K (10), p-mTOR (8), and p-Src (11), thus indicating
that we detect both membrane and cytoplasmic proteins. All 11
assayed proteins were detected in the single-cell experiments, ex-
cept that p-PDGFRβ was detected only at low levels (S∕N ¼ 2)
or not at all; p-JNK2 was only detected consistently in U87
EGFRvIII cells (S∕N of 2–6 for single-cell assays); p-mTOR and
p-70S6K were not detected for U87 EGFRvIII PTEN cells with
erlotinib þ EGF.
Fig. 1.
stitutively activated EGFRvIII. All proteins in light blue with central yellow
background were assayed. Orange background proteins were expressed in
the cell lines U87 EGFRvIII or U87 EGFRvIII PTEN. The oval, yellow background
components are the investigated molecular perturbations.
The PI3K pathway activated by EGF-stimulated EGFR or by the con-
Fig. 2.
layer (red) and the control valve layer (blue) are delineated with food dyes. A
photograph (B) and a drawing (C) of a single microchamber, with critical parts
labeled. A cell is isolated in the cell chamber by the valves. The neighboring
chamber contains cell lysis buffer. The duplicate DNA barcode copies are con-
verted into an antibody array prior to cell loading, counting, and lysis. Also
see SI Appendix, Fig. S1.
The single-cell barcode chip. (A) A photograph of an SCBC. The flow
420
∣
www.pnas.org/cgi/doi/10.1073/pnas.1110865109Shi et al.

Page 3

Fig. 3B shows protein assays measured from a population of
EGF-stimulated U87 EGFRvIII PTEN cells. These assays used
similar cell lysis and assay protocols as the SCBC assays. Compar-
ison across Fig. 3 A and B reveals that the bulk assays and SCBC
measurements are self-consistent. Comparisons between bulk cell
assays (SI Appendix, Fig. S3), SCBC single-cell measurements,
and literature results (18, 20–23) were also done to detect distinct
phosphorylation states of EGFR under the influence of EGFand
erlotinib stimulation. Those results again formed a self-consistent
dataset.
Data, such as shown in Fig. 3, were first averaged to recapitu-
late measurements of proteins from cell populations for compar-
ison with known biology. It was then more fully analyzed to yield a
statistical representation of fluctuations at the single-cell level.
Bulk-Like Protein Profiles Collected from SCBC Data. Fig. 4A presents
the protein abundances (averaged over all three-cell experi-
ments), measured for each cell line and for all conditions (mean
intensities and standard deviations are presented in SI Appendix,
Table S3). We compared these SCBC results with literature find-
ings that used conventional bulk cell assays, as well as with our
own Western blot assays (Fig. 4). In the following discussion,
literature citations following the protein names provide valida-
tion of our SCBC results.
At basal level, U87 cells (Fig. 4A, Top) showed low EGFR
phosphorylation (23) (SI Appendix, Fig. S3) and modest activa-
tion of signaling proteins, including p-Src (22), p-mTOR (24),
p-p70S6K (23, 25), p-GSK3α/β (23, 26, 27), p-p38α (24), and
p-ERK (18, 22, 26), whereas p-JNK2 was not detected (23). U87
EGFRvIII cells (Fig. 4A, Middle) exhibited increased baseline le-
vels of phosphorylation compared with cells expressing wild-type
EGFR, including p-Src (22), p-mTOR, p-p70S6K (25), p-ERK
(18), and p-JNK2 (28). In U87, EGFRvIII PTEN cells (Fig. 4A,
Bottom), PTEN coexpression diminished baseline phosphoryla-
tion of p-Src, p-mTOR, p-p70S6K (25), p-ERK (18), and p-JNK2
compared with U87 EGFRvIII.
EGF stimulation induced EGFR phosphorylation (18, 27, 28)
(SI Appendix, Fig. S3) and promoted downstream pathway acti-
vation in all three cell lines, irrespective of PTEN status, includ-
ing activation of p-p70S6K (25) and p-ERK (18). The increase of
levels of p-ERK in response to EGF stimulation in U87 EGFR-
vIII and U87 EGFRvIII PTEN cells is demonstrated by the Wes-
tern blots shown in Fig. 4A. The level of p-GSK3α/β in response
Fig. 3.
maps of SCBC protein level assays. Each column represents one microchamber
assay; each row represents a protein. (B) Protein assays from a population of
U87 EGFRvIII PTEN cells under EGF stimulation. The contrast of these images
has been equally adjusted, and the intensity of EGFR is divided by five in the
heat maps.
SCBC and bulk cell measurements of U87 EGFRvIII PTEN cells. (A) Heat
Fig. 4.
EGF exposures. (A, Left) Measured protein expression profiles, in fluorescent
intensity units, averaged over the three-cell measurements (n of approxi-
mately 20). The signal of EGFR/EGFRvIII is divided by 10. Results are shown
as mean ? SD. SD represents the combined experimental error and intrinsic
biological variation, but isdominated by thebiological variation. (Right) Wes-
tern blot analysis of p-EGFR, p-ERK, p-mTOR, p-Src, p-p70S6K, p-GSK3β, and
p-Akt expression in U87 EGFRvIII and U87 EGFRvIII PTEN cell lines at the basal,
EGF stimulation, and erlotinib þ EGF treatment states. (B) Heat map of rela-
tive expression fold changes of proteins, normalized by unperturbed U87
cells. (C, Left) Mean fold change of phosphorylation levels of p-ERK and
p-mTOR in different cell lines and conditions, relative to unperturbed U87
cells. (E, EGF; e þ E, erlotinib þ EGF). (Right) Western blot analysis of p-ERK
and p-mTOR expression in response to EGF stimulation in U87 EGFRvIII
and U87 EGFRvIII PTEN cell lines. These cells were cultured in DMEM medium
containing 10% FBS for 24 h, then in serum-free medium for 24 h or (+) er-
lotinib (10 μM) treatment in serum-free medium, followed by stimulation (+)
with EGF (20 ng∕mL) for 15 min. Cells were lysed and the listed proteins were
detected by Western blotting.
Averaged responses for all three cell lines to EGF and erlotinib ðebÞ þ
Shi et al.PNAS
∣
January 10, 2012
∣
vol. 109
∣
no. 2
∣
421
ENGINEERING
BIOPHYSICS AND
COMPUTATIONAL BIOLOGY

Page 4

to EGF stimulation was increased in U87 (27) and U87 EGFR-
vIII cells, but remained relatively unchanged in U87 EGFRvIII
PTEN cells (consistent with Western blots in Fig. 4A).
Erlotinib inhibition þ EGF stimulation diminished phosphor-
ylation of both EGFR and EGFRvIII (18) (SI Appendix, Fig. S3)
relative to EGF stimulation. It led to decreased phosphorylation
levels in U87, although those levels are higher than in the unsti-
mulated cells. One previously identified example of this effect is
p-p70S6K (14, 18). Erlotinib þ EGF showed little impact on U87
EGFRvIII cells, indicating that PTEN loss confers resistance to
EGFR tyrosine kinase inhibitors (14, 21). The phosphoprotein
expression levels decrease, but are above the unstimulated levels.
Representative proteins include p-Src (22) and p-p70S6K (14, 18,
25). Erlotinib significantly diminished phosphorylation levels of
p-ERK, p-p70S6K (14, 18), p-mTOR, and p-Src only for the
U87 EGFRvIII PTEN cells. Those phosphorylation levels are be-
low those observed for unperturbed cells; p-p70S6k and p-mTOR
drop to below the detection limit. These results are consistent
with previous findings that coexpression of EGFRvIII and PTEN
protein by GBM cells is associated with clinical response to
EGFR kinase inhibitor therapy (14).
Fig. 4B shows the heat map of relative mean fold changes in
the expression levels of proteins and phosphoproteins for the dif-
ferent cell lines and conditions, normalized by the protein levels
measured from unperturbed U87 cells. This plot was calculated
as follows. For a microchamber i containing n cells, the fluores-
cence levels recorded from the two barcode assays for a given
protein ρ were averaged to yield ρi;n. The fluorescence intensity
for ρ, averaged over all zero-cell measurements, was subtracted
as background:ðρi;n− ¯ ρ0Þ. This value was then normalized against
the background-subtracted, fluorescence levels of ρ averaged
over all n cell measurements for unperturbed U87:ðρi;n− ¯ ρ0Þ ·
ð¯ ρnU87Þ−1. These fold changes were then averaged over all micro-
chambers containing two to five cells, and were combined to pro-
duce the heat map of Fig. 4B. This map provides a relative
comparison of the pathway activation states in different cell lines
and conditions, but it also emphasizes that the phosphorylation of
p-ERK (representative of MAPK signaling) exhibits correlation
with the phosphorylation of mTOR (PI3K signaling).
Recent work suggests cross-talk between the Rat scarcoma
(Ras)/MAPK and PI3K signaling pathways (3, 29). Recent work
has also uncovered a negative regulatory feedback loop by which
mTOR complex 1 signaling through S6K1 suppresses PI3K-
mediated activation of MAPK activity, so that inhibition of
mTOR signaling through S6K1 can activate MAPK (30, 31). This
implied correlation between PI3K and MAPK signaling can be
estimated by comparing the phosphorylation levels of ERK and
mTOR in varying genetic contexts that regulate PI3K signaling
and in response to ligand stimulation and/or inhibition. The mean
fold changes of p-ERK and p-mTOR in U87 EGFRvIII and U87
EGFRvIII PTEN cell lines are shown in Fig. 4C, Left. In U87
EGFRvIII cells, the fold change of p-ERK under basal level,
EGF stimulation, and erlotinib þ EGF treatment are statistically
lower than that of p-mTOR (SI Appendix, Table S3). However, in
U87 EGFRvIII PTEN cells, the situation is reversed. Obviously,
PTEN expression sensitizes GBM cells to MAPK signaling stimu-
lated by EGF. This preferential activation of MAPK signaling
pathways in response to EGF activation in GBM cells containing
PTEN was validated by immunoblot analysis (Fig. 4C, Right) and
is consistent with recent findings that minimal levels of ERK sig-
naling are required for optimal EGFRvIII-mediated tumor cell
growth in PTEN null glioblastomas (15). These data demonstrate
that SCBC measurements can uncover feedback loops and path-
way cross-talk in situations where the connectivity is less well
defined.
Single-Cell Protein Profiles, Protein–Protein Correlations, and Correla-
tion Networks. Profiles that reveal the relative importance of the
measured biological fluctuations versus the experimental errors
are shown in SI Appendix, Fig. S4 A and B. Two points are rele-
vant for comparing bulk cell assays and single-cell measurements.
SI Appendix, Fig. S4A, which plots P53 intensity versus experi-
ment number, for the sets of one, two, and three cell experiments,
illustrates how a small fraction of cells can dominate an assay. SI
Appendix, Fig. S4B provides histograms of the number of p-ERK
molecules detected, versus frequency of detection, for single U87
EGFRvIII PTEN cells under all three conditions. Those histo-
grams may be compared against the averaged p-ERK intensities
presented at the bottom of Fig. 4A. According to Fig. 4A, the
p-ERK level for the unperturbed cells is only slightly higher than
for the EGF þ erlotinib exposed cells. However, the coefficient of
variation of p-ERK levels is much larger (57%) than in the EGF þ
erlotinib perturbed cells (28%). This effect, which is not captured
in bulk assays, may represent an increased amount of regulation
for p-ERK in the EGF þ erlotinib perturbed cells (32).
The levels of several proteins associated with PI3K signaling
should exhibit coordinated behaviors (6). A typical protein–
protein positive correlation (p-mTOR vs. p-p70S6K for unstimu-
lated U87 EGFRvIII cells) and an anticorrelation (p-GSK3α/β vs.
p-ERK for unstimulated U87 EGFRvIII PTEN) are shown in SI
Appendix, Fig. S4 C and D. The positive correlation is indepen-
dent of the numbers of cells per microchamber assay, whereas the
negative correlation begins to be masked for populations as low
as three cells.
Fig. 5 provides nine SCBC-derived protein correlation net-
works. The line weight defines the strength of the correlation
(see key). We used the Bonferroni method (33), which limits cor-
relations to those that exhibit a p value ≤ 0.05; correlation coef-
ficients above 0.4 or below −0.4 are significant. Perturbation by
ligand stimulation and/or receptor inhibition reveal new relation-
ships and the genetic context of those relationships. EGF stimu-
lation of EGFRvIII-expressing GBM cells greatly enhances
network connectivity in a way that is very different from what
would beexpected from simply summing the effects ofEGF treat-
ment (U87 þ EGF, Fig. 5, Top Center) and EGFRvIII expression
(U87 EGFRvIII, Middle Left). This represents a clinically and
biologically relevant result because wild-type EGFR is always
present in EGFRvIII-expressing cells (14). The greatly enhanced
network interconnectivity for the EGF-stimulated U87 EGFR-
vIII cells may suggest a mechanism underlying the difficulty of
inhibiting downstream signaling in EGFRvIII-expressing, PTEN
null tumor cells, potentially providing one mechanism for their
striking tumorigenicity and their established role in promoting
therapeutic resistance. This observation is consistent with the
clinical failure and the lack of p70S6K inhibition observed in
EGFRvIII-expressing, PTEN-deficient GBM patients treated
with erlotinib (14), and suggests that clinically relevant insights
may potentially be derived from these types of single-cell ex-
periments.
Classical genetics is also often used to combine perturbations
and phenotypic responses to infer functional relationships be-
tween genes (34), but specific interactions are difficult to extract
because intermediate interacting partners may contribute combi-
nations of positive and/or negative interactions.
Discussion
The SCBC provides certain advantages for assaying cytoplasmic
proteins. The ability to normalize protein levels to numbers of
cells permits for the SCBC data torecapitulate qualitative protein
measurements from bulk cell populations, but in a quantitative
fashion. One example relates toward interrogating cross-talk be-
tween the Ras/MAPK and RTK/PI3K signaling in GBM (3, 29,
30). Using the SCBC, we found that, for U87 EGFRvIII PTEN
cells, stimulation with EGF (associated with RTK/PI3K signaling)
led to a sharp increase in levels or p-ERK (associated with the
Ras/MAPK pathway), a result that was confirmed using Western
422
∣
www.pnas.org/cgi/doi/10.1073/pnas.1110865109Shi et al.

Page 5

blot analysis of the bulk cell lines. Exposure of those same cells to
erlotinib þ EGF kept the p-ERK levels near the level of unstimu-
lated U87 EGFR vIII PTEN cells.
A second advantage relates to the assessment of the single-cell
fluctuations, defined by the distribution of the levels of a given
protein, measured across many SCBC assays. The measured bio-
logical variation that arises from the functional heterogeneity of a
genetically identical cell population is significantly higher than
the experimental error and varies across proteins. These fluctua-
tions provide a gauge of the heterogeneity of the cell population
and can be used to predict the thermodynamic stability of specific
proteins toward perturbations (32).
The SCBC barcodes could potentially be expanded to 35–40
proteins, depending upon the availability of antibody pairs, but
even for just 11 intracellular proteins, the correlation networks
extracted from SCBC data already provide interesting parallels
with the tumorigenecity and therapeutic resistance of EGFRvIII
positive, PTEN null tumors. Expanding the protein panel will
permit a more complete mapping of the connectivity between
known GBM signaling pathways and how that connectivity may
be influenced by molecular (i.e., therapeutic) or physical (i.e.,
hypoxia) perturbations. A further significant challenge will be
to extend this platform toward the analysis of clinical specimens.
Materials and Methods
Cell Lines, Antibodies and Regents. The human GBM cell line U87 was pur-
chased from American Tissue Culture Collection. U87 EGFRvIII and U87 EGFR-
vIII PTEN cells were constructed as previously described (14, 18). Cell lines
were routinely maintained in DMEM (American Type Culture Collection) con-
taining 10% fetal bovine serum in a humidified atmosphere of 5% CO2, 95%
air at 37 °C. See SI Appendix, Table S1 for DNA and antibody reagents. Other
reagents were obtained as follows: phosphatase inhibitor cocktail, bovine
serum albumin, and n-dodecyl-β-D-maltoside, Sigma-Aldrich; Cy5-conjugated
straptavidin, eBioscience; human EGF, Prospec; cell lysis buffer, Cell Signaling;
complete protease inhibitor cocktail, Roche.
Microchip Fabrication. The SCBCs were assembled from a DNA barcode micro-
array glass slide and a polydimethylsiloxane (PDMS) slab containing the mi-
crofluidic circuit, as fully described in SI Appendix, Methods II. The PDMS SCBC
chip was fabricated using a two-layer soft lithography, with a control layer
and a flow layer (11). The control layer PDMS chip was aligned onto the flow
layer and bonded for 60 min at 80°C. The two-layer PDMS chip was then cut
off, access holes were drilled, and then it was thermally bonded onto the
barcoded glass slide to yield an SCBC.
Cell Stimulation and Erlotinib Treatment. For EGF stimulation, cells were serum
starved for 24 h and then stimulated by EGF at 50 ng∕mL for 10 min before
harvest. For erlotinib treatment, serum-starved cells were treated with 10 μM
erlotinib for 24 h, followed by EGF stimulation (50 ng∕mL) for 10 min before
harvest. The treated cells were dissociated with trypsin and EDTA and sus-
pended in cold PBS with a concentration of 1,000 cells per microliter prior
to loading to the device.
Cytoplasmic Protein Measurement Using SCBCs. All SCBC microchannels were
blocked with blocking buffer for 60 min. A cocktail of all DNA-antibody con-
jugates was flowed through the channels for 60 min, transforming the DNA
barcode microarrays into antibody microarrays. Unbound conjugates were
removed with washing buffer. Then 3× lysis buffer was loaded into the lysis
buffer chambers, and cells were loaded in the cell chamber while keeping the
valves between these chambers closed. The valves were opened to allow on-
chip diffusion of lysis buffer to the neighboring cell chambers for 30 min on
ice. The SCBC was then incubated 30 min on ice and 1 h at room temperature
with gentle shaking. Cell lysate was quickly removed by washing buffer, fol-
lowed by flowing biotin-labeled detection antibodies and fluorescent dye-
labeled streptavidin for visualization. The barcoded glass slide was then de-
tached for scanning.
Data Analysis and Statistics. Axon GenePix 4400A (Molecular Devices) was
used to obtain the fluorescence images at laser power 80% (635 nm) and
10% (532 nm), optical gains 600 (635 nm) and 400 (532 nm), brightness 80,
and contrast 83. Fluorescence line profiles were generated by ImageJ (Na-
tional Institutes of Health). A custom Excel macroextracted average fluores-
cence signal for all bars within a given barcode and the barcode profiles were
compared to the number of cells using the same program. Heatmap profiles
were generated using Treeview (Stanford), and R software was used for com-
puting Pearson correlation coefficients. Comparisons in phosphorylation le-
vels of mTOR and ERK were performed with a one-tailed t test. P < 0.05 was
considered statistically significant.
Fig. 5.
(p ¼ 0.05). Underlined proteins are below the detection limit.
Protein correlation maps under different genetic and environmental perturbations. All indicated correlations pass a Bonferroni corrected p-value test
Shi et al.PNAS
∣
January 10, 2012
∣
vol. 109
∣
no. 2
∣
423
ENGINEERING
BIOPHYSICS AND
COMPUTATIONAL BIOLOGY