Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Murakawa et al
1
Fast-spiking neurons in monkey orbitofrontal
cortex underlie economic value computation
Tomoaki Murakawa1, Takashi Kawai2, Yuri Imaizumi 3, Hiroshi Yamada4*
Short title: Economic value computations in monkey orbitofrontal cortex
1: Academic service office for the medical science area, University of Tsukuba,
1-1-1 Tenno-dai, Tsukuba, Ibaraki 305-8577, Japan
2: The Picower Institute for Learning and Memory, Department of Biology and
Department of Brain and Cognitive Sciences, Massachusetts Institute of
Technology, Cambridge, MA 02139, USA.
3: College of medical sciences, University of Tsukuba, 1-1-1 Tenno-dai,
Tsukuba, Ibaraki 305-8577, Japan
4: Division of Biomedical Science, Institute of Medicine, University of Tsukuba,
1-1-1 Tenno-dai, Tsukuba, Ibaraki 305-8577, Japan.
*Correspondence to Hiroshi Yamada, Ph.D.
Division of Biomedical Science, Institute of Medicine, University of Tsukuba
1-1-1 Tenno-dai, Tsukuba, Ibaraki, 305-8577 Japan
Tel: 81-29-853-6013; e-mail: h-yamada@md.tsukuba.ac.jp
Acknowledgements
The authors express their appreciation to Yoshiko Yabana, Rika Akitake, and Shiho
Nishino for their technical assistance. We appreciate Yasuhiro Tsubo for his valuable
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
2
comments. Monkey FU was provided by NBRP “Japanese Monkeys” through the
National Bio Resource Project of MEXT, Japan. This study was supported by JSPS
KAKENHI Grant Number JP:15H05374, 24K02135, JST Moonshot R&D JPMJMS2294
(H.Y.).
Author Contributions
H.Y. designed the study; Y.I. and H.Y. conducted the experiment; T.M., T.K., and H.Y.
analyzed the data; H.Y. wrote the manuscript. All authors approved the final manuscript.
Conflict of interest: The authors declare no competing interests.
Data availability: All data used in this study are presented in the manuscript.
Keywords
Orbitofrontal cortex, inhibitory interneuron, monkey, economic behavior
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
3
ABSTRACT (149/150)
Inhibitory interneurons are fundamental constituents of cortical circuits that process
information to shape economic behaviors. However, the role of inhibitory interneurons in
this process remains elusive at the core cortical reward-region, orbitofrontal cortex (OFC).
Here, we show that presumed parvalbumin-containing GABAergic interneurons (fast-
spiking neurons, FSNs) cooperate with presumed regular-spiking pyramidal neurons
(RSNs) during economic-values computation. While monkeys perceived a visual lottery
for probability and magnitude of rewards, identified FSNs occupied a small subset of
OFC neurons (12%) with high-frequency firing-rates and wide dynamic-ranges, both are
key intrinsic cellular characteristics to regulate cortical computation. We found that FSNs
showed higher sensitivity to the probability and magnitude of rewards than RSNs.
Unambiguously, both neural populations signaled expected values (i.e., probability times
magnitude), but FSNs processed these reward’s information strongly governed by the
dynamic range. Thus, cooperative information processing between FSNs and RSNs
provides a common cortical framework for computing economic values.
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
4
INTRODUCTION
Activity of inhibitory interneurons regulates information flow in the cortical and subcortical
structure (1-5). This computational process is thought to rely on circuit structures that
regulate the economic behavior of animals. Indeed, cortical inhibitory dysfunction results
in various diseases including mental disorders (6, 7). Since excitatory neurons constitute
the majority of neurons at the core cortical center, the orbitofrontal cortex (OFC), they
have been well examined in relation to economic behavior to obtain rewards (8-14).
However, it remains unclear how OFC inhibitory interneurons are involved in shaping
economic behaviors, especially in macaque monkeys, close relatives to humans.
Parvalbumin-containing GABAergic interneurons have been identified as fast-
spiking neurons (FSNs) in the brain based on their narrow spike waveform (15-18).
Although cortical excitatory activity is regulated by inhibitory interneurons (3, 4, 19-21),
FSN activity in the cortical brain region, especially in monkeys, has only been examined
in a small number of studies of cognitive and motor task performance (1, 22-26). To our
best knowledge, very few studies has examined the role of FSNs in the OFC during
economic behavior in both monkeys and rodents. This is largely because FSNs
constitute a minority of neurons; thus, only a small amount of sample data can be
obtained in a single study. Given this limitation, it is challenging to elucidate the inhibitory
mechanism of FSNs at monkey OFC, which process economic-value computations as
suspected from the inhibitory dysfunction (6, 27).
In the present study, we aimed to understand how FSNs regulate OFC activity during
gambling behavior in monkeys. We differentiated FSNs from presumed regular-spiking
pyramidal neurons (RSNs) based on spike waveforms recorded extracellularly from the
OFC of behaving monkeys. We addressed two critical issues in examining the role of
FSNs: 1) How are FSNs in the OFC of behaving monkeys involved in perceiving
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
5
expected values, i.e., probability multiplied by magnitude of reward?; 2) How does the
activity of FSNs differ from that of RSNs in the OFC when computing expected values?
Our results suggest that FSNs compute expected values in coordination with RSNs in
the OFC, governed by the dynamic range.
RESULTS
Identification of FSNs and their basic firing properties
We studied total 377 neurons in the OFC of behaving monkeys during a single cue task
(Figure 1A). We previously reported monkey behavior during a choice task (10) (Figure
1B-D). In short, monkeys chose the option with higher expected value (i.e., probability
times magnitude). While the monkeys looked at a visual lottery (Figure 1A), neuronal
activity was recorded from the OFC (Figure 1E; medial [mOFC]; 14O, central [cOFC],
13M). We classified the neurons into FSNs and RSNs based on the spike waveforms
(Figure 2A), according to the procedure previously used in rat and monkey studies (2,
19). A scatter plot of peak width (i.e., width at the half maximum of the negative peak
amplitude) against peak-to-valley width (i.e., time from negative peak to valley) for all
neurons formed two clusters (Figure 2A). We classified the FSNs as neurons in one
cluster that exhibited narrow spike waveforms (Figure 2A, green; see insets). The
identified FSNs accounted for approximately 12% (42/377; cOFC, n = 25; mOFC, n= 17)
of the recorded OFC neurons. We previously reported the activity of RSNs (10, 12, 13)
but not the activity of FSNs during the cued lottery task. We note that we did not record
the OFC activity during choice task.
Typical FSN activity recorded from the cOFC showed tonic firing of >10 Hz in most
of the task periods, with a phasic increase in discharges for some task events (Figure
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
6
2B, top). Another example of FSN activity recorded from the mOFC showed a phasic
increase in discharges during the start of a trial, and an increase and decrease in activity
throughout the trial was observed (Figure 2B, bottom). We first examined these firing
rate changes through a trial before and after the visual cue for probability and magnitude
appeared (Figure 1A, see gray bars for seven analysis periods). Quantitative comparison
of the average firing rates between 42 FSNs and 335 RSNs among seven task periods
(Start1, Start2, Cue1, Cue2, Cue3, Cue4, and Pre-fb, see gray bars in Figure 2B)
demonstrated higher firing-rates in FSNs than RSNs throughout a task trial (Figure 2C,
two-way ANOVA, n = 377, neuron type, F(1,363) = 97.9, P < 0.001, task period, F(6,363) =
1.94, P = 0.0731, interaction, F(6,363) = 1.01, P = 0.420). Thus, the FSNs identified in the
OFC of the behaving monkeys showed a typical characteristic commonly observed in
FSNs.
Specifically, FSNs changed their activity at different times during the task trials
(Figure 2D). Approximately 60% of FSNs demonstrated peak activity during cue
presentation (Figure 2D, top, 59.5%, 25/42), whereas a similar proportion of RSNs
showed peak activity during cue presentation (Figure 2E, 44.5%, 149/335, Chi-square
test, n = 42, P = 0.093, X2 = 2.82, df = 1). Peak activity with short latencies was observed
in the FSNs (Figure 2F, latency: Wilcoxon rank-sum test, n = 174, P < 0.001, W = 2671.5,
df = 1) with higher magnitudes of activity (Figure 2G, peak firing rate: n = 174, P < 0.001,
W = 896, df = 1). The speed of activity changes were similar between the two types of
neurons (Figure 2H, half-peak width: Wilcoxon rank-sum test, n = 174, P = 0.160, W =
2190.5, df = 1). In addition, the dynamic ranges (see Material and Methods) in FSNs
were wider than those in RSNs (Figure 2I, n = 377, P < 0.001, W = 262337.5, df = 1),
which is the critical characteristics to process computation (28, 29). We also confirmed
that the baseline firing rates during the inter-trial interval were higher in FSNs (Figure 2J,
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
7
n = 377, P < 0.001, W = 2794, df = 1), similar to activity during the task trials (Figure 2C).
Collectively, high-frequency activity with short latency occurred in FSNs, in contrast
to lower firing rates in RSNs. Unambiguously, the dynamic range of FSNs was wider than
that of RSNs, indicating that the identified FSNs showed characteristics that matched
those of parvalbumin-containing GABAergic interneurons (30-32).
Coordinated coding of expected values in FSNs and RSNs
Next, we examined how individual FSNs and RSNs processed the probability and
magnitude of rewards during the expected values computation. First, after the cue
appearance, 40% to 50% of the FSNs encoded the probability and magnitude of rewards
until the outcome appeared (Figure 3A, left). We identified four coding types: probability,
magnitude, expected values, and risk-return types. For example, neurons signaling the
expected value were found (Figure 3B, see Figure 4E, Cue1), whose activity increased
if either probability or magnitude of rewards becomes larger (i.e., EV+ type, see also
Figure 3A, reddish). In addition, the probability (Figure 3C, P- type, see also Figure 4E,
Cue1 and Figure 3A, bluish) and magnitude (Figure 3D, M+ type, see also Figure 4E,
Cue1 and Figure 3A, greenish) types were found for both positive and negative coding
type. Indeed, the neural signals carried by FSNs and RSNs were composed of a mixture
of these signals (Figure 3A, left and right), such as the signals for the expected value
and its components (i.e., probability and magnitude). Thus, Both FSNs and RSNs signal
information for the expected value computations.
Next, we compared the encoded information between FSNs and RSNs at the
population level. Both neural populations encoded the expected values after cue
presentation, as observed in the regression slopes close to 45° angle (Figure 3E, Cue1).
This expected value code evolved immediately after the appearance of the cues (Figure
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
8
3E, see Cue1, gray line, general linear model, n = 1885, coefficient, F = 257.6, P < 0.001,
df = 1). Thereafter, they gradually lost these expected value signals throughout the trial,
as indicated by changes in the regression slopes (Figure 3G, task period, F = 2.83, P =
0.023, df = 4). The signal change to probability code (regression slop close to 0° angle)
occurred concurrently in both types, although the regression slopes were consistently
larger in FSNs than in RSNs throughout a trial (cell type, F = 10.4, P = 0.001, df = 1),
indicating FSNs are closer to the expected value code compared to RSNs (i.e., RSNs
are closer to probability code). Thus, FSNs and RSNs may share some single
computational process at the population level or at the local circuit level, suggesting the
existence of common cortical computation in the OFC.
In addition to these comparisons, we compared the amount of carried information in
FSNs and RSNs. Carried information by the FSNs was larger compared to RSNs,
irrespective of probability or magnitude information (Figure 3F, Four-way ANOVA, n =
1700, neuron type, F(1,1660) = 229.4, P < 0.001, coding type, F(1,1660) = 2.23, P = 0.135,
task period, F(4,1660) = 9.51, P < 0.001). Thus, information carried by the FSNs is larger
than RSNs, while the carried information changed through a task trial.
Dynamic range and carried information during expected value computation
Finally, we examined how the cortical local circuit structure rely on the expected value
computations between FSNs and RSNs. For this purpose, we analyzed the influence of
the dynamic range on the extent of carried information by FSNs and RSNs, which is one
of the key factors regulating cortical computation according to the local circuit structure
(Figure 4A) (30). We found that the dynamic range affected amount of carried information
in both FSNs and RSNs (Figure 4B, dynamic range, F = 1109.5, P < 0.001, df = 1).
Unambiguously, a wider dynamic range of FSNs co-occurred with stronger neural
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
9
modulations, hence larger carried information (Figure. 4B, top) than in RSNs (Figure. 4B,
bottom) (cell type, F = 41.6, P < 0.001, df = 1), whereas there was no significant
difference between the probability and magnitude information (Fig. 4B, green and blue,
coefficient type, F = 2.08, P = 0.150, df = 1). Thus, both FSNs and RSNs process
expected value computations under the influence of dynamic range, suggesting the local
circuit inhibition (Figure. 4A) may control the computational process.
We also made model selection approach to explore the factors that best explained
the encoded information in each of FSNs and RSNs. We found that the combination of
the average firing rates in each task period (FR) and the dynamic range (DR) best
explained the information processing in both FSNs and RSNs (Figure 4C, see the best
model, red) (log-likelihood ratio test, P < 0.001 for all conditions). The same model best
explains the amount of carried information in both types. Thus, FSNs and RSNs
cooperate in the OFC circuit with the slightly different dynamic range during the expected
values computation for economic behavior. We note that the instantaneous firing rate
(FR) predominantly affect the amount of the carried information (See Figure 4C, x-label
for the selected models in rank order), while coefficient type (CT, i.e., probability or
magnitude) only model was the worst one (Worse than null model).
DISCUSSION
In the present study, we analyzed the activity of OFC neurons recorded during economic
behavior in monkeys. We differentiated FSNs from other neurons (i.e., RSNs) based on
their spike waveforms. Thereafter, we found two properties inherent to FSNs compared
to RSNs. First, FSNs displayed high frequency firing rates and wide dynamic range, in
contrast to RSNs. Second, the neural representation of the probability and magnitude of
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
10
rewards (i.e., carried information for economic behavior) was similar but quantitatively
different between the two classes; while FSNs encoded reward information similar to
RSNs in terms of the proportion of neurons (Figure 3A) and of the regression coefficient
(Figure 3E), signals carried by FSNs were more selective to the probability and
magnitude of rewards to signals expected values (Figure 3E and F). Furthermore, we
found that dynamic range is a key factor explaining these information processing in both
FSNs and RSNs, with a significantly stronger dependence in FSNs on dynamic range
(Figure 4B and C). These findings suggest that FSNs regulate information processing to
compute expected values in coordination with RSNs via local circuit inhibition during
economic behavior in monkeys.
Identification of FSNs with the spike waveform in the primate OFC
In the in vivo cortical structure, FSNs have been identified based on extracellularly
recorded spike waveforms in other neurons in rodents (19). In Bartho et al., neurons in
the rat prefrontal cortex were identified based on a narrow spike waveform recorded
extracellularly, which reflects the intracellular properties of the action potential (33). Most
of the identified neurons showed inhibitory effects on neighboring neurons, while none
of these neurons showed excitatory effects (Figure 4 in Bartho et al., 2004), indicating
that these narrow spike-waveform neurons were inhibitory interneurons. Accumulating
evidence from in vivo and in vitro studies of cortical and subcortical structures supports
the hypothesis that narrow spike-waveform neurons are parvalbumin-containing
GABAergic interneurons (FSNs) in rodents (4, 5, 20, 21, 34) and monkeys (1, 2, 35). The
electrophysiological and neurochemical properties of the cortical and subcortical
structures are similar between primates and rodents (15, 18), and it is generally agreed
that FSNs recorded from behaving monkeys are parvalbumin-containing GABAergic
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
11
interneurons.
In the present study, we identified FSNs based on spike waveforms, similar to
previous rodent studies on cortical and subcortical structures (19, 34). The identified
FSNs exhibited high-frequency firing rates during the task period (approximately 10 Hz)
compared to RSNs (Figure 2C). However, the average firing rate of FSNs in this study
was lower than that in other monkey’s studies in the visual (>30 Hz) (24) and prefrontal
cortices (>20 Hz) (36). This discrepancy may arise from differences in cortical regions as
well as behavioral tasks performed by the monkeys, because neural firing rates depend
on input to the local circuit (Figure 4A, gray), although cortical areas share a six-layer
structure composed of different types of interneurons (18).
The spike waveform is one of the predominant characteristics used to identify neuron
types in vivo; however, it cannot differentiate between all neuron types. RSNs appear to
be comprised of multiple neuron types. In addition, spike waveforms are strongly
dependent on the amplifier filter settings: the frequency of the low-pass and high-pass
filters and the type of filter (e.g., Butterworth, Bessel, or Chebyshev) (2); hence, the
characteristics of the spike waveform must be compared in the same experimental
settings. Thus, we reliably identified FSNs in the present study.
Dynamic range, firing rates, and information conveyed during economic behavior
In the present study, we found a similar but slightly stronger neural modulation in FSNs
than in RSNs at the neuronal population level (Figure 3E-F and Figure 4B). This finding
contrasts with that in the striatum, where FSNs are less selective than output neurons
(2). While the local circuit structures differed between the cortical and subcortical
structures (37-39), both FSNs in the cortical and subcortical structures consistently
showed high-frequency baseline firing rates. The reason for the higher firing rates in
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
12
FSNs must be their intrinsic membrane properties of FSNs, such as high input resistance
(30, 32). If the input resistance is high, the neurons are easy to become active to
excitatory inputs. As a result, high-frequency firing rates (Figure 2C and J), and larger
changes in task-related activity (Figure 2G), and hence, wider dynamic range (Figure 2I)
were observed. These neural properties might be related to the larger changes in carried
information as a function of firing rates and dynamic range (Figure 4B, compare FSNs
and RSN regression slopes, Figure 4C, red). As a result, the output neurons in cortical
(9, 10, 12, 13) and subcortical (40-43) structures becomes active via feedforward
inhibition (Figure 4A) during economic behavior.
We found that relations in reward processing for probability and magnitude were
similar between the FSNs and RSNs (Figure 3), but a stronger dependency in FSNs on
the dynamic range was observed (Figure 4B). These similarity and difference between
FSNs and RSNs should de derived from local circuit structure: mutual inhibition between
FSNs and RSNs as well as feedforward inhibition from FSNs to RSNs (Figure 4A, green).
The mutual inhibition determines mean firing rates of the circuitry neurons according to
the excitatory inputs level, while the feed forward inhibition determines output level of
circuit, i.e., RSN’s activity. While the excitatory inputs was not able to be observed in this
study, these two key properties of local circuit are possible to regulate expected value
computations. Indeed recurrent inhibition is known to control circuit dynamics (44). Thus,
the inhibition of FSNs on RSNs may yield the expected value computation for economic
behavior.
Coordinated coding of reward probability and magnitude information by FSNs and
RSNs
Our data suggest that FSNs may regulate discharge selectivity of RSNs in the OFC
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
13
according to the higher selectivity to the reward information (Figure 3E and F), which
was related to the wider dynamic range (Figure 4B). Functional role of the inhibitory
interneurons to regulate neural selectivity has been suggested in the cerebral cortex.
FSNs with parvalbumin immunoreactivity in visual area V1 of mice have been shown to
be selectively involved in shaping orientation tuning and enhancing the directional
selectivity of neighboring neurons (45). Furthermore, an inhibitory role of FSNs in
improving various cognitive functions in distinct cortical regions has been suggested as
follows. For example, FSNs in the monkey prefrontal cortex have demonstrated a
relationship with the learning and performance of cognitive tasks (26, 46). FSNs in visual
area V4 showed modulation in their control of attention (24), suggesting that the reliability
of the output neuron’s response is increased by reducing response variability. Thus, feed-
forward inhibition (Figure 4A) could be a general mechanism for improving output
selectivity, while the input structure is the key factor in driving a local network.
An unambiguous finding of this study was that the coding of the probability and
magnitude of rewards by FSNs was similar to that by RSNs (Figure 4). Previous monkey
studies of other prefrontal regions have also indicated that FSN activity is selective for
reward cues (25, 36). Why and how does this coordinated coding of reward information
occur in the local circuit (Figure 4A), thereby producing similarities and differences
between FSNs and RSNs? If common inputs excite neighboring FSNs and RSNs
simultaneously in the cortical circuit (Figure 4A, grays), neural selectivity would be similar.
Neighboring RSNs must be suppressed by the inhibition of FSNs (Figure 4A, green), and
the balance between excitatory and inhibitory effects must determine neural selectivity
for reward information. In contrast, if divergent inputs drive these adjacent cortical
neurons, both FSNs and RSNs might sometimes be selective for the probability and
magnitude of rewards, but the neural selectivity could be different among these two
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
14
neuron types because of the input difference. Further studies are required to elucidate
the local circuit dynamics as input–output structures produced by local inhibition in the
cortices.
One limitation of our study was that we did not examine the activity of directly
connected FSN–RSN pairs. Therefore, we could not directly test the possibilities
mentioned above. Previous studies have mostly been performed in the prefrontal cortex,
striatum, and hippocampus, but no study has identified FSNs in the monkey OFC, which
is involved in economic behavior.
Materials and Methods
Subjects and experimental procedures
Two rhesus monkeys were used in this study (Macaca mulatta, SUN, 7.1 kg, male;
Macaca fuscata, FU, 6.7 kg, female). All experimental procedures were approved by the
Animal Care and Use Committee of the University of Tsukuba (protocol no 23-057) and
performed in compliance with the US Public Health Service’s Guide for the Care and Use
of Laboratory Animals. Each animal was implanted with a head restraint prosthesis. Eye
movements were measured using a video camera system at 120 Hz. Visual stimuli were
generated using a liquid-crystal display at 60 Hz, placed 38 cm from the monkey’s face
when seated. The subjects performed the cued lottery task 5 days a week. The subjects
practiced the cued lottery task for ten months, after which they became proficient in
choosing lottery options. We have previously reported the activity of RSNs but have not
reported the activity of FSNs during this task.
Behavioral task
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
15
Cued lottery tasks. The animals performed one of two visually cued lottery tasks: single
cue task or choice task. Neuronal activity was only recorded during the single cue task.
Single cue task: At the beginning of each trial, the monkeys had 2 s to align their gaze
within 3º of a 1º-diameter gray central fixation target. After fixating for 1 s, an 8º pie chart
providing information about the probability and magnitude of the rewards was presented
for 2.5 s at the same location as that of the central fixation target. The pie chart was then
removed and 0.2 s later, and a 1 kHz and 0.1 kHz tone of 0.15 s duration indicated the
reward and no-reward outcomes, respectively. The animals received a fluid reward, for
which the magnitude and probability were indicated by green and blue pie charts,
respectively; otherwise, no reward was delivered. A high tone preceded the reward by
0.2 s. A low tone indicated that no reward was delivered. An intertrial interval of 4 to 6 s
followed each trial.
Choice task: At the beginning of each trial, the monkeys had 2 s to align their gaze
within 3º of a 1º-diameter gray central fixation target. After fixing for 1 s, two peripheral 8º
pie charts providing information on the probability and magnitude of rewards for each of
the two target options were presented for 2.5 s, at 8º to the left and right of the central
fixation location. Gray 1° choice targets appeared at the same locations. After a 0.5 s
delay, the fixation target disappeared, cueing saccade initiation. The animals were free
to choose for 2 s by shifting their gaze to either target within 3º of the choice target. A 1
kHz and 0.1 kHz tone of 0.15 s duration indicated reward and no-reward outcomes,
respectively. The animals received a fluid reward, indicated by the green pie chart of the
chosen target, with the probability indicated by the blue pie chart; otherwise, no reward
was delivered. An intertrial interval of 4 to 6 s followed each trial.
Pay-off and block structure. Green and blue pie charts indicated reward magnitudes from
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
16
0.1 to 1.0 mL, in 0.1 mL increments, and reward probabilities from 0.1 to 1.0, in 0.1
increments, respectively. A total of 100 pie charts were used in this study. In the single
cue task, each pie chart was presented once in random order. In the choice task, two pie
charts were randomly assigned to the two options. During one session of
electrophysiological recording, approximately 30 to 60 trial blocks of the choice task were
sometimes interleaved with 100 to 120 trial blocks of the single cue task.
Calibration of the reward supply system. A precise amount of liquid reward was controlled
and delivered to the monkeys using a solenoid valve. An 18-gauge tube (0.9 mm inner
diameter) was attached to the tip of the delivery tube to reduce variation across trials.
The reward amount in each payoff condition was calibrated by measuring the weight of
water with a precision of 0.002 g (2 L) on a single-trial basis. This calibration method
was the same as that described previously (9).
Electrophysiological recordings
Conventional techniques were used to record single-neuron activity in the cOFC and
mOFC. Monkeys were implanted with recording chambers (28 mm × 32 mm) targeting
the OFC and striatum, centered 28 mm anterior to the stereotaxic coordinates. The
locations of the chambers were verified using anatomical magnetic resonance imaging
(MRI). At the beginning of the daily recording sessions, a stainless-steel guide tube was
placed within a 1-mm spacing grid, and a tungsten microelectrode (1-3 M, FHC) was
passed through the guide tube. To record neurons in the mOFC and cOFC, the electrode
was lowered until it approximated the bottom of the brain after passing through the
cingulate cortex, dorsolateral prefrontal cortex, or between them. Electrophysiological
signals were amplified, bandpass filtered, and monitored. Single-neuron activity was
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
17
isolated based on spike waveforms. We recorded from the two brain regions of a single
hemisphere of each of the two monkeys (179 in monkey SUN and 198 in monkey FU):
42 SFNs (cOFC, 25, mOFC, 17), and 335 RSNs (cOFC, 182, mOFC, 153) for FSNs and
RSNs. The activity of all individual neurons was sampled when the activity of an isolated
neuron demonstrated a good signal-to-noise ratio (>2.5). Blinding was not performed.
The sample sizes required to detect effect sizes (number of recorded neurons, number
of recorded trials in a single neuron, and number of monkeys) were estimated according
to previous studies (9, 40, 47, 48). Neural activity was recorded during 100-120 trials of
the single cue task. During the choice trials, neural activity was not recorded.
Classification of neuron type.
In the analysis, FSNs (presumed to be parvalbumin-containing GABAergic interneurons)
were differentiated from RSNs (presumed to be pyramidal neurons) by their spike width
(i.e., the width at the half maximum of the negative peak amplitude and the width of the
spike from peak to valley), according to a previous study (19). We classified the FSNs
as neurons in one cluster that exhibited narrow spike waveforms. In our previous reports
(10, 12, 13, 43), we reported the activity of RSNs but not of FSNs. The number of
reported RSNs in this study differed from that in previous studies because we did not
perform a quantitative classification of these neurons based on the waveform in those
studies.
Statistical analysis
Statistical analyses were performed using the R statistical software package
(http://www.r-project.org/). All statistical tests for behavioral and neural analyses were
two-tailed.
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
18
Effects of units on statistical analysis. In the present study, we used two variables for
analysis: probability and magnitude. We defined the probability of the reward from 0.1 to
1.0, and the magnitude of the reward from 0.1 to 1.0 mL. Under this unit definition, the
effects of probability and magnitude on the data were equivalent.
Behavioral analysis
No new behavioral results were included; however, the procedure for the behavioral
analysis was as follows:
We previously reported that monkey behavior depends on expected values defined
as the probability time magnitude (10). We described the analysis steps to check whether
the monkey’s behavior reflected task parameters, such as reward probability and
magnitude. Importantly, we showed that the monkeys’ choice behavior reflected the
expected values of the rewards, i.e., the probability multiplied by the magnitude. For this
purpose, the percentage choosing the right option was analyzed in the pooled data using
a general linear model with a binomial distribution:
PchoosesR = 1 / (1 + e-z) (3)
where the relationship between PchoosesR and Z is given by the logistic function in each
of the following three models: number of pie segments (M1), probability and magnitude
(M2), and expected values (M3).
M1: Z = b0 + b1NpieL + b2NpieR (4)
where b0 is the intercept, and NpieL and NpieR are the number of pie segments contained
in the left and right pie chart stimuli, respectively. The values of b0 to b2 are free
parameters and were estimated by maximizing the log likelihood.
M2: Z = b0 + b1PL + b2PR + b3ML + b4MR (5)
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
19
where b0 is the intercept; PL and PR are the probabilities of rewards for the left and right
pie chart stimuli, respectively; and ML and MR are the magnitudes of rewards for the left
and right pie chart stimuli, respectively. The values of b0 to b4 are free parameters and
are estimated by maximizing the log likelihood.
M3: Z = b0 + b1EVL + b2EVR (6)
where b0 is the intercept and EVL and EVR are the expected values of rewards as
probability multiplied by magnitude for the left and right pie chart stimuli, respectively.
The values of b0 to b2 are free parameters and were estimated by maximizing the log
likelihood. We identified the best model to describe the monkeys’ behavior by comparing
their goodness-of-fit based on Akaike’s information criterion (AIC) and Bayesian
information criterion (BIC) (49).
Neural analysis.
Peri-stimulus time histograms were drawn for each single neuron activity aligned at the
onset of a visual cue. The average activity curves were smoothed using a 50-ms
Gaussian kernel (σ = 50 ms). We analyzed neural activity during a 2.5-s period of pie
chart stimulus presentation in the single cue task, including baseline activity before the
presentation of cues during a 1.0 s fixation period. The firing rates of each neuron during
the 0.5 s time window were estimated every 0.5 s for a total of seven analysis periods
named Start1, Start2, Cue1, Cue2, Cue3, Cue4, and Pre-fb (feedback). A Gaussian
kernel was not used for statistical analyses.
Basic firing properties, such as peak firing rates, peak latency, duration of peak
activity (half-peak width), and dynamic range, were compared among the four brain
regions using parametric or nonparametric tests, with a statistical significance level of P
< 0.05. The dynamic range (DR) was defined as the firing rate difference between the
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
20
maximum and minimum among the seven task periods after cue presentation: Start1,
Start2, Cue1, Cue2, Cue3, Cue4, and Pre-fb. Baseline firing rates 1 s before the
appearance of the central fixation targets were also compared, with a statistical
significance level of P < 0.05. Yamada et al. (2021) also analyzed the basic firing
properties of RSNs, but not for FSNs.
Linear regression to detect firing modulations in each individual neuron. Neural discharge
rates (F) were fitted using the following variables:
F = b0 + bp Probability + bm Magnitude (8)
where Probability and Magnitude are the probability and magnitude of the rewards
indicated by the pie chart, respectively. b0 is the intercept. If bp and bm are not zero at P
< 0.05, the discharge rates were regarded as being significantly modulated by that
variable.
On the basis of the linear regression, activity modulation patterns were categorized
into several types: “Probability” (P) type with a significant bp and without a significant bm;
“Magnitude” (M) type without a significant bp and with a significant bm; “Expected value”
(EV) type with significant bp and bm with the same sign (i.e., positive bp and positive bm
or negative bp and negative bm); “Risk-Return” (RR) type with significant bp and bm with
both having opposite signs (i.e., negative bp and positive bm or positive bp and negative
bm) and “non-modulated” type without significant bp and bm. The risk–return types reflect
high-risk high returns (prefer low probability and large magnitude) or low-risk low returns
(prefer high probability and low magnitude).
We compared the basic firing properties and activity modulations between the FSNs
and RSNs as follows: 1) proportion of neuron types using the chi-square test; 2) average
firing rates using ANOVA, Kruskal-Wallis test, or Wilcoxon rank-sum test with Bonferroni
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
21
correction for multiple comparisons; and 3) regression coefficients using a general linear
model, such as ANOVA and linear regression.
Linear regression to detect firing modulations at the level of population. The regression
coefficients for reward magnitude (R) were fitted using the following variables:
R = b0 + b1 Rp + b2 Task period + b3 Cell type (9)
where Rp denotes the regression coefficient of the reward probability. Task period was a
categorical variable composed of Cue1, Cue2, Cue3, Cue4, and Pre-fb. Cell type is a
categorical variable comprising FSNs and RSNs. If b1 to b3 are not zero at P < 0.05, the
discharge rates were regarded as being significantly modulated by that variable.
Dynamic range and neural modulations. To analyze the influence of basic firing
properties on the regression coefficients for the probability and magnitude of rewards,
we modeled how the dynamic range and average firing rates affect neural modulation as
follows:
R = b0 + b1 DR + b2 FR + b3 CT (10)
Where R is the absolute value of the regression coefficients for the probability and
magnitude of the rewards, bp and bm in Eq. 8. b0 is the intercept. DR is the dynamic range.
FR is the average firing rates in each of the five task periods after cue presentation: Cue1,
Cue2, Cue3, Cue4, and Pre-fb. CT is the regression coefficient type (i.e., probability or
magnitude) as a categorical parameter. If b1 is not 0 at P < 0.05, neural modulation by
the probability and magnitude of rewards was regarded as significantly affected by the
dynamic range of neurons. If b2 is not 0 at P < 0.05, neural modulation by the probability
and magnitude of rewards was regarded as significantly affected by the average firing
rate in each neuron. If b3 is not 0 at P < 0.05, neural modulations by the probability and
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
22
magnitude of rewards were different among the probability and magnitude of rewards,
i.e., the regression coefficient types.
Model comparisons. To identify the best structural model to describe neural modulation,
as described above, we applied a model selection approach based on all possible
combinations of variables in Eq. 10. We sought a combination of best-fit parameters to
explain the neural modulation based on the probability and magnitude of rewards. We
compared the goodness of fit based on AIC and BIC (49).
AIC (Model) = −2L + 2k (10)
BIC (Model) = −2L + k log n (11)
where, L is the maximum log-likelihood of the model, k is the number of free parameters,
and n is the sample size. After estimating the best-fit parameters for each model, the
model that exhibited the smallest AIC and BIC values was selected. To evaluate the
model fit, we estimated the difference between these values and the null model’s AIC or
BIC, which is the log likelihood under the assumption that all free parameters are zero in
the model, except the intercept, b0. We used the log-likelihood ratio test for each of the
selected models to the null model at P < 0.05.
REFERENCES
1. G. Gonzalez-Burgos, L. S. Krimer, N. V. Povysheva, G. Barrionuevo, D. A.
Lewis, Functional properties of fast spiking interneurons and their synaptic
connections with pyramidal cells in primate dorsolateral prefrontal cortex.
J
Neurophysiol
93, 942-953 (2005).
2. H. Yamada
et al.
, Characteristics of fast-spiking neurons in the striatum of
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
23
behaving monkeys.
Neurosci Res
105, 2-18 (2016).
3. R. Wilbers
et al.
, Structural and functional specializations of human fast-
spiking neurons support fast cortical signaling.
Sci Adv
9, eadf0708 (2023).
4. N. Giordano
et al.
, Fast-Spiking Interneurons of the Premotor Cortex
Contribute to Initiation and Execution of Spontaneous Actions.
J Neurosci
43,
4234-4250 (2023).
5. S. Chamberland
et al.
, Brief synaptic inhibition persistently interrupts firing
of fast-spiking interneurons.
Neuron
111, 1264-1281 e1265 (2023).
6. R. Hattori, K. V. Kuchibhotla, R. C. Froemke, T. Komiyama, Functions and
dysfunctions of neocortical inhibitory neuron subtypes.
Nat Neurosci
20,
1199-1208 (2017).
7. P. Allami, N. Yazdanpanah, N. Rezaei, The role of neuroinflammation in PV
interneuron impairments in brain networks; implications for cognitive
disorders.
Rev Neurosci
, (2025).
8. E. L. Rich, J. D. Wallis, Decoding subjective decisions from orbitofrontal
cortex.
Nat Neurosci
19, 973-980 (2016).
9. H. Yamada, K. Louie, A. Tymula, P. W. Glimcher, Free choice shapes
normalized value signals in medial orbitofrontal cortex.
Nat Commun
9, 162
(2018).
10. H. Yamada, Y. Imaizumi, M. Matsumoto, Neural Population Dynamics
Underlying Expected Value Computation.
J Neurosci
41, 1684-1698 (2021).
11. A. Pastor-Bernier, A. Stasiak, W. Schultz, Reward-specific satiety affects
subjective value signals in orbitofrontal cortex during multicomponent
economic choice.
Proc Natl Acad Sci U S A
118, (2021).
12. Y. Imaizumi, A. Tymula, Y. Tsubo, M. Matsumoto, H. Yamada, A neuronal
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
24
prospect theory model in the brain reward circuitry.
Nat Commun
13, 5855
(2022).
13. H. Chen
et al.
, Stable Neural Population Dynamics in the Regression
Subspace for Continuous and Categorical Task Parameters in Monkeys.
eNeuro
10, (2023).
14. C. Padoa-Schioppa, Neuronal origins of choice variability in economic
decisions.
Neuron
80, 1322-1336 (2013).
15. Y. Kawaguchi, C. J. Wilson, S. J. Augood, P. C. Emson, Striatal interneurones:
chemical, physiological and morphological characterization [published
erratum appears in Trends Neurosci 1996 Apr;19(4):143].
Trends Neurosci
18,
527-535 (1995).
16. B. Cauli
et al.
, Molecular and physiological diversity of cortical nonpyramidal
cells.
J Neurosci
17, 3894-3906 (1997).
17. J. DeFelipe, Types of neurons, synaptic connections and chemical
characteristics of cells immunoreactive for calbindin-D28K, parvalbumin and
calretinin in the neocortex.
J Chem Neuroanat
14, 1-19 (1997).
18. Y. Kawaguchi, S. Kondo, Parvalbumin, somatostatin and cholecystokinin as
chemical markers for specific GABAergic interneuron types in the rat frontal
cortex.
J Neurocytol
31, 277-287 (2002).
19. P. Bartho
et al.
, Characterization of neocortical principal cells and
interneurons by network interactions and extracellular features.
J
Neurophysiol
92, 600-608 (2004).
20. S. F. Owen, J. D. Berke, A. C. Kreitzer, Fast-Spiking Interneurons Supply
Feedforward Control of Bursting, Calcium, and Plasticity for Efficient
Learning.
Cell
172, 683-695 e615 (2018).
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
25
21. R. J. Hatch, G. D. C. Mendis, K. Kaila, C. A. Reid, S. Petrou, Gap Junctions
Link Regular-Spiking and Fast-Spiking Interneurons in Layer 5
Somatosensory Cortex.
Front Cell Neurosci
11, 204 (2017).
22. B. Wang
et al.
, Firing Frequency Maxima of Fast-Spiking Neurons in Human,
Monkey, and Mouse Neocortex.
Front Cell Neurosci
10, 239 (2016).
23. N. V. Povysheva
et al.
, Properties of excitatory synaptic responses in fast-
spiking interneurons and pyramidal cells from monkey and rat prefrontal
cortex.
Cereb Cortex
16, 541-552 (2006).
24. J. F. Mitchell, K. A. Sundberg, J. H. Reynolds, Differential attention-
dependent response modulation across cell classes in macaque visual area V4.
Neuron
55, 131-141 (2007).
25. T. Kawai, H. Yamada, N. Sato, M. Takada, M. Matsumoto, Preferential
Representation of Past Outcome Information and Future Choice Behavior by
Putative Inhibitory Interneurons Rather Than Putative Pyramidal Neurons
in the Primate Dorsal Anterior Cingulate Cortex.
Cereb Cortex
29, 2339-2352
(2019).
26. C. Constantinidis, P. S. Goldman-Rakic, Correlated discharges among
putative pyramidal neurons and interneurons in the primate prefrontal
cortex.
J Neurophysiol
88, 3487-3497 (2002).
27. T. Prevot, E. Sibille, Altered GABA-mediated information processing and
cognitive dysfunctions in depression and other brain disorders.
Mol
Psychiatry
26, 151-167 (2021).
28. H. Zhang
et al.
, Neural-WDRC: A Deep Learning Wide Dynamic Range
Compression Method Combined With Controllable Noise Reduction for
Hearing Aids.
Trends Hear
29, 23312165241309301 (2025).
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
26
29. W. L. Shew, H. Yang, T. Petermann, R. Roy, D. Plenz, Neuronal avalanches
imply maximum dynamic range in cortical networks at criticality.
J Neurosci
29, 15595-15600 (2009).
30. Y. Kawaguchi, Physiological subgroups of nonpyramidal cells with specific
morphological characteristics in layer II/III of rat frontal cortex.
J Neurosci
15, 2638-2655 (1995).
31. N. V. Povysheva
et al.
, Parvalbumin-positive basket interneurons in monkey
and rat prefrontal cortex.
J Neurophysiol
100, 2348-2360 (2008).
32. Y. Kawaguchi, Physiological, morphological, and histochemical
characterization of three classes of interneurons in rat neostriatum.
J
Neurosci
13, 4908-4923 (1993).
33. D. A. Henze
et al.
, Intracellular features predicted by extracellular recordings
in the hippocampus in vivo.
J Neurophysiol
84, 390-400 (2000).
34. G. J. Gage, C. R. Stoetzner, A. B. Wiltschko, J. D. Berke, Selective activation
of striatal fast-spiking interneurons during choice execution.
Neuron
67, 466-
479 (2010).
35. J. Kunimatsu, S. Yamamoto, K. Maeda, O. Hikosaka, Environment-based
object values learned by local network in the striatum tail.
Proc Natl Acad Sci
U S A
118, (2021).
36. H. Fan, X. Pan, R. Wang, M. Sakagami, Differences in reward processing
between putative cell types in primate prefrontal cortex.
PLoS One
12,
e0189771 (2017).
37. H. Inokawa, N. Matsumoto, M. Kimura, H. Yamada, Tonically Active Neurons
in the Monkey Dorsal Striatum Signal Outcome Feedback during Trial-and-
error Search Behavior.
Neuroscience
446, 271-284 (2020).
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
27
38. Y. Tsubo, Y. Isomura, T. Fukai, Neural dynamics and information
representation in microcircuits of motor cortex.
Front Neural Circuits
7, 85
(2013).
39. Y. Isomura, R. Harukuni, T. Takekawa, H. Aizawa, T. Fukai, Microcircuitry
coordination of cortical motor information in self-initiation of voluntary
movements.
Nat Neurosci
12, 1586-1593 (2009).
40. H. Yamada
et al.
, Coding of the long-term value of multiple future rewards in
the primate striatum.
J Neurophysiol
109, 1140-1151 (2013).
41. H. Yamada, H. Inokawa, N. Matsumoto, Y. Ueda, M. Kimura, Neuronal basis
for evaluating selected action in the primate striatum.
Eur J Neurosci
34,
489-506 (2011).
42. H. Yamada, N. Matsumoto, M. Kimura, History- and current instruction-
based coding of forthcoming behavioral outcomes in the striatum.
J
Neurophysiol
98, 3557-3567 (2007).
43. H. Chen
et al.
, Formation of brain-wide neural geometry during visual item
recognition in monkeys.
iScience
28, 111936 (2025).
44. M. B. Lynn
et al.
, Nonlinear recurrent inhibition through facilitating
serotonin release in the raphe.
Nat Neurosci
, (2025).
45. S. H. Lee
et al.
, Activation of specific interneurons improves V1 feature
selectivity and visual perception.
Nature
488, 379-383 (2012).
46. X. L. Qi, C. Constantinidis, Correlated discharges in the primate prefrontal
cortex before and after working memory training.
Eur J Neurosci
36, 3538-
3548 (2012).
47. H. Yamada, N. Matsumoto, M. Kimura, Tonically active neurons in the
primate caudate nucleus and putamen differentially encode instructed
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
28
motivational outcomes of action.
J Neurosci
24, 3500-3510 (2004).
48. K. Enomoto, N. Matsumoto, H. Inokawa, M. Kimura, H. Yamada, Topographic
distinction in long-term value signals between presumed dopamine neurons
and presumed striatal projection neurons in behaving monkeys.
Sci Rep
10,
8912 (2020).
49. K. Burnham, D. Anderson, Multimodel inference: understanding AIC and
BIC in model selection.
Sociol. Method Res.
33, 261–304 (2004).
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
29
Figures
Figure 1. Task, behavior, and recording sites.
A Sequence of events during the single cue task. A single pie chart with green and blue
segments was presented visually to the monkeys. B Choice task. Two pie charts were
presented visually to the monkeys on the left and right sides of the center. After visual fixation
on the central point, it disappeared, and the monkeys chose either of the targets by fixating
on it. A block of choice trials was sometimes interleaved between the single cue trial blocks.
During the choice trials, neural activity was not recorded. C Percentages of right target
choices during the choice task plotted against the expected values (EVs) of the left and right
target options. Aggregated choice data were used. D Percentage of right target choices
estimated in each recording session (gray lines) plotted against the difference in expected
values (right minus left). The choice data were segmented by seven conditions of the
difference in the expected values: -1.0 ~ -0.5, -0.5 ~ -0.3, -0.3 ~ -0.1, -0.1 ~ 0.1, 0.1 ~ 0.3,
0.3 ~ 0.5, and 0.5 ~1.0. The black plots indicate the mean values. E Illustration of neural
recording areas based on sagittal MR Neurons were recorded from the medial (mOFC, 14O,
orbital part of area 14) and central parts of the orbitofrontal cortex (cOFC, 13M, medial part
of area 13) at the A31-A34 anterior-posterior (A-P) level. These figures are taken from
Yamada et al. (2021).
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
30
Figure 2. Classification of FSNs and their basic activity properties during task trials.
A Scatter plots of mean spike waveform durations (x, width at the half maximum of the
negative peak amplitude; y, width from peak to valley; see inset) for OFC neurons. FSNs
were defined as neurons in one cluster that exhibited narrow spike waveforms (green).
Neurons in clusters with wider spike waveforms were classified as RSNs. B Two examples
of FSN activity recorded from the cOFC of monkey SUN and the mOFC of monkey FU during
the single cue task. Rasters and histograms were aligned for each behavioral event. The
seven gray bars indicate the 0.5 s analysis periods. All histograms (50-ms bins) were
smoothed using a Gaussian kernel (50 ms). C Average firing rates of 42 FSNs and 335 RSNs
during seven analysis periods. D Color map histograms of FSN and RSN activity. Each
horizontal line indicates the neural activity aligned with cue onset averaged for all lottery
conditions. Neuronal firing rates were normalized to peak activity. E Percentage of neurons
showing an activity peak during cue presentation. F Peak activity latency after cue
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
31
presentation. G Firing rates of peak activity observed during cue presentation. H Half-peak
width, indicating the phasic nature of activity changes. I Dynamic range defined as the
difference between maximum and minimum firing rates. J Box plots of baseline firing rates
during the 1 s time period before the presentation of the central fixation target. In E-J,
asterisks indicate statistical significance between the two neural populations (Wilcoxon rank-
sum test, *P < 0.05, **P < 0.01).
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
32
Figure 3. Probability and magnitude modulations in FSNs and RSNs.
A Percentages of neural modulation types for FSNs and RSNs during the seven analysis
periods. Probability (P), magnitude (M), expected value (EV), and risk–return (RR) types
were detected based on the significance of the positive and negative regression coefficients.
B-D Examples of FSNs for EV+, M+, and P- are shown. Reward probability (P) is
differentiated among low, middle, and high conditions. Reward magnitude is also
differentiated among low, middle, and high conditions. The gray-hatched time windows are
the analysis periods, Cue1. E Regression coefficients for the probability and magnitude of
rewards during a task trial. The gray lines indicate the regression slopes. F Box plots of the
regression coefficient for the probability and magnitude of rewards among positive- and
negative-coding type. Asterisks indicate statistical significance between the two neural
populations (Wilcoxon rank-sum test, *P < 0.05, **P < 0.01).
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
Murakawa et al
33
Figure 4. Dynamic rage of firing rates differed between FSN and RSN in neural
modulation.
A Schematic depiction of the cortical circuit for the presumed parvalbumin-containing
GABAergic interneurons (FSN) and the presumed output pyramidal neurons (RSN). Below
indicates information processing via inhibition is shown from 1 to 4. B Plots of regression
coefficients for probability (blue) and magnitude (green) of rewards against dynamic range
for FSNs (left) and RSNs (right). Gray lines show the regression slopes from the general
linear model. C Plots of the difference in Bayesian information criterion values between the
top seven models and the null model. The X-axis labels indicate the selected models in rank
order. In A and B, DR, dynamic range; FR, firing rate; CT, regression coefficient type (i.e.,
probability or magnitude). In B, red indicate the best model.
.CC-BY-NC-ND 4.0 International licensemade available under a
(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is
The copyright holder for this preprintthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.06.647503doi: bioRxiv preprint
