The development and introduction of genetic methods in rodents to label and manipulate the major IN populations in cortex (PV, SST, VIP and ID2 or Lamp5) have been transformative, and have resulted in major advances in our understanding of cortical circuits ^70,71. It is now evident that these IN groups are heterogeneous and are comprised of discrete subtypes that often have important differences in connectivity and function. For example, PV INs encompass both basket cells and chandelier cells, populations that share similarities in molecular gene expression and intrinsic electrophysiological properties but that have different output connectivity: basket cell terminals target the soma and proximal dendrite of PCs whereas axo-axonic cells target the PC axon initial segment. Recent work has underscored the importance of this next level of organization, as illustrated by the monumental single-cell transcriptomic analyses of cortical neuron types by the Allen Institute ^33–35,39, the analysis of the diversity of PV basket cells ^72–74, of chandelier cells ⁷⁵, and of SST INs ^76–81. Linking the transcriptomic diversity of IN subtypes to their functional properties, circuit function and in vivo activity has begun to be achieved via remarkable post-hoc efforts ^33,81–85, but this remains a dauntingly complex endeavor. Considering that no single non-overlapping set of genes have been identified for the vast majority of IN subtypes, progress in understanding the functional specialization of IN subpopulations has benefitted from the development of new genetic approaches that give experimental access to this diversity in a practical and reasonable way.

In this study we applied intersectional genetic methods to reveal clearly distinct functional populations of VIP INs (Figure 1). We found that VIP INs can be subdivided into four main groups that exhibit distinct laminar organization, stereotypical patterns of axonal and dendritic arborization and differences in input and output connectivity (Figures 2&4). The two types of VIP/CCK INs, which account for about 30% of L2-4 VIP neurons both provide inhibition onto PCs (Figure 4 C), but had interesting differences in their synaptic properties (Figure 5) as well as their transcriptomic profiles (e.g., Ca²⁺ related genes such as Cadps2; Supplementary Figure 1). They are predicted to mediate inhibition on PCs while the VIP/CR cells and the non-CCK non-CR VIP INs predominantly target SST INs and are predicted to be disinhibitory (Figure 4 C). In vivo experiments exploring the effect on the cortical network of light stimulation of ChR2 expressing VIP/CCK or VIP/CR neuron subtypes in freely moving mice provided support to these predictions (Figure 7).

Interestingly, although both the VIP/CR and the non-CCK non-CR VIP INs predominantly target SST INs, we found that they differed in the populations of L2/3 SST INs that receive strong innervation from each subtype. VIP/CR INs mainly target SST INs in deep L2/3, while the non-CCK non-CR VIP INs equally targeted superficial and deep L2/3 SST INs (Figure 4). Distinct SST IN subtypes, with different localization in the cortical column have been shown to inhibit different types of PCs and/or different compartments of the same PC ⁸⁰. This predicts that distinct VIP IN subtypes may disinhibit distinct subsets of PCs. Moreover, we found that non-CCK non-CR VIP neurons but not VIP/CR cells also target PV INs, albeit much less than SST INs, suggesting that inputs targeting these VIP INs produce disinhibition of both the soma and the distal dendrite of PCs, and thus could facilitate both feedforward and feedback inputs (Figure 4).

Examination of the inputs to VIP INs in vS1 by long-range projections also indicated that there is differential targeting of VIP subtypes. Glutamatergic inputs from vM1 preferentially targeted non-CCK non-CR VIP INs, whereas these VIP neurons received the weakest input from cholinergic axons from the basal forebrain (Figure 6 A-K). Lastly, we found strong noradrenergic responses on both VIP/CCK and non-CCK non-CR VIP INs but not on VIP/CR cells (Figure 6 L-M). Given their differential efferent connectivity, this indicates that the functional effect of any stimulus that activates VIP INs depends on which VIP IN subpopulations are targeted ⁸⁶. Together, the results suggest that VIP IN subtypes form highly specific circuits, as has been proposed for SST IN subtypes ⁸⁰. While the molecular basis for the differential input and output connectivity of VIP IN subtypes is not yet clear, there are some notable cell surface molecules that show selective expression in specific populations (e.g., the protocadherin Pcdh11x in non-CCK non-CR VIP INs; Supplementary Figure 1). Considering recent evidence for distinct VIP synaptic dynamics on different SST IN subtypes⁸⁷, it will be interesting to elucidate how the molecular heterogeneity of VIP and SST subpopulations results in specialized disinhibitory subcircuits.

L2-4 VIP INs project a significant amount of axon to deep cortical layers (Figure 2 and Supplementary Figures 2 C-E, 3 F-I, K), where a diverse population of SST INs reside ^3,76–80. Furthermore, paired recordings have shown that L2/3 VIP cells strongly connect to deep layer SST cells ^9,88,89. It will be of great interest to investigate whether L2-4 VIP/CR and non-CCK non-CR VIP INs differentially target distinct L5-6 SST subpopulations. This requires the development of methods to identify specific SST subtypes that are compatible with the genetics used to select distinct VIP IN subtypes for stimulation ⁹⁰. The recent discovery of enhancers labeling specific SST IN subpopulations is promising in this regard ⁹¹. These studies will also require the ability to express optogenetic actuators in VIP INs in superficial cortical layers without contamination from VIP INs in deep layers.

The VIP-SST disinhibitory circuit motif has emerged as a key circuit mechanism for top-down modulation and context-dependent cortical processing (see recent reviews by ^{3,5,15,92–99}. The existence of VIP IN subtypes with different output connectivity may explain some results that appear to contradict the view that increased activity of VIP neurons produces disinhibition of excitatory neurons ¹⁰⁰. For instance, while in V1, locomotion increases sensory responses ⁹⁴, in auditory cortex (A1) it does the opposite, although in both cases VIP INs are recruited. The VIP disinhibitory circuit has been implicated as the cause of increased responses in V1 ⁹⁴. A recent study, while confirming that movement suppresses responses in auditory cortex, shows that in contrast, optogenetic activation of VIP INs in this cortical area also increases stimulus-evoked spike rates ¹⁰¹. One possibility is that while activating VIP INs optogenetically excites all VIP subtypes, in A1, movement might preferentially recruit inhibitory VIP/CCK cells. In another example Batista-Brito (2017) ²¹ showed that deletion of the schizophrenia-related gene ErbB4 specifically from VIP INs impaired VIP IN function and produced dramatic effects on visual responses and severe deficits in cortical activity, visual responses, sensory perceptual function and other behavioral changes characteristic of models of schizophrenia. However, deletion of ErbB4 from VIP-INs also led to a marked increase in the activity of cortical excitatory neurons. One possibility is that ErbB4 deletion in VIP cells primarily affected VIP/CCK INs, and that it is the impairment of VIP IN-mediated inhibition that is responsible for the effects.

There is increasing evidence for heterogeneity of VIP IN activity in vivo. A recent study investigating stimuli encoding in visual cortex reported the presence of distinct subpopulations of VIP INs (∼5) encoding specific combinations of novel stimuli, omissions of familiar stimuli or behavioral outcomes in an image change detection task ¹⁰². It is possible that this observed heterogeneity is related to the diversity of VIP IN populations revealed by this study or via transcriptomic studies. Differences in intrinsic electrophysiology, connectivity, and response to neuromodulators, as observed in this study, could all affect the respective in vivo activities of VIP subpopulations ¹⁰³. In sum, it is increasingly clear that the heterogeneity of VIP INs observed in molecular profiling and genetic studies is functionally important, and provides additional flexibility and modularity to disinhibitory circuitry in the brain.