Strength and Diversity of Inhibitory Signaling Differentiates Primate Anterior Cingulate from Lateral Prefrontal Cortex

Dendritic topology of L3 pyramidal neurons

Somatodendritic morphology, which influences the capacity for synaptic integration and filtering (Spruston, 2008), was assessed in ACC and LPFC L3 neurons (Fig. 1), the somata of which were located within 250–500 μm from the pia. Mean soma-to-pia distance (ACC, 426 ± 29 μm; LPFC, 379 ± 31 μm; p = 0.26, Student's t test) and soma surface area (ACC, 488 ± 105 μm²; LPFC, 389 ± 68 μm²; p = 0.42) were similar in the two areas. Neurons in ACC and LPFC did not differ in mean total, apical, and basal dendritic length or branch number (Fig. 1A–D), but basilar branch density was higher in LPFC than in ACC neurons (p = 0.00002; Fig. 1E). The dendritic span of basal arbors was wider in ACC than in LPFC neurons, having greater horizontal (p = 0.00005) and vertical extents (p = 0.0002; Fig. 1B,F). For apical arbors, horizontal (p = 0.04), but not vertical (p = 0.78) extent, was significantly wider in ACC than LPFC neurons (Fig. 1B,F). Sholl analyses (Sholl, 1953) revealed that dendritic lengths of ACC neurons were significantly longer than LPFC neurons in selected distal apical (160, 300, and 320 μm) and basal (140–180 μm from the soma) dendritic segments (*p < 0.05; Fig. 1G), whereas complexity (branch points) did not differ (Fig. 1G). The dendritic arbors of L3 pyramidal neurons in ACC are modestly but significantly larger than those of LPFC neurons.

Excitatory synaptic responses and spines in ACC versus LPFC pyramidal neurons

Spontaneous EPSCs of L3 pyramidal neurons in ACC and LPFC (Fig. 2A,B) did not differ in frequency, amplitude, or kinetics (Fig. 2A,B). However, the density of dendritic spines—the sites of excitatory synapses—was ∼1.3× higher on ACC neurons than on LPFC neurons (p < 0.001; Fig. 2D). Analyses of spine subtypes revealed a significantly higher number, density, and proportion of thin spines (*p < 0.03; Fig. 2C,D) and a lower proportion of stubby spines (*p = 0.01, apical; *p = 0.009, basal; Fig. 2D) in ACC versus LPFC neurons. The relative distribution of spines along the extent of dendritic arbors did not differ in neurons from ACC and LPFC (Fig. 2E,F, top), but total spine numbers in select distal apical (160 and 320 μm from the soma) and basal (140–180 μm) dendritic segments were higher in ACC versus LPFC neurons (*p < 0.03; Fig. 2E). Sholl analyses of spine distribution by subtype revealed that, compared with LPFC neurons, ACC neurons had a greater number and density of thin spines and a lower number and density of stubby spines across the proximal–distal extent of both apical and basal arbors (*p < 0.05; Fig. 2E). The size of spine heads was smaller across the proximal–distal extent of basal (*p < 0.01; Fig. 2G) and, to a lesser extent, apical arbors of ACC neurons compared with LPFC neurons (#p = 0.06, *p = 0.05; Fig. 2G). Separating spines into large versus small (<0.52 μm, determined by cluster analysis; Medalla and Luebke, 2015) showed that ACC neurons have a higher proportion of small spines compared with LPFC neurons (*p < 0.04; Fig. 2H).

Higher density but smaller size of excitatory synapses in ACC versus LPFC

Most excitatory asymmetric synapses in L3 neuropil were located on dendritic spines (∼81–89%), with the rest located on dendritic shafts likely belonging to aspiny/sparsely-spiny interneurons (Fig. 3A). Asymmetric axospinous synapses in ACC neuropil were ∼2.2× more dense than in LPFC (p = 0.0003; Fig. 3B), but the mean surface area of PSDs was ∼42% smaller in ACC (p = 0.004; Fig. 3D). Consistent with PSD area findings, the volume of spines of ACC synapses were ∼50% smaller than in LPFC (p = 0.001; Fig. 3D). There was a significant leftward shift (higher frequency of small synapses and spines) of cumulative distribution of synapse and spine size in ACC relative to that in LPFC [Kolmogorov–Smirnov (K-S) test, p < 0.01; Fig. 3E]. In both areas, most axospinous synapses (∼63–78%) were nonperforated and the proportion of perforated synapses did not differ (ACC ∼22–32%, LPFC ∼32–38%; Fig. 3C). The proportion of spines with spine apparatus (SA) or smooth endoplasmic reticulum (SER), a structure associated with calcium reuptake efficacy (for review, see Bourne and Harris, 2008) was higher in LPFC (60–81%) than in ACC (46–63%, p = 0.02; Fig. 3C). Subpopulations of axospinous excitatory synapses (e.g., with perforated PSDs, SA, or SER and/or with an accompanying inhibitory synapse) were each larger in LPFC than in ACC (*p < 0.02; Fig. 3D). In both ACC and LPFC, PSD area correlated positively with spine volume (data not shown) and with presynaptic bouton volume (ACC, r = 0.54; LPFC, r = 0.62; p < 0.01; Fig. 3F,H). Bouton volume in turn correlated positively with number of synaptic vesicles (ACC, r = 0.81; LPFC, r = 0.81; p < 0.01; Fig. 3F,H).

More frequent, larger, and longer duration inhibitory synaptic currents in ACC versus LPFC neurons

In contrast to EPSCs, IPSCs were very different in ACC and LPFC pyramidal neurons (Fig. 4A,B). The mean frequency of spontaneous IPSCs was ∼6× higher in ACC than in LFPC neurons (p = 0.0005; Fig. 4B). Moreover, IPSCs of ACC neurons also exhibited larger mean amplitude (p = 0.01) and area (p = 0.003), and longer mean decay time constant (tau, p = 0.001) and half-width (p = 0.00002) than those of LPFC neurons. Therefore, the population distribution of events based on frequency (interevent interval) was left shifted, whereas those based on decay times were right shifted in ACC relative to LPFC neurons (K-S test, p < 0.01; Fig. 4C). Although mean rise times did not differ between the two areas (p = 0.13), the population distribution of event rise times was also right shifted in ACC relative to LPFC neurons (K-S test, p < 0.01; Fig. 4C).

Higher density of VGAT⁺ inhibitory boutons apposed to ACC versus LPFC neurons

Quantification of inhibitory boutons immunostained with VGAT⁺ apposed to labeled L3 pyramidal neurons in ACC versus LPFC revealed that most VGAT⁺ appositions were on dendritic shafts (∼65–75%), with fewer on dendritic spines (Fig. 4D–H). There was a significantly higher density of VGAT⁺ appositions onto apical dendritic shafts in ACC than LPFC neurons (*p = 0.04; Fig. 4H), whereas the density of appositions on basal dendrites did not differ. The density of appositions on spines was similar between the two areas, but apical dendrites of ACC neurons had a higher proportion of VGAT⁺ appositions on spine necks versus heads compared with LPFC neurons (*p = 0.008; Fig. 4I). In both areas, ∼13–22% of all spines were apposed to VGAT⁺ boutons (Fig. 4J) and these occurred primarily on the most prevalent thin spines. However, thin spines with VGAT⁺ appositions comprised only a small (∼14%) proportion of the total population of thin spines (Fig. 4J). In contrast, mushroom spines were selectively targeted, because 27–44% of this subtype had VGAT⁺ appositions (Fig. 4J).

The peak density of VGAT⁺ appositions on dendritic shafts occurred within the first 40 μm from the soma, corresponding to the proximal spine-free areas of pyramidal neuron dendrites, followed by a sharp decline to a steady state in the more distal spiny dendritic segments (Fig. 4K). ACC neurons had a higher density of VGAT⁺ appositions in these proximal segments of apical (p = 0.047) and basal (p = 0.02) dendrites and a portion of the most distal apical segments (320–380 and 400 μm from the soma, respectively, p < 0.02) than LPFC neurons (Fig. 4K). In neurons from both areas, low densities of VGAT⁺ appositions on spines were found in the proximal ∼60 μm, increasing to peak at ∼100 μm and then maintaining a relatively constant density (Fig. 4K). VGAT⁺ appositions on spines of distal basal segments were higher in density on ACC than on LPFC neurons (*p < 0.01; Fig. 4K).

Neurochemical diversity of perisomatic inhibitory boutons in ACC versus LPFC

Similar to dendritic compartments, the somata (p = 0.03) and AIS (p = 0.007) of ACC neurons had a higher density of VGAT⁺ appositions compared with LPFC neurons (Fig. 5A,B). Consistent with data on individual filled neurons, there was a higher optical density of VGAT⁺ puncta surrounding dendritic compartments in the neuropil (p = 0.001) and somata (p = 0.0004, nested ANOVA) in ACC compared with LPFC (Fig. 5C,D). The higher density of inhibitory inputs onto dendrites in ACC versus LPFC is consistent with the higher proportion of dendritic-targeting CB⁺ inhibitory neurons in ACC than in LPFC (Dombrowski et al., 2001). However, the finding of a higher density of perisomatic inhibitory inputs in ACC compared with LPFC is unexpected because the density of somatic-targeting PV⁺ inhibitory neurons is lower in ACC than in LPFC, as shown here qualitatively (Fig. 5E) and as shown quantitatively in previous studies (Dombrowski et al., 2001). Given that pyramidal neuron density is also lower in ACC than in LPFC (Dombrowski et al., 2001), the PV-to-pyramidal neuron ratio is similar between ACC and LPFC, but each PV⁺ neuron may issue more boutons to each pyramidal neuron. Alternatively, there could be other sources of perisomatic inhibition in ACC that are absent or present at lower density in LPFC. To address these possibilities, we investigated the specific sources of VGAT⁺ perisomatic inhibition in ACC and LPFC.

Consistent with the VGAT⁺ particle analyses, the density of VGAT⁺ boutons apposed to L3 pyramidal neuron somata was ∼1.5× greater in ACC than LPFC (nested ANOVA, p = 0.00006; Fig. 5F). Next, we assessed the density and proportion of these VGAT⁺ perisomatic boutons double labeled with PV⁺ or CCK⁺, which label two different populations of somatic-targeting interneurons (Bartos and Elgueta, 2012; Fig. 5E). The density of perisomatic VGAT⁺ PV⁺ boutons per pyramidal neuron was similar in the two areas (p = 0.4), whereas the density of perisomatic VGAT⁺ CCK⁺ boutons was higher in ACC than LPFC (p = 0.0005; Fig. 5F), consistent with previous qualitative observations (Oeth and Lewis, 1990). Of the total perisomatic VGAT⁺ boutons on each ACC pyramidal neuron, ∼45% were colocalized with PV⁺ and ∼32% with CCK⁺. In LPFC, PV⁺ input made up a significantly higher proportion of perisomatic VGAT⁺ appositions compared with ACC (p = 0.000003), with ∼72% of VGAT⁺ perisomatic boutons colocalized with PV⁺ and only ∼20% with CCK⁺ (Fig. 5G). Concomitantly, the proportion of perisomatic CCK⁺ boutons was lower in LPFC than in ACC (p = 0.0002). Perisomatic boutons that colocalized with neither CCK nor PV were observed in ∼23% of the VGAT⁺ appositions in ACC, but only ∼8% in LPFC. Due to the high labeling efficacy of PV and CCK used here and in other studies (Oeth and Lewis, 1990; DeFelipe, 1997) and the fact that we observed highly consistent regional specificity in labeling across cases, these VGAT⁺/PV⁻/CCK⁻ appositions likely represent a third population of perisomatic input that, similar to CCK⁺ input, is more abundant in ACC than in LPFC.

Higher density of inhibitory synapses in ACC versus LPFC

These confocal microscopy analyses of inhibitory appositions onto pyramidal neurons represent only an estimate of synaptic input. Therefore, we used electron microscopy to quantify the density of inhibitory synapses in ACC versus LPFC neuropil. The density of inhibitory symmetric synapses was ∼3.3× higher in L3 neuropil of ACC than in LPFC (207 ± 49 vs 64 ± 2 × 10⁶ synapses/mm³; p = 0.04, t test; Fig. 6A–C), consistent with the VGAT⁺ confocal data. However, the mean PSD surface area of symmetric synapses was similar between the two areas (p = 0.68; Fig. 6D). The majority of symmetric synapses targeted dendritic shafts (∼69–89% in ACC; 75–88% in LPFC), with the remainder targeting spines; both synapse types were present at higher densities in ACC than in LPFC neuropil (*p < 0.05; Fig. 6C). Symmetric synapses on shafts were similar in size, whereas symmetric synapses on spines had larger PSDs in LPFC compared with ACC (p = 0.02; Fig. 6D). The density of symmetric synapses on somata and proximal apical trunks of L3 pyramidal neurons in the two areas was assessed (Fig. 6E–H), revealing a higher density in both compartments of ACC neurons than LPFC neurons (*p = 0.004, proximal apical #p = 0.059; Fig. 6H, somata).

E:I ratio and physiological properties of ACC versus LPFC neurons

We further investigated how functional differences in inhibitory signaling in the two prefrontal areas relate to excitatory signaling and intrinsic biophysical properties of pyramidal neurons. There were no significant relationships between EPSC and IPSC frequency or area (Fig. 7A) or any other synaptic variable (data not shown). Due to the higher frequency and larger area of IPSCs in ACC, the mean E:I ratio (EPSC to IPSC frequency × area) was significantly lower in ACC than in LPFC (*p = 0.02, t test; Fig. 7A), indicative of a higher inhibitory tone in ACC.

To determine whether the specific population of ACC neurons with robust inhibition share similar intrinsic electrophysiological features, we further examined relationships between the properties of IPSCs (frequency and decay time constant) and intrinsic membrane and firing properties. Assessments of intrinsic membrane properties revealed that the mean passive membrane time constant (tau_m) of ACC neurons was ∼1.6× longer than that of LPFC neurons (p = 0.0000005, t test; Fig. 7B,C). V_r of pyramidal neurons in the ACC (−64.4 ± 0.6 mV) and LPFC (−64.3 ± 0.8 mV; p = 0.9) and R_n (p = 0.13) did not differ significantly. Compared with LPFC neurons, ACC neurons displayed a significantly lower rheobase (p = 0.00005; Fig. 7B,D). Individual action potentials in ACC neurons had slightly lower amplitude (p = 0.00003) and slower kinetics (AP duration; p = 0.0000007) compared with those in LPFC neurons (Fig. 7B,E). However, neurons in the two areas were similar in terms of repetitive AP firing frequency responses to prolonged depolarizing current injections, especially at the lower current steps between 80 and 180 pA (p = 0.2–0.8; Fig. 7F,G). Firing frequency was positively correlated with R_n in neurons from both areas, but firing frequency was positively correlated with tau_m only in LPFC neurons (p < 0.05 for all correlations; Fig. 7G).

The population distributions of neurons based on IPSC frequency and tau_m showed a positively skewed unimodal distribution in LPFC neurons for both variables, with a high proportion of neurons having low IPSC frequency and short tau (Fig. 7H). In contrast, ACC neurons showed a somewhat bimodal distribution with higher variability and larger range, specifically due to second peak in the proportion of neurons with higher IPSC frequency and also longer tau (Fig. 7H). Cumulative histograms of tau_m and IPSC frequency for individual neurons revealed a significant rightward shift of the ACC population curve relative to that of LPFC (K-S test, p < 0.01).

Scatter plots of IPSC frequency versus tau_m and versus R_n show that ACC and LPFC neurons formed two largely nonoverlapping groups, with ACC neurons forming a cluster with relatively higher IPSC frequency, longer tau, and higher R_n compared with LPFC neurons (Fig. 7I, top). IPSC frequency was negatively correlated with tau_m (r = 0.42, p < 0.05) and R_n (r = 0.48, p < 0.05) in ACC neurons; neurons with higher IPSC frequencies tend to have shorter tau_m and lower R_n (Fig. 7I). A similar trend was observed for LPFC neurons, but the correlation was weaker and not significant. A significant positive correlation was found between IPSC decay time constant and tau_m across all neurons (r = 0.47, p < 0.05), in which neurons with long IPSC decay time constants also exhibited long tau_m (Fig. 7I, bottom). Neither IPSC frequency nor IPSC decay time constant showed significant correlations with rheobase or repetitive AP firing properties (data not shown).

Relationship between inhibitory synaptic physiology and morphology

We then assessed whether functional differences in inhibitory signaling are related to structural differences in inhibitory input onto pyramidal neurons in the two prefrontal areas. There was a positive exponential relationship between IPSC frequency and VGAT⁺ apposition density (Fig. 7J), specifically on the apical dendrites (r = 0.7, p < 0.01) and on the AIS (r = 0.82, p < 0.01). Within-group regression analyses confirmed this relationship specifically in the population of ACC neurons (r = 0.8, p < 0.05), but not in LPFC neurons, all of which had low IPSC frequencies and low VGAT⁺ apposition densities. Consistent with the relationship between IPSC properties and tau_m, regression analyses in ACC neurons revealed a negative correlation between VGAT⁺ apposition density on apical dendrites and tau_m that best fit an exponential function (for ACC, r = 0.71, p < 0.05; Fig. 7J). No significant relationships were found between any IPSC property and VGAT⁺ apposition density on basal arbors, somata, or proximal dendrites.

Distinct temporal dynamics of inhibition in ACC and LPFC

Given that fast and slow IPSCs have distinct effects on network behavior (Bartos et al., 2002; Kopell et al., 2010), we examined the proportion of fast (decay times = 2–14 ms) and slow (>14 to 70 ms) IPSCs in each cell (Fig. 8A,B). The mean proportion of IPSCs with slow decay times was higher in ACC compared with LPFC neurons (37 ± 5 vs 16 ± 4%, p < 0.001) and their mean maximal decay time was also greater (∼70 vs 30 ms; Fig. 8B). These findings are consistent with our anatomical results showing that, compared with LPFC, neurons in the ACC receive denser inputs from both dendritic-targeting interneurons and somatic-targeting CCK⁺ interneurons, which have been reported to elicit long-duration IPSCs in rodents (Bartos and Elgueta, 2012). Conversely, inhibitory currents in L3 pyramidal neurons in LPFC on the other hand, were characterized by fast decay kinetics, consistent with a higher proportion of inputs from PV inhibitory neurons, which elicit short duration IPSCs (Bartos and Elgueta, 2012). The schematic shown in Figure 8C summarizes the key differences in the anatomical and functional characteristics of inhibitory inputs onto L3 pyramidal neurons in the two prefrontal areas. The higher density of inhibitory synapses and higher frequency of IPSCs, as well more numerous IPSCs with both larger areas and longer decay times, all contribute to a higher total inhibitory conductance (g_i) in ACC compared with LPFC pyramidal neurons.

The strength and temporal dynamics of inhibition have significant effects on network oscillatory behavior. A PING computational model that simulates gamma oscillations in the hippocampus predicts that the period of the oscillatory rhythm depends predominantly on the decay time constant of the GABA_A-mediated inhibitory synapses (Börgers and Kopell, 2005; Gloveli et al., 2005; Tort et al., 2007; Kopell et al., 2010). Using this model, we tested whether the different IPSC properties in ACC versus LPFC pyramidal neurons alone were capable of altering the frequency of network oscillatory behavior (Fig. 8D,E). Setting IPSC decay time constant to the means for LPFC (12 ms) and ACC (17 ms), together with the EPSC decay time constant set to the mean value for each area (6.4 ms for both), resulted in synchronous firing of pyramidal neurons at frequencies of ∼31 Hz in LPFC and ∼24 Hz in ACC, both within the mid-beta to low-gamma range (for review, see Benchenane et al., 2011; Fig. 8D). To model the combined effect of the longer IPSC decay time and 6× higher frequency in ACC neurons, we also scaled the total maximal conductance of inhibitory synapses to excitatory neurons (g_IE) by 6× (Fig. 8E). These conditions reduced ACC oscillation frequency to ∼14 Hz in the low-beta range. Without the increase in g_IE, the model predicted that IPSC decay times of 50–70 ms are required to obtain oscillations within the 5–10 Hz in the theta-α range (Fig. 8F). In LPFC pyramidal neurons, the maximum decay times of spontaneous IPSCs was ∼30 ms (<1% of events), whereas in ACC pyramidal neurons, a significant proportion of IPSCs had longer decay times, falling between 30 and 70 ms (∼9%). Therefore, together, the empirically observed differences in inhibition and network modeling predict that slow oscillations are generated more readily in ACC networks than in LPFC.

Distinct inhibitory modulation of ACC versus LPFC neurons: implications for synaptic balance and temporal dynamics

The E:I ratio in ACC is significantly lower than in LPFC due to the greater frequency and size of IPSCs but similar EPSC properties in ACC versus LPFC neurons. There is a higher density of both inhibitory and excitatory synaptic inputs onto ACC neurons, which would predict a higher frequency of both IPSCs and EPSCs. That this is not the case suggests a higher level of shunting of excitatory by inhibitory currents (for review, see Markram et al., 2004; Kubota et al., 2016) and a higher degree of filtering of distal excitatory inputs in the larger ACC neurons (Rall, 1964). The lower E:I ratio in ACC neurons does not lead to reduced excitability, however, because these neurons exhibit a lower rheobase and similar steady-state AP firing rates compared with those in LPFC. This demonstrates that E:I ratio, although it may differ broadly, does not necessarily lead to altered firing frequency in vitro (Barral and D Reyes, 2016; Denève and Machens, 2016).

Inhibitory synapses impinging on pyramidal neuron somata are favorably situated to influence AP generation (for review, see Freund and Katona, 2007; Kubota et al., 2015). We found that ACC neurons possess a higher density and diversity of perisomatic inhibitory boutons and synapses compared with LPFC neurons. The density of PV⁺ boutons is the same in the two areas, but the density of CCK⁺ boutons is higher in the ACC. Therefore, the proportions of perisomatic PV⁺ and CCK⁺ inputs differ dramatically, with ratios of ∼1:1 versus ∼3:1 PV:CCK in ACC and LPFC, respectively. PV⁺ and CCK⁺ inhibitory neurons have very different physiological properties and also play distinct roles in network oscillatory dynamics in the hippocampus of rodents (Klausberger et al., 2005). PV⁺ fast-spiking interneurons fire at much higher frequencies than CCK⁺ interneurons (for review, see Freund and Katona, 2007; Lewis et al., 2011; Bartos and Elgueta, 2012). GABA_A receptors found at PV⁺ synapses possess the α1 subunit, which produce IPSCs with fast decay kinetics (Klausberger et al., 2002; Szabadics et al., 2007; Doischer et al., 2008). In contrast, GABA_A receptors at CCK⁺ synapses possess the α2 subunit, resulting in slower IPSC decay kinetics (Klausberger et al., 2002). Moreover, CCK⁺ neurons have a higher rate of asynchronous GABA release that prolongs the overall inhibitory response (Hefft and Jonas, 2005). In summary, kinetically rapid PV⁺ neurons are well suited for rapid input–output transformations and for the generation of fast-gamma oscillations (Bartos et al., 2002; for review, see Kopell et al., 2010; Buzsáki and Wang, 2012). Kinetically slower CCK⁺ neurons are capable of mediating long-lasting inhibition and slower theta-type oscillations (for review, see Bartos and Elgueta, 2012). Because “fast” PV⁺ inputs predominate on LPFC neurons, IPSC decay times are nearly all rapid; in contrast, because “fast” PV⁺ and “slow” CCK⁺ inputs are similar in density on ACC neurons, the range of IPSC decay times is broader and includes a significant proportion (9%) of very slow events. In summary, the near equivalent distribution of PV and CCK inhibition in ACC is consistent with a microcircuit suited for both rapid and long-lasting inhibition and diverse timescales for signal integration.

Slower inhibitory dynamics in ACC versus LPFC neurons: implications for network behavior

Distinct inhibitory networks underlie temporal dynamics of signal processing in vivo (Wang et al., 2004; Wilson et al., 2012), which differs between ACC and LPFC (Johnston et al., 2007; Voloh et al., 2015; Stoll et al., 2016). The correlated fluctuation of spontaneous AP activity decays more slowly in the ACC than in the LPFC in vivo, indicating that the intrinsic timescales of signal processing differ in the two areas (Murray et al., 2014). Our finding of higher inhibitory tone and longer membrane time constant in ACC neurons likely contribute to lengthening the time window for signal summation in this brain area.

The oscillatory frequencies of cortical networks are dependent on the timescale of inhibitory events (Börgers and Kopell, 2005; Vierling-Claassen et al., 2010). Computational modeling of a pyramidal interneuron network (Gloveli et al., 2005; Tort et al., 2007; Kopell et al., 2010) predicts that the prolonged time course of IPSCs in ACC will generate slower oscillatory activity in this region (∼24 Hz, mid-beta) compared with LPFC (∼31 Hz, low-gamma/high-beta). Incorporation into the model of the 6-fold higher frequency of IPSCs in ACC further reduces the oscillation frequency to ∼14 Hz in the low-beta range. Gamma and beta oscillations associated with rapid sensory–motor processing and attentional allocation have been observed in both LPFC and ACC in vivo (Rothé et al., 2011; for review, see Engel and Fries, 2010; Benchenane et al., 2011; Helfrich and Knight, 2016). The ACC also processes information that occurs at slower timescales, such as integration of past reward context (Kolling et al., 2016). The ACC and LPFC have coordinated but distinct contributions to theta-gamma phase-amplitude coherence during attentional tasks, with slower theta activity being more prominent in ACC than in LPFC (Tsujimoto et al., 2010; Voloh et al., 2015). Indeed, unlike the LPFC, the ACC can access motivational signals from long-term memory through strong connections with the medial temporal lobe, where theta activity is robust (Paz et al., 2008; Takehara-Nishiuchi et al., 2011; Jutras et al., 2013; for review, see Barbas et al., 2013; Likhtik and Paz, 2015). Moreover, compared with LPFC, the ACC receives denser cholinergic innervation (Mesulam et al., 1984; Lewis, 1991; Ghashghaei and Barbas, 2001), an important neuromodulator of theta oscillations (for review, see Heys et al., 2012). Our data further suggest that theta can be induced more readily in ACC than in LPFC at the local circuit level based on the following: (1) the higher density of CCK⁺ boutons in ACC compared with LPFC (in the rodent hippocampus, CCK and PV neurons play key roles in setting the theta cycle; Klausberger et al., 2005) and (2) the higher proportion of very slow IPSCs with 50–70 ms decay times, which, if selectively activated, are predicted to reduce oscillation frequency to 5–10 Hz in the theta-α range. These prolonged inhibitory events likely result from asynchronous GABA release by inhibitory neurons, a mechanism associated with theta activity in rodents (Klausberger et al., 2005). Whether this same mechanism applies to monkey ACC will be important to assess in future empirical and computational studies. In the rodent cortex, long-lasting inhibition of pyramidal neurons by non-fast-spiking interneurons is necessary to drive low-frequency oscillations (Vierling-Claassen et al., 2010). In addition to inputs from fast-spiking PV⁺ interneurons, we show that ACC pyramidal neurons receive a higher proportion inputs from “slow,” non-fast-spiking interneuron types (dendritic-targeting and CCK⁺ perisomatic-targeting) compared with those in LPFC. These two temporally distinct sources of inhibition endow the ACC network with a large dynamic range of synaptic timing and network oscillations. These circuits are well suited for the highly integrative processing in ACC that spans both rapid and long behavioral timescales, allocating attention and monitoring reward outcomes and error to guide ongoing and future action (Ito et al., 2003; Johnston et al., 2007; Kolling et al., 2016); review (Lee et al., 2007; Rushworth et al., 2011).

The ACC–PFC network and inhibitory neurons are selectively vulnerable in schizophrenia (for review, see Allen et al., 2008; Lewis et al., 2011). Multiple lines of evidence suggest that GABA neurotransmission from PV⁺ neurons is decreased, whereas that from CCK⁺ neurons is increased, resulting in a reduced ratio of PV:CCK inhibition in schizophrenia (for review, see Benes, 2000; Reynolds et al., 2001; Curley and Lewis, 2012). Our findings predict that diminished PV⁺ inhibition associated with schizophrenia will remove strong perisomatic inhibition almost completely in LPFC, but only partially in ACC neurons, because the abundant inhibitory drive from CCK neurons will remain intact or even be enhanced within ACC. These region-specific alterations in inhibitory circuitry can result in robust disinhibition of the LPFC, whereas ACC circuits remain relatively inhibited. Moreover, inhibition of ACC activity can lead to further LPFC disinhibition through weakening of the ACC pathway poised to engage inhibition within LPFC (Medalla and Barbas, 2009). A weakening of ACC relative to LPFC activity is consistent with the hypoactivation of the ACC and concomitant hyperactivation of the LPFC observed in schizophrenic patients (Kerns et al., 2005; Allen et al., 2008; Fornito et al., 2009; Leicht et al., 2010; Bersani et al., 2014; for review, see Honey and Fletcher, 2006; Allen et al., 2008). The diverse intrinsic features of excitatory and inhibitory circuits reported here provide a compelling basis to test hypotheses on prefrontal areal dynamics and behavior in normal and neuropathological states.