Generating sparse and selective third-order responses
in the olfactory system of the fly
Sean X. Luoa, Richard Axela,b,c,1, and L. F. Abbotta,d,1
aDepartment of Neuroscience,bDepartment of Biochemistry and Molecular Biophysics,cHoward Hughes Medical Institute, anddDepartment of Physiology
and Cellular Biophysics, Columbia University, New York, NY 10032
Contributed by Richard Axel, April 28, 2010 (sent for review March 4, 2010)
In the antennal lobe of Drosophila, information about odors is
transferred from olfactory receptor neurons (ORNs) to projection
of the protocerebrum (LHNs) and to Kenyon cells (KCs) in the mush-
room body. The transformation from ORN to PN responses can be
described by a normalization model similar to what has been used
in modeling visually responsive neurons. We study the implications
of this transformation for the generation of LHN and KC responses
under the hypothesis that LHN responses are highly selective and
therefore suitable for driving innate behaviors, whereas KCs pro-
vide a more general sparse representation of odors suitable for
forming learned behavioral associations. Our results indicate that
thetransformationfrom ORNtoPNfiringratesin theantennallobe
equalizes the magnitudes of and decorrelates responses to differ-
ent odors through feedforward nonlinearities and lateral suppres-
sion within the circuitry of the antennal lobe, and we study how
these two components affect LHN and KC responses.
antennal lobe|decorrelation|lateral horn|normalization|olfaction
receptor neurons (ORNs) located in the antennae and maxillary
palps synapse onto projection neurons (PNs) within the glomeruli
of the antennal lobes (Fig. S1). ORNs expressing a given receptor
converge on an anatomically invariant glomerulus. PNs innervate
a single glomerulus and thus receive their primary input from
ORNs expressing the same olfactory receptor (1), although cross-
glomerular interactions mediated by interneurons are also present
(2–5). The PNs send their axons to two distinct regions of the fly
brain, the lateral horn and the mushroom body. By constructing
models of neurons in these regions, we study the effects of the
transformations arising from circuitry within the antennal lobe on
to represent and discriminate odors.
Any functional interpretation of the transformation from ORN
to PN responses depends on the nature of the processing being
performed by the third-order neurons that receive PN input. The
lateral horn and mushroom body appear to be involved in dif-
ferent forms of sensory processing. The lateral horn is believed to
be important in generating innate behaviors (6), including those
elicited by pheromones (7). PN projections to lateral horn neu-
rons (LHNs) are spatially stereotyped (8, 9) and clustered (10),
suggesting that circuits in this region may be “hardwired” for the
detection of specific odors that elicit innate behavioral responses.
The mushroom body is implicated in decision making (11) and in
the formation of associative memories (12, 13). Both calcium
imaging (14) and electrophysiology (15) indicate that the re-
sponses of Kenyon cell (KCs) in the mushroom body are sparse,
with each odorant eliciting responses in a few percent of the KCs
and individual KCs responding to a small number of tested
odorants. Electrophysiological experiments suggest that con-
nections from PNs to Kenyon cells may be random (16, 17), al-
though zonal specificity of PN projections and KC dendrites in
the mushroom bodies has been reported (10, 18). The distinct
n the olfactory system, as in other sensory systems, signals from
primary receptors are processed and transformed before being
characteristics of LHNs and KCs lead us to consider the impli-
cations of the ORN-to-PN transformation with respect to two
different downstream tasks: odor discrimination in the lateral
horn and general-purpose representation of a wide variety of
odors in the mushroom body. We begin by discussing how model
PN responses are constructed, then show how LHN and KC
responses with the desired properties can be generated, and fi-
nally analyze the roles played by different elements of the model.
We build our third-order neuron models by taking measured
ORN firing rates (19, 20) (Fig. 1A) and then determining the
corresponding PN responses on the basis of an experimental
characterization of the transformations occurring within the an-
tennal lobe (see below) (21, 22). Finally, we use the computed PN
responses to drive model LHNs and KCs. To model the responses
of LHNs or KCs, the computed firing rate for each PN in re-
sponse to a particular odor is multiplied by a weight representing
the strength of a synapse from that PN onto the third-order
neuron being modeled. The weighted PN rates are then summed,
and the total is compared to a fixed threshold (23) to determine
whether or not the third-order neuron generates a response to
that odor. Although selectivity is likely to arise from some form of
input summation and thresholding, the computations performed
by neurons in the lateral horn and the mushroom body are un-
doubtedly more complex than those of our model. Using a mini-
mal model allows us to identify the problems that more complex
cellular and circuit mechanisms must solve, and it allows us to
focus on the role of the antennal lobe circuitry.
Transformation from ORN to PN Firing Rates. The transformation
from ORN to PN responses consists of two functional compo-
nents: a feedforward nonlinearity and lateral suppression. The
relationship between the firing rate of an ORN and that of a PN
is nonlinear with strong saturation at high ORN rates (21), a re-
sult due at least in part to short-term depression at ORN-PN
synapses (24). We refer to this transformation as a feedforward
nonlinearity, although lateral circuitry may also contribute to
this effect. This nonlinearity was described using an exponential
function, but it can also be fit by an equation of the form
rates of a corresponding PN and ORN pair, and Rmaxand σ are
constants. The advantage of using this latter form is that it fits in
well with more recent work in which the effects of lateral sup-
pression have been incorporated (22). Lateral suppression, which
is the dominant effect of ORNs that do not drive a given PN di-
rectly, can be described by adding a term proportional to the sum
ORNÞ, where rPNand rORNare the firing
Author contributions: S.X.L., R.A., and L.F.A. designed research, performed research, con-
tributed analytic tools, analyzed data, and wrote the paper.
The authors declare no conflict of interest.
Freely available online through the PNAS open access option.
1To whom correspondence may be addressed. E-mail: firstname.lastname@example.org or lfa2103@
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
| June 8, 2010
| vol. 107
| no. 23
of the firing rates of all of the ORNs to the denominator of the
expression for rPN. Defining sORNas the sum of all ORN firing
rates (independent of the glomerulus to which they project), the
where m is another constant (see Methods for parameter values),
provides an accurate description of both the feedforward non-
linearity and lateral suppression generated by the antennal lobe
circuitry (22). Eq. 1 refers to average responses, but trial-to-trial
variability is included in the PN responses we use (Fig. 1B) by
adding randomly generated fluctuations modeled after the data
Before modeling LHNs and KCs, it is instructive to examine
how the transformation of Eq. 1 modifies olfactory responses.
The sum of all ORN firing rates, sORN, introduced above and the
analogous sum over all PN firing rates, sPN, are measures of the
total magnitude of the response to a given odor, which has
a strong impact on LHN and KC performance. A third-order
neuron tuned to respond to an odor that elicits a weak overall
PN response (small sPN) will tend to respond nonselectively to
odors that elicit stronger total responses (large sPN). This prob-
lem can be avoided if the magnitudes of PN responses are
equalized, meaning that the total activity across the population
of PNs (i.e., sPN) is roughly the same for each odor. We use the
term equalization, rather than normalization, because normali-
zation has been used to refer to the entire transformation of
Eq. 1 in the visual system literature where such a transformation
was first proposed and has been used extensively (25, 26). The
transformation of Eq. 1 has a strong magnitude-equalizing effect;
the sum of ORN rates (sORN, Fig. 1C) is much more variable
across odors than the sum over PN rates (sPN, Fig. 1E). To iso-
late the origin of this equalization of the response magnitude, we
also computed sPNwhen lateral suppression was eliminated by
setting the parameter m in Eq. 1 to zero. The results indicate that
much, but not all, of the response magnitude equalization arises
from the feedforward nonlinearity (Fig. 1D).
Selectivity is also difficult to achieve if the patterns of PN ac-
tivation across odors are too similar, that is, too highly correlated.
We performed a principal component analysis (PCA) of the
ORN and computed PN firing rates to compare their degree of
correlation. For totally correlated data, a single principal compo-
nent accounts for 100% of the data variance, whereas for com-
pletely uncorrelated data, all of the components account for an
equal percentage of the variance. For the olfactory responses, >40%
of the ORN response variance is accounted for by a single prin-
cipal component projection, indicating a high degree of correla-
tion (Fig. 1F). The maximal principal component projection for
the PN responses accounts for only 15% of the total variance
(Fig. 1H), so PNs are significantly decorrelated by the trans-
formation of Eq. 1. This decorrelation is a consequence of lateral
suppression in the antennal lobe, because it is almost entirely
eliminated if we set the parameter m in Eq. 1 to zero (Fig. 1G). A
previous mathematical characterization of PN responses did not
show decorrelation (21, 27) because it did not include lateral
suppression. In agreement with more recent data (22), the more
accurate characterization we use here does decorrelate.
Discrimination from Odor-Specific Weights (Lateral Horn). We hy-
pothesize that neurons in the lateral horn are connected in a
manner that generates highly selective responses to specific odors,
allowing them to evoke innate behaviors. To study selectivity in
the lateral horn from the data we are using, we constructed 110
LHNs, each selective for a different one of the 110 odorants in
the dataset. In other words, we assigned a single target odor from
this set to each LHN and attempted to make it respond exclu-
sively to this target odor. We are not suggesting that each of these
odors generates an innate behavior. Instead, the model LHNs are
used as a proof of principle to demonstrate how highly specific
odor-selective LHNs can be constructed.
We assign synaptic weights for the LHNs using Fisher’s linear
discriminant, a standard technique (Methods). Importantly, this
approach uses only PN responses to the targeted odor and general
statistical properties of the PN responses across odors such as the
mean and covariance to determine the weights, and avoids over-
fitting. The weights of inputs onto LHNs produced in this way are
from inhibitory input to the LHN from unmodeled interneurons
driven by the corresponding PN. Using these weights, we can
generate LHNs tuned to respond almost exclusively to a single
odorant for all 110 odors in the dataset (Fig. 2).
LHNs will respond exclusively to their target odor if the input
for this odor is consistently higher than the inputs for any other
odors. Histograms of the sum of 20 weighted PN inputs (Fig. 2A)
for targeted (red) and nontargeted (blue) odors for all 110 LHNs
over repeated trials reveal two roughly Gaussian distributions
that are well separated. This allows us to divide the inputs into
those producing and those not producing a response with a fixed
threshold (dashed vertical line in Fig. 2A). Using this threshold,
0 50 100
0 10 20
0 50 100
0 10 20
0 10 20
Fi ring Ra te (H z)
20 406080 100
20 4060 80 100
as input to our models from Hallem and Carlson (20). Colors (scale bar) show
the firing rates of ORNs expressing 20 different olfactory receptors (hori-
zontal axis) to 110 different odorants (vertical axis). From the full dataset,
we used responses generated by 20 nonspecialized receptors (we do not
include pheromone receptors) to 110 tested odors. (B) The same data as in A
in the same format, but converted to model PN responses through the
transformation of Eq. 1. (C–H) Response magnitude equalization and
decorrelation. ORN, NL, and PN indicate that ORN, PN with no lateral sup-
pression, and PN response were used for the corresponding column. (C)
Sums of ORN responses to 110 odorants from the dataset plotted relative to
their average across odors. (D) Sums of PN responses to 110 odors generated
by Eq. 1, but with no lateral suppression (m = 0 in Eq. 1). (E) Sums of PN
responses to 100 odors generated by Eq. 1. (F–H) Percentage of variances
from a PCA analysis of the response used in C–E.
ORN and PN responses. (A) A representation of the dataset we use
| www.pnas.org/cgi/doi/10.1073/pnas.1005635107Luo et al.
we computed response probabilities across trials (PN responses
are different in each trial due to the response variability built
into our PN model) for 110 differently tuned LHNs, each labeled
by the odor to which it is targeted (LHN index in Fig. 2B). The
diagonal elements of the matrix of response probabilities (Fig.
2B) correspond to each LHN responding to its targeted odor,
and the nondiagonal entries represent responses to nontargeted
odors. For perfect performance of the LHNs, response proba-
bilities would be 100% along the diagonal and zero off the di-
agonal, meaning that the LHNs respond every time to the odor
for which they are targeted and never to any other odor. The
selectivity of the LHNs is quite close to being perfect in this
sense (Fig. 2B). This excellent performance is in agreement with
a theoretical analysis of the capacity of a system of this type (28).
Performance of the LHNs can also be characterized by the
fraction of untargeted odors that generate a response (the false-
positive rate) and the fraction of targeted odors that fail to
generate a response (the false-negative rate). These quantities
depend on the threshold, but if we adjust this so that the false-
positive and false-negative rates are equal, the value they take is
0.4%. The area under a receiver operating characteristic (ROC)
curve provides a threshold-independent measure of discrimina-
tion performance, and this area is 0.999 for model LHNs driven
by 20 PN inputs.
We also considered what happens if the number of PN inputs
to the model LHNs is reduced to five, a number suggested from
analysis of KCs (15). In this case, the weights are constructed
again using Fisher’s linear discriminant, but for a reduced set of
inputs (Methods). With only five PN connections per LHN, the
input distributions for targeted and nontargeted odors are
slightly more overlapping (Fig. 2C), and there are more nonzero
response probabilities for nontargeted odors (Fig. 2D). Never-
theless, the overall performance is still good. The equal false-
positive and false-negative probabilities are 2%, and the area
under the ROC curve is 0.987. Thus, acceptable discrimination
can be obtained with as few as five PN inputs per LHN.
Sparse KC Responses from Random Weights (Mushroom Body). Our
modeling work for KCs is based on the assumption that they
provide a sparse representation across a wide variety of odorants.
Of course the LHNs, being highly selective, also provide a sparse
representation, but not across nearly as broad a range of odors as
the KCs do. Our goal is to construct a population of 2,500 model
KCs, the number estimated in the mushroom body (29), such that
on average 5% of the KCs respond to each odor, and each KC
responds to ∼5% of the odorants presented, corresponding to
experimental estimates (14, 15). We allow fluctuations around
this 5% value from cell to cell or from odor to odor, but we as-
Our procedure for constructing a KC response is to randomly
choose PNs from n glomeruli and to set the strengths of the
synapses connecting them to a KC randomly and independently
from a uniform distribution over positive values (Methods). Good
performance can be achieved with such random connections (Fig.
3A) even though we used the same threshold for all of the KCs
and did not individually adjust the threshold to achieve a precise
5% sparseness in each KC. The scattering of color shows that the
sparse KC responses are well distributed across all 2,500 cells and
110 odorants. Fig. 3 shows KCs receiving excitatory input from
PNs connected to five different glomeruli, along with global in-
hibition (30) (Methods). In this case, all 110 odors elicit responses
above the 95% threshold in, on average, 125 cells and, at a min-
imum, at least two KCs.
One measure of the quality of the sparse representation for
different numbers of PN inputs is the number of missed odors
(Fig. 3B Upper). Such an odor would not be represented in the
mushroom body and, presumably, could not be detected by the fly
unless it was represented separately in the lateral horn. We define
the 110 odors is missed for most values of n, the number of PNs
connected to each KC (Fig. 3B Upper). The best performance is
for n < 15. A high-quality sparse representation should also
10 20 30 40 50 60 70 80 90 100110
10 20 30 40 50 60 70 80 90 100110
1.51 0.50 0.51 1.5
1.51 0.50 0.51 1.5
targeted odors the other (red). The green dashed line indicates the response threshold. Input is in units that set the mean input for the targeted odor equal to
1. (B) Response probabilities for 110 model LHNs to 110 odors, with 20 PN inputs per LHN. Each row shows response probabilities for LHNs tuned to respond to
a specific odor (indicated by the LNH index), and each column shows the responses of all LHNs to a particular presented odor. The diagonal gives the response
of each LHN to its targeted odor, and the off-diagonal elements are responses to nontargeted odors. Colors denote the fraction of trials producing a response
(color bar). (C) Input histogram for a reduced connectivity model in which each LHN receives input from only 5 PNs. (D) Response probabilities for 5 PN
connections per LHN.
Responses of model LHNs. (A) Histograms of the distribution of LHN inputs from 20 PNs. Nontargeted odors produce one distribution (blue) and
Luo et al.PNAS
| June 8, 2010
| vol. 107
| no. 23
efficiently use its neuronal resources, which means there should
not be a large number of KCs that never respond to any odor.
Whereas 1 or 2 PN inputs are sufficient to obtain a good sparse
representation, this leads to a large number of silent KCs (Fig. 3B
Lower). With only 1 PN input, about 800 of the 2,500 KCs do not
respond to any odor, according to the 50% criterion. The number
of nonresponding KCs drops progressively as the number of PN
to KC connections increases. Combining the results of Fig. 3B
Upper and Lower, the best performance seems to arise for be-
tween 5 and 15 PN inputs per KC, numbers consistent with esti-
mates obtained from electrophysiology (15).
The synaptic weights connecting PNs to our model KCs are all
positive, representing pure excitation, but the model KCs also
receive global inhibition (Methods). Inhibition is useful because
it allows the KCs to take advantage of the decorrelation of the
PN inputs. However, we also examined the performance of
model KCs when all inhibition was removed and they were
driven solely by PN input (Fig. S2). Performance is still good with
regard to missed odors (Fig. S2 Upper), but there are significantly
more silent KCs when inhibition is eliminated (Fig. S2 Lower).
Concentration Dependence. Changing odor concentration is a par-
ticularly stringent test of a model like ours in which responses are
determined by a fixed threshold. Without response magnitude
concentrations will be lost at high concentrations or, alternatively,
neurons that are selective at high concentrations will fail to re-
spond at low odor concentrations. Imaging and electrophysiolog-
ical experiments (14, 15) indicate that at least some of the KCs
retain theirselectivitytothesame odorsacrossa range ofdifferent
concentrations.Itisreasonable toassumethat atleastsome LHNs
share this feature, as this would have obvious behavioral advan-
tages. Hallem and Carlson measured ORN firing rates at four
different concentration levels for10 odorants(20).We studied the
the previous sections with this collection of 40 response sets,
representing 10 odors presented at four different concentrations,
10−2, 10−4, 10−6, and 10−8dilutions.
For the model LHNs, we generalized Fisher’s linear discrimi-
nant for multiple stimuli across different concentrations (Meth-
ods) and used five PN inputs per LHN. The performance of these
model LHNs is shown as four response probability matrices (Fig.
recapitulated four times, illustrating that these cells respond re-
liably to their target odors at all of the different concentrations.
The off-diagonal response probabilities, representing responses
to nontargeted odors, remain small, indicating that the false-
positive rate is low across concentrations. The main error comes,
not surprisingly, from discrimination at low concentration, where
the false-positive rate increases somewhat. Over all odors and
concentrations, the false-positive and false-negative rates are 7%
and the area under the ROC curve is 0.92.
To construct KCs, we again used random positive synaptic
weights with global inhibition and adjusted the single threshold
used for all KCs so that 5% of the neurons respond, averaged
across concentrations. Note that in contrast to previous models
of concentration invariance (31), the KCs in our model do not
receive compensatory inputs or change their thresholds adap-
tively for different concentrations. These fixed-threshold KCs
respond to odors sparsely at both low (10−8SV) and high (10−2
SV) concentrations, and an analysis of the results shown in Fig.
4B indicates that, collectively, the population responds to every
odor in the dataset regardless of its concentration. In other
words, none of the odors is missed across this wide range of
concentrations, so a full representation of the odors is retained.
Role of the ORN-to-PN Response Transformation. To study the effect
of the nonlinear transformation that takes place between the
ORN and PN responses, we repeated our analysis of model
LHNs and KCs but driven either by direct ORN input or by PN
input with lateral suppression removed (m = 0 in Eq. 1).
The results are summarized in Fig. 5, and further details can
be found in Figs. S3–S5.
ForLHNs driven by either 20 (Fig.5A) or 5 (Fig. 5B) inputs,the
false-positive and false-negative error rates (when the threshold is
adjusted to make them equal) rise only slightly when the effect of
lateral suppression is removed (bars marked “NL” in Fig. 5 A and
B), but there is a larger increase in error percentage when the
feedforward nonlinearity is also eliminated (equivalent to using
ORN rather thanPNinput,barsmarked“ORN”inFig.5AandB;
also Fig. S3). KCs driven by PN input lacking lateral suppression
perform basically as well as those using full PN input, as in neither
case are any odors missed from the KC representations with 5
inputs per KC (Fig. 5C). Performance degrades significantly in the
absence of the feedforward nonlinearity when ORN input is used
(Fig. 5C and Fig. S4). In this case, ∼30 odors are missing from
Lateral suppression, and hence decorrelation, plays a larger
role in buffering the concentration dependence of LHN respon-
ses. The equal false-positive and false-negative error percentages
rise more significantly when lateral interactions are eliminated,
although most of the decrement in performance associated with
using ORN input still arises from the absence of the feedforward
nonlinearity (Fig. 5D and Fig. S5A). KCs driven by ORN rather
than PN input (removing the effects of Eq. 1) fail to respond to
many of the tested odors at low concentrations (Fig. S5B) and
thus lose the ability to represent odors across the wide range of
concentrations seen in Fig. 4A.
Number of PN to KC Connections
20 40 6080 100
02468 101214 16 1820
02468 101214 16 18 20
elicit responses. (B) The number of missed odors (odors producing no response across the entire KC population, Upper) and the number of silent KCs (KCs
responding to none of the 110 odorants, Lower).
Model KC responses. (A) Response probabilities of 2,500 model KCs, each receiving five random PN connections and global inhibition. All of the odors
| www.pnas.org/cgi/doi/10.1073/pnas.1005635107Luo et al.
In summary, the most important factor in generating selective
(LHNs) and sparse (KCs) third-order responses is the response
magnitude equalization provided by the feedforward nonlinearity
of the antennal lobe. The effects of lateral suppression, which are
responsible for response decorrelation, are also important, al-
though smaller than those of the feedforward nonlinearity, for
retaining selectivity over a range of odor concentrations.
A transformation in the representation of odors occurs in the an-
tennal lobe that equalizes the magnitudes of and decorrelates the
responses to different odors. Equalizing sensory response magni-
tudes might appear to be an unwise strategy because it diminishes
a difference that can distinguish sensory stimuli. For this reason,
discussions of sensory processing often stress the expansion of
representations rather than the compression imposed by the com-
pressive nonlinearity of the ORN-to-PN transformation. We have
shown, however, that this compressive nonlinearity allows highly
selective and sparse responses to be generated in higher-order
neurons through the application of a fixed threshold. Response
magnitude equalization is crucial for all of the results we have
concentration invariant LHN and KC responses. Given the im-
portance of selectivity and sparseness in sensory representations,
magnitude equalization may be a widespread coding strategy.
Decorrelation has also been postulated to be an important as-
pect of sensory processing (32–34). Decorrelation within the an-
tennal lobe contributes less significantly to our results than does
response magnitude equalization. This is, in part, because our
on their inputs. The Fisher linear discriminant we used to set the
specific connections from PNs to LHNs can perform this function,
as can the global inhibition we introduced to model KCs. Global
inhibition can decorrelate ORN responses because their largest
variance principal component (Fig. 2B), like the global inhibition
(Methods), is aligned with their odor-averaged response (20).
The normalization model (25) that describes the transfor-
mation from ORN to PN responses (Eq. 1) has been used to ac-
count for an impressive range of properties of visually responsive
neurons (25, 26, 35, 36). It has been recently noted that normal-
ization affects the population coding of orientation in the visual
system in ways related to what we have found here for olfaction
(37). The fact that the normalization model accounts for the
effects of both feedforward nonlinearities and lateral suppression
in olfactory responses supports the idea that it describes a funda-
mental computation in sensory processing.
The models we have generated to explore how olfactory rep-
resentation may result in either innate or learned behaviors in-
voke anatomically distinct circuits for the two types of behavior.
Innate behaviors, we postulate, result from determined con-
nections from PNs to LHNs that elicit stereotyped responses in
downstream circuits. This is in accord with the observation that
a topographic map of olfactory information is retained in the
lateral horn, but the character of the map differs from that of the
antennal lobe (8, 9). In our model of the mushroom body, we
propose a second transformation that invokes random, feedfor-
Odor Presented (Concentration)
15 1015 1015 101
5 1015 1015 1015 101
Odor Presented (Concentration)
(A) Response probabilities for LHNs receiving input from five
PNs and targeted to 10 different odors at four different con-
centrations. Four 10 × 10 matrices are shown, corresponding
to the four concentrations, 10 trial odors, and LHNs targeted
to each of the 10 odors. As in Figs. 2 and 3, each row shows
a particular LHN, labeled by its target odorant, and each col-
umn gives the responses across LHNs to a specific odor. The
bright diagonals repeated four times indicate that the LHNs
respond reliably to their targets at different concentrations.
Off-diagonal elements show false-positive responses. (B) Re-
sponse probabilities of 2,500 KCs with five PN inputs each to
10 odorants at four concentrations. Four 10 × 2,500 blocks are
shown, corresponding to the four concentrations, 10 odors,
and 2,500 KCs. Otherwise, each block is as in Fig. 3A.
LHN and KC responses for different concentrations.
Luo et al.PNAS
| June 8, 2010
| vol. 107
| no. 23
ward, convergent connectivity between the glomeruli and the
KCs. Neurons responsive to a given odor would thus be randomly
the ultimate behavioral output of a representation in the mush-
room body would therefore be imposed by experience, rendering
this a purely associative structure. If the mushroom body is indeed
an associative olfactory center, spatial order is conceptually su-
perfluous. In contrast, innate behavioral responses mediated by
the lateral horn are the consequence of stereotyped and de-
termined inputs imposed by evolution rather than experience.
Recent data suggest that a similar organizational logic may be
operative in mammals, where information from the olfactory bulb
bifurcates to project to limbic structures and the piriform cortex.
Stereotyped connections to the amygdala may mediate innate ol-
factory responses and random inputs to the piriform cortex (38)
may mediate more measured or learned olfactory behavior.
general sensors, receptors Or33b, Or47b, Or65a, and Or88a were removed
from the dataset of Hallem and Carlson (20) (SI Text). PN responses were
generated from the ORN firing rates using Eq. 1 with Rmax= 165 Hz, σ = 12 Hz,
and m = 0.05 Hz (22). Values of these parameters differ between PNs, so we
have used values that provide the best overall fit. Noise was included by
making the transformation rPN→rPNþ δ tanhðαrPNÞη, where η is a random
variable with zero mean and unit variance, δ = 10 Hz, and α = 0.025 Hz. We
used Fisher’s linear discriminant to set the weights from PNs to LHNs (SI Text).
For model KCs, we chose the n synaptic connections for the KCs randomly
and drew their weights, denoted by the vector w,from a uniform distri-
bution between 0 and 1. Model KCs received a global inhibition constructed
as described in SI Text.
ACKNOWLEDGMENTS. We thank R. Wilson, S. Olsen, E. Hallem, and J. Carlson
for extremely valuable discussion concerning their data; L. Vosshall and S. Fusi
for insights and encouragement; and R. Datta, V. Ruta, S. Ganguli, T.
Toyoizumi, Z.Pitkow, and M.Vidnefortheirhelp.This researchwassupported
by a National Institutes of Health Director’s Pioneer Award, part of the Na-
tional Institutes of Health Roadmap for Medical Research, through Grant 5-
1. Vosshall LB, Amrein H, Morozov PS, Rzhetsky A, Axel R (1999) A spatial map of
olfactory receptor expression in the Drosophila antenna. Cell 96:725–736.
2. Olsen SR, Bhandawat V, Wilson RI (2007) Excitatory interactions between olfactory
processing channels in the Drosophila antennal lobe. Neuron 54:89–103.
3. Shang Y, Claridge-Chang A, Sjulson L, Pypaert M, Miesenböck G (2007) Excitatory local
circuits and their implications for olfactory processing in the fly antennal lobe. Cell
4. Root CM, et al. (2008) A presynaptic gain control mechanism fine-tunes olfactory
behavior. Neuron 59:311–321.
5. Olsen SR, Wilson RI (2008) Lateral presynaptic inhibition mediates gain control in an
olfactory circuit. Nature 452:956–960.
6. Tanaka NK, Awasaki T, Shimada T, Ito K (2004) Integration of chemosensory pathways
in the Drosophila second-order olfactory centers. Curr Biol 14:449–457.
7. Datta SR, et al. (2008) The Drosophila pheromone cVA activates a sexually dimorphic
neural circuit. Nature 452:473–477.
8. Marin EC, Jefferis GS, Komiyama T, Zhu H, Luo L (2002) Representation of the
glomerular olfactory map in the Drosophila brain. Cell 109:243–255.
9. Wong AM, Wang JW, Axel R (2002) Spatial representation of the glomerular map in
the Drosophila protocerebrum. Cell 109:229–241.
10. Jefferis GS, et al. (2007) Comprehensive maps of Drosophila higher olfactory centers:
Spatially segregated fruit and pheromone representation. Cell 128:1187–1203.
11. Zhang K, Guo JZ, Peng Y, Xi W, Guo A (2007) Dopamine-mushroom body circuit
regulates saliency-based decision-making in Drosophila. Science 316:1901–1904.
12. de Belle JS, Heisenberg M (1994) Associative odor learning in Drosophila abolished by
chemical ablation of mushroom bodies. Science 263:692–695.
13. Davis RL (2005) Olfactory memory formation in Drosophila: From molecular to
systems neuroscience. Annu Rev Neurosci 28:275–302.
14. Wang Y, et al. (2004) Stereotyped odor-evoked activity in the mushroom body of
Drosophila revealed by green fluorescent protein-based Ca2+ imaging. J Neurosci 24:
15. Turner GC, Bazhenov M, Laurent G (2008) Olfactory representations by Drosophila
mushroom body neurons. J Neurophysiol 99:734–746.
16. Masuda-Nakagawa LM, Tanaka NK, O’Kane CJ (2005) Stereotypic and random
patterns of connectivity in the larval mushroom body calyx of Drosophila. Proc Natl
Acad Sci USA 102:19027–19032.
17. Murthy M, Fiete I, Laurent G (2008) Testing odor response stereotypy in the
Drosophila mushroom body. Neuron 59:1009–1023.
18. Lin HH, Lai JS, Chin AL, Chen YC, Chiang AS (2007) A map of olfactory representation
in the Drosophila mushroom body. Cell 128:1205–1217.
19. Hallem EA, Ho MG, Carlson JR (2004) The molecular basis of odor coding in the
Drosophila antenna. Cell 117:965–979.
20. Hallem EA, Carlson JR (2006) Coding of odors by a receptor repertoire. Cell 125:
21. Bhandawat V, Olsen SR, Gouwens NW, Schlief ML, Wilson RI (2007) Sensory processing
in the Drosophila antennal lobe increases reliability and separability of ensemble
odor representations. Nat Neurosci 10:1474–1482.
22. Olsen SR, Bhandawat V, Wilson RI (2010) Divisive normalization in olfactory
population codes. Neuron 66:287–299.
23. Huerta R, Nowotny T, García-Sanchez M, Abarbanel HD, Rabinovich MI (2004) Learning
classification in the olfactory system of insects. Neural Comput 16:1601–1640.
24. Kazama H, Wilson RI (2008) Homeostatic matching and nonlinear amplification at
identified central synapses. Neuron 58:401–413.
25. Heeger DJ (1992) Normalization of cell responses in cat striate cortex. Vis Neurosci 9:
26. Carandini M, Heeger DJ, Movshon JA (1997) Linearity and normalization in simple
cells of the macaque primary visual cortex. J Neurosci 17:8621–8644.
27. Abbott LF, Luo SX (2007) A step toward optimal coding in olfaction. Nat Neurosci 10:
28. Itskov V, Abbott LF (2008) Pattern capacity of a perceptron for sparse discrimination.
Phys Rev Lett 101:018101.
29. Crittenden JR, Skoulakis EM, Han KA, Kalderon D, Davis RL (1998) Tripartite
mushroom body architecture revealed by antigenic markers. Learn Mem 5:38–51.
30. Tanaka NK, Tanimoto H, Ito K (2008) Neuronal assemblies of the Drosophila
mushroom body. J Comp Neurol 508:711–755.
31. Assisi C, Stopfer M, Laurent G, Bazhenov M (2007) Adaptive regulation of sparseness
by feedforward inhibition. Nat Neurosci 10:1176–1184.
32. Barlow HB (1961) Possible principles underlying the transformation of sensory
messages. Sensory Communication, ed Rosenblith WA (MIT Press, Cambridge, MA).
33. Uttley AM (1979) Information Transmission in the Nervous System (Academic,
34. Atick JJ, Redlich AN (1990) Towards a theory of early visual processing. Neural
35. Simoncelli EP, Schwartz O (1999) Modeling non-specific suppression in V1 neurons
with a statistically-derived normalization model. Advances in Neural Information
Processing Systems, eds Kearns MS, Solla SA, Cohn DA (MIT Press, Cambridge, MA),
Vol 11, pp 153–159.
36. Reynolds JH, Heeger DJ (2009) The normalization model of attention. Neuron 61:
37. Ringach DF (2010) Population coding under normalization. Vision Res, 10.1016/j.
38. Stettler DD, Axel R (2009) Representations of odor in the piriform cortex. Neuron 63:
PN NL ORN
PN NL ORN
PN NL ORN
PN NL ORN
and KC responses. PN indicates that PN input was used with the full trans-
formation of Eq. 1, NL indicates that no lateral suppression was included by
setting m = 0 in Eq. 1, and ORN means that ORN rather than PN input was
used, eliminating both components of the antennal lobe transformation.
Percentage of error indicates the equal false-positive and false-negative
error rates. (A) Percentage of errors for LHNs receiving 20 inputs and tuned
as in Fig. 2. (B) Same as A, but for 5 inputs per LHN. (C) The number of odors
to which 2,500 randomly tuned KCs (constructed as in Fig. 3) are un-
responsive. (D) Percentage of errors for LHNs responding to odors at a vari-
ety of concentrations as in Fig. 4A.
Effect of feedforward nonlinearity and lateral suppression on LHN
| www.pnas.org/cgi/doi/10.1073/pnas.1005635107Luo et al.