Three major non-overlapping classes of interneurons expressing parvalbumin, somatostatin and vasoactive intestinal peptide (henceforth denoted PV, SST and VIP respectively) make up more than 80% of GABAergic cells of mouse cortex (Rudy et al., 2011). These neurons show cell-type-specific connectivity among themselves and with excitatory (E) neurons (Pfeffer et al., 2013; Jiang et al., 2015) forming a canonical microcircuit in the cortex. This microcircuit motif, initially proposed theoretically (Wang et al., 2004), has been the subject of numerous recent experimental studies using optogenetic tools applied to behaving mice (Lee et al., 2012; Saleem et al., 2013; Kepecs and Fishell, 2014; Hawrylycz et al., 2016) as well as computational studies (Lee and Mihalas, 2015; Lee and Mihalas, 2017; Lee et al., 2017; Yang et al., 2016; Yang and Wang, 2017). However, we still do not fully understand the mechanisms that underlie the behavior of this microcircuit which are often complex and counterintuitive.

A notable observation was that pyramidal neurons and VIP interneurons concomitantly increase their activities in the primary visual cortex V1 during locomotion in comparison with immobility (Niell and Stryker, 2010), even in the complete absence of visual input (Keller et al., 2012). Moreover, optogenetically activating (respectively inactivating) VIP interneurons mimics (respectively eliminates) the effect of running (Fu et al., 2014). Since VIP cells primarily target SST cells, a natural explanation for this phenomenon is disinhibition (Wang et al., 2004; Lee et al., 2013): activation of VIP cells suppresses SST cells, therefore neurons targeted by the SST population are disinhibited, enhancing the overall activity of excitatory neurons. However, recent experiments show that the network behavior might be more complex. Namely, in darkness the activation of VIP cells results in an average decrease of SST population activity (Fu et al., 2014), whereas in the presence of visual stimulation the response of SST cells is reversed and its firing rate increases during locomotion compared to immobility (Pakan et al., 2016). These findings, which have been further confirmed in a recent preprint (Dipoppa et al., 2017), appear to challenge the disinhibition hypothesis, suggesting that the nature of the interaction between VIP and SST could be stimulus dependent.

These experimental results raise two questions: First, the external activation of a population that directly inhibits a second population can trigger a positive response of the latter. What is the mechanism behind this apparently paradoxical behavior? Second, the same top-down modulation can trigger both a positive response and a negative response of certain populations of the circuit depending on the sensory input. Under which conditions can we expect one response or the other?

In this study, we model cortical activity and provide a comprehensive answers to these two questions. We show that these counterintuitive phenomena rely on two basic features of cortical networks: (i) the presence of multiple populations of interneurons and (ii) nonlinear responses to input. Finally, we use our model to predict complex behaviors that have not yet been experimentally tested. Beyond the mechanistic explanation for the observed behavior in mice V1, our work provides a very general and powerful framework to explain the dynamics of neural networks with multiple interneuron types, their context-dependent interactions, and the emsergence of counterintuitive effects that may occur across different cortical structures and animals.

We simulate microcircuit activity using a four population firing rate model. The average rate of each population is given by a nonlinear function of its input that we refer to as the f-I curve (Abbott and Chance, 2005). The f-I curve is such that when the input is low (below threshold), cells are little responsive to changes in external input. Instead for high input (above threshold) small changes in the input can drive substantial changes in the response (Miller and Troyer, 2002) (see Figure 1b). This nonlinearity has been analyzed experimentally and theoretically (Murphy and Miller, 2003; Phillips and Hasenstaub, 2016) and as we will show later, it is a key feature of the model.

Figure 1

Download asset Open asset

Populations are connected according to the microcircuit scheme in Figure 1a which contains the connections reported in both Jiang et al., 2015 and Pfeffer et al. (2013). We also consider three sources of input: (i) top-down modulation that targets VIP cells (ii) local recurrent input and (iii) constant background input set so that the populations have some fixed baseline activity (see Materials and methods for details).

Circuit behavior explained by response matrix

In order to formally characterize the steady state response of a population to external input we introduce the response matrix $M$. The intuition behind the response matrix is that if we change the input to population $j$ (where $j=E,P,S,V$ for excitatory, PV, SST and VIP populations respectively) by a small amount $\delta I_j$, then the change in rate of the population $i$ will be $\delta r_i=\delta I_j{M}_{ij}$. If $M_{ij}$ is positive (negative), an increase of the external excitation to $j$ will result in an increase (decrease) of the rate of population $i$ (see Materials and methods and Table 3 for details). In contrast to the connectivity matrix, which takes into account only the direct path from population $j$ to $i$, the response matrix contains information about all the possible ways in which population $j$ can affect population $i$, namely through indirect connections $j$-$h$-$i$. Due to the complexity of these indirect pathways, for different values of the connectivity matrix (but preserving the excitatory/inhibitory structure) $M_{ij}$ can be positive or negative irrespective of whether the connection from $j$ to $i$ is inhibitory or excitatory. Furthermore, due to the nonlinearities in the f-I curve, the response depends on the baseline rate of each of the populations and, as shown before, it can reverse its sign.

As an example, we analyze in detail the response of the SST population to external input to VIP cells. As we show in the Materials and methods section, this term of the response matrix is given by:

$M_{SV}=C{w}_{SV}\left((w_{EE}-d_E)(w_{PP}+d_P)-w_{EP}{w}_{PE}\right),$

where $w_{ij}$ are the absolute values of the connection weights and therefore are positive by definition and for the system to be stable $C$ has to be positive (see Materials and methods for details). The terms $d_i$ are proportional to the inverse of the first derivative of the f-I curves and are always positive. In particular, $d_E$ becomes arbitrarily large when the input is very low and tends monotonically to a positive constant $d_E^{\mathrm{\infty }}$ for high input. Therefore, if $w_{EE}\le d_E^{\mathrm{\infty }}$ then $M_{SV}$ will always be negative. However, for $w_{EE}>d_E^{\mathrm{\infty }}$ the behavior is much richer: if input is high then $d_E$ will be close to its minimum $d_E^{\mathrm{\infty }}$ and $w_{EE}>d_E$ allowing for $M_{SV}$ to be positive (provided that the product $w_{EP}{w}_{PE}$ is small enough). Instead if the input is low, $d_E$ will become very large and $M_{SV}$ will be negative.

It is remarkable that this change in the interaction between VIP and SST populations depends on the activation level of E: modifying the state of one population has a impact in the interactions between other populations. The heuristic explanation is that if the recurrent excitation is strong enough and the E population is already strongly excited (above threshold), a small decrease in the inhibition from SST to the E population can boost its activity and therefore strongly drive the whole microcircuit. If instead, the E population is in a low activation state the change in inhibition will have a weak effect that will not be able to reverse the response of SST.

This observation provides an explanation to the reversal of the response of SST to VIP activation when the baseline activity is changed: as we show in Figure 2a and c for low baseline activity, $M_{SV}$ is negative and the presence of an external excitatory current targeting VIP cells will result in a negative response of SST cells and positive response of E, PV and VIP cells, conforming to the disinhibitory hypothesis. On the other hand, for high baseline activity (panels 2b and 2d), the response of the SST population to input to VIP cells becomes positive leading to the response reversal regime.

Figure 2

Download asset Open asset

A similar analysis can be conducted for all terms in $M$. For example, another case of response reversal in this circuit is that of $M_{EE}$ which can have different signs for different baseline activity levels, meaning that the excitatory population can have a negative response to excitatory input to itself. Intuitively, if an external excitatory current targets the E population, its rate will increase transiently and thus the excitation that SST and VIP receive will also increase. If this effect is stronger in SST than in VIP the rate of the VIP population will decrease and therefore the inhibition that SST receives will decrease as well resulting in stronger inhibition to E cells. Note that for this to happen both SST and VIP have to be in the high activity baseline (i.e. $d_S$, $d_V$ have to be small) and $w_{SV}$, $w_{VS}$ have to be strong. The explicit expression of $M_{EE}$ (see Table 3) reveals that if the SST-VIP-SST loop is not strong enough or if $d_S$, $d_V$ are large $M_{EE}$ will always be positive.

Random network model

Experimental recordings showed a great diversity across neural responses even when recording from the same class of cells (Pyramidal, SST, PV or VIP) (Pakan et al., 2016). Although this diversity can have many origins, such as intrinsic heterogeneity in the cells within the same class, we proposed that random connectivity alone is sufficient to explain it. To do so we develop an extension of our model where each population is composed of multiple identical randomly connected rate units and where the probability that one connection exists from one unit to another depends on the populations of the presynaptic and postsynaptic units according to data extracted from Jiang et al. (2015); Pfeffer et al. (2013) (see Materials and methods for details).

For each unit, we measure the rate modulation (rate during top-down modulation minus baseline activity) for the different baselines. If the rate modulation is positive it means that the neuron is more active in the presence of the modulatory current and vice versa. In Figure 3, we show scatter plots of the rate modulation under the low baseline condition versus the rate modulation under the high baseline condition for each unit. These simulations reveal that the behavior of individual neurons can be quite variable while the population average still corresponds to the behavior of the population-based model. Since all units of each population are identical, variability in the response has to be due to the heterogeneity in the connectivity. This variability can result in cells within the same population having responses with opposite sign, as has been observed to be the case in mouse V1 (Reimer et al., 2014; Pakan et al., 2016) and A1 (Kuchibhotla et al., 2017). In addition, variability might also have further implications for gating of signals, since variability in inhibitory cells has been proposed to modulate the response gain of neural circuits (Mejias and Longtin, 2014).

Figure 3

Download asset Open asset

Model of mouse V1 accounts for experimental measurements

Our framework allows us to easily understand the counterintuitive behavior of V1 during locomotion. In the experiments mice with their head fixed face a screen where different visual stimuli are presented and can run freely on a treadmill (Fu et al., 2014; Pakan et al., 2016). Different visual stimuli result in different baseline activities in V1 and top-down modulation is triggered when the mice start running.

To model visual input we use external currents. In the case of size-varying gratings, this input has two sources: thalamic input that targets excitatory cells and cortical input that targets SST cells. In order to reproduce the surround suppression effect (Ozeki et al., 2009; Adesnik et al., 2012), excitatory cells have a small receptive field and therefore receive center input and SST cells have a large receptive field and receive surround input (see Materials and methods for details).

Figure 4b shows the response reversal phenomenon when a weak visual stimulus is presented. Before the visual stimulation, the SST has higher activity for immobility than for locomotion, by contrast, when the visual stimulus is presented, the activity of the SST population is higher for locomotion. In Figure 4c, we show the experimental data from Pakan et al. (2016) for three different experimental conditions (darkness, gray screen and grating) and in Figure 4d our simulations of V1 under the same conditions. Figure 4—figure supplement 1 shows the experimental data from the preprint (Dipoppa et al., 2017) for gratings of different sizes alongside with the behavior of our model.

Our simulations of this V1 circuit model reproduce the phenomena described in the literature: in the presence of visual stimulation, the activities of all populations, including SST, increase during locomotion (Pakan et al., 2016). In darkness, the activities of excitatory, PV and VIP populations increase during locomotion while the activity of SST decreases as reported in Fu et al. (2014) and in the preprint (Dipoppa et al., 2017). In Pakan et al. (2016), the response of SST to locomotion in darkness is weakly positive but this result is not statistically significant while the other two are.

To show that our results do not rely on a fine tuning of the connectivity parameters or even on certain details of the microcircuit structure, we have run the model with several connectivity matrices and perturbations of them (Figure 4—figure supplement 2) and we find that different connectivity parameters can reproduce the same circuit behavior as has been shown before in other systems (Marder et al., 2015). We have also considered other microcircuit structures to account for the differences between studies ([Pfeffer et al., 2013] reports projections from PV to VIP and (Jiang et al., 2015) from PV to SST) and we also consider thalamic input to PV (Figure 4—figure supplement 3). In all these cases, the results were consistent with our original findings showing that the phenomenon and the analysis are robust and not a peculiarity of one specific circuit.

Figure 4 with 3 supplements see all

Download asset Open asset

We have developed a theoretical model of cortical circuit with multiple interneuron types that accounts for newly identified complex interactions between cell types. The model has been used to reproduce and explain two counterintuitive phenomena observed in mouse cortex. First, in certain cases the activation of VIP cells results in an overall positive response of the SST population (Pakan et al., 2016). Second, the sign of the SST population response to excitation of VIP cells depends on the baseline activity of the circuit (Fu et al., 2014). Two features of the system lead to this behavior: the presence of multiple interneuron populations and the nonlinearity of f-I curves.

We explained heuristically the response reversal by closely looking at transient dynamics of the circuit. One experimentally testable prediction of our analysis is that, as Figure 1d and our calculations of the transient behavior show, in the response reversal regime, the overall SST population response to top-down modulation should initially decrease and later increase until reaching a higher rate than the baseline.

Based on our model, we introduced the response matrix $M$, which is a comprehensive framework to understand counterintuitive steady state responses. It provides explicit information about the contribution of each individual connection. For example by looking at the elements in $M_{SV}$ (see Table 3), one can readily see that if the recurrent excitation between pyramidal cells is not large enough, $M_{SV}$ can only be negative and therefore response reversal of SST would not happen. This statement can be easily tested by repeating the experiments while suppressing the activation of the E population. As we discussed before, another example is that if both SST and VIP populations have high baseline activities and if the SST-VIP-SST loop is strong enough, $M_{EE}$ can be negative, that is the excitatory population can have a negative response to excitatory input (see Table 3 for the explicit expression of $M_{EE}$). If the connections between the SST and the VIP populations are removed (or weakened) or if their baseline activities are sufficiently lowered $M_{EE}$ will always be positive. This constitutes another interesting prediction that can be experimentally tested.

Our calculations also revealed sign correlations between entries of $M$, for example $M_{SV}$ and $M_{SS}$ have opposite signs for any connectivity matrix (given the microcircuit) and for any baseline activity. This predicts that in the regime where SST activity has a positive response to excitatory input targeting VIP, SST has to have a negative response to external input targeting SST. In addition, our results are in line with experimental studies that show that VIP interneurons play an important role in cortical activity modulation (Mesik et al., 2015; Ibrahim et al., 2016; Jackson et al., 2016).

Our approach constitutes a general conceptual framework in which previous work regarding complex cortical interactions can be better understood (Tsodyks et al., 1997; Ozeki et al., 2009; Litwin-Kumar et al., 2016). The analysis of the response matrix shows that for the given microcircuit structure all terms of the matrix can be positive or negative. This is not the case in E-I networks (networks with one excitatory (E) population and only one inhibitory (I) population) (Tsodyks et al., 1997; Ozeki et al., 2009). In that case $M_{EE}$ and $M_{IE}$ are always positive, $M_{EI}$ is always negative and only $M_{II}$ can have both signs (see Materials and methods). In this sense, having more than one inhibitory population results in a much more versatile network. Another important point that can be derived from our calculations is the relationship between response reversal and inhibition stabilized networks (ISN) (Ozeki et al., 2009). Looking at the terms of the response matrix for an E-I network, we can see that the condition to have response reversal and the condition to be an ISN is the same: $W_{EE}$ has to be larger than $d_E^{\mathrm{\infty }}$. When analysing networks with more than one inhibitory population the relationship is not necessarily bidirectional any more. In the network that we analyzed, we found that in the high baseline activity the network is in the ISN regime and $M_{SV}$ is positive (as observed in [Litwin-Kumar et al., 2016), whereas in the low baseline activity the network is not in the ISN regime and $M_{SV}$ is negative, so in this case there is a clear relationship between being an ISN and exhibiting response reversal. However, the condition for other cases of response reversal such as $M_{EE}$ do not involve $W_{EE}$ and therefore do not require the network to be an ISN.

Finally, this study provides a parsimonious yet powerful explanation to striking observations of interneuronal circuits in V1 (Fu et al., 2014; Pakan et al., 2016; Lee et al., 2017) without requiring the assumption of top-down excitatory inputs explicitly targeting SST or PV neurons. Both our computational neural network model and the approach presented here (the response matrix analysis) go beyond circuit dynamics in mice V1 and can be easily applied to other species and cortical areas. By extending previous works (Tsodyks et al., 1997; Ozeki et al., 2009), it naturally explains the response reversal observed in cat visual cortex (Ozeki et al., 2009). It could also be applied to explain similar phenomena observed in mouse primary auditory cortex (Seybold et al., 2015; Kuchibhotla et al., 2017). In particular, in Kuchibhotla et al. (2017), the authors find that locomotion reduces the activity of excitatory cells. Assuming that the main modulation in the circuit is mediated by VIP cells this observation implies that $M_{EV}<0$ which is the case when the connections $W_{EP}$ and $W_{P}S$ are strong enough. In mouse somatosensory cortex, activating VIP neurons results in an intuitive decrease in SST activity, instead of a response reversal (Lee et al., 2013). As our results suggest, this qualitative difference between V1 and somatosensory cortex may be explained by the quantitative difference between their circuit architectures: in a recent study the authors showed that cell densities of different types of interneurons differ substantially across cortical areas resulting in counterintuitive impacts on circuit responses (Kim et al., 2017). These responses can be readily understood using the response matrix.

In this work, we mainly focused on steady-state responses. However, neural responses in many cortical areas, including primary auditory cortex, are largely transient and dynamical (Wehr and Zador, 2003). In addition, synaptic connections to and from interneurons are often subject to short-term plasticity (Reyes et al., 1998). Understanding transient dynamics in nonlinear, multi-type interneuronal circuits would be an important topic for future research.

We have shown that similar to the now well-known paradoxical effect that the presence of a single inhibitory neuron type can cause (Tsodyks et al., 1997; Ozeki et al., 2009), the presence of multiple types of interneurons has an even stronger impact on the activity of neural circuits. We have also exposed the effect of nonlinearity of the f-I curve. Our analysis suggests that in a circuit with multiple populations, the most interesting circuit behavior is found when spontaneous baseline activity is close to threshold since in that regime responses will change the most with small changes in population rates. These two features significantly broaden the richness of the dynamics of cortical circuits and enhance their usefulness for cognitive and behavioral computations. We conclude that computational models and mathematical analysis are critical to fully understand the dynamics of neural circuits underlying behavior (Gjorgjieva et al., 2016), especially when several types of interneurons are involved as intuition alone may be misleading and provide erroneous predictions on such circuits.

1. 1. Abbott LF
   2. Chance FS

   (2005) Drivers and modulators from push-pull and balanced synaptic input

   Progress in Brain Research 149:147–155.

   https://doi.org/10.1016/S0079-6123(05)49011-1

   • PubMed
   • Google Scholar
2. 1. Adesnik H
   2. Bruns W
   3. Taniguchi H
   4. Huang ZJ
   5. Scanziani M

   (2012) A neural circuit for spatial summation in visual cortex

   Nature 490:226–231.

   https://doi.org/10.1038/nature11526

   • PubMed
   • Google Scholar
3. Preprint

   1. Dipoppa M
   2. Ranson A
   3. Krumin M
   4. Pachitariu M
   5. Carandini M
   6. Harris KD

   (2017) Vision and locomotion shape the interactions between neuron types in mouse visual cortex

   bioRxiv.

   https://doi.org/10.1101/058396

   • Google Scholar
4. 1. Fu Y
   2. Tucciarone JM
   3. Espinosa JS
   4. Sheng N
   5. Darcy DP
   6. Nicoll RA
   7. Huang ZJ
   8. Stryker MP

   (2014) A cortical circuit for gain control by behavioral state

   Cell 156:1139–1152.

   https://doi.org/10.1016/j.cell.2014.01.050

   • PubMed
   • Google Scholar
5. 1. Gerstner W

   (2000) Population dynamics of spiking neurons: fast transients, asynchronous states, and locking

   Neural Computation 12:43–89.

   https://doi.org/10.1162/089976600300015899

   • PubMed
   • Google Scholar
6. 1. Gjorgjieva J
   2. Drion G
   3. Marder E

   (2016) Computational implications of biophysical diversity and multiple timescales in neurons and synapses for circuit performance

   Current Opinion in Neurobiology 37:44–52.

   https://doi.org/10.1016/j.conb.2015.12.008

   • PubMed
   • Google Scholar
7. 1. Hawrylycz M
   2. Anastassiou C
   3. Arkhipov A
   4. Berg J
   5. Buice M
   6. Cain N
   7. Gouwens NW
   8. Gratiy S
   9. Iyer R
   10. Lee JH
   11. Mihalas S
   12. Mitelut C
   13. Olsen S
   14. Reid RC
   15. Teeter C
   16. de Vries S
   17. Waters J
   18. Zeng H
   19. Koch C
   20. MindScope

   (2016) Inferring cortical function in the mouse visual system through large-scale systems neuroscience

   PNAS 113:7337–7344.

   https://doi.org/10.1073/pnas.1512901113

   • PubMed
   • Google Scholar
8. 1. Ibrahim LA
   2. Mesik L
   3. Ji XY
   4. Fang Q
   5. Li HF
   6. Li YT
   7. Zingg B
   8. Zhang LI
   9. Tao HW

   (2016) Cross-modality sharpening of visual cortical processing through layer-1-mediated inhibition and disinhibition

   Neuron 89:1031–1045.

   https://doi.org/10.1016/j.neuron.2016.01.027

   • PubMed
   • Google Scholar
9. 1. Jackson J
   2. Ayzenshtat I
   3. Karnani MM
   4. Yuste R

   (2016) VIP+ interneurons control neocortical activity across brain states

   Journal of Neurophysiology 115:3008–3017.

   https://doi.org/10.1152/jn.01124.2015

   • PubMed
   • Google Scholar
10. 1. Jiang X
    2. Shen S
    3. Cadwell CR
    4. Berens P
    5. Sinz F
    6. Ecker AS
    7. Patel S
    8. Tolias AS

    (2015) Principles of connectivity among morphologically defined cell types in adult neocortex

    Science 350:aac9462.

    https://doi.org/10.1126/science.aac9462

    • Google Scholar
11. 1. Keller GB
    2. Bonhoeffer T
    3. Hübener M

    (2012) Sensorimotor mismatch signals in primary visual cortex of the behaving mouse

    Neuron 74:809–815.

    https://doi.org/10.1016/j.neuron.2012.03.040

    • PubMed
    • Google Scholar
12. 1. Kepecs A
    2. Fishell G

    (2014) Interneuron cell types are fit to function

    Nature 505:318–326.

    https://doi.org/10.1038/nature12983

    • PubMed
    • Google Scholar
13. 1. Kim Y
    2. Yang GR
    3. Pradhan K
    4. Venkataraju KU
    5. Bota M
    6. García Del Molino LC
    7. Fitzgerald G
    8. Ram K
    9. He M
    10. Levine JM
    11. Mitra P
    12. Huang ZJ
    13. Wang XJ
    14. Osten P

    (2017) Brain-wide maps reveal stereotyped cell-type-based cortical architecture and subcortical sexual dimorphism

    Cell 171:456–469.

    https://doi.org/10.1016/j.cell.2017.09.020

    • PubMed
    • Google Scholar
14. 1. Kuchibhotla KV
    2. Gill JV
    3. Lindsay GW
    4. Papadoyannis ES
    5. Field RE
    6. Sten TA
    7. Miller KD
    8. Froemke RC

    (2017) Parallel processing by cortical inhibition enables context-dependent behavior

    Nature Neuroscience 20:62–71.

    https://doi.org/10.1038/nn.4436

    • PubMed
    • Google Scholar
15. 1. Lee SH
    2. Kwan AC
    3. Zhang S
    4. Phoumthipphavong V
    5. Flannery JG
    6. Masmanidis SC
    7. Taniguchi H
    8. Huang ZJ
    9. Zhang F
    10. Boyden ES
    11. Deisseroth K
    12. Dan Y

    (2012) Activation of specific interneurons improves V1 feature selectivity and visual perception

    Nature 488:379–383.

    https://doi.org/10.1038/nature11312

    • PubMed
    • Google Scholar
16. 1. Lee S
    2. Kruglikov I
    3. Huang ZJ
    4. Fishell G
    5. Rudy B

    (2013) A disinhibitory circuit mediates motor integration in the somatosensory cortex

    Nature Neuroscience 16:1662–1670.

    https://doi.org/10.1038/nn.3544

    • PubMed
    • Google Scholar
17. 1. Lee JH
    2. Mihalas S

    (2015) Cell-type specific connectivity accounts for diverse in vivo functional roles of inhibitory neurons in V1

    BMC Neuroscience 16:P165.

    https://doi.org/10.1186/1471-2202-16-S1-P165

    • Google Scholar
18. 1. Lee JH
    2. Koch C
    3. Mihalas S

    (2017) A computational analysis of the function of three inhibitory cell types in contextual visual processing

    Frontiers in Computational Neuroscience 11:28.

    https://doi.org/10.3389/fncom.2017.00028

    • PubMed
    • Google Scholar
19. 1. Lee JH
    2. Mihalas S

    (2017) Visual processing mode switching regulated by VIP cells

    Scientific Reports 7:1843.

    https://doi.org/10.1038/s41598-017-01830-0

    • PubMed
    • Google Scholar
20. 1. Litwin-Kumar A
    2. Rosenbaum R
    3. Doiron B

    (2016) Inhibitory stabilization and visual coding in cortical circuits with multiple interneuron subtypes

    Journal of Neurophysiology 115:1399–1409.

    https://doi.org/10.1152/jn.00732.2015

    • PubMed
    • Google Scholar
21. 1. Marder E
    2. Goeritz ML
    3. Otopalik AG

    (2015) Robust circuit rhythms in small circuits arise from variable circuit components and mechanisms

    Current Opinion in Neurobiology 31:156–163.

    https://doi.org/10.1016/j.conb.2014.10.012

    • PubMed
    • Google Scholar
22. 1. Mejias JF
    2. Longtin A

    (2014) Differential effects of excitatory and inhibitory heterogeneity on the gain and asynchronous state of sparse cortical networks

    Frontiers in Computational Neuroscience 8:107.

    https://doi.org/10.3389/fncom.2014.00107

    • PubMed
    • Google Scholar
23. 1. Mesik L
    2. Ma WP
    3. Li LY
    4. Ibrahim LA
    5. Huang ZJ
    6. Zhang LI
    7. Tao HW

    (2015) Functional response properties of VIP-expressing inhibitory neurons in mouse visual and auditory cortex

    Frontiers in Neural Circuits 9:22.

    https://doi.org/10.3389/fncir.2015.00022

    • PubMed
    • Google Scholar
24. 1. Miller KD
    2. Troyer TW

    (2002) Neural noise can explain expansive, power-law nonlinearities in neural response functions

    Journal of Neurophysiology 87:653–659.

    https://doi.org/10.1152/jn.00425.2001

    • PubMed
    • Google Scholar
25. 1. Murphy BK
    2. Miller KD

    (2003)

    Multiplicative gain changes are induced by excitation or inhibition alone

    Journal of Neuroscience 23:10040–10051.

    • PubMed
    • Google Scholar
26. 1. Niell CM
    2. Stryker MP

    (2010) Modulation of visual responses by behavioral state in mouse visual cortex

    Neuron 65:472–479.

    https://doi.org/10.1016/j.neuron.2010.01.033

    • PubMed
    • Google Scholar
27. 1. Ozeki H
    2. Finn IM
    3. Schaffer ES
    4. Miller KD
    5. Ferster D

    (2009) Inhibitory stabilization of the cortical network underlies visual surround suppression

    Neuron 62:578–592.

    https://doi.org/10.1016/j.neuron.2009.03.028

    • PubMed
    • Google Scholar
28. 1. Pakan JM
    2. Lowe SC
    3. Dylda E
    4. Keemink SW
    5. Currie SP
    6. Coutts CA
    7. Rochefort NL

    (2016) Behavioral-state modulation of inhibition is context-dependent and cell type specific in mouse visual cortex

    eLife 5:e14985.

    https://doi.org/10.7554/eLife.14985

    • PubMed
    • Google Scholar
29. 1. Pfeffer CK
    2. Xue M
    3. He M
    4. Huang ZJ
    5. Scanziani M

    (2013) Inhibition of inhibition in visual cortex: the logic of connections between molecularly distinct interneurons

    Nature Neuroscience 16:1068–1076.

    https://doi.org/10.1038/nn.3446

    • PubMed
    • Google Scholar
30. 1. Phillips EA
    2. Hasenstaub AR

    (2016) Asymmetric effects of activating and inactivating cortical interneurons

    eLife 5:e18383.

    https://doi.org/10.7554/eLife.18383

    • PubMed
    • Google Scholar
31. 1. Reimer J
    2. Froudarakis E
    3. Cadwell CR
    4. Yatsenko D
    5. Denfield GH
    6. Tolias AS

    (2014) Pupil fluctuations track fast switching of cortical states during quiet wakefulness

    Neuron 84:355–362.

    https://doi.org/10.1016/j.neuron.2014.09.033

    • PubMed
    • Google Scholar
32. 1. Reyes A
    2. Lujan R
    3. Rozov A
    4. Burnashev N
    5. Somogyi P
    6. Sakmann B

    (1998) Target-cell-specific facilitation and depression in neocortical circuits

    Nature Neuroscience 1:279–285.

    https://doi.org/10.1038/1092

    • PubMed
    • Google Scholar
33. 1. Rudy B
    2. Fishell G
    3. Lee S
    4. Hjerling-Leffler J

    (2011) Three groups of interneurons account for nearly 100% of neocortical GABAergic neurons

    Developmental Neurobiology 71:45–61.

    https://doi.org/10.1002/dneu.20853

    • PubMed
    • Google Scholar
34. 1. Saleem AB
    2. Ayaz A
    3. Jeffery KJ
    4. Harris KD
    5. Carandini M

    (2013) Integration of visual motion and locomotion in mouse visual cortex

    Nature Neuroscience 16:1864–1869.

    https://doi.org/10.1038/nn.3567

    • PubMed
    • Google Scholar
35. 1. Seybold BA
    2. Phillips EA
    3. Schreiner CE
    4. Hasenstaub AR

    (2015) Inhibitory actions unified by network integration

    Neuron 87:1181–1192.

    https://doi.org/10.1016/j.neuron.2015.09.013

    • PubMed
    • Google Scholar
36. 1. Tsodyks MV
    2. Skaggs WE
    3. Sejnowski TJ
    4. McNaughton BL

    (1997)

    Paradoxical effects of external modulation of inhibitory interneurons

    Journal of Neuroscience 17:4382–4388.

    • PubMed
    • Google Scholar
37. 1. Wang X-J
    2. Tegner J
    3. Constantinidis C
    4. Goldman-Rakic PS

    (2004) Division of labor among distinct subtypes of inhibitory neurons in a cortical microcircuit of working memory

    PNAS 101:1368–1373.

    https://doi.org/10.1073/pnas.0305337101

    • Google Scholar
38. 1. Wehr M
    2. Zador AM

    (2003) Balanced inhibition underlies tuning and sharpens spike timing in auditory cortex

    Nature 426:442–446.

    https://doi.org/10.1038/nature02116

    • PubMed
    • Google Scholar
39. 1. Yang GR
    2. Murray JD
    3. Wang XJ

    (2016) A dendritic disinhibitory circuit mechanism for pathway-specific gating

    Nature Communications 7:12815.

    https://doi.org/10.1038/ncomms12815

    • PubMed
    • Google Scholar
40. 1. Yang GR
    2. Wang X

    (2017)

    A disinhibitory motif and flexible information routing in the brain

    Current Opinion in Neurobiology In press.

    • Google Scholar

Author details

1. Luis Carlos Garcia del Molino

   Center for Neural Science, New York University, New York, United States

   Contribution

   Conceptualization, Formal analysis, Investigation, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-9934-9461

2. Guangyu Robert Yang

   Center for Neural Science, New York University, New York, United States

   Contribution

   Conceptualization, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8919-4248

3. Jorge F Mejias

   Center for Neural Science, New York University, New York, United States

   Present address

   Swammerdam Institute for Life Sciences, Center for Neuroscience, Faculty of Science, University of Amsterdam, Amsterdam, Netherlands

   Contribution

   Conceptualization, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8096-4891

4. Xiao-Jing Wang

   Center for Neural Science, New York University, New York, United States

   Contribution

   Conceptualization, Funding acquisition, Writing—review and editing

   For correspondence

   xjwang@nyu.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-3124-8474

• 3,014

  views

• 561

  downloads

• 56

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.