A Role for Somatostatin-Positive Interneurons in Neuro-Oscillatory and Information Processing Deficits in Schizophrenia

Abstract

Alterations in neocortical GABAergic interneurons (INs) have been affiliated with neuropsychiatric diseases, including schizophrenia (SZ). Significant progress has been made linking the function of a specific subtype of GABAergic cells, parvalbumin (PV) positive INs, to altered gamma-band oscillations, which, in turn, underlie perceptual and feedforward information processing in cortical circuits. Here, we review a smaller but growing volume of literature focusing on a separate subtype of neocortical GABAergic INs, somatostatin (SST) positive INs. Despite sharing similar neurodevelopmental origins, SSTs exhibit distinct morphology and physiology from PVs. Like PVs, SSTs are altered in postmortem brain samples from multiple neocortical regions in SZ, although basic and translational research into consequences of SST dysfunction has been relatively sparse. We highlight a growing body of work in rodents, which now indicates that SSTs may also underlie specific aspects of cortical circuit function, namely low-frequency oscillations, disinhibition, and mediation of cortico-cortical feedback. SSTs may thereby support the coordination of local cortical information processing with more global spatial, temporal, and behavioral context, including predictive coding and working memory. These functions are notably deficient in some cases of SZ, as well as other neuropsychiatric disorders, emphasizing the importance of focusing on SSTs in future translational studies. Finally, we highlight the challenges that remain, including subtypes within the SST class.

GABA, cortex, theta, gamma, oscillations, parvalbumin

Introduction

Schizophrenia (SZ) is a polygenic disease with a complex and currently undetermined biological basis.^1,2 Novel genetic, molecular, and circuit-level tools developed over the past decade continue to provide transformative insight into neurobiological mechanisms, but these technologies are largely limited to animal studies. Since recapitulating a complex human disease like SZ in animal models is not possible, an emphasis on simpler disease features and “biomarkers,” which may be more translatable to mouse research, represents a promising strategy.³

A dysregulation of gamma-band oscillations in electroencephalography (EEG) recordings is among the best replicated biomarkers of SZ.^4–7 Given current limitations of noninvasive neuroimaging, the neurobiological basis of such gamma-band dysregulation is not immediately clear from surface-level EEG alone. Postmortem brain samples from individuals diagnosed with SZ consistently identify abnormalities in neocortical GABAergic interneurons (INs).^8,9 Studies in animal models have provided myriad evidence linking parvalbumin (PV) positive INs, a subset of affected GABAergic INs in SZ, to the generation of synchronous gamma-band oscillations.^10,11 Because certain cognitive and perceptual aberrations in SZ correlate with altered gamma-band oscillations,⁵ these findings together establish a translational roadmap for linking molecular to behavioral features of SZ.¹²

Given that other GABAergic INs are found to be abnormal in postmortem human brain tissue,^8,9 and animal models of neuropsychiatric risk,^13,14 it is important to understand how other subtypes contribute to pathophysiology. The current review focuses on a smaller but growing body of work concerning the function of somatostatin (SST) positive INs in neocortex and their relationship with disease features common in neuropsychiatric diseases, including SZ. Interestingly studies investigating PV and SST messenger RNAs (mRNAs) and cell counts have identified mostly equivalent reductions in the neocortex of SZ,⁸ yet PVs have received more attention in basic neurobiological investigations, perhaps due to their narrow/fast-spiking phenotype, making them easier to differentiate from excitatory pyramidal neurons (PYRs) in extracellular electrophysiology. The development of a suite of transgenic mice-expressing Cre-recombinase in genetically identified cell types,¹⁵ together with enhanced optical techniques,¹⁶ has opened the door for direct recording and manipulation of SSTs in awake, behaving animals. Here, we review such studies and hypothesize the potential consequences of SST dysfunction for SZ neurophysiology and information processing distinct from the PV domain: low-frequency oscillations and cortical feedback integration.

Properties of SST INs

SZ is a disease with significant cortical pathology.^8,17,18 While the principal neuron type of cortical circuits is the glutamatergic PYR, a smaller portion of neurons are inhibitory INs (15%–35%), which synapse locally and release GABA. INs densely target the local circuit and have high firing rates,¹⁹ playing a disproportionately large role in controlling local computations and cortical circuit outputs.²⁰

GABAergic INs in mammalian neocortex show remarkable diversity. Detailed classification is an area of active research and discussion.²¹ Nevertheless, developmental origin and expression of calcium-binding proteins, neuropeptides, and receptors together support a widely accepted primary segregation into 3 major nonoverlapping IN subtypes: PVs, SSTs, and 5HT3a-positive Ins.^22,23 The morphology and electrophysiology of these cell types, their distribution across cortical laminae, connectivity profiles, and patterns of neurochemical innervation further confirm this categorization,^24,25 forming the basis for their differential roles in circuit motifs (feedforward inhibition, disynaptic inhibition, or disinhibition; figure 1), brain oscillations (gamma and theta), and computation (eg, orientation selectivity).²⁶ While this review focuses on SSTs, the reader is referred to available reviews on other subtypes.^22,23

Fig. 1.

Three major neocortical interneuron categories. Interneuron groups are depicted, with subtypes and their laminar (middle) and major postsynaptic cellular (eg, pyramidal neurons) and subcellular (eg, dendrites) targets are graphically depicted (left and right).

The distinctive features of SSTs are well defined in cortical circuits. SSTs release GABA synaptically and target apical dendrites of PYRs and other INs (but rarely other SSTs), directly modulating excitatory inputs through GABA_A and GABA_B receptor-mediated inhibition.^25,27,28 SSTs exhibit more moderate intrinsic firing rates (6–8 Hz) than fast-spiking PVs,^29–31 correlating with local field potential (LFP) oscillatory power in theta/alpha (4–14 Hz; figure 2) and visually induced oscillations in the beta (15–30 Hz) range (figure 3).^30,32,33 Neurotransmission through slower GABA_B receptors³⁴ produces slow inhibitory responses that are well suited to regulate lower-frequency oscillations, including the theta band.^32,33 Such oscillations may support temporally longer integration windows, a known feature of inputs in distal dendrites,²⁷ where SSTs synapse, altogether facilitating long-range coherence in distributed brain networks.³⁵ Compared to PVs, SSTs receive less feedforward excitatory drive from thalamus or layer 4 (L4), and more local and feedback/top-down inputs.³⁶

Fig. 2.

Martinotti-type somatostatin-positive (SST) interneurons in cortical circuits. Left: Cortical-column schematic depicting known roles of SSTs in dendritic inhibition and integration of top-down feedback (top) and disynaptic inhibition between pyramidal neurons (PYRs; triangles). Top-right: Hypothetical computational, circuit, and physiological signatures of layers 2/3 and 5 Martinotti SSTs. A theoretical set neuron-type raster plots and local field potential (LFP) recording is drawn from the hypotheses of this paper. Bottom-right: Noninvasive biomarkers hypothetically related to Martinotti SST motifs.

Fig. 3.

Layer 4 X94 somatostatin-positive (SST) interneurons in cortical circuits. Left: Cortical-column schematic depicting known roles of SSTs (ovals) in inhibiting parvalbumin-positive (PV) interneurons (circles) to effectively disinhibit feedforward inputs of layer 4 excitatory cells (diamonds) to layer 2/3 pyramidal neurons (triangles). Top-right: Hypothetical computational, circuit, and physiological signatures of layer 4 X94-SSTs. A theoretical set neuron-type raster plots and local field potential recording is drawn from the hypotheses of this paper. Bottom-right: Noninvasive biomarkers hypothetically related to X94 SST-motifs.

Another key property of SSTs that may relate to this role in broader spatiotemporal coherence and integration is facilitating synapses. Whereas PYRs and PVs show rapidly depressing postsynaptic responses to repeated inputs,³⁶ SSTs show enhanced responses,^19,37 potentially gating in top-down feedback.¹⁹ SSTs are largely considered “regular spiking,” sharing electrophysiological features with PYRs (though some exhibit low-threshold-spiking, fast-spiking, or stuttering subtypes). SSTs play a significant role in gating plasticity of local circuits in both motor and sensory areas,³⁸ which may be mediated by cholinergic mechanisms.³⁹ SST inhibition is strongly modulated by cholinergic,^40,41 and feedback inputs, suggesting a role of SSTs for contextual, attentional, and cognitive modulation of fine-scale circuit computations.

Together, SSTs help regulate neural activity within and across different cortical laminae, through both inhibition and disinhibition, in accord with feedback and lateral inputs. The spatial and temporal properties of their innervation may optimally enable SSTs to serve as modulators of long-range cortical integration, tolerating longer conduction delays from distributed neural ensembles.^42,43

As a class of neurons, SSTs also show discontinuous heterogeneity in numerous properties, indicating SST subclasses serving differential inhibitory roles within cortical circuits (figure 1). SSTs are divisible into at least 2 subcategories based on morphology and targeting: Martinotti cells (figure 2) and non-Martinotti cells (figure 3).⁴⁴ Martinotti cells constitute the majority of SSTs, and their somata reside mainly in layer 2/3 and layer 5a/b (L2/3; L5) and can arborize both locally and in L1.⁴⁵ Martinotti cells generally connect to other INs, such as PVs and vasoactive intestinal peptide (VIP) cells (a 5HT3a subtype that mutually inhibits SSTs), as well as densely targeting PYR apical dendrites in L1^46,47 but never other Martinotti cells. Martinotti neurons in L5 also subserve disynapatic inhibition, wherein neighboring L5 PYRs effectively inhibit each other by recruiting nearby SSTs.⁴⁸

In contrast, non-Martinotti cells in L4 target L4 PVs preferentially (figure 1), effectively disinhibiting excitatory neurons and local responses to thalamic inputs and tend to be fast-spiking (like PVs).^46,49 L4 non-Martinotti SSTs may play a functionally distinct, disinhibitory role compared to L2/3 SSTs. It has been shown that L5 non-Martinotti cells target L4 excitatory neurons to provide translaminar inhibitory feedback.⁵⁰ An additional non-Martinotti cell primarily present in L6 (~7% of SSTs overall) also disinhibits local PYRs by suppressing PVs,³⁹ sends long-distance cortically and subcortically, and expresses a nicotinic receptor modulator LYPD6,⁵¹ which may selectively gate cortical plasticity.³⁹

Although studying these subtypes separately is important, most recent basic and clinical research into SSTs have not differentiated. GIN and X94 mouse lines fluorescently label Martinotti and L4 non-Martinotti SST subtypes, respectively,⁵² and have supported the notion of different morphologies and functional roles. The development of mice-expressing Cre-recombinase restricted to SST-expressing neurons made testing the causal roles of SSTs in cortical circuits in vivo more tractable (eg, with optogenetic manipulation and Cre-dependent viral expression). However useful, this Sst-IRES-Cre line⁵³ effectively labels all SSTs irrespective of subtype, leaving important questions regarding the functional roles of SSTs. Unless otherwise stated, the majority of the findings reviewed herein combine SST subtypes (though may likely reflect L2/3 or L5 Martinotti cells, given that they are the most numerous).³⁶ Future work is needed to study SST subtypes in isolation.^50,54

By definition, SSTs express the neuropeptide SST, but the functional role of SST released by SSTs is less studied in cortical circuits than the faster, synaptic release of GABA from SSTs.^24,29 Most research on SSTs in rodent or human cortex refers to SST as a marker for a GABAergic IN subtype rather than a functionally relevant neuropeptide.³⁶ The peptide SST has a net inhibitory effect on local circuits, like GABA, activating 5 distinct G-coupled protein receptors⁵⁵ and is released under overlapping yet slightly different conditions (eg, long-duration high-frequency stimulation)⁵⁶ from different compartments of the cell (dendrites). SST peptides may serve a role in inhibiting pathological overactivity, eg, as observed in an epileptic seizure.³⁶ More research is needed to understand the functional conditions and effects of SST release, as well as how it may compliment or diverge from SST-neuron synaptic GABA release with regards to circuit roles and disease relevance. The current review focuses on SST as a distinguishing marker²⁴ rather than a functionally informative molecule (for more, see reference ⁵⁷).

SST INs and Postmortem Brain Studies of SZ

Although it is not currently possible to directly and noninvasively assess SST function in humans, multiple lines of evidence from postmortem brain samples suggest that SSTs are affected in people with SZ (table 1; see Supplementary Material). A reliable reduction in the number of SSTs and PVs, as well as the levels of SST and PV mRNAs have been reported in the prefrontal cortex (PFC)²⁴ and hippocampus⁵⁸ in brains from people with SZ. SST abnormalities in SZ span across all layers except the superficial L1 and deep L6,⁵⁹ where few SSTs reside. It is unlikely that SST IN aberrations are due to antipsychotic medications,⁶⁰ suggesting a relationship to disease pathophysiology. Still, whether SSTs are affected differently in throughout the course of SZ and whether SSTs are affected in relatives of patients with SZ and, thus, the relationship of SST IN aberrations to underlying etiology in SZ remain unclear.

Table 1.

SST-related biomarkers in human cortical tissue

Authors	SST-related marker	Disorder(s) examined	Direction of SST effect	Brain region(s) examined	Markers of other INs (effect in SZ)
Roberts et al (1983)66	SST peptide	SZ	Decreased	Hippocampus, amygdala, and temporal cortex	CCK (decreased), VIP (unaffected)
Nemeroff et al (1983)67	SST peptide	SZ and Huntington’s chorea	Decreased in SZ and increased in Huntington’s chorea	Caudate, nucleus accumbens, amygdala, hypothalamus, and Brodmann areas 12, 24, and 32	None
Gabriel et al61	SST peptide levels	SZ and Alzheimer’s disease	Decreased in SZ and AD	Frontal, temporal, and occipital cortices	CCK and NPY (reduced). VIP (no effect)
Nakatani et al (2006)68	mRNA	SZ and bipolar I disorder	No effect (n = 7)	DLPFC	None
Hashimoto et al (2008)60	mRNA	SZ	Decreased	DLPFC	Neuropeptide Y (decreased)
Hashimoto et al (2008)8	mRNA	SZ	Decreased	DLPFC, anterior cingulate cortex, primary motor cortex, and primary visual cortex	PV (decreased). calretinin (no effect)
Morris et al59	mRNA	SZ and SZA	Decreased in SZ and SZA	DLPFC	None
Mellios et al62	mRNA	SZ	Decreased (trend, P = .086)	PFC (frontal pole)	NPY and PV (decreased)
Fung et al21	mRNA	SZ	Decreased	DLPFC	PV, calretinin, CCK, VIP, and NPY (decreased). Calbindin (increased)
Konradi et al58	mRNA; SST-positive cells	SZ	Decreased	Hippocampus	PV (decreased)
Wang et al63	SST-positive cells	SZ, bipolar I disorder	Decreased in entorhinal cortex; no effect in subiculum	Entorhinal cortex, subiculum	PV (no effect in subiculum and decreased in entorhinal cortex). Calbindin (no effect)
Volk et al9	mRNA	SZ	Decreased	PFC	PV (decreased) and calretinin (increased)
Fung et al22	mRNA	SZ, BP1	Decreased in SZ and BP1	DLPFC, orbitofrontal cortex	PV and calretinin (no effect), VIP (decreased), and calbindin (increased in SZ only)
Volk et al24	mRNA	SZ, BP1, SZA	Decreased in SZ, BP1, and SZA	DLPFC	PV (decreased)
Tsubomoto et al64	mRNA	SZ	Decreased in all areas	DLPFC, PPC, V1, and V2	PV (decreased) and VIP (decreased only in V1)

Note: BP1: Bipolar I disorder; CCK: cholescystokinin; DLPFC: dorsolateral-prefrontal cortex; PPC, posterior parietal cortex; V1/V2, primary/secondary visual cortex; IN, interneuron; mRNA, messenger RNA; PFC, prefrontal cortex; PV, parvalbumin; SST, somatostatin; SZ, schizophrenia; SZA, schizoaffective disorder; VIP, vasoactive intestinal peptide.

Although SZ is traditionally associated with PFC pathology, observations of SST markers in other neocortical regions reveal similar aberrations. Reductions of SST and PV mRNA are present in sensory and motor cortices,⁸ as well as caudal entorhinal cortex.⁶⁵ While SST and PV IN abnormalities are often identified in the same studies,⁸ PV and SST mRNA expression are inversely correlated across brain regions.⁶⁹ There is even evidence that SST mRNA reductions are more dramatic than PV-related abnormalities in the PFC.^8,22 Although the third major IN subtype, 5HT3a-containing, has not been studied in SZ en masse, VIP INs appear less affected than SST or PVs.^22,61,64 Other markers of INs such as calbindin (CB), caretinin (CR), and neuropeptide Y (NPY) mRNA have been shown to be altered in SZ as well.^{8,9,22,60,63,65} These effects are less consistent and not well correlated with SST effects (table 1). As there is significant overlap and heterogeneity in CR, CB, and NPY expression among the 3 cortical IN subtypes,⁴² their reductions could be imperfect proxies for SST-neuron deficits.

Postmortem studies indirectly suggest SST dysfunction in the cortices of patients. Cell-wise expression of SST mRNA is activity dependent,^25,70 so links between SST function and SZ-related biomarkers (identified in animal studies) may be valuable for clarifying the core disease pathophysiology/ies. The nature of SST pathology in SZ remains unclear given difficulties in distinguishing between missing SSTs and SSTs with undetectable levels of SST mRNA. mRNA reductions are the most consistent, but a failure of SSTs to migrate to cortex has been evinced as well.⁷¹ While the cause of SST dysfunction in SZ is currently unclear, SSTs and PVs share a developmental origin distinct from other cortical INs. Abnormal expression of Lhx6 (a key transcription factor expressed during this period in PVs and SSTs) suggest altered prenatal migration in SZ.⁹ This could implicate a role of the established risk posed by maternal infection occurring contemporaneously.^71,72 Furthermore, microRNA-195 dysregulation, leading to reduced Brain Derived Neurtrophic Factor (BDNF), has been proposed.⁶²

Although reductions in SST markers are consistent across studies, cluster analysis suggests dorsolateral PFC aberrations in SST mRNA are only present in a subset of patients.⁹ Interestingly, a separate study found that low-frequency oscillations and cognitive deficits are also mostly present only in a subset of psychosis patients: ie, a psychosis “biotype.” ⁷³ Future work in larger patient samples, correlating behavior, and neurophysiology is critical to test whether these patient clusters overlap, suggesting that SST IN function could be a biotype-specific abnormality.

Brain Oscillations and the Time-Frequency Subspace of Cortical Circuits

EEG recordings demonstrate that the brain exhibits ongoing “waves” or oscillations spanning slow (delta/theta [1–7 Hz]), moderate (alpha/beta [8–24 Hz]), and fast (gamma [25–100 Hz]) frequency bands. At the local level, oscillations temporally group spikes from populations of individual neurons to modulate network synchrony in fast (<25 ms) and slow (1 s) time packets, temporally filtering inputs, segregating neuronal ensembles, and modulating synaptic plasticity.⁷⁴ Broadly, oscillations can be viewed as an organizational principle of the brain that routes distinct streams of information processing, supporting and/or reflecting dynamic synchrony among proximal and distal circuitries. The frequency of an oscillation within a given ensemble or circuit often scales with the distance between the cells involved due in part to aggregate conduction delays⁷⁵; slower oscillations (delta and alpha) involve distributed networks, whereas higher-frequency oscillations (gamma) reflect local synchrony. Furthermore, feedback in hierarchically organized cortico-cortical circuits, typically originating in infragranular layers and synapsing in L1, is reflected in alpha- or beta-band oscillations, whereas feedforward information propagation, typically originating in supragranular layers and terminating in granular L4, is reflected in gamma-band oscillations.^76,77 Thus, separate, simultaneously present oscillatory activity reflects, and perhaps supports, the integration of information from distinct channels in the brain dedicated to certain processes.

Brain oscillations are an emergent phenomenon of neural networks, arising from the confluence of mechanisms across multiple scales, including intrinsic properties of individual receptors and membranes, ephaptic coupling, cell morphology, and circuit interactions.^75,78 They may arise due to a single pacemaker cell or ensemble or due to distributed intrinsic properties in multiple interconnected ensembles.⁷⁸ In cortical circuits, inhibitory INs play a significant role in generating oscillations due to their dense local innervation and fast tonic firing.⁷¹

Oscillations are just one aspect of the complex spatiotemporal activity in a neural circuit. A thorough study of pathological neural systems may require the consideration of space, time, and frequency domains together, relating circuit firing to ongoing rhythms that gate and organize information flow.⁵ Regardless of whether the status of network rhythms or responses in the time-frequency subspace is directly informative regarding underlying circuit properties, studying cortical oscillations may serve as a translational bridge, as oscillatory signals could be used to relate noninvasive human EEG recordings to intracortical investigations in animal models.⁷⁹

High- vs Low-Frequency EEG Oscillations and SZ

Within the taxon of SZ, irregularities across multiple frequency bands have been reported with scalp-level neurophysiology (EEG and magnetoencephalography [MEG]).^5,33,80,81 SZ effects in a particular frequency band depend on the signal of interest (resting, stimulus induced, and phase locked), paradigm (attended or passive), and brain region of interest (sensory vs motor). Variability in particular time-frequency measures is common across studies, which may be attributable to paradigmatic variables (eg, stimulus characteristics and attention),⁸² data analysis,⁸³ and medication status,⁸⁴ for example.

Even when these variables are held constant, deviations in particular frequency bands do not present uniformly across patients. For instance, augmented background gamma activity during passive sensory stimulation has been reported in some^83,85 but not all studies in SZ.⁸² A study of a large, multidiagnostic sample demonstrated that increased baseline or nonspecific gamma-band power in the EEG may be characteristic of one-third of people with psychotic disorders, suggestive of a cortically hyperactive “biotype.” ⁷³ Increased resting activity (eg, 5 min of recording with eyes open or closed with no explicit stimulation or task) in delta–theta ranges and decreased power in alpha ranges in SZ is consistently reported, while aberrations in the gamma range are a less consistent finding for resting-state studies.^81,86,87 Furthermore, decreased passive stimulus-evoked responses in the delta–theta bandwidth show larger effect sizes and greater reliability than stimulus-induced gamma-band responses.^4,88 Abnormalities in stimulus-evoked and resting low-frequency activity may also be characteristic of a subset of psychosis patients, indicating a different biotype than augmented gamma biotype (above). These low-frequency responses show significant heritability in psychosis families⁸⁹ and associations with psychosis-relevant risk genes.⁸⁶

Alternatively, evidence suggests a primacy of deficits in low-frequency oscillating circuits in SZ cortical pathology. Even with the same paradigms, inconsistent findings differ between augmented,^85,90–95 and attenuated,^5,96,97 gamma-band activity in SZ. Again, this may be due to gamma abnormalities that are present in only a subset of psychosis patients,⁷³ though, within this subset, a general dysregulation of gamma-band dynamics may best explain such findings.⁸¹ In cortical systems, separate frequency bands interact, as the phase of low-frequency oscillation is known to group gamma-range activity (eg, amplitude modulation),⁹⁸ termed “cross-frequency coupling” (CFC; figure 2). For instance, in direct human neocortical recordings, gamma power (80 Hz) is modulated in time with theta phase such that higher gamma amplitude occurs during theta peaks vs troughs.⁹⁹ This CFC relationship correlates directly with working memory processes, suggesting theta–gamma coupling reflects how local neuronal ensembles are bound with global information processing. Furthermore, alpha frequency oscillations bind together appropriately timed high-frequency oscillations, like gamma, for accurate perception of stimuli.⁷⁸

Altogether, differences in heritability, genetic associations, patterns, and reliabilities of effects in passive or resting contexts suggest that EEG signals in low vs high oscillatory bands, if both equally present, are at least differentially related to underlying neuropathology. These trends are reviewed at length elsewhere.⁸¹ An important caveat is that patterns of oscillatory aberrations may change across the course of an illness and medication status, especially with regards to high-frequency abnormalities.¹⁰⁰ Still, identifying these bands as potentially separate neural biomarkers represents an important inroad for neuropsychiatric studies as a better mechanistic understanding of these distinct signals could help refine diagnoses and treatment approaches.¹⁰¹

A Role of SST INs in Low-Frequency Oscillations

A growing body of literature has implicated INs in the generation of oscillations in neural circuits brain-wide.^33,102 Aberrations in EEG signals in a given frequency band measured at the scalp could represent a number of distinct underlying circuit mechanisms and cortical activities.³³ That is, eg, 40-Hz gamma-band stimulus-induced power (from a given electrode or scalp distribution) likely does not arise from the same or even overlapping neural circuits as baseline 40-Hz power, even though both can be referred to as “gamma.” In this sense, identifying the mechanisms of a given oscillatory band in EEG may be overdetermined. On the other hand, logically proceeding in the forward direction, a given circuit motif may exhibit a relatively stable oscillatory signature.³³ For instance, PVs in sensory cortex provide feedforward inhibition, which synchronizes local ensembles and enhances responses to sensory inputs in a 40-Hz rhythm,¹⁰³ while alteration of excitatory inputs to these cells dysregulates the background gamma-band power.¹⁰⁴ In general, the fast-spiking PV IN subtype is responsible for generating and influencing gamma-band activity,^{8,10,26,32,105,106} which facilitates fast local processing and feedforward information flow.⁷⁷ Multiple reviews have been written on the idea that PV IN deficits in SZ lead to the myriad of gamma-range EEG biomarkers.^106,107 Although no direct evidence exists evincing disrupted PV IN activity in SZ patients, the pattern of gamma-band abnormalities across a number of SZ studies, in the context of known brain-wide PV mRNA aberrations, supports this hypothesis.

Although less studied than the link between PVs and gamma, evidence suggests that SSTs may play a significant role in low-frequency oscillations and, by extension, related deficits in SZ. SSTs are affected in SZ postmortem brain samples and show lower intrinsic firing rates (in the 7-Hz range).³¹ Some direct links have been identified between SSTs and lower-frequency bandwidths in cortical circuits.¹⁰² In visual cortex, rhythmic optogenetic stimulation of SSTs during LFP recordings suggests an intrinsic resonance in the 5–30-Hz range (theta to beta), while the suppression of SSTs directly disrupts visually induced beta (20 Hz; figure 2).³² As expected, the same manipulations of PVs affected gamma-range oscillations (30–80 Hz).

The distinct physiology and anatomy of SST circuitry suggests a role in slower neuronal processing and/or oscillations. SSTs and PVs may serve different roles in regulating information integration across the somatodendritic axis. For instance, Martinotti SSTs synapse on distal, apical dendrites of PYRs (figures 1 and 2), where time windows of synaptic integration are longer than in proximal dendritic compartments.²⁷ In contrast, PV inhibition targets dendritic and somatic regions proximal to the axon hillock and thus may be better positioned for modulation in shorter integration windows (gamma oscillations).¹⁰ At the synaptic level, SSTs also show facilitating presynaptic inputs and less adaptation of postsynaptic outputs to PYRs.^19,102 In contrast, PVs exhibit rapid adaptation in their inhibition of PYRs, a dynamic that matches the transient gamma response better. The dynamics of low-frequency oscillations to sensory innervation in cortical circuits, both beta and alpha, match the slower activity of SSTs.^89,108

Finally, non-Martinotti SSTs in L4 directly inhibit and modulate PV activity. It remains an intriguing but untested hypothesis whether this SST subcircuit could support theta–gamma CFC (figure 3). EEG and MEG studies in SZ suggest that, although theta and gamma oscillations are both affected, albeit differently, theta–gamma CFC remains intact.¹⁰⁹

SST-Positive INs and Broad-Scale Spatiotemporal Integration

Here, we posit low-frequency oscillatory activity in cortical circuits indexes the integration of local information processing with broader spatial or temporal context as mediated and modulated by local SSTs. Cortico-cortical feedback inputs, which integrate local processing with higher-level sensory processing and behavioral goals, tend to synapse in supragranular layers of cortex,¹¹⁰ wherein L2/3 SSTs are well positioned to influence local integration of such inputs.¹¹¹ In contrast, PVs are most prevalent in L4, suggesting a greater role in feedforward cortical processing. Beta- and alpha-range oscillations, which may be more of the domain of SSTs, correlate strongly with cortico-cortical feedback, while gamma-band oscillations, which are linked to PVs, correlate with feedforward processing,⁷⁶ further emphasizing that the difference in oscillatory bandwidths of SST and PV circuits may relate to their difference in information processing.

What direct evidence exists supporting this tripartite relationship, involving SSTs, low-frequency activity, and broad spatiotemporal integration and top-down feedback in cortical circuits? Studies relating cortical INs to both local computations and oscillations are somewhat sparse. In sensory cortices, SSTs respond to different physical stimulus features than other INs. In contrast to PVs, SSTs in primary visual cortex (V1) exhibit size tuning (figure 2, top), in that they are most responsive to visual stimuli, which occupy more retinotopic space. In this manner, SSTs support “surround suppression” of PYRs by integrating excitatory input from intralaminar horizontal axons.¹¹² A similar finding was reported in auditory cortex with regard to “spectral” space.¹¹³ Although LFP oscillatory activity was not recorded in these studies, V1 SSTs have been shown to display intrinsic entrainment to stimulation in the 5–30-Hz range (theta to beta).³²

SSTs may also support the integration of sensory processing with broader (>2 s) temporal context as well.¹¹⁴ The variability and nature of cortical responses to sensory stimuli have been described in a “predictive coding” framework, wherein cortical regions higher in an information processing hierarchy build a generative model of the environment based on experience. Model predictions are projected downward (eg, to sensory cortex) to both reduce responses to predictable external stimuli and generate “prediction error” responses to stimuli not predicted.^76,115 Sensory “oddball” paradigms have been employed to study these phenomena in humans and animals.¹¹⁴ Here, a series of the same stimulus (ie, “redundant”) is rapidly presented (0.5–2 Hz) to attenuate cortical responses to the stimulus (so-called “stimulus specific adaptation,” SSA). A different stimulus is then rarely interspersed (ie, “deviant”) to generate a cortical response that is above normative response levels to that stimulus (ie, deviance detection, DD), suggestive of a cortical prediction error. DD is strongly correlated with theta-band oscillatory responses in both humans and mice.^79,116 One study found that chemicogenetic silencing of SSTs in the visual cortex reduced DD, but not SSA, in PYRs. Similar findings in auditory cortex suggest a different or absent role of PVs when compared to SSTs.^117,118 Suppressing SSTs selectively altered deviance-related theta–beta activity while leaving gamma-band activity intact.³⁰ Future work is needed to determine what role SST subtypes play in predictive coding, with a particular focus on the frequency-domain signature of these subcircuits.

Beyond sensory cortex, higher cortical SSTs are involved in the complex integrative process of working memory, which requires a distributed network to integrate information processing while maintaining behavioral goals (figures 2 and 3). Temporal coordination between the ventral hippocampus and PFC is critical,¹¹⁹ and the functional coupling between the 2 regions is strongest in the theta band (3–8 Hz in humans and 4–10 Hz in rodents).^28,120,121 The strength of this theta-band coupling directly correlates to spatial and contextual memory performance.^{120,122–124} Abbas et al²⁸ determined that SSTs and low-frequency oscillatory dynamics play a crucial role in modulating hippocampal-prefrontal synchrony. Optogenetic inhibition of medial prefrontal cortex (mPFC) SSTs, but not PVs, during the encoding phase of a task impaired working memory performance due to decreased hippocampal-mPFC synchrony in the theta band. Additionally, SSTs can shape temporal dynamics of mPFC ensembles and contribute to the maintenance of working memory when an animal is in the “delay” portion of a working memory task.¹²⁵ Together, these studies suggest that SSTs play an integral part in working memory through the low-frequency synchronization of neural communication across distal brain regions.

Low-Frequency Oscillations Mark Information Integration and Context-Processing Deficits in SZ

Although SZ involves global deficits in many forms of information processing, some of the most consistent findings in SZ have come from EEG studies, which highlight intrinsic dysregulation of low-frequency oscillations in contexts where sensory and behavior information must be integrated across space and time.

Long-range neural synchrony is globally impaired in SZ.¹²⁶ Specifically, dysfunction between the medial temporal lobe (mTL) and PFC is implicated in SZ^127,128 and demonstrated behaviorally through impairments during spatial memory tasks in individuals with SZ.^129–132 This is consistent with previous findings that SZ involves reductions in evoked theta/alpha and beta oscillatory activity during the encoding phase of working memory tasks (figures 2 and 3).¹³³ Additionally, mPFC–mTL (primarily left anterior mTL) dysconnectivity is attributed to theta band (1–8 Hz) decreases, and loss of phase coupling has been observed during memory retrieval in individuals with SZ.¹²⁰

Individuals with SZ show consistent deficits in spatial context processing (figure 2).¹³⁴ Integration of visual motion,¹³⁵ and contour information,¹³⁶ across space are uniquely aberrant in the disease, while simpler components of neural processing are spared. Regarding temporal context, studies using sensory oddball sequences have identified reduced mismatch negativity (MMN) and P300 EEG potentials in SZ (indices of DD), which together constitute two of the most consistent physiological markers of SZ to date.^137,138 In healthy human subjects and mice, the MMN co-occurs with oscillatory activity in the theta range, and both MMN and theta depend on cortical SSTs.³³ In SZ, MMN deficits correlate to theta reductions.^{116,139–141}

These context processing deficits in sensory processing have been interpreted in a predictive coding framework,^115,142 wherein individuals with SZ show aberrant integration of top-down (feedback) predictive information with bottom-up (feedforward) sensory information in basic sensory cortices. This leads to perceptual sequalae and delusions.^115,142 Altogether, a disruption of SSTs may disrupt cortical feedback integration of broad contextual information, manifesting in altered low-frequency oscillations and the high-level perceptual and cognitive deficits fundamental to SZ.¹⁴³

Conclusion

Here, we review what is known regarding the GABAergic cortical IN class containing SST, as well as various potential subtypes therein. Postmortem studies demonstrate that SSTs are altered in the cerebral cortex in SZ. We draw links between known aberrations in SSTs in SZ and the distinct patterns of low-frequency oscillatory aberrations and complex information integration seen in the disease. We hypothesize that deficiencies in SSTs in the cortex, potentially specific to L2/3 or L5 Martinotti-type SSTs, disrupt the alpha/beta oscillations that facilitate long-range communication, as well as information integration processes, such as the integration of local processing with temporal context and behavioral goals (ie, working memory; figure 2). Furthermore, we hypothesize that SSTs in L4 (X94 subtypes) may play a key role in modulating PV gamma-band activity in the theta range, supporting the generation of CFC known to underlie working memory and to be disrupted in SZ EEG studies, as well as disinhibiting feedforward information flow, eg, in accord with attentional demands (figure 3). Though tools to directly test this association in humans are lacking, evidence of decreased SSTs,^{8,9,22,24,25,58,63,65,70,144–146} disrupted theta/alpha oscillations,^{4,5,33,78,80,81,88,116,120,139–141,147–150} and slow-time course information processing, as well as working memory deficits,^{81,89,120,126,129–133,151} have all been observed in individuals with SZ, and the connection between these phenomena has also been observed in animal models. Thus, we propose tripartite models,³³ which link specific cell-, systems-, and cognitive-level pathology in SZ (figures 2 and 3). Future work should investigate whether in vivo measurements of SST (eg, via cortical spectroscopy) correlates with oscillatory biomarkers (eg, theta band sensory responses; MMN) and distinct cognitive deficits (eg, working memory or visuospatial integration).

The aforementioned hypotheses leave some questions unanswered. Firstly, IN subtypes are not evenly distributed across the cortical laminae¹⁵² and include SST subclasses, which may play distinct roles in controlling circuit dynamics.⁵⁴ However, layer-specific aberrations in SZ have not been currently explored in postmortem studies. Additionally, the distinct roles of each IN subtype and their interconnections across cortical layers in coordinating theta vs beta oscillations remain untested, as well as how these electrophysiological signals relate to working memory and context-dependent information processing in both mice and SZ patients.

Overall, the hypothetical framework presented serves to highlight a promising link between basic and clinical findings, suggesting important follow-up work required to better understand this link. Characterizing the distinct neurochemical channels for which circuits selectively engage or inhibit SSTs may lead to the development of precision treatments for related cognitive deficits in SZ.

Funding

This work was supported by the National Institutes of Mental Health (F32MH125445 to J.M.R.; R00MH115082 to J.P.H.) and the Whitehall Foundation (2019-05-44 to J.P.H.).