Like other layers, L6 contains several distinct classes of spiny, glutamatergic principal cells (Thomson, 2010 for review). The birth-dates of two broad groups are distinguished by their expression of latexin (carboxypeptidase-A inhibitor). Corticortical cells, which express latexin are born after corticothalamic cells which do not: E15 cf E14 (Arimatsu and Ishida, 2002). (Thomson and Lamy, 2007, Figure 5, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2518047/figure/F5/).

Only earlier born, corticothalamic pyramids receive direct thalamic input. One subclass of these upright cells, with apical dendritic tufts in L4, send narrow, ascending axonal arbors to L4 (and sometimes lower L3) where it terminates with characteristic short, drumstick-like side branches. These neurones project subcortically to “specific,” or primary sensory thalamic nuclei and to nucleus reticularis thalami (nRT, the thalamic inhibitory nucleus) (Zhang and Deschenes, 1998). All L6 corticothalamic pyramids fire with minimally accommodating/adapting, near tonic discharge and preferentially innervate GABAergic cells with consistently facilitating patterns of transmitter release (West et al., 2006). In neocortex, >90% of their synaptic boutons contact dendritic shafts of non-spiny neurones (White and Keller, 1987), including L4 (Tarczy-Hornoch et al., 1999; Beierlein et al., 2003) and L5 interneurones (Staiger et al., 1996). Despite their frequent innervation of parvalbumin (PV) interneurones (which receive depressing inputs from all other pyramidal classes), L6 corticothalamic pyramids elicited facilitating EPSPs (excitatory postsynaptic potentials) in all cell types studied, including ventroposterior, posterior medial thalamic and nRT neurones. This contrasts with the depressing EPSPs elicited by L5 pyramids in posterior medial thalamic nucleus (Reichova and Sherman, 2004).

The second corticothalamic subclass, more commonly found in deep L6, projects to both specific and non-specific thalamic regions such as PO (posterior thalamic group) The apical dendrites of these short, upright pyramids and their ascending axons typically terminate in upper L5. Neither subgroup of corticothalamic cells has long horizontal axon collaterals in the infragranular layers, all branches turn toward the pial surface.

In cats and primates, where L4 subdivisions are morphologically and functionally distinct, subclasses of corticothalamic cells are found, each with its apical dendritic branches and axonal ramifications restricted to a specific L4 sublayer (Lund, 1987; Wiser and Callaway, 1996). Further specificity was demonstrated by a study combining in vivo physiology and morphology in cat V1 (Hirsch et al., 1998b,a). L6 pyramidal simple cells (“simple” implying significant direct input from lateral geniculate nucleus, LGN and resembling “specific” thalamocortical pyramids) targeted L6 and/or L4, layers rich in simple cells. L6 complex cells (receiving integrated, rather than “specific” corticothalamic inputs), targeted L2/3 and L5, layers rich in complex cells.

In striking contrast to corticothalamic pyramidal cells, are the rapidly adapting, corticocortical L6 pyramids, which preferentially innervate other pyramidal cells with “depressing” synapses and display an array of morphologies (Mercer et al., 2005), short, upright pyramids whose apical dendrites terminate in L5, bipolar cells and inverted pyramids. All have long, horizontal axons confined to L5/6 (Zhang and Deschenes, 1998; Mercer et al., 2005), some crossing areal boundaries. Their pronounced spike accommodation/adaptation cannot be overcome by injecting larger square-wave current pulses; these only result in more rapid and profound soma/initial segment Na⁺ channel inactivation. However, a ramp-shaped current superimposed on the original threshold square-wave pulse, activates tonic firing of overshooting APs, probably originating at more distant axonal locations and propagated, or reflected passively, back to the soma (unpublished; Stuart et al., 1997; Colbert and Pan, 2002; Clark et al., 2005, for axonal spike-initiation).

The near tonically firing claustrum-projecting cells form the third major L6 pyramidal class, with long slender apical dendrites that reach L1 without forming a tuft there and a broad, axonal arbor confined to L5 and L6 (Katz, 1987). Like L6 corticocortical cells, claustrum-projecting pyramids preferentially innervate pyramids locally, with “depressing” synapses (Mercer et al., 2005).

L6 is often perceived as a predominantly thalamo-recipient layer, but only corticothalamic pyramids receive powerful, direct thalamic input. Nor do corticocortical cells receive powerful excitation from neighboring thalamo-recipient corticothalamic cells. Some descending inhibitory projections from L4 ramify in L6, but excitatory projections from superficial layers are often narrow and sparse. Binzegger et al. (2004, cat V1) estimated the numbers of synapses supplied to each layer by cortical and LGN relay neurones. When compared with estimates based on stereological analysis (Beaulieu and Colonnier, 1985), the estimates for excitatory synapses were within 10% for L2/3 and L5, but differed by 32% for L4 and 70% for L6. Many additional corticocortical, or subcortical inputs are required to account for the boutons in these thalamo-recipient layers. (Thomson, 2010, Figure 4; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2885865/figure/F4/).

In cats and primates most subplate (SP) neurones disappear during development; a few remaining in the underlying white matter as interstitial neurones (Kostovic and Rakic, 1980; Luskin and Shatz, 1985; Valverde et al., 1989; Naegele et al., 1991). However, in rodents, degeneration in SP is less dramatic and the SP becomes L6b (or L7) (Valverde et al., 1989; Ferrer et al., 1992).

The principal inputs to L5 (and to L5 pyramidal dendrites in L3) are local and more distant corticocortical projections. In turn, large L5 pyramidal cells project to many subcortical targets, including “non-specific” thalamic nuclei, superior colliculus, pons and spinal cord (targets depending on cortical region). Smaller L5 pyramids project to other cortical and subcortical regions and transcallosally, to contralateral neocortex.

Upper L5 contains the largest neocortical pyramids (only these approaching the size and spine densities of CA pyramids). In cat V1, the large cells that project to the colliculi, and/or the pons (Hallman et al., 1988), have thick apical dendrites with well-developed apical tufts in L1/2 and substantial basal dendritic arbors largely contained within L5. The largest, Betz cells (Betz, 1874), are found in motor cortex and project via the corticospinal tract to the spinal cord. Large L5 cells display a stereotypical “intrinsically burst-firing” behavior (Connors et al., 1982); the burst of two or more, high frequency spikes being superimposed on a well-developed depolarizing envelope, due to activation of a dendritic Ca²⁺ spike (Purpura and Shofer, 1965; Llinas, 1975; Larkum et al., 1999). The short interspike-interval train of two or more spikes, typical of rapidly adapting/accommodating neurones (smaller L5-, L6 corticocortical-, and some L4 pyramids), should not be confused with stereotypical bursts (though it often is); it is not superimposed upon, or triggered by a stereotypical depolarizing envelope and does not occur repetitively if the cell is held near spike threshold. Quasi burst-firing can also be elicited in adapting cells electrically coupled to intrinsic bursters (Mercer et al., 2006). The local axons of large L5 pyramids arborize almost exclusively within the deep layers while the smaller pyramids also project to the superficial layers (Larsen and Callaway, 2006).

A significant input to these large, intrinsically burst-firing pyramids comes from smaller L5 pyramids. Whether these smaller pyramids project to non-specific thalamus, or to other cortical regions was not determined. In adult rat L5, small adapting pyramids were 10 times more likely to innervate large burst-firing pyramids than vice versa (unpublished data: Thomson and West, 1993; Deuchars et al., 1994). Large L5 pyramidal cells that are close neighbors are, however, relatively densely interconnected (hit rate of 1:10: Markram, 1997). Large, but not small L5 pyramids, also receive a dense, highly focussed, input from deep L3 pyramids (hit rate > 1:4, Hübener et al., 1990; Thomson and Bannister, 1998, for reconstructions of cat L5 pyramids).

The apical dendrites of small-medium corticothalamic and corticocortical L5 pyramids are slender, rarely extend beyond L2/3 and have no significant apical tuft. L5 corticothalamic pyramids provide large boutons to non-specific thalamic regions from which they receive no reciprocal inputs, in contrast to L6 corticothalamic cells which are reciprocally connected with “specific” thalamic nuclei and deliver small boutons (Van Horn and Sherman, 2004). A separate population of smaller, shorter L5 pyramids projects transcallosally (Hübener et al., 1990; Kasper et al., 1994). Transcallosal cells are found in all layers except L1 (Kasper et al., 1994).

If we can assume that three layered cortices in some non-mammalian species do a perfectly good job, as far as the requirements of those animals are concerned, receiving e.g., sensory information in one layer (equivalent to mammalian L6) and integrating that information with signals from elsewhere, in the adjacent layer (equivalent to L5), which then sends instructions to other brain regions, we could ask why a presumed need for more complex and sophisticated organization and integration of that information could not have been achieved simply by expanding these two existing layers. Whether this was attempted in some long lost evolutionary dead-end, we may never know, we can only assume that such an attempt did not survive. Instead, a new germinal zone, SVC and three new layers (L2-L4) to which the SVC contributes spiny cells, were added. Interestingly, these new layers repeat the pattern established in deeper layers: peripheral input into L4, with integration within, and distribution from L2/3.

Layer 4 contains two broad classes of spiny excitatory cells. Typically, the basal dendrites of L4 pyramidal cells are contained within L4 with apical oblique dendrites in L2/3 (though they receive little or no input from local L3 pyramids) and an apical dendrite extending into L1, often forming a small tuft there. Spiny stellate cells lack an apical dendrite, most or all of their dendrites are confined to L4 (Lund, 1973). Perhaps the most striking distinguishing feature of L4 spiny neurones in rat and cat, especially when compared with the “chunkier” pyramids in adjacent layers, is their simple (Rojo et al., 2016) and delicate appearance (Bannister and Thomson, 2007).

Despite being a major thalamo-recipient layer, thalamocortical inputs to L4 contribute only 6% of the synapses onto spiny stellate neurones in cat V1 and up to 22.9% in mouse (Benshalom and White, 1986), terminating predominantly on dendritic spines via large en-passant boutons. In contrast, ascending L6 corticothalamic pyramidal axons, form synapses with small boutons, but provide 45% of the excitatory inputs onto L4 spiny cells (cat, primate, Lund et al., 1988, for review). In primate V1 axons from area MT terminate in L1, L4B, and L6. This contrasts with other so called “feedback” connections from “higher” visual areas terminating in L1; projections that might more meaningfully be termed “cognitive” or “attentional” feed-forward. In V2 they terminate primarily in L1 and L5 or L6 (Rockland and Knutson, 2000). A further 28% of the excitatory input to L4 spiny cells originates from within L4 (cat, Ahmed et al., 1994). Despite the small numbers of thalamocortical inputs, their large boutons provide secure, faithful transmission of early presynaptic spikes, albeit followed by pronounced presynaptically mediated depression. Thalamocortical synapses have three times more release sites than those of local circuit axons, with higher release probabilities, making the average thalamocortical connection several times more effective (Gil et al., 1999, mouse S1), at the start of a spike train.

The axons of L4 spiny neurones make dense, topographically precise ascending projections to L3 and sparse descending projections to upper L5 (rat, cat) (Valverde, 1976; Parnavelas et al., 1977; Feldman and Peters, 1978; Gilbert, 1983; Burkhalter, 1989) where they innervate pyramids and (less commonly) interneurones (Thomson et al., 2002). Pyramids and spiny stellates contribute to these projections and both provide relatively narrow, horizontal arbors within L4. In cat V1, some spiny cell axons make most of their synapses within L4, others form a larger proportion in L3 (Binzegger et al., 2004), a finding supported by morphometric analysis coupled with paired recordings in rat barrel cortex (Lübke et al., 2003) and one that correlates with distinct electrophysiological classes (below). The sparse projection, from L4 to L6, appears to originate predominantly with pyramids (unpublished).

The firing patterns of adult L4 pyramidal and/or spiny stellate cells correlated with distinct connectivity patterns. Stereotypical intrinsic bursts were rare. Around 60% displayed rapid spike accommodation and frequency adaptation (recoverable with a superimposed ramp) and innervated other L4 spiny cells. The remaining 40% produced a short train of 3–5 short interspike interval spikes, followed by a brief afterdepolarization, then a slow afterhyperpolarization upon which a spike-train of increasing interspike interval was superimposed. These cells preferentially innervated L3 pyramids (Bannister and Thomson, 2007). This raises interesting questions about patterns of synaptic input in vivo and how they might interact with the cells' inherent firing characteristics; tonic input to L3, phasic to neighboring L4 pyramids in reponse to maintained depolarization.

Connectivity ratios for pairs of L4 pyramids were relatively low (1:18 adult rat; 1:14 cat V1), with no selection for firing characteristics, and fell off extremely rapidly with increasing horizontal somatic separation. All identified synaptic contacts onto spiny cells (L4 and L3, rat and cat) were onto proximal primary, secondary, and tertiary, electrotonically compact basal dendrites, and all EPSPs were brief and depressing (Bannister and Thomson, 2007).

Layer 3 pyramidal cells are “typical” pyramids, with adapting firing patterns, well developed basal and apical oblique dendrites and an apical dendrite forming a tuft in L1. The largest are close to the L4 border. More superficial L2 cells are, naturally, very short with almost no apical dendrite. L3 pyramidal axons ramify densely in L2/3 delivering depressing inputs to other L3 pyramids with a hit rate of 1:3 that falls off only gradually with distance. Their main descending axons typically pass through L4 without branching to ramify in L5, in rat (Lorente de Nó, 1922; Burkhalter, 1989), cat (O'Leary, 1941; Gilbert and Wiesel, 1983; Kisvárday et al., 1986), and primate (Spatz et al., 1970; Lund et al., 1993; Yoshioka et al., 1994; Kritzer and Goldman-Rakic, 1995; Fujita and Fujita, 1996), where they innervate large L5 pyramids;. Somewhat surprisingly some deep L3 pyramids in mice have substantial axonal arbors within L4 (Larsen and Callaway, 2006). Their targets in L4 are of interest, because although L3 pyramids innervate L4 interneurones that have dendrites projecting into L3, they rarely, if ever excite L4 spiny cells in adult rat or cat (Bannister and Thomson, 2007).

An additional, distinctive firing pattern has been described in cat V1—chattering cells (Gray and McCormick, 1996). These neurones generate extremely fast intrinsic spike-bursts, with an intraburst firing rate up to 800 s⁻¹ and a repeat rate of 20–70 s⁻¹, in response to visual stimuli or suprathreshold current injection. During visual stimulation these cells exhibit pronounced oscillations in membrane potential that are largely absent at rest. All chattering cells recovered after dye-filling were typical L2/3 pyramidal neurones.

L3 receives a substantial trans-callosal input, larger than that to L5 (Porter and White, 1986) and pyramids in both layers project trans-callosally, often to topographically related cortical areas. In L2/3 of rat barrel cortex, 97% of the connections, both local and distant, made by L3 pyramidal axons are onto dendritic spines. This is a striking target preference (seen in all layers), when only 80% of all asymmetrical synapses in L3 are onto spines (White and Czeiger, 1991) and L6 corticothalamic cells preferentially innervate aspiny dendritic shafts in L4 and L6 (Elhanany and White, 1990). In primate visual, motor and somatosensory cortex, L3 (and to a lesser extent L5) cells also provide dense innervation of patches of cortex a few 100 μ wide and up to a few millimeters from the injection site within L1–3 (Levitt et al., 1993, 1994). In prefrontal cortex, a narrow stripe-like, rather than a patchy pattern is apparent (Levitt et al., 1993).

Deep L3 pyramids can also receive thalamocortical inputs from primary sensory thalamus, again, largely to proximal basal dendrites, though the further they are from the L4 border, the weaker this input becomes (White and Hersch, 1982). A major part of the projection from the pulvinar (the most caudal thalamic group, with roles in attention and oculomotor behavior) also terminates in L3 extrastriate visual areas (Rockland et al., 1999). The dense, focussed, excitatory input from L4 spiny cells onto L3 pyramids also terminates proximally, on first, second, or third order basal dendrites (Thomson et al., 2002; Thomson and Bannister, 2003), while the many inputs from other L3 pyramids are located more distally (mean 97 μm cf. 69 μm: Feldmeyer et al., 2002), on both basal and apical oblique dendritic branches. Proximal basal synapses result in taller, narrower EPSPs (excitatory postsynaptic potentials) than more distal inputs.

This relationship is partly due to the smoothing of current transfer over the length of a cable with resistance and capacitance (Rall, 1962) and partly to the activation of voltage-gated ion channels distributed with unique patterns of surface expression across somata, axons and dendrites of each class of neurone (Nusser, 2009, 2012, for reviews). Amongst the conductances whose density increases with distance from the soma, perhaps the most studied has been the rapidly inactivating K⁺ current, I_A. The I_A α-subunit Kv4.3 clusters in neocortical pyramidal dendrites and dendritic spines (Burkhalter et al., 2006) and I_A density increases as I_Na decreases more distally in large L5 pyramidal basal dendrites (Kampa et al., 2006). In CA1 pyramidal dendrites, the density of Kv4.2 also increases along the soma-dendritic axis (Kerti et al., 2012), although the gradient was shallower than expected from dendritic recordings of I_A (e.g., Hoffman et al., 1997; Sun et al., 2011; Nestor and Hoffman, 2012); a discrepancy that might result from involvement of other I_A α-subunits, auxiliary subunits, or modulators of channel conductance.

In mouse L3 pyramids, selective blockade of Kv4.2/4.3 enhanced glu-EPSPs activated by focal glutamate-uncaging at single spines. It also promoted activation of fast, dendritic spikes by summed glu-EPSPs at proximal dendritic locations and of slower, all-or-none, stereotypical, depolarizing events at proximal-intermediate dendritic locations (A Biro, A Bremaud, A. Ruiz, unpublished). Without channel-blockers these additional events required near simultaneous activation at 7–8 closely neighboring locations. Such events would enhance responses of L3 cells to thalamocortical inputs. However, two excitatory synapses provided by any one presynaptic axon rarely, if ever innervate the same pyramidal dendrite, let alone 7 or 8. They distribute across the dendritic tree on different branches, albeit at similar electrotonic distances from the soma. How frequently 7–8 presynaptic terminals, each from 7 to 8 different presynaptic neurones, all impinging on a single dendritic compartment, are activated simultaneously in life, is difficult to predict. More distal inputs e.g., from other L3 neurones or cortical regions may lower the threshold for such events (Branco and Häusser, 2011), perhaps when attention to a behaviourally important input is required. In both mouse A1 and V1 L4, local circuit activation amplified and prolonged thalamocortical responses, without altering frequency or direction selectivity and with spectral range and tuning (auditory), or with frequency and direction selectivity (visual) preserved (Li et al., 2013a,b).

There is considerable evidence that epigenetic cues are required for the final differentiation of neocortical neurones, still somewhat multi-potent on arrival. Obvious candidates for such cues are in-growing axons, particularly, perhaps, thalamocortical axons in primary sensory regions (López-Bendito and Molnár, 2003, for review).

Connections between the neocortex and subcortical structures course through the internal capsule, a thick fiber tract lying between the caudate nucleus and thalamus. Subplate neurones, diverse in site of origin, birth date, survival and gene expression, exhibit a range of morphologies and axonal projection patterns (Hoerder-Suabedissen and Molnár, 2013), including pioneer axons to the emerging internal capsule and commissural fibers of the early hippocampus (Sarnat and Flores-Sarnat, 2002). The subplate zone becomes a “waiting compartment” in which thalamocortical-, basal forebrain cholinergic-, callosal, commissural, and ipsilateral corticocortical-afferents cease growing until an appropriate environment or signal emerges.

Early born GABAergic cells migrate tangentially from their germinal zone in the LGE (lateral ganglionic eminence) and into the MGE (medial ganglionic eminence), forming a stream of cells between the MGE and globus pallidus (E11.5 to E14). These “corridor” cells form a permissive pathway through which thalamocortical axons can grow (López-Bendito et al., 2006) (Molnár et al., 2012: Figures 1 and 2; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4370206/figure/F1/; https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5040712/figure/F2/). Otherwise chemical repellents and the structure of the PSPB (pallial-subpallial boundaries): high cell-density, and a radial glial fascicle running across the trajectory of thalamocortical axons, would hinder their onward growth toward the cortex. Cortigofugal axons may also assist the forward growth of thalamocortical axons through this barrier (Molnár et al., 1998; Molnár and Butler, 2002).

Like many cortical neurones, most thalamic neurones are born between E13 and E19 (rat), the LGN, for example, between E12 and E14. By E16/E17, nuclear differentiation in thalamus has begun and both neocortex and dorsal (specific) thalamus have started to generate prospective reciprocal connections. To reach their destinations, these axons must overcome and cross several emerging barriers, or boundary zones: the diencephalic-telencephalic (DTB) and (PSPB) form transient barriers to axon growth, but interestingly, also a route for early born migrating neurones that form the permissive corridor. A largely transient population of pioneering subplate neurones sends the first projections to the internal capsule (IC) and beyond; though the axons of other cortical neurones actually invade and innervate specific thalamic nuclei first. However, without subplate projections, thalamocortical axons cannot traverse the PSPB to enter the telencephalon. Moreover, subplate ablation at this time, prevents formation of ocular dominance columns, inhibition in L4 does not mature, barrels are disrupted and spindle activity abolished (Hoerder-Suabedissen and Molnár, 2015).

By P0, axons from L6 have reached the ventrobasal thalamic nucleus (primary somatosensory) and over the next 4 days they invade and form a barreloid pattern. Corticothalamic fibers do not, however, ramify within LGN until the eyes open and spontaneous activity begins. By E16-19 thalamocortical axons have accumulated in the subplate, but they also wait, extending horizontal collaterals that may facilitate reorganization of maps at a later date, until peripheral afferents innervate the appropriate dorsal thalamic nucleus (Molnár et al., 2012).

Here we see an important change in the forward growth of thalamocortical axons from the external route seen in lower vertebrates lacking a six layered cortex, where they run over the developing cortex, to the internal route of mammals, via the corpus callosum. The midline repellent, Slit2, redirects the migration of corridor neurones, switching thalamic axons from an external to a mammalian-specific internal path (Bielle et al., 2011). It is proposed that this switch allowed the neocortex to grow radially. Interestingly, the hippocampus is deep in the brain and bounded by dense fiber tracts. Perhaps it was not able to grow in this way.

Having accumulated in the subplate, the growth of thalamocortical axons into neocortex is prevented if SNARE-complex proteins, essential for AP-driven, Ca²⁺-dependent transmitter release (though not spontaneous, “miniatures,” Ramirez and Kavalali, 2011) are knocked out. With the arrival of thalamocortical axons, transient circuits form between thalamic axons, subplate and L4 neurones. Multiple interactions now control the growth of- and connections made by- incoming axons and the development of cortical neurones and circuits. For example, two extracellular molecules: NRN1 (Neuritin-1, a GPI-anchored neuronal protein that modulates neurite outgrowth) and VGF (a nerve growth factor), both manufactured by thalamic cells and transported to their neocortical terminals, promote L4 spiny stellate dendritic growth—selectively (Sato et al., 2012). Lhx2 promotes activity-dependent L4 dendritic growth toward thalamic afferents, by inducing the transcription factor BBtbd3 (Wang et al., 2017), while several neurotrophins are implicated in the critical stages during which precise thalamocortical connections are made (Ma et al., 2002; Yamamoto and Hanamura, 2005). As L4 and its thalamic inputs mature postnatally, spiny stellate cells receive a transient input from SOM interneurones in L5b, which themselves receive thalamic input. Development of thalamic input to spiny stellates is delayed in the absence of this transient input (Marques-Smith et al., 2016), while thalamic afferents are misdirected to inappropriate barrels when Proteoglycan-2 (PRG-1, a phospholipid- interacting molecule) is knocked out (Cheng et al., 2016).

As in neocortex, expression patterns demonstrate that pyramidal classes are predestined at E15.5 while they are still in IZ. For example, SCIP (POU domain transcription factor), is present in future CA1 pyramids (Frantz et al., 1994; Tole et al., 1997), while KA1 (GluR subunit) is expressed in future CA3 neurones (Wisden and Seeburg, 1993; Tole et al., 1997) and many regulators that control neurogenesis in neocortex also act here (Urbán and Guillemot, 2014).

From LII and LIII of the entorhinal cortex information from many subcortical structures is relayed to the hippocampus via the perforant path, providing powerful input to the molecular layer of the dentate gyrus and to distal apical dendritic tufts of CA1-3 pyramidal cells in stratum lacunosum moleculare. Mossy fibers project from dentate granule cells to CA3 stratum lucidum, innervating the most proximal apical dendrites of CA3 pyramids with huge boutons. In turn, CA3 pyramidal axons (Schaffer collaterals) project to stratum radiatum and oriens of CA1. The hippocampus also sends information to and receives inputs from subcortical regions: medial septum, cingulate gyrus, mammillary bodies, thalamus and amygdala as well as regions of association cortex.

The precise position of CA2 in this unidirectional trisynaptic pathway has been unveiled more recently (Chevaleyre and Piskorowski, 2016; Dudek et al., 2016; for reviews). LII of the entorhinal cortex provides strong, proximal excitation to CA2 pyramidal cells via dentate and mossy fiber synapses in stratum lucidum (Kohara et al., 2014). CA2 is also thought to receive direct input from LIII of the entorhinal cortex in stratum radiatum and lacunosum moleculare (Chevaleyre and Siegelbaum, 2010) in addition to Schaffer collaterals (Chevaleyre and Siegelbaum, 2010; Jones and McHugh, 2011). In turn, CA2 pyramids project preferentially to calbindin-negative, deep CA1 pyramids which lie adjacent to stratum oriens (Kohara et al., 2014). CA2 pyramids also project “back” to the supramammillary nucleus (Tamamaki et al., 1988; Cui et al., 2013) and in some cases, back to LII of the medial enthorinal cortex (Rowland et al., 2013).

During development, a projection from CA1 non-pyramidal cells to the medial septum, (hippocampo-septal pathway) (Supèr and Soriano, 1994) develops before the reverse, septo-hippocampal projection: E15 vs. E17 (mouse) (for parallel studies in rat and involvement of Cajal-Retzius cells: Ceranik et al., 1999, 2000). Chemo-repulsive semiphorins repel septal axons, promote growth cone collapse and may contribute to target selection; GABAergic septo-hippocampal fibers terminate preferentially on sema3C-expressing GABAergic interneurones, while cholinergic septo-hippocampal fibers terminate on sema3E- and sema3A-expressing CA pyramidal and dentate granule cells (Pascual et al., 2005).

By E17, LIII entorhinal axons are ramifying densely and exclusively in stratum lacunosum moleculare. Invasion of the dentate comes later, but by E19, the first entorhinal axons begin to ramify there, predominantly in the outer molecular layer (Supèr and Soriano, 1994). Commissural fibers first enter the contralateral hippocampus at E18 and arborize in stratum radiatum and oriens, along with the Schaffer collaterals. The earliest commissural fibers to enter the dentate gyrus are seen even later, at P2, terminating in the inner zone of the molecular layer and the hilus. Thus, as in neocortex, incoming pathways do not meander indiscriminately; they invade their ultimate target layers and regions from their earliest appearance, some following paths marked by early born, non-pyramidal neurones.

Again, a host of genes selectively expressed at different times, in different locations and in different cell classes in the developing hippocampus appear to contribute to its normal development. For examples, see Fazzari et al. (2010) for signaling with Nrg1(Neuregulin1, a ligand for ERBB3 and 4) and ErbB4 (receptor tyrosine kinase, an epidermal growth factor receptor); Silva et al. (2015), for LGI1 (Leucine-rich, glioma inactivated 1) in both neocortex and hippocampus; (Mingorance et al., 2004), for tempero-spatial patterns of Nogo expression and its debated involvement in perforant path development (Urbán and Guillemot, 2014).

Large numbers of calretinin-expressing (CR), bipolar or multipolar Cajal-Retzius neurones appear in the molecular layer of the developing CP, becoming distributed through all layers. Collaterals of their thick primary axon make synaptic contact first with pyramidal cells in emerging L6, then sequentially with pyramids in L5 to L2 (del Rio et al., 1995). In hippocampus they become densely innervated by afferent axons from entorhinal cortex, whose ramification in CA stratum lacunosum moleculare and dentate outer molecular layer is severely reduced if Cajal-Retzius cells are ablated. Up to 90% of these cells disappear during development; the remainder form a sparse population in adult neocortical L1 (del Rio et al., 1995), hippocampal stratum lacunosum moleculare and the dentate gyrus outer molecular layer (Del Río et al., 1997). Cajal-Retzius neurones produce GABA, possibly ACh, calmodulin, PV(parvalbumin) and CR and neuropeptides. They express important mediators of radial neuroblast migration and lamination of the cortical plate: Reelin (a secreted extracellular matrix protein, essential for the normal “inside-out” development of neocortical layering), Lis1 (a motor protein Dynein-regulator), and Dscam (Down syndrome cell adhesion molecule). In addition to forming the first intrinsic synaptic circuits of the cortical plate and its first afferent and efferent connections with subcortical structures, Cajal-Retzius neurones may contribute to ocular dominance column-formation, to regulation of neurogenesis, and to cortical repair (Sarnat and Flores-Sarnat, 2002, for review).

Many recent studies have used rodents—young enough for many neuronal properties still to be maturing. Neonatal voltage gated channels, transporters and receptors are replaced during the first few postnatal weeks, resulting in a dramatic—up to four fold—reduction in the time course of many electrophysiological events. This “juvenile” period is also a time of synapse proliferation and pruning, and the speed and complexity of short term synaptic dynamics (Thomson, 2000a,b, 2003) increase in parallel. Some of the ambiguity encountered in attempts to classify cortical interneurones could result from cells at different stages of maturity; a day or two at these ages could make quite a difference: Kvα1 (Butler et al., 1998); SK2 (Cingolani et al., 2002); Kv3.2 (Tansey et al., 2002); Kv3.1b (Du et al., 1996); speeding of AMPA-R-EPSPs, P8 cf P35; shortening of synaptically released glutamate waveform, P8-P18 (Cathala et al., 2003, 2005); GABA_A-R α6-subunit expression, P7 cf P30 (Tia et al., 1996; time course of NMDA-R mediated EPSCs (Hestrin, 1992; Cathala et al., 2000); NMDA-R subunits (Farrant et al., 1994); switch from FLIP to FLOP GluR splice variants P8-14 (Monyer et al., 1991). (Batista-Brito and Fishell, 2009, Figure 5, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4465088/figure/F5/).

Moreover, reconstructions of “juvenile” cells typically demonstrate rather limited axonal ramification.

Studying a more restricted developmental stage might, therefore result in a “tidier” picture. However, cortical interneurones have to establish their own territories within a field of already established cortical layers, sublayers and regions; environments, moreover, that continue to change throughout development and in ways not entirely prescribed genetically. Following detailed studies of the crab stomatogastric ganglion, Marder and Prinz (2002) concluded that “…similar neuronal and network outputs can be produced by a number of different combinations of ion channels and synapse strengths. This suggests that individual neurons of the same class may each have found an acceptable solution to a genetically determined pattern of activity, and that networks of neurons in different animals may produce similar output patterns by somewhat variable underlying mechanisms….” It is perhaps not surprising, therefore, that while many properties are common to all members, where a given clearly definable subclass exists, others may be subject to variation and modification by the existing environment. It is only necessary to study the convoluted trajectories of interneuronal axons (and pyramidal dendritic branches) to appreciate how thoroughly they explore their environment for appropriate synaptic partners.

Two broad classes of interneurones target somata/proximal dendrites and axon initial segments of pyramidal cells, respectively. Many of their axonal branches become significantly—if sporadically—myelinated and their synaptic boutons are large and contain mitochondria; facilitating the fast, precisely timed, proximal inhibition they provide.

Basket cells “Baskets” of axons bearing large synaptic boutons that surrround principal cell somata were first described in cerebellum (Golgi, 1883, 1906) then elsewhere (Cajal, 1888, 1906; Lorente de Nó, 1922; Kritzer and Goldman-Rakic, 1995; Buhl et al., 1997; Ali et al., 1998; Tamás et al., 1998). A hippocampal interneurone destined to inhibit pyramidal somata has little choice but to innervate stratum pyramidale. Similarly, to sample all excitatory inputs controlling activity in its target cells, it extends its dendrites across all layers, from stratum oriens to lacunosum moleculare—an easily identifiable, classical CA basket cell. Some basket cell axons can also extend to proximal stratum oriens and stratum radiatum (wide arbor basket cells, Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/basket).

Three types of CA1 basket cells, the majority otherwise fairly similar in their appearance, are distinguished by immunoreactivity for PV, CCK/VIP; or CCK/VGLUT3 (Katona et al., 1999; Somogyi et al., 2004) and CB1R (type-1 cannabinoid receptor: Takács et al., 2015) (Pawelzik et al., 2002, for distributions of CA1 PV and CCK interneurones). CA1 CCK basket cells receive less synaptic input than PV baskets, with proportionally more inhibition, suggesting that they do indeed subserve different rôles (Mátyás et al., 2004) and unlike PV basket cells, whose cell bodies lie predominantly in stratum pyramidale, CCK basket somata are also found in stratum oriens and radiatum, i.e. their sampling of incoming information also has a different bias.

For neurones with such a similar overall structure and specific target preference, it is surprising perhaps that PV and CCK basket cells originate in different subpallial regions: PV interneurones in vMGE, CCK interneurones in POA or CGE and may be born later. PV interneurones (devoid of other common markers) are typically fast spiking (FS) and deliver fast IPSPs mediated by α1β2/3γ2-GABA_ARs to pyramids, while many CCK basket cells display adapting firing patterns, have broader action potentials (Pawelzik et al., 2002) and activate α2β2/3γ2-GABA_ARs on pyramids. The different pharmacologies of these receptors (hippocampus: Pawelzik et al., 1999, 2003; Thomson et al., 2000; neocortex: Ali and Thomson, 2008) and the behavioral effects of manipulating their efficacy (Möhler et al., 2002) suggest that PV baskets mediate pharmacological sedation and contribute to anti-convulsant therapies, CCK basket cells (and possibly axo-axonic cells, Nusser et al., 1996) promote anxiolysis (Möhler et al., 2002), while certain dendrite-preferring interneurones, acting on α5β1γ2-GABA_ARs (Pawelzik et al., 1999, 2003; Ali and Thomson, 2008) influence cognition (Rudolph and Möhler, 2014).

Chandelier, or Axo-axonic cells innervate pyramidal axon initial segments in deep stratum pyramidale and proximal oriens. Their cartridge bouton arrays are only partially coincident with basket cell axonal arbors (Buhl et al., 1994b). For chandeliers with somata in stratum pyramidale this and the often distinctive claw-like appearance of their apical dendritic terminal branches as they extend into stratum lacunosum moleculare assist their identification (Pawelzik et al., 2002) (Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/axo-axonic). For stratum oriens axo-axonic cells with horizontal dendrites, see Ganter et al. (2004).

At least nine classes of CA1 interneurones preferentially innervate pyramidal dendrites. Their termination zones suggest that each class selectively innervates dendritic regions also receiving a particular afferent pathway, or combination thereof. The names they have acquired often reflect this preference (Klausberger et al., 2003; Klausberger and Somogyi, 2008; Bezaire and Soltesz, 2013, for reviews). As a gross generalization, dendrite-targeting interneurones have finer, unmyelinated axons and smaller, mitochondria-poor synaptic boutons than proximally targeting cells. They display a range of firing patterns, but are rarely classical FS. Those that have horizontally oriented dendrites (OLM cells being a prime example), be they in stratum oriens, radiatum, or lacunosum moleculare, in CA1 or CA2, often display an adapting firing pattern and a pronounced “sag” current in responses to large hyperpolarizing current pulses, which can elicit rebound firing and many receive facilitating EPSPs from pyramids.

Perforant path associated cells Perforant path associated cells whose axons and dendrites are restricted to stratum lacunosum moleculare respond to perforant path input by inhibiting pyramidal apical dendritic tufts that are also in receipt of perforant path input. (CCK) (Vida et al., 1998; Pawelzik et al., 2002) (Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/ppa).

Bistratified cells have axonal arbors ramifying in stratum oriens and radiatum, but not in stratum pyramidale or lacunosum moleculare. Bistratified cells with somata in stratum pyramidale, have dendrites that span stratum oriens and radiatum. Those with cell bodies in stratum oriens, have horizontal dendrites confined to stratum oriens.The axons of both subtypes are associated with Schaffer collateral/commissural inputs to intermediate pyramidal dendrites via α5β1γ2-GABA_ARs (Pawelzik et al., 1999; Thomson et al., 2000; Thomson and Jovanovic, 2010 for review). (Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/bistratified).

(SOM, PV, CCK) (Buhl et al., 1994a, 1996; Halasy et al., 1996; Pawelzik et al., 2002; Klausberger et al., 2004; Baude et al., 2007).

Schaffer collateral-associated cells innervate the same regions as bistratified cells, but receive a different combination of inputs. Their somata lie close to the stratum radiatum-lacunosum moleculare border and their dendrites span both these layers and stratum oriens. In addition to Schaffer collateral and commissural input, therefore, these interneurones receive proximal input from perforant path, but restrict their influence to the termination regions of the Schaffer/commissural inputs (CCK: Vida et al., 1998; Pawelzik et al., 2002). (Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/sca).

Apical dendrite-innervating cells have axonal and dendritic spans similar to those of the Schaffer collateral-associated cells, but innervate the main apical dendritic trunks of pyramids, rather than their apical oblique branches (Klausberger et al., 2005; Klausberger, 2009) (CCK).

Oriens-lacunosum moleculare, or OLM cells (Cajal, 1911; Lacaille et al., 1987; Lacaille and Williams, 1990; Buckmaster et al., 1994; Blasco-Ibáñez and Freund, 1995), have horizontal thorny dendrites restricted to stratum oriens (in CA1) where they receive their most powerful drive from CA1 pyramids (Blasco-Ibáñez and Freund, 1995) with facilitating EPSPs (Ali and Thomson, 1998). In CA3, OLM dendrites also project into stratum radiatum, where local CA3 pyramidal axons also ramify. OLM cells do not, however, innervate stratum oriens, pyramidale, or radiatum. They send one or more long axons to stratum lacunosum moleculare, where they form a dense arbor in the perforant path termination zone and deliver fast IPSPs, almost invisible at the soma, but apparent in distal apical dendritic recordings (Hannelore Pawelzik, 1960-2004; Hannelore Pawelzik, unpublished) (Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/olm).

(SOM: Morrison et al., 1982; Kosaka et al., 1988; Kunkel and Schwartzkroin, 1988). (mGluR1α: Ferraguti et al., 2004) (up to one third express PV weakly: Ferraguti et al., 2004; Varga et al., 2012) one (metabotropic glutamate receptor 7, mGluR7, selectively expressed in excitatory boutons contacting OLM cells: Shigemoto et al., 1996).

(Jinno, 2009, Figure 1, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2718779/figure/F1/).

These are perhaps the group most difficult to classify and one of the smallest 4% of CA1 interneurones. Since the majority of reported cells in the following four classes have horizontally oriented dendrites confined to stratum oriens, it is probable that, like OLM cells, they receive strong excitatory input from CA1 pyramids and relay information about activity here to other regions. In addition to long distance projections, they have local axonal arbors in stratum oriens and radiatum.

(SOM; Jinno et al., 2007; Katona et al., 2017) PV possible: Ferraguti et al., 2004).

Oriens-retrohippocampal projection cells project to the subiculum. (SOM/Cb: Jinno et al., 2007; Klausberger and Somogyi, 2008). A range of subtypes project to subiculum, including an mGluR8-decorated, M2R-expressing, SOM-negative trilaminar cell (Ferraguti et al., 2005).

Double projection cells (Klausberger and Somogyi, 2008) project to the septum and subiculum (SOM/Cb, or CR). Some also express mGluR1α and/or NPY and up to 30% express PV weakly.

Back-projection cells (Sik et al., 1994; Katona et al., 2017, for in vivo filled cells) project to CA3 and/or dentate gyrus, sometimes crossing the fissure, which appears to be an impenetrable barrier to other neuronal processes. (PV, SOM, Cb-negative). (Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/backprojection).

Cb-septal projection cells project to the septum (SOM, Cb) (Gulyás et al., 2003).

Amygdala-projecting interneurones project from ventral CA1 stratum oriens, pyramidale and radiatum, to the amygdala (Lübkemann et al., 2015). (PV, Cb, SOM, NPY and/or CCK).

The neurogliaform family (Overstreet-Wadiche and McBain, 2015, for review).

Two classes have been described, which differ predominantly in the inputs they receive and the subcellular compartments they inhibit.

Neurogliaform cells are often found at the stratum radiatum-lacunosum moleculare border, with short, fine, often highly convoluted dendrites and a dense and spatially restricted axonal arbor, positioned to inhibit distal apical dendrites of pyramidal cells; (nNOS (neuronal nitric oxide synthase), NPY, α-actinin-2, COUP-TFII) (Price et al., 2005; Fuentealba et al., 2010).

Ivy cells are structurally similar to neurogliaform cells, but lie close to the stratum radiatum-pyramidale border where they inhibit proximal pyramidal compartments. Although the GABA_ARs activated by ivy cells demonstrate rapid kinetics via receptors also utilized at PV basket synapses (α1β2/3γ2, unpublished results), the proximal IPSPs elicited by ivy cells are very slow. This may be due to non-synaptic, as well as synaptic release of GABA, since what appear to be synaptic vesicles in these axons are not always apposed to postsynaptic specializations (Fuentealba et al., 2008a; see also Oláh et al., 2009; Armstrong et al., 2012, for review) (nNOS, NPY). (Supplementary Figure 1, http://uclsop.net/interneuron-reconstruction/ivy).