Polyfunctional responses by human T cells result
from sequential release of cytokines
Qing Hana,1, Neda Bagherib,1, Elizabeth M. Bradshawc, David A. Haflerd,e, Douglas A. Lauffenburgerb,e,
and J. Christopher Lovea,e,2
aDepartment of Chemical Engineering, Koch Institute for Integrative Cancer Research, andbDepartment of Biological Engineering, Massachusetts Institute
of Technology, Cambridge, MA 02139;cBrigham and Women’s Hospital, Boston, MA 02115;dDepartment of Neurology and Immunobiology, Yale University,
New Haven, CT 06520; andeThe Eli and Edythe L. Broad Institute, Cambridge, MA 02142
Edited by Robert L. Coffman, Dynavax Technologies, Berkeley, CA, and approved November 14, 2011 (received for review October 18, 2011)
The release of cytokines by T cells defines a significant part of their
functional activity in vivo, and their ability to produce multiple
cytokines has been associated with beneficial immune responses.
To date, time-integrated end-point measurements have obscured
whether these polyfunctional states arise from the simultaneous
or successive release of cytokines. Here, we used serial, time-
dependent, single-cell analysis of primary human T cells to resolve
the temporal dynamics of cytokine secretion from individual cells
after activation ex vivo. Weshowthatmultifunctional,Th1-skewed
cytokine responses (IFN-γ, IL-2, TNFα) are initiated asynchronously,
but the ensuing dynamic trajectories of these responses evolve
programmatically in a sequential manner. That is, cells predomi-
nantly release one of these cytokines at a time rather than maintain
active secretion of multiple cytokines simultaneously. Furthermore,
these dynamic trajectories are strongly associated with the various
activities of many individual T cells contribute to sustained, popula-
tion-level responses. The trajectories of responses by single cells
may also provide unique, time-dependent signatures for immune
monitoring that are less compromised by the timing and duration
of integrated measures.
diseases (1). Determining their characteristic diversity remains
a central goal for defining immunological signatures that indicate
the status of human diseases or responses to interventions like
vaccines (2). T cells are typically classified by their state of dif-
ferentiation based on surface-expressed glycoproteins (e.g., CD3,
CD8, CD45RA, CCR7) (3) and then assigned a functional state
(e.g., Th1, Th2, Th17) based on their ability to produce one or
more cytokines within specific groups (4). Efforts to improve
immune monitoring have focused on understanding the pheno-
types and functions that reflect effective T-cell responses to dis-
eases and clinical interventions, but these correlations have
remained imperfect thus far.
Both the magnitude and quality of a T-cell response are con-
sidered important metrics in evaluating the efficacy of an immune
the magnitude, whereas the nature and diversity of the functional
responses has been associated with measures of quality. These
functions include releasing one or more cytokines that induce
proliferation, modulate inflammation, mediate cytolysis of other
cells, and inhibit viral replication (1). The production of multiple
cytokines by T cells has been associated with productive immune
responses to infectious diseases (5–7) and to vaccines (8–10).
The manner in which polyfunctional responses by individual
cells contribute to the evolution of an immune response at a
population level is not well understood. The types and concen-
trations of cytokines in the extracellular milieu, and percentages of
cells play a significant role in adaptive immune responses to
infectious diseases and in the pathogenesis of inflammatory
Indeed, the production of both IL-2 and IFN-γ by CD4+T cells
in vivo has been shown to begin within hours of stimulation and
wane after 16–18 h (13, 14). It has not been possible, however,
to determine whether cells release multiple cytokines simulta-
neously, or sequentially, in time because techniques such as in-
tracellular cytokine staining (ICS) and multiparametric ELISpot
provide only integrative, endpoint measures (15–18). Therefore,
resolving when activated T cells initiate the release of cytokines,
and how their responses evolve in time, should provide funda-
mental insight into how individual cells dynamically modulate in-
tercellular signals to affect population-level responses toward
pathological conditions or clinical interventions.
Here we examine how the synchrony and evolution of secreted
cytokines varies upon activation among different subsets of pri-
mary human CD3+T cells isolated from peripheral blood. Using
a combination of imaging cytometry and quantitative single-cell
analysis of secreted cytokines, we monitored the release of three
Th1-skewed cytokines (IFN-γ, IL-2, and TNFα) over time. We
find that T cells initiate the release of cytokines at different
points in time upon stimulation. Furthermore, most of these cells
initiate secretion in a monofunctional state. Computational
analyses of these data indicate that the simultaneous release of
the measured cytokines is short-lived and that cells follow pro-
grammatic, rather than random, patterns of release. Moreover,
T-cell receptor (TCR)-dependent activation does not change the
nature of these trajectories. Finally, we present evidence that
these trajectories, rather than initial time of secretion or the
overall integrated response, associate closely with the differen-
tiated state of the cell. Together, observations of distributed
activation and evolving release suggest how single T cells may
use time-dependent mechanisms to evolve population-level
responses and how dynamic monitoring of immune cells may
improve profiling functional responses associated with immune
status relative to integrated, endpoint measurements (19).
Serial Microengraving Quantifies Dynamic Rates of Cytokine Secretion
from Single T Cells. We used a dense array of subnanoliter wells
(nanowells) to isolate CD3+T cells from the peripheral blood of
approximately onecellperwell, we imaged each wellby automated
fluorescence microscopy to determine the occupancy, viability, and
Author contributions: Q.H., N.B., and J.C.L. designed research; Q.H., N.B., and E.M.B.
performed research; Q.H., N.B., D.A.L., and J.C.L. analyzed data; and Q.H., N.B., D.A.H.,
D.A.L., and J.C.L. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
See Commentary on page 1359.
1Q.H. and N.B. contributed equally to this work.
2To whom correspondence should be addressed. E-mail: email@example.com.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
| January 31, 2012
| vol. 109
| no. 5
CCR7) (Fig. 1A and Fig. S1). Each array comprised 84,672 wells
and yielded over 10,000 individual cells per experiment.
The cells in the array were then activated by a combination of
phorbol 12-myristate 13-acetate (PMA) and ionomycin (20). We
chose this TCR-independent stimulation initially to probe a broad
range of responses. We used serial microengraving—a nonde-
structive, quantitative method for sampling the supernatants from
each cell in the array (SI Text) (21)—to measure the paracrine
rates of secretion for three cytokines associated with Th1-skewed
responses (IFN-γ, IL-2, and TNFα) every 2 h during sustained
stimulation (Fig. 1A). This process generates 30 data points per
cell over 16 h that describe the temporal release of cytokines from
eight distinct CD3+T-cell subsets (Fig. 1B and Figs. S1 and S2).
The data collected from activated, viable single cells revealed
that the secretion of cytokines occurred in a dynamic and het-
erogeneous manner (Fig. 1C). Aggregation of the responses
showed the numbers of cells secreting IL-2 and IFN-γ increased
over time, whereas those secreting TNFα diminished (Fig. 2A
and Fig. S3 A–D). The secretion of IL-2 was also more prevalent
among CD8−T cells than CD8+T cells. Together, these basic
trends were consistent with those determined by bulk analysis of
the order and timing of cytokine responses using ICS and ELISA
(12, 22). The mean rates of secretion measured for each cytokine
by microengraving were also consistent with those estimated
from a combination of ICS and Luminex (Fig. S4A).
Initial Release of Cytokines Varies Temporally Among Activated T
Cells. One interesting observation was that new cohorts of T cells
sharply transitioned into active secretion during each sampling
period. These asynchronous, apparently stochastic, events were
biphasic, concentrated at ≈2–6 h and 12–16 h (Fig. 2B). In-
terestingly, there was no statistical association between the tim-
ing of initiation and specific subsets, although memory T cells
(CD45RA−) tended to respond most often during the first burst
of activity (Fig. S3E and Table S1).
To confirm that these variations were not influenced by isolated
activation, we also measured secretion from cells stimulated in
a bulk culture. The frequencies of secreting cells were consistent
whenmeasuredbymicroengraving6–7 h after stimulationeither in
bulk (10.7%) or on-chip (13.6%). These frequencies were also
smaller than the accumulated numbers of cells measured by ICS
over 7 h, further supporting temporal variance in activation (Fig.
S4B). The diversity of differentiated subsets present ex vivo could
also impact the observed timing of initial release so we also mea-
sured the secretion from an in vitro-expanded human T-cell clone
(Fig. S5). These cells exhibited similar asynchronous release of
levels of kinases, transcription factors, and other signaling proteins
(ERK, NFAT, SHP-1) (23, 24), along with slow epigenetic events
such as chromatin remodeling near transcription factor binding
sitesthat promote production ofcytokines (25, 26), may contribute
to these observed temporal distributions of initial release.
Simultaneous Release of Multiple Cytokines Is Transient. Although
ICS enumerates cells that produce multiple cytokines over time,
it cannot reveal the lifetime, persistence, or concomitance of
these productions. Our experiments here demonstrated that
most cells (≈90%) first initiate secretion in a monofunctional
manner, releasing only a single cytokine. Further, the frequency
single T cells in time. (B) Representative micrographs of data evaluating viability (Calcein and SYTOX); phenotype (CD8, CD45RA, and CCR7); and TNFα (blue),
IL-2 (red), and IFN-γ (green) secretion. Blue squares outline positive events. (C) Cytokine secretion kinetics of 3,015 viable T cells. Each row within each block
reflects the dynamic activity of an individual T cell over time. The color wheel illustrates the type and relative magnitude of secreted cytokines; inactivity is
black. Block columns and block rows organize cytokine profiles by initial time of activity and differentiated cell types, respectively. Kinetic profiles are ordered
within each block according to cytokine function. These data are representative from three independent experiments.
Cytokine secretion dynamics for individual T cells upon activation. (A) Illustration of serial microengraving to monitor cytokine secretion by viable
| www.pnas.org/cgi/doi/10.1073/pnas.1117194109Han et al.
of multifunctional states during any single sampling period was
significantly lower than that seen by integrating these data across
time (Fig. 2A and Fig. S3F). These observations suggest T cells
are more likely to secrete multiple cytokines sequentially than
We found that cells secreting multiple cytokines simulta-
neously were more likely to change their functional states than
those releasing individual cytokines (Fig. 2C). Only cells that
secreted IFN-γ or IL-2 showed significant persistence of their
functional states. The average lifetimes of states in which two or
more cytokines were secreted simultaneously were 1.5- to 2 fold
shorter than those of IFN-γ or IL-2 alone (Fig. 2D). Cells that
initiated secretion within 4 h of stimulation were more likely to
produce multiple cytokines in total, either simultaneously or
sequentially (Fig. S3G). Together, these results demonstrate that
the simultaneous secretion of two or more Th1-associated
cytokines likely occurs as a transition between states, and that
the secretory responses by T cells evolve dynamically during
sustained, TCR-independent activation.
Multifunctional Th1 Responses Evolve with Time. To identify the
most common transitions among functional states, we quantified
the likelihood that a cell in a secretory state at time tNwould
transition to another state 2 h later, tN+2h(Fig. 3A). The most
probable outcomes observed here were either retaining the
current state or downgrading the number of cytokines secreted.
For example, the release of TNFα in combination with either
IFN-γ or IL-2 commonly resolved to the secretion of IFN-γ or
IL-2 alone. These analyses further confirm that cytokine secre-
tion by individual cells occurs in a predominantly sequential
manner, with multifunctional release arising as a transient state.
We then computed the corresponding Z scores for these state
transitions relative to randomly permuted datasets to evaluate
whether certain transitions occurred more or less commonly than
expected by chance (Fig. 3B). As anticipated, persistence of in-
dividual secretory states was significant, confirming that cells
actively sustain specific functional states. Other transitions were
significantly underrepresented. For instance, observed transi-
tions between IFN-γ and IL-2 occurred less frequently than
expected by chance. This result is consistent with observations
that IFN-γ expression, controlled by the transcription factor T-
bet, suppresses the bulk production of IL-2 by lymphoma cells
activated by PMA/ionomycin (27). We anticipate that identifying
dominant individual-cell secretory transitions may offer new
insights on the regulation of cytokine signaling and provide a
means to predict T-cell responses.
T Cells Exhibit Programmatic Trajectories of Cytokine Secretion. The
global transition matrices suggested that the trajectories of se-
cretory states among cells evolve with identifiable, deterministic
programs, rather than stochastic or idiosyncratic courses. That is,
the set of trajectories observed is small relative to the number of
possible trajectories for the three cytokines (>107). To test this
hypothesis, we investigated the cytokine trajectories that emerged
from clustering the first three time-aligned data points by self-
organizing maps (SOMs) (Figs. S6 and S7). For each CD8−T-cell
subset, the optimal number of clusters was determined by eval-
uating the explained variance (elbow criterion) (28) (Fig. S7A).
Metaclusters were determined by further SOMs and qualitative
alignment of similar clusters (Fig. 3C and Fig. S7 B and C).
The dominant trajectories exhibited either persistent secretion
of individual cytokines (e.g., IL-2, IFN-γ) or a transition from
a single functional state to another (e.g., TNFα → IL-2; TNFα →
IFN-γ). Memory T cells (CD45RA−) used the most diverse
sets of states, with a small, but significant, bias toward TNFα
secreting states among the effector memory (CCR7−) cells,
whereas CD45RA+cells predominantly exhibited a short burst
of IFN-γ. These results support models for T-cell differentiation
where T cells maintain transient memory for gene expression
resulting from chromatin remodeling (26), and also suggest that
some subsets of T cells from all differentiated populations can
release limited bursts of IFN-γ within 2 h of initial activation.
Dynamic Cytokine Trajectories Can Discriminate T-Cell Subsets. We
next considered whether the observed sequential cytokine tra-
jectories could distinguish different subsets of cells (effector
memory, central memory, effector, and naïve) more effectively
than time-integrated data, which may fail to resolve differences
in how multifunctional responses are reached. Using principal
component analysis (PCA), in combination with feature selec-
tion, we identified unique subspaces that best segregated subsets
in specified training data. These subspaces were subsequently
used to classify cells based on their dynamic cytokine profiles.
Using raw CD8−T-cell data, we could discriminate among the
four subsets more accurately (41 ± 1%, percent correct classifi-
cation) than random assignment (25%) (Fig. 3D). Integrating the
data over 6 h (i.e., approximating ICS) reduced the accuracy of
classification to 33 ± 1%. Remarkably, aligning the trajectories
total percentage of actively secreting cells as a function of time and class
(CD3+CD8−/CD3+CD8+), and their corresponding state distributions. The
rightmost bar represents the integrated responses over 2–16 h. n is the total
number of cells in the given dataset. (B) Histograms of the total percentage
of cells that initiate cytokine secretion at each time point, and their corre-
sponding state distributions. The early and late responses were fit to
Gaussian curves (t, mean time). (C) Bar graph of the probability that a given
cell preserves its functional state in 16 h. The secretory states for cells at 4, 6,
and 8 h were used as independent reference points. Error bars indicate SD.
(D) Heatmap indicating lifetime of secretory states. Solid orange lines reflect
mean lifetime; orange dots reflect the mean plus 1 SD; white dashed line
reflects the lower limit of detection, 2 h. All subtypes of T cells were included
in C and D.
Distribution and stability of secretory states. (A) Histograms of the
Han et al. PNAS
| January 31, 2012
| vol. 109
| no. 5
of individual cells according to the initial time of cytokine release
dramatically improved the accuracy of classification to 58 ± 4%.
We also found that classification using time-aligned data im-
proved monotonically with the temporal length of the trajecto-
ries, especially for naïve and effector cells (Fig. 3E and Fig. S8A).
Sensitivity analysis for the binary classification of subsets (based
on receiver operating characteristic curves) confirmed that ef-
fector memory and central memory cells were challenging to
discriminate based on their functional profiles (Fig. S8B), sug-
gesting that there are limited differences between the ranges of
dynamic cytokine responses for these two subsets, and that local
microenvironments along with receptor-mediated signaling likely
modulate context-specific responses. Further resolution of the
phenotypic diversity within memory cells may also improve their
TCR-Dependent Activation Induces Similar Programmed Responses.
Whereas the stimulation of T cells in a TCR-independent
manner provided a view of the accessible trajectories of secretory
states, activation of T cells in vivo occurs through the engage-
ment of TCRs with cognate antigens presented in class I or II
major histocompatibility complexes (MHC) and costimulatory
molecules such as CD28 (29). To determine whether the dy-
namics of cytokine secretion after PMA/ionomycin stimulation
were consistent with TCR-dependent stimulation, we compared
the responses of primary T cells subjected to each condition
during the period in which all functional states and transitions
were most prevalent (2–10 h). We coincubated CD3+T cells
with beads bearing anti-CD3 and anti-CD28 as a homogeneous
surrogate for antigen-presenting cells (APCs) (4 ± 2 beads per
well), and monitored the dynamic evolution of their secretory
states; T cells predominantly contacted beads in the wells within
1 h (Movie S1). Qualitatively, the responses measured were
similar to those observed during TCR-independent stimulation
(Fig. 4A). Different populations of cells again initiated secretion
of cytokines throughout the period of observation, and most cells
did not begin in a multifunctional state.
To compare the differences in responses between stimuli di-
rectly, we monitored the cytokine responses from T cells iso-
lated from the same subjects. TCR-dependent stimulation
produced more cells secreting IL-2, and fewer secreting TNFα,
than those stimulated with PMA/ionomycin (Fig. 4B). TCR-
dependent activation also favored fast, limited bursts of secre-
tion rather than sustained release, despite persistent stimulation
(Fig. 4C). This response is qualitatively consistent with the finite
temporal persistence of phosphorylated ERK observed by flow
cytometry in mouse T cells after activation (30). Regulation of
persistent TCR-dependent signals to allow only transient release
of cytokines suggests another mechanism for limiting the effects
of indiscriminate activation and supports in vivo observations
that multiple serial encounters are often required to induce
For both stimulations, cells secreting a single cytokine were
more likely to preserve that functional response than cells re-
leasing multiple cytokines (Fig. 4D). The number of cells pre-
serving functions was generally higher after TCR-dependent
stimulation and may be a consequence of the limited bursts of
activity (Fig. 4E). Surprisingly, both stimuli induced similar tra-
jectories, differing only in the frequencies of observed states (Fig.
4F).Although stimulation via the TCR appears toalter the timing
and persistence of specific secretory states, the programmatic
each row to show the likelihood that a cell in any given state would transition to a new state in the next sampling period. The adjacent gray scale bar reflects
the relative frequency of each state over 16 h. (B) Z scores highlight transitions that occur with significantly more/less frequency than expected by chance. Z
scores indicate significant transitions (>2, red; <2, blue); insignificant values (within ±2) are white. (C) Common time-aligned cytokine secretion profiles of
CD8−T cells (determined by SOM) and their relative percent distribution within each differentiated subset of T cells. Colorimetric representation of cytokine
states are consistent with Fig. 1C. (D) Matrices depicting the accuracy of classification of CD8−T-cell subtypes based on dynamic secretion profiles. These
“confusion matrices” quantify the percent of accurately classified cell types (defined by surface cell markers) computed by PCA relative to the true subtypes
(identified by their cell surface markers). The color bar reflects the percentage of cells classified as a certain subtype; uniform random assignment (25%) and
below is denoted in black and gray scale, respectively. (E) Average percent correct classification (PCC) of CD3+CD8−T-cell subtypes over 10 independent,
randomly sampled iterations as a function of the length of dynamic trajectory. Error bars indicate SD.
Functional evolution and T-cell discrimination. (A) Transition matrices for CD8−and CD8+T cells. Frequencies of transitions were normalized across
| www.pnas.org/cgi/doi/10.1073/pnas.1117194109 Han et al.
trajectories of secretory states after activation do not appear to
depend strongly on TCR-mediated signaling.
We have presented an intensive experimental characterization of
the dynamic evolution of cytokine secretion exhibited by individual
human T cells. Together, these data reveal that cytokine responses
evolve along a limited set of largely deterministic courses regard-
less of the initiating cue, and the accessibility of the particular
courses vary among differentiated subsets of T cells. These dy-
namic responses more accurately distinguish the subsets than ei-
ther the timing of activation or the cumulative cytokine response.
The temporal diversity in functional phenotypes observed here
have implications for both monitoring immune responses and the
behaviors of the cellular networks comprising the immune system.
The asynchronous activation and programmatic evolution of
responses by single T cells have practical consequences for im-
mune monitoring. These time-dependent attributes represent
sources of intrinsic noise for common integrative measurements
like ELISpot or ICS. Such assays are likely influenced by both
the timing and duration of sampling, and the degree of func-
tionality assigned to a cell may be misclassified within a specific
window of time. Because distinct classes of trajectories for se-
cretion were consistently identified across samples from multiple
donors, the frequencies of these trajectory classes in patient
samples may provide time-dependent cellular signatures associ-
ated with particular clinical conditions.
More broadly, our findings suggest another means for enabling
robustness within the network of cells comprising the immune
system. The maintenance of homeostatic stability, while con-
currently monitoring environmental perturbations, is an essential
property of the immune system—one that is believed to derive
from modularity of components and incorporation of control via
feedback (32, 33). The capability of T cells to access multiple
programs for cytokine release allows diverse and dynamically
evolvable contributions to the sustained accumulation of cyto-
kines in the global milieu, and such behavior represents an in-
triguing manifestation of modularity, cell-to-cell communication,
and feedback. The adaptability of multicellular systems to tem-
poral and environmental fluctuations is yet another quality that
improves when the collective population comprises phenotypi-
cally diverse individuals (34–36). Dynamically evolving func-
tional responses by T cells, therefore, suggest an alternate form
of functional plasticity (37) and may provide yet another means
for the immune system to adaptively respond to perturbations.
Temporal phenotypic diversity among T cells resonates with
other recent experimental and computational observations. For
example, the activation of individual T cells does not occur ho-
mogeneously upon stimulation—even isogenic ones—but rather
activate in a digital (on/off) manner, resulting from variations in
transcriptional activity or gene accessibility (25, 26), the numbers
and dynamic expression of signaling receptors (e.g., CD8, IL-2R),
and regulators of both positive and negative feedback loops (e.g.,
ERK, SHP-1, and RAS) (23, 30, 38, 39). Expression of IL-2 and
IFN-γ in human CD4+T cells has also been shown to depend on
the categorical translocation of the transcription factor NFATc2
(24). Our results revealing time-dependent variations in secretion
do not directly resolve the relationship between this functional
outcome and variances within the molecular signaling networks
that regulate the production and subsequent release of cytokines;
further characterization of this linkage could inform the regula-
tory mechanisms involved in the distinct trajectories measured.
A report by Feinerman et al. (39) supports the notion that the
transient nature of cytokine release discerned here can offer
opportunities for feedback and modulation of inputs from prox-
imal cells. Incorporating APCs in a single-cell coculture system
would further refine our dynamic single-cell measures to reveal
theeffects ofboth antigen–specificity andcell–cell interactionson
the evolution of paracrine responses. We anticipate that dynamic
single-cell analysis of cellular functional responses should help
over 10 h with anti-CD3/CD28. Colors, intensities, and rates of secretion are consistent with Fig. 1C. (B) Comparison of the frequencies of cytokine-secreting
cells from the same subject stimulated with PMA/ionomycin (Left) and anti-CD3/CD28 (Right). (C) Bar graph of the normalized average distributions of
sustained (>2 h) or burst-like (2 h) secretions by activated T cells as a function of stimulation. Error bars indicate SD. (D) Bar graph depicting the probability
that a given cell preserves its functional state in the 10-h period upon stimulation with PMA/ionomycin (filled bars) and anti-CD3/CD28 (open bars). Error bars
indicate SD. (E) Matrix reflecting the differential normalized probabilities of T-cell state transitions upon PMA/ionomycin (red, P) and anti-CD3/CD28 (blue, T)
stimulation. Cells were randomly sampled from each dataset to compare equal population sizes (n = 797). Gray scale bars reflect the relative frequencies of
states within each dataset. (F) Common trajectories that are accessed by PMA/ionomycin and TCR-dependent stimulated T cells. An equal number of cells (n =
478) were randomly sampled from each population, combined, and used for SOM clustered. The distributions of accessibility for cells to these trajectories are
shown in adjacent histograms. Data are representative from experiments for three different donors. All T-cell subtypes were included in A–E; only CD3+CD8−
T cells were included in F.
T-cell receptor-dependent stimulation and corresponding dynamic T-cell responses. (A) Temporal cytokine profiles from 797 viable T cells stimulated
Han et al. PNAS
| January 31, 2012
| vol. 109
| no. 5
evaluate the nature and evolution of intercellular interactions
present in biological systems such as lymphatic tissues, tumor
microenvironments, and stromal niches for stem cells.
Materials and Methods
Dynamic Single-Cell Analysis of Cytokine Release. Arrays of 1-mm-thick pol-
ydimethylsiloxane (PDMS) nanowells (50 μm or 30 μm) were manufactured
by using a custom-built mold and adhered directly to a 3-inch × 1-inch
glass slide. Primary human CD3+T cells were isolated from peripheral
blood of healthy donors, stained for CD8, CD45RA, CCR7, and viability
(Calcein violet), then loaded into nanowells at a density of ≈1 cell per well.
The array of nanowells was imaged by automated epifluorescence mi-
croscopy (Zeiss). After imaging, the array was cultured in serum-free HL-1
complete media supplemented with 10 ng/mL PMA and 1 μg/mL ion-
omycin, at 37 °C with 5% CO2. Cytokine secretion was measured 2 h after
initial stimulation and repeated every 2 h (21). During each cycle (2 h), IFN-
γ, IL-2, and TNFα from each well was captured by microengraving with
a glass slide supporting corresponding antibodies for 1 h, followed by
culturing the nanowells in media for another hour. After the kinetic
measurements, cells were stained in situ with a viability marker (Calcein)
and a dead cell marker (SYTOX green) and reimaged. Glass slides with
captured cytokines were probed with fluorescence-labeled detection
antibodies and imaged. TCR-mediated responses were measured by
coculturing T cells with anti-CD3/CD28 Dynabeads.
Data Analysis. Secretion data and cell phenotypes were extracted from col-
lected images and then combined according to the unique ID for each
nanowell. Wells occupied with single live cells before and after serial
microengraving that yielded cytokine secretion at any of the measured time
points were selected for analysis.
Computational and Statistical Analysis. SOMs were used to cluster similar
dynamic profiles from individual T-cell responses. Principal component
analysis (PCA) was used to classify T-cell subsets based on their functional
profiles. Receiver operating characteristic curves were computed to evaluate
the sensitivity of classification.
Additional Methods. Additional descriptions for all methods are available in SI
Materials and Methods, including fabrication of nanowells, functionaliza-
tion of glass, postprocessing of printed arrays, data processing, and
ACKNOWLEDGMENTS. We thank A. Chakraborty, D. Irvine, and H. Eisen for
comments and discussion. This work was supported by the W. M. Keck
Foundation, the Charles A. Dana Foundation, and National Institute of
Allergy and Infectious Diseases Grants 1U19AI089992, 5P01AI045757,
1R21AI088590, and 1RC1AI086152. The content is solely the responsibility
of the authors and does not necessarily represent the official views of the
National Institute of Allergy And Infectious Diseases or the National Insti-
tutes of Health.
1. Sallusto F, Lanzavecchia A (2009) Heterogeneity of CD4+ memory T cells: Functional
modules for tailored immunity. Eur J Immunol 39:2076–2082.
2. Seder RA, Darrah PA, Roederer M (2008) T-cell quality in memory and protection:
Implications for vaccine design. Nat Rev Immunol 8:247–258.
3. Appay V, van Lier RAW, Sallusto F, Roederer M (2008) Phenotype and function of
human T lymphocyte subsets: Consensus and issues. Cytometry A 73:975–983.
4. Zhou L, Chong MMW, Littman DR (2009) Plasticity of CD4+ T cell lineage differenti-
ation. Immunity 30:646–655.
5. Betts MR, et al. (2006) HIV nonprogressors preferentially maintain highly functional
HIV-specific CD8+ T cells. Blood 107:4781–4789.
6. Duvall MG, et al. (2008) Polyfunctional T cell responses are a hallmark of HIV-2 in-
fection. Eur J Immunol 38:350–363.
7. Caccamo N, et al. (2010) Multifunctional CD4(+) T cells correlate with active Myco-
bacterium tuberculosis infection. Eur J Immunol 40:2211–2220.
8. Precopio ML, et al. (2007) Immunization with vaccinia virus induces polyfunctional
and phenotypically distinctive CD8(+) T cell responses. J Exp Med 204:1405–1416.
9. Darrah PA, et al. (2007) Multifunctional TH1 cells define a correlate of vaccine-me-
diated protection against Leishmania major. Nat Med 13:843–850.
10. Lindenstrøm T, et al. (2009) Tuberculosis subunit vaccination provides long-term
protective immunity characterized by multifunctional CD4 memory T cells. J Immunol
11. Lagrelius M, Jones P, Franck K, Gaines H (2006) Cytokine detection by multiplex
technology useful for assessing antigen specific cytokine profiles and kinetics in
whole blood cultured up to seven days. Cytokine 33:156–165.
12. Mascher B, Schlenke P, Seyfarth M (1999) Expression and kinetics of cytokines de-
termined by intracellular staining using flow cytometry. J Immunol Methods 223:
13. Rogers WO, et al. (1997) Visualization of antigen-specific T cell activation and cyto-
kine expression in vivo. J Immunol 158:649–657.
14. Sojka DK, Bruniquel D, Schwartz RH, Singh NJ (2004) IL-2 secretion by CD4+ T cells
in vivo is rapid, transient, and influenced by TCR-specific competition. J Immunol 172:
15. Perfetto SP, Chattopadhyay PK, Roederer M (2004) Seventeen-colour flow cytometry:
Unravelling the immune system. Nat Rev Immunol 4:648–655.
16. Harriman WD, et al. (2009) Multiplexed Elispot assay. J Immunol Methods 341:
17. Gazagne A, et al. (2003) A Fluorospot assay to detect single T lymphocytes simulta-
neously producing multiple cytokines. J Immunol Methods 283:91–98.
18. Ma C, et al. (2011) A clinical microchip for evaluation of single immune cells reveals
high functional heterogeneity in phenotypically similar T cells. Nat Med 17:738–743.
19. Pantaleo G, Harari A (2006) Functional signatures in antiviral T-cell immunity for
monitoring virus-associated diseases. Nat Rev Immunol 6:417–423.
20. Chatila T, Silverman L, Miller R, Geha R (1989) Mechanisms of T cell activation by the
calcium ionophore ionomycin. J Immunol 143:1283–1289.
21. Han Q, Bradshaw EM, Nilsson B, Hafler DA, Love JC (2010) Multidimensional analysis
of the frequencies and rates of cytokine secretion from single cells by quantitative
microengraving. Lab Chip 10:1391–1400.
22. Sandberg JK, Fast NM, Nixon DF (2001) Functional heterogeneity of cytokines and
cytolytic effector molecules in human CD8+ T lymphocytes. J Immunol 167:181–187.
23. Feinerman O, Veiga J, Dorfman JR, Germain RN, Altan-Bonnet G (2008) Variability
and robustness in T cell activation from regulated heterogeneity in protein levels.
24. Podtschaske M, et al. (2007) Digital NFATc2 activation per cell transforms graded T
cell receptor activation into an all-or-none IL-2 expression. PLoS ONE 2:e935.
25. McKarns SC, Schwartz RH (2008) Biphasic regulation of Il2 transcription in CD4+ T
cells: Roles for TNF-alpha receptor signaling and chromatin structure. J Immunol 181:
26. Mariani L, et al. (2010) Short-term memory in gene induction reveals the regulatory
principle behind stochastic IL-4 expression. Molec Sys Biol 6:359.
27. Szabo SJ, et al. (2000) A novel transcription factor, T-bet, directs Th1 lineage com-
mitment. Cell 100:655–669.
28. Ketchen DJ, Shook CL (1996) The application of cluster analysis in strategic man-
agement research: An analysis and critique. Strateg Manage J 17:441–458.
29. Cantrell DA (2002) T-cell antigen receptor signal transduction. Immunology 105:
30. Das J, et al. (2009) Digital signaling and hysteresis characterize ras activation in
lymphoid cells. Cell 136:337–351.
31. Henrickson SE, et al. (2008) T cell sensing of antigen dose governs interactive behavior
with dendritic cells and sets a threshold for T cell activation. Nat Immunol 9:282–291.
32. Kitano H, Oda K (2006) Robustness trade-offs and host-microbial symbiosis in the
immune system. Mol Syst Biol, 2006.0022.
33. Zak DE, Aderem A (2009) Systems biology of innate immunity. Immunol Rev 227:
34. Acar M, Mettetal JT, van Oudenaarden A (2008) Stochastic switching as a survival
strategy in fluctuating environments. Nat Genet 40:471–475.
35. Balaban NQ, Merrin J, Chait R, Kowalik L, Leibler S (2004) Bacterial persistence as
a phenotypic switch. Science 305:1622–1625.
36. Pearl S, Gabay C, Kishony R, Oppenheim A, Balaban NQ (2008) Nongenetic in-
dividuality in the host-phage interaction. PLoS Biol 6:e120.
37. Zhu J, Paul WE (2010) Heterogeneity and plasticity of T helper cells. Cell Res 20:4–12.
38. Busse D, et al. (2010) Competing feedback loops shape IL-2 signaling between helper
and regulatory T lymphocytes in cellular microenvironments. Proc Natl Acad Sci USA
39. Feinerman O, et al. (2010) Single-cell quantification of IL-2 response by effector and
regulatory T cells reveals critical plasticity in immune response. Mol Syst Biol 6:437.
| www.pnas.org/cgi/doi/10.1073/pnas.1117194109 Han et al.