Cytochrome oxidase “blobs”: a call for more anatomy

Abstract

An ordered relation of structure and function has been a cornerstone in thinking about brain organization. Like the brain itself, however, this is not straightforward and is confounded both by functional intricacy and structural plasticity (many routes to a given outcome). As a striking case of putative structure–function correlation, this mini-review focuses on the relatively well-characterized pattern of cytochrome oxidase (CO) blobs (aka “patches” or “puffs”) in the supragranular layers of macaque monkey visual cortex. The pattern is without doubt visually compelling, and the semi-dichotomous array of CO+ blobs and CO− interblobs is consistent with multiple studies reporting compartment-specific preferential connectivity and distinctive physiological response properties. Nevertheless, as briefly reviewed here, the finer anatomical organization of this system is surprisingly under-investigated, and the relation to functional aspects, therefore, unclear. Microcircuitry, cell type, and three-dimensional spatiotemporal level investigations of the CO+ CO− pattern are needed and may open new views to structure–function organization of visual cortex, and to phylogenetic and ontogenetic comparisons across nonhuman primates (NHP), and between NHP and humans.

Graphic abstract

Introduction

Few cortical features are as visually compelling as the pattern of cytochrome oxidase (CO) blobs in the primary visual cortex of humans and nonhuman primates (NHP). The distinctness together with the relatively accessible supragranular location has made this pattern an important guide and facilitator of functional-connectivity investigations. This very feature, however, has too often tended to obscure the large continuing gaps in our understanding, both at the smaller spatial scale (< 100 μm) and the more global, whole area level. Investigations of the CO network in humans and NHP commonly state that the function remains unclear (e.g., Adams et al. 2015). Less frequent is the statement that their anatomical organization is also not clear, at the level of cellular diversity, axonal convergence and divergence, and microcircuitry. Recent physiological results suggest that the CO pattern is associated with a functional architecture more complex than previously appreciated (e.g., Garg et al. 2019; Chatterjee et al. 2021), making the need for higher resolution anatomical data all the more urgent.

In this mini-review, discussion is limited to the strongly expressed anatomical periodicity in layer 3, with some comparisons to the related but smaller scale structures (“honeycomb”) in layer 4A and at the border of layers 1 and 2. Among more extensive reviews, treating issues of functional specialization and intrinsic and extrinsic connectivity, are (Merigan and Maunsell 1993; Callaway 1998; Sincich and Horton 2005; Nassi and Callaway 2009; Angelucci et al. 2017; Vanni et al. 2020).

Brief background

Early interest in CO in the visual cortex was as a mitochondrial enzyme that might be a useful proxy for relative levels of neuronal activity (reviewed in Horton 1984; Wong-Riley 1994, 2010). The actual discovery of CO blobs in the supragranular layers of macaque V1 was quasi-serendipitous, as credited to Margaret Wong-Riley, but was quickly followed by multiple investigations establishing their dimensions, density, and spatial regularity, especially in relation to retinotopic organization (Horton 1984; Wong-Riley 1994), co-localization with other markers, developmental trajectories, selective connectivity, and physiological response properties, at the level of single units (Livingstone and Hubel 1984a; Tootell et al. 1988a,b; Born and Tootell 1991). The CO pattern has been mapped in relation to more macro-level populational responses to ocularity or orientation preference, by optical or intrinsic imaging (Livingstone and Hubel 1984b; Yoshioka et al. 1996; Stettler et al. 2002; Li et al. 2019).

The CO blobs in layer 3 are visualized, in addition to cytochrome oxidase, by higher levels of other metabolic enzymes, consistent with heightened levels of neural activity (lactate and succinate dehydrogenase; (Horton 1984; Wong-Riley 1994). They co-localize with higher levels of acetylcholinesterase and of the neuropeptide tachykinin (Hendry et al. 1988); of GABAergic inhibitory terminations, but not GABAergic neurons (Hendrickson et al. 1981; Fitzpatrick et al. 1987; Wong-Riley 1994; Nie and Wong-Riley 1996; Adams et al. 2015; but see Beaulieu et al. 1992); and of nitric oxide (NOS) + terminations, but not NOS + cell bodies (equivalent to Neuropeptide Y + interneurons, (Kuljis and Rakic 1989; Horton and Adams 2005). Higher levels of myelin correspond to CO + blobs (Horton 1984; Rockoff et al. 2014).

Despite the obviousness of the CO patterning at low magnifications, borders are not easy to delineate (e.g., Purves and LaMantia 1990; Farias et al. 1997). Pyramidal cell basal dendrites are promiscuous in relation to CO + compartments, and do not respect the position of the cell body. For those neurons at the edge of a CO + blob, dendrites extend in both CO+ and CO− compartments (Hubener and Bolz, 1992; Malach. 1994). Similarly, neurons with soma in the upper half of layer 3 will still have basal dendrites extending into the lower half; and those in the lower half will have dendrites extending into layer 4A or even 4B.

At the electron microscopic level, CO reactive mitochondria can be visualized preferentially in post-synaptic dendrites within CO+ blobs, consistent with higher spontaneous activity and associated greater metabolic demand of neurons in these domains (Wong-Riley 1994, 2010). Axons in the white matter exhibit low levels of CO activity; but there may be variable energy demand at the distal axon terminals (Wong-Riley 1994). Excitatory axon terminals (not further identified) in blobs contain about three times more dark CO + mitochondria, and presumably are synaptically more active (Nie and Wong-Riley 1996). Dense CO activity might reside in thalamocortical afferents and/or in dendrites that directly receive denser or more efficacious thalamic input (Takahata 2016; Yao et al. 2021).

The two CO+ /CO− compartments are not symmetric, in that the CO− “matrix” (a.k.a. interblobs) occupies up to 80% or more of the V1 area (estimates vary according to studies and according to species: Fig. 10, Marcondes et al. 2017). This asymmetry has been widely acknowledged, but the implications perhaps less so, in evaluating the functional organization, local connectivity, and upstream connectivity of the two compartments. The larger CO− interblob territory can be presumed to have more extrastriate projecting neurons and give rise to more synapses, both intrinsic and extrinsic. Moreover, as discussed below, there is evidence for further intra-compartment subdivisions (i.e., more than two) and cross-compartmental dendritic overlap (i.e., gradientwise organization). A well-established feature and possible clue, still needing further investigation, is that CO+ blobs as a group are variable in shape and size, with curious “anomalies,” such as CO-to-CO “bridges” (Horton, 1984; Wong-Riley, 1994; Marcondes et al. 2017).

Functional segregation

An early observation was that the CO+ blobs in macaque are centered over the ocular dominance domains and might be related to basic aspects of visual field topography and point-image size, and to basic visual properties, such as color and shape (Livingstone and Hubel 1984a; Tootell et al. 1988a,b). A strong interpretation of a binary specialization for color (CO blobs) or form (interblobs) has over the years evolved to a more nuanced view, with the realization that segregation of response properties is not complete (e.g., Levanthal et al. 1995; Economides et al. 2011, and reviews in Merigan and Maunsell 1993; Sincich and Horton 2005; Nassi and Callaway 2009). Recent studies using two-photon calcium imaging discuss alternatives, where orientation and color might be processed by overlapping circuits (Garg et al. 2019), or color representation might switch between blobs and a combined blob/interblob system (Chatterjee et al. 2021).

Anatomical duality?

As reviewed in this section, several markers are preferentially associated with CO blobs or interblobs. The appearance of a compartmental dissociation, however, may be deceptive. On the one hand, there are likely to be significant subcompartments. The CO+ blobs have often been remarked as non-homogeneous and “mottled” (Horton 1984; “reticulated interior, like a puff of cotton,” Wong-Riley 1994, p.150). On the other, dendrites extend across compartments, as mentioned above, so as potentially to effect differential combination, rather than any sharp segregation of inputs.

Layer 3 compartments: CO+ blobs

A major distinguishing anatomical feature is that the CO+ blobs receive direct input from the lateral geniculate nucleus (LGN), as established by multiple criteria. Immunocytochemisty for VGLUT2, a specific marker for thalamocortical terminations, localizes with CO + blobs (Bryant et al. 2012; Garcia-Marin et al. 2013; Yao et al. 2021). This input is further specified as from the koniocellular LGN, demonstrated in macaques by cortical injections of retrograde tracer, or anterograde injections in the LGN (Livingstone and Hubel 1982; Fitzpatrick et al. 1983 (squirrel monkey); Hendry and Yoshioka 1994; Casagrande et al. 2007). Koniocellular layers selectively receive inputs from the superior colliculus and the cholinergic parabigeminal nucleus (reviewed in Hendry and Reid 2000; Casagrande et al. 2007). The koniocellular LGN also projects to extrastriate area MT (Sincich et al. 2004; Vanni et al. 2020). There is no evidence that axons branch to both V1 and MT, but a short multisynaptic pathway could be possible; for example, via feedback projections from MT to apical dendrites of layer 3 “blob” neurons in layer 1 of V1 or basal dendrites extending into layer 4B.

Receptor architecture in the cholinergic system is differentially correlated with the CO pattern. As shown by Tigges et al. (1997), muscarinic receptors differentially interdigitate (m² receptors) or overlap (m³ receptors) with the CO+ (VGluT2+) blobs. M2 receptors can be pre- or postsynaptic, and are multifunctional depending on density, among other factors (Disney and Aoki 2008). In the upper layers, m2 receptors are preferentially located on inhibitory interneurons, and have been proposed as overall inhibitory in effect (Disney et al. 2012). An interesting possibility is that there are cholinergic modulatory compartments, operating at small spatial and fine temporal scales, a configuration that might allow task relevant cortical circuitry to interact with and influence its own modulation (Coppola et al. 2016). How this cholinergic CO+/CO− compartmental dissociation relates to visual processing remains obscure. Serotonin (Watakabe et al. 2009) and GABA receptors (Zilles and Palomero-Gallagher 2017) also have a regular organization within the visual cortex, investigated in terms of laminar specificity but less so in relation to the CO compartmental pattern.

Layer 3 compartments: CO−interblobs

CO− interblobs do not receive direct LGN input, but have preferential input from zinc+ (Zn+) cortical terminations (Dyck and Cynader 1993) (Fig. 1). Synaptic zinc is a pre-synaptic activity- and calcium-dependent neuromodulator, known to interact with a variety of receptors, ion channels, and neurotrophic factors (reviewed in Dyck et al. 2003; Nakashima and Dyck 2009; Sensi et al. 2011). Zn+ terminations are prominent in limbic areas (e.g., hippocampal mossy fibers have the highest amount of zinc), but occur at reduced density in association and sensory areas. Given the greater tissue volume of CO− interblobs, the Zn+ terminations can be inferred to outnumber the LGN input to the CO+ blobs, although how this impacts “synaptic weights” (in terms of synaptic numbers and postsynaptic action) is unknown.

Fig. 1.

A, B Tangential sections through layer 3 of macaque V1. CO+ blobs are typically spaced ~ 500 μm apart (A) within a matrix of CO− interblobs (B). The latter are demarcated by zinc+ (Zn+) terminations. Arrows point to corresponding blood vessels in adjacent sections. C Coronal section stained for Zn+ terminations (area V1). Layer 4C is conspicuously sparse in Zn, and layer 3 has a regular Zn+ and Zn− periodicity (corresponding, respectively, to CO− and CO+ domains). Arrow points to a Zn+ band in layer 1b. D Coronal section (50 μm thick) from area V1 of a 2.5-month-old macaque, stained for calbindin (CB). Note periodic pattern, where CB+ domain surrounds “empty” zones, which correspond to CO+ blobs, three being indicated by pointers. E Tangential section through layer 4A honeycomb (macaque V1), the hollows of which are redrawn in F. F Superimposition of the layer 4A honeycomb in relation to the CO+ blobs (shaded) in overlying layer 3. Both patterns are periodic, but the layer 4A honeycomb is only quasi-regular. Circles within the CO+ blobs indicate blood vessels. E, F Adapted with permission from Peters and Sethares (1991a)

Specific sources of the Zn+ terminations have not been identified, but candidates are: intrinsic Zn+ neurons within V1 and/or the subpopulations of layer 6 feedback neurons in V2, V4, and MT that use synaptic zinc (Ichinohe et al. 2010). Since Zn+ terminations are only a subset of corticocortical terminations, other, Zn-cortical synapses are not necessarily compartment specific. The Zn + terminations may be another layer of complexity, but not necessarily of sharp segregation.

The Zn+ (CO−) domain has a higher density of calbindin+ (CB+) terminations (Fig. 1; and Celio et al. 1986). Again, the source has not been identified, but candidates include: terminations from the small number of CB+ pyramidal neurons in extrastriate areas (Kondo et al. 1999) or, more likely, from CB+ inhibitory neurons in V1, especially from the CB dense layer 2 (van Brederode et al. 1990) or layer 4A (Preuss and Coleman 2002; Bryant et al. 2012; Garcia-Marin et al. 2013). The CO− interblob domain is the metabolically less active and, although identifiable by CB+ terminations, has a lower density of GABAergic inhibitory terminations than the CO+ blobs (Fitzpatrick et al. 1987; Wong-Riley 1994; Nie and Wong-Riley 1996).

Heightened density of parvalbumin + (PV +) neuropil is associated with thalamocortical terminations, in the CO + blobs and in layers 4A and 4C (van Brederode et al. 1990; Johnson and Casagrande 1995; Adams et al. 2015). How these very different neuropils, CB+ or PV+ , are organized in a functional microcircuitry needs to be further investigated.

Structural unknowns important to know

Much of the structural data on the functional features and connectivity of CO+ blobs and CO− interblobs has been at a macro level, not single neuron. This approach has been conspicuously successful in inspiring concepts, such as cortical processing streams, although the complicated compartmental connectivity of V1 to V2 is still controversial and will only be briefly discussed here (see Vanni et al. 2020 for a recent review and current references). My main goal here is to highlight, in a positive sense, the surprising gaps that still remain at the cellular and microcircuitry level and the need for continued investigations, especially with the advent of increasingly sensitive analysis tools that allow for both better resolution and larger sample size. For the sake of brevity and greater focus, the treatment is necessarily selective, and inhibitory connections are largely ignored.

Cell types

Early work of Wong-Riley and collaborators described three general types of neurons in the CO+ blobs: Type A (small pyramidal and nonpyramidal; the most prevalent); Type B, medium to large pyramidal neurons; and type C, medium-sized nonpyramidal neurons (reviewed in Wong-Riley 1994, 2010). In aggregate, there are more CO+ neurons in the blobs than interblobs (16.4 vs. 12.1 per 20,000μm²), but no evidence for differential cell density, although a high degree of inter-individual variability has been remarked (Wong-Riley 1994). The relative proportion of inhibitory to excitatory neurons is not established for either the CO+ or CO− domain, nor how this may differ across the visual field representation.

Pyramidal cells in layer 3 (not distinguished according to CO+ or CO−) that project to extrastriate area V2 are morphologically heterogeneous (Fig. 2; Rockland 1992). Soma size distinguishes small and medium neurons (respectively, 10 × 10 μm or 15 × 20 μm), intermingled without any perceptible size gradient within upper and lower layer 3. There is variability of apical dendritic morphology. The apical dendrite of most neurons (n = 35 of 50) fanned into a terminal tuft (70–110 μm in diameter) at the level of layer 2; but some (n = 15 of 50) bifurcated in a simple fork that ascended to layer 1 without forming a terminal tuft (Fig. 2, and Rockland 1992).

Fig. 2.

A Drawings of two pyramidal neurons in layer 3A of macaque area V1, retrogradely labeled by injections of PHA-L (A) or biocytin (B) in area V2. Neuron in A has a bifurcated, forked apical dendrite (arrow); the neuron in B has a more typical, single apical dendrite with a terminal tuft in layers 1, 2. Ax axon. C Photomicrographs of pyramidal neurons in macaque V1 retrogradely labeled by an injection of PHA-L in area V2. Two neurons are designated, with forked apical dendrite (solid arrow) or single apical dendrite fanning into a terminal tuft (hollow arrow). Scale bar = 50 μm. D Photomicrograph of a semi-tangential section through layers 2, 3 of area V1 (squirrel monkey). Clusters of projection neurons are retrogradely labeled by an HRP injection in area V2. Scale bar = 200 μm. E Coronal section with projection neurons, concentrated within layer 3A of macaque V1, retrogradely labeled by a biocytin injection in area V2. Scale bar = 200 μm. Arabic numbers refer to cortical layers. Adapted with permission from Rockland (1992)

From intracellular biocytin fills of layer 3B pyramidal neurons in vitro, Sawatari and Callaway (2000) distinguished between those with (7 of 20 neurons) or without (13 of 20) expanded apical tufts. The former, “projecting” neurons, were found to have axons descending into the white matter while the latter, “local pyramids,” did not.

The relation of connectivity and physiological response properties to cell types remains under active investigation, with continuing technical advances. Transcriptomic analyses, for example, strongly suggest that finer classification is a common feature within pyramidal cell projection groups: e.g., mouse subiculum: Cembrowski and Spruston (2019); layer 6 of macaque V1: Hawken et al. (2020); mouse visual cortex: Kim et al. (2020). Several genes are known to be differentially expressed in area V1 and the CO+ blobs (Takahata et al. 2009). These are related to part of the extracellular matrix, and may be a hint of the important role of innate factors in the functional architecture of area V1 (i.e., the heightened expression of myelin in the blobs; and separately, Crowley and Katz 2002). Further data can be expected on the arrangement of defined cell types in relation to the CO blobs, interblobs, and border zones, as well as to the micromaps reported in recent physiological studies (Chatterjee et al. 2021).

Another source of neuronal diversity is the early developmental timeline. For example, in the striatum, striosomal spiny projection neurons are generated earlier than those with matrix identity. The neurogenesis of the two populations is mediated through two distinct types of intermediate progenitors with limited or expanded capacity, where the former give rise to the numerically smaller striosome compartment (Kelly et al. 2018). In the case of NHP visual cortex, layer 3, and by extension the CO blobs, cannot be regarded as uniform in depth. According to the known inside-out cortical progression, neurons deeper in layer 3 are born earlier than those located more superficially. Deeper layer 3 is also known to be less cell dense than layer 3A (Lund and Boothe 1975; Rockland 1992). Further information is lacking about tangential, compartment specific differences in neurogenesis.

Microcircuitry

At the cellular and subcellular level, detailed microcircuitry of macaque visual cortex is still in its infancy, despite the importance of those data for further understanding of structural substrates and functional mechanisms. Four outstanding issues are briefly discussed below as pertaining to the CO compartments.

First, LGN input, known to originate from koniocellular neurons (K), has only been investigated at the presynaptic and light microscopic level. From anterograde labeling of K-axons we know there are two or more subtypes (Hendry and Reid 2000; Casagrande et al. 2007). One has a terminal arbor that is relatively simple and preferentially targets layer 1 (postsynaptic target = ?) or, sporadically, layer 1 together with the upper subdivision of CO+ blobs (Fig. 3). The second has a more elaborate terminal arbor that tends to concentrate in CO+ blobs only (Fig. 3; and Casagrande et al. 2007). On this basis, the group of CO+ blobs may be further subdivided by direct K-input (1) to proximal dendrites only, (2) to proximal dendrites AND distal dendritic tufts, or (3) mainly to distal dendritic tufts only. Among the questions that arise:

• What is the postsynaptic pool of the K-axons? Does this, presumably, include dendrites from neurons in layer 5 or from interblob neurons?

• Are there neurons in the CO+ blobs which preferentially receive numerically dense or sparse LGN input, and how does this relate to other inputs to the proximal dendrites or distal apical dendrites in layer 1?

• What is the synaptic distribution along postsynaptic dendrites, and the spatial relationship to other inputs? Spatial clustering of synaptic inputs has been reported at dendritic targets (macaque V1: Ju et al. 2020; and mouse:Karimi et al. 2020; Scholl et al. 2017). The emerging evidence for complex post-synaptic functional synaptic architecture supports the idea that individual neurons have a broad and dynamic range of operations (reviewed in Scholl and Fitzpatrick 2020).

Fig. 3.

Top: four koniocellular LGN axon segments (2–5) in area V1, labeled by an anterograde tracer injection spanning LGN layers K3–K6 (insert) of macaque monkey. Shaded circles represent CO + blobs in cortical layer 3; dashed lines demarcate cortical layers 1–6, where pia (*) is at the top. Bottom: five koniocellular axon segments (2–6) labeled by an anterograde tracer injection spanning LGN layers K1–K2. Note preferential arborisation in layer 1. Conventions as in the top image. D, dorsal; L, lateral; M, medial; V, ventral. Adapted with permission from Casagrande et al. (2007)

Second, there are questions about differential input to layer 1 apical dendritic tufts in relation to cell type (e.g., pyramidal cells with forked dendrites and sparse tufts vs. those with more standard apical tufts, among others) and in relation to parent cell location (e.g., potential variability for neurons with soma within, bordering, or outside the CO + blobs). The excitatory input pool to layer 1 includes (Fig. 4) thalamocortical inputs, primarily to layer 1a (from lateral pulvinar and LGN; Fitzpatrick et al. 1983 in squirrel monkey; Rockland et al. 1999; Bryant et al. 2012; Garcia-Marin et al. 2013; Moore et al. 2019); feedback cortical inputs, of which a Zn + subpopulation originating from layer 6 preferentially targets layer 1b (Ichinohe et al. 2010); amygdalocortical inputs to layer 1b (Freese and Amaral 2005); and cholinergic fibers from the basal forebrain (Mesulam et al. 1992). Inhibitory inputs also target layer 1, from zona incerta (Lin et al. 1990) and an assortment of local inhibitory interneurons (Lund 1988; DeFelipe 2011). It is unknown whether these layer 1 inputs preferentially target neurons whose somata are in CO+ or CO− domains, and how they might be spatially clustered on distal dendrites in layer 1.

Fig. 4.

Excitatory inputs to layer 1 (macaque V1). A and B Darkfield photomicrographs from area V1 with A amygdalocortical terminations (in layer 1b) and B feedback corticocortical terminations, anterogradely labeled by injections of tritiated amino acids. ccs = calcarine sulcus. Scale bar = 1 mm. C Pulvinocortical projections to layer 1a (macaque V1), anterogradely labeled by an injection of BDA in the pulvinar. Scale bar = 200 μm. Adapted with permission from Freese and Amaral (2005) (A); Shipp and Zeki (1989) (B); Moore et al. (2018) (C)

Third, the intrinsic connectivity of layer 3 pyramidal neurons, with soma either in blobs or interblobs, has not been definitively addressed at the level of single axons and their postsynaptic targets within V1. Most of the results on intrinsic connectivity of layer 3 pyramidal neurons have been from relatively macroscopic injections (> 0.1 mm in diameter), which are best suited to reveal an averaged convergence of multiple pooled neurons. The resulting patchiness of terminal arrays broadly supports the idea of like-to-like connectivity in terms of CO+ or CO− compartments and their functional response properties (Livingstone and Hubel 1984b; Stettler et al. 2002; Vanni et al. 2020; but see Yoshioka et al. 1996). Higher resolution data from single intracellularly filled neurons, however, are suggestive of a more trendwise architecture. Analysis of twenty intracellularly filled neurons, for example, showed a trend, but only a trend, for like-to-like connectivity (Yabuta and Callaway 1998); and a similar, trendwise relationship was more recently reported for cat visual cortex, when referenced to orientation domains (Koestinger et al. 2017).

The idea of “like-to-like” intrinsic connectivity undergoes further revision as research more and more addresses the subcellular level of organization. Individual synaptic inputs exhibit a high degree of functional diversity compared to somatic output, blurring to some extent the distinction of “like” or “unlike.” Along with other evidence (reviewed in Scholl and Fitzpatrick 2020), the proposal is that individual neurons have a broad range of operations and encode multiple sensory features and contextual modulation. This is a significant modification of older views of a binary organization, although there still may well be distinctive operational modes correlated with the CO+/CO− landscape.

Perhaps not surprising, given the relatively small sample size of intracellular fills, we still have only estimates for the average number and spatial distribution of intrinsic collaterals and their terminations (e.g., McGuire et al. 1991). One early in vitro study in macaque (Yabuta and Callaway 1998) distinguished a subpopulation of pyramidal neurons (n = 5 of 20, with soma 130–200 μm from blob centers) with only short horizontal axon collaterals, not extending beyond 600–800 μm from the cell body, in contrast with the more typical number of 5–12 collaterals extending for several millimeters. Intriguingly, Koestinger et al. 2017 (in vivo, in cat) distinguished some layer 3 axons with a “hybrid” morphology (n = 5 neurons), where intrinsic collaterals of a single neuron bear synapses arranged in clusters or along a more linear array. The linear segments of the same neuron can be myelinated or unmyelinated. In the CO+ /CO− context, how heterogeneous are pyramidal neurons in terms of number of intrinsic collaterals, synaptic density, and compartmental targeting? Photostimulation-based mapping of functional input to layer 3 pyramidal neurons (n = 31) already revealed highly diverse input sources to individual cells (“No 2 cells…received detectable input from the same combination of layers,” Sawatari and Callaway 2000).

Fourth, extrinsic connectivity has been largely investigated at the macro level, typically guided by the concept of like-to-like selectivity (reviewed in Vanni et al. 2020). Retrograde tracer injections in extrastriate areas can demonstrate a preference for V1 projections specific to CO+ blobs or interblobs, but issues remain about the proportion of different cell types (themselves, still to be established) and the numerical fall off of labeled projecting neurons across a spatially distributed array of blobs. Single axon reconstructions, V1 neurons projecting to V2 often have multiple spatially dispersed arbors (Rockland and Virga, 1990), but the organization of the arbors within the CO−defined domain in V2 has not been mapped.

Smaller scale periodicity

Layer 4A

In the macaque and other NHP, a smaller scale CO+ / CO− “honeycomb” is situated in Layer 4A (Fig. 1) (Hendrickson et al. 1981; Horton 1984; Peters 1994). The pattern has an average center-to-center diameter of 60 μm, but is more irregular than the overlying CO+ blobs (variable range: 30–100 μm; Peters and Sethares 1991a). The components, in the CO+ “walls”: are LGN terminations, but from parvocellular not koniocellular LGN, PV+ terminations (probably corresponding to the LGN terminations), and higher density of GAD+ inhibitory terminations (Hendrickson et al. 1981; Blasdel and Lund 1983; Fitzpatrick et al. 1983, 1987; Casagrande et al. 2007; Bryant et al. 2012; Garcia-Marin et al. 2013). The CO− “hollows,” as for the overlying CO interblob zone in layer 3, have a greater density of m2 receptors (Mrzljak et al. 1996; Tigges et al. 1997; Krueger and Disney 2019) and Zn+ terminations (Dyck and Cynader 1993; Dyck et al. 2003). In addition, small clusters (“cones”) of pyramidal cells are located in the honeycomb hollows (Peters and Sethares, 1991a, b; Peters 1994). Their apical dendrites extend toward the pia, presumably indiscriminately through both the overlying CO+ and CO− domains.

The small scale layer 4A honeycomb is anatomically conserved across NHP and humans, but with distinct changes (Preuss and Coleman 2002; Bryant et al. 2012; Garcia-Marin et al. 2013). In humans, the honeycomb, as in NHP, can be recognized by location (above layer 4B) and the configuration of small patches (Peters and Sethares 1991a; Peters, 1994). Labeling by VGLUT2 or CO, however, is absent, while the walls are instead labeled by CB+ neuropil. This is opposite from the features of layer 3 CO+ blobs, and opposite from the layer 4A honeycomb walls in macaque, which are CO+ and VGLUT2+ but CB− (similar to the layer 3 CO+ blobs). Bryant et al. (2012) have proposed that parvocellular LGN projections to layer 4A have been markedly reduced or lost in hominid evolution.

The layer 4A honeycomb is mentioned here for several reasons. First, the structure is even more compelling than the overlying blobs/interblobs in layer 3, but the function has remained obscure. Second, cellular level responses in layer 4A, until now relatively unexplored, are addressable with the increasingly common use of 2-photon in-depth imaging. Third, the relatively clear identity but small size of the honeycomb makes it an attractive target for quantitative evaluations. For example, in macaque, LGN terminations in the honeycomb walls, visualized by VGLUT2, are 5–6 times less dense specifically in the foveal representation (Garcia-Marin et al. 2015). The authors suggest that this may indicate a functionally significant reduced density of S-cone blue/yellow opponent projecting thalamic neurons, in correlation with the low number of blue/yellow opponent S-cones in the fovea and the reduced blue perceptual ability for small fields in the center of gaze.

Layers 1, 2

Another small scale “honeycomb” has more recently been identified at the border of layers 1 and 2, where a heightened density of m2 muscarinic receptors constitutes a complementary array to CO + blobs (Fig. 5; Mrzljak et al. 1996; Tigges et al. 1997; Ji et al. 2015; sFig. 4). It is unclear whether this represents the uppermost extension of the layer 3 pattern or whether this might be another separate structure, analogous to the honeycomb in layer 4A.

Fig. 5.

Tangential section through the border of layers 1,2 of macaque area V1 (160 μm below the surface). A Periodic M2 expression; B complementary pattern of M2 (in red) and CO+ blobs (in green). Scale bars = 1.0 mm. Arrows indicate matching locations. Adapted with permission from Ji et al. 2015 (Supp. Fig. 4)

Rodent honeycomb

The layer 3 CO+ CO− pattern does not occur in rodents; but there are several reports of a small-scale honeycomb periodicity at the border of layers 1 and 2 in rodents, possibly homologous to the small scale honeycomb patterns in NHP (Fig. 6), and worth mentioning for that reason. In both rats (Ichinohe et al. 2003) and mice (Ji et al. 2015; D’Sousa et al. 2019), terminations from the LGN form patches that coincide with a greater density of m2 receptors (mice: Ji et al. 2015; rats; sFig. 3). The layer 1, 2 honeycomb has been best described functionally in mice, where physiological recordings report that neurons aligned with m2 + patches have high spatial acuity, but those in m2-zones have high temporal acuity (Ji et al. 2015).

Fig. 6.

Periodic patterns in layers 1,2 of rodent area V1. A (upper row): Tangential section (mouse V1) showing patchy distribution of anterogradely labeled geniculocortical projections, at low (left) and, from box, higher magnification (right). A (lower row) Same section stained with antibody against m2, which reveals a similar patchy pattern, at low (left) and, from box, higher magnification (right). Overlays (not shown) indicate overlap of the LGN and m2 receptor patterns. B Three adjacent tangential sections (A′, B′, C′) showing a parvalbumin + (PV +) honeycomb at the border of layers 1,2 of rat visual cortex. D(ISP CHK ALL)’ Higher magnification. E′, F′ PV-hollows are cell sparse (arrowheads), as seen by comparing PV (E′) and Nissl (F′). C Coronal section through mouse V1, stained with antibody against m2. The m2 expression is strongest in layer 4 and lower 3, and moderate in layers 1, 2, where the m2 expression is distinctly patchy (white arrowheads) Adapted with permission from Ji et al. (2015) (A, C); and Ichinohe et al. (2003) (B)

Dendritic neuropil

A fine-scale modularity of apical dendrites has been reported in at least two systems. One is the “minicolumn” (diameter ~ 30 μm, Peters and Sethares 1991b) proposed to consist of a central core of apical dendrites of layer 5 pyramidal cells, surrounded, as these ascend toward layer 1, by apical dendrites of layer 3 pyramidal neurons (Peters and Sethares 1991b; Peters 1994). The bundles are best visualized in sections tangential to layer 1. The overview image is compelling, but as with the CO pattern, there are other structural features which do not it; for example, not all apical dendrites form bundles (Peters and Sethares 1991b), and the topography has not been investigated for the more complicated neuropil of basal dendrites, oblique dendrites, and apical dendritic tufts. The bundles are likely to be heterogeneous. This has not been investigated for visual cortex, but a quantitative study (Gabbott 2003) of apical dendritic bundles in the frontal cortex of humans reported a diameter range of 0.5–4.0 mm (average 2.3 μm).

Two-photon dendritic imaging with a genetically encoded glutamate sensor has enabled visualization of spatiotemporal maps along an identified dendrite. Synaptic inputs are found to spatially cluster locally by stimulus feature (e.g., color or orientation specificity) but functionally scatter in multidimensional feature space. Apical dendritic inputs had larger receptive fields and longer response latencies than basal dendritic inputs (Ju et al. 2020; and related: Scholl et al. 2017).

Other dendritic arrangements

In layer 2, as mentioned above, the short apical dendrites of pyramidal neurons cluster as part of a distinct honeycomb, rather than, as originally proposed, forming the outer circumference of the layer 5 originating minicolumns. The pyramidal cell cones in layer 4A organize as another dendritic neuropil. The infragranular Meynert cells are reported to distribute preferentially in relation to the CO interblob domain (Payne and Peters 1989), and presumably have apical dendrites that avoid the CO+ blobs. These can be supposed as “solitary” (i.e., non-bundled) in keeping with the sparse distribution of the parent Meynert cells.

A regular topography of dendritic neuropil is strikingly obvious in the non-sensory limbic cortical areas. One example is the granular retrosplenial cortex (in the rat, not mouse), where apical dendrites of layer 2 pyramidal neurons form conspicuous bundles (Wyss et al. 1990). These originate from callosally projecting neurons, selectively receive thalamic input from the anteroventral nucleus, and interdigitate with interbundle zones (30–200 μm wide), preferentially receiving intracortical projections and occupied by ascending dendrites of layer 5 neurons (Ichinohe and Rockland 2002). Much of the entorhinal cortex is organized into prominent CO+ cell islands in layer 2, which are interspersed with apical dendrites from pyramidal neurons in the deeper layers (Hevner and Wong-Riley 1992).

Conclusion

The CO pattern is conspicuous at a macro-scale (i.e., visible by eye or low magnification), and much of the research on the connectivity and physiological properties has been referenced to this. A next stage of investigations might directly interrogate the known variability in size and shape of CO+ blobs in three-dimensional space, and their dynamic interactions across distributed space, as a function of defined behavioral responses, extending beyond strictly visual properties. The blobs occur in the representation of the far visual field periphery (Horton 1984), a low acuity region that receives extra-visual input from auditory (Rockland and Ojima 2003) and parietal cortices (Borra and Rockland 2011). Pyramidal neurons elaborate apical dendritic tufts in layer 1, where they are accessible to visual corticocortical feedback and thalamocortical inputs, but also to amygdalocortical and cholinergic inputs, and inhibitory circuits from zona incerta (Lin et al. 1990) and local inhibitory interneurons (Lund, 1988). Cholinergic input is commonly linked to attentional signaling (Herrero et al. 2008), and amygdala inputs are associated with emotional valence.

At the microcircuitry level, there is a complex intertwining of neuropil: basal dendrites extend beyond CO defined borders, apical tufts of pyramidal cells extend in layer 1 tangential to the pia surface beyond defined CO borders, undefined numbers of intrinsic axon collaterals arborize according to parameters that are still incompletely defined. Important but less immediately accessible issues are synaptic and subcellular components, such as cytoskeleton, myelin, and mitochondria. As super-resolution light microscopy, light sheet microscopy, and volumetrically dense electron microscopy become increasingly routine, one can look forward to reinvestigations of the CO pattern at this level, where the relationship of structure and function may be more complex, as well as stranger than originally postulated. Extended to finer level microcircuitry investigations, the CO+ CO− pattern may be an effective window not only to the deeper structure and function of visual cortex, but also to phylogenetic and ontogenetic comparisons across NHP, and between NHP and humans.