The world of the identified or digital neuron

Abstract

In general, neurons in insects and many other invertebrate groups are individually recognizable, enabling us to assign an index number to specific neurons in a manner which is rarely possible in a vertebrate brain. This endows many studies on insect nervous systems with the opportunity to document neurons with great precision, so that in favourable cases we can return to the same neuron or neuron type repeatedly so as to recognize many separate morphological classes. The visual system of the fly’s compound eye particularly provides clear examples of the accuracy of neuron wiring, allowing numerical comparisons between representatives of the same cell type, and estimates of the accuracy of their wiring.

Keywords:

• Drosophila
• Ciona intestinalis
• photoreceptor
• synapse

Prelude

The first paper I read as a graduate student in Scotland was one entitled “A further study of the paradox phenomenon of Crustacean muscle” by Harold Atwood and the late Graham Hoyle, who had then recently moved to the University of Oregon. It introduced me to the study of nervous systems with identified neurons, and in it, the paradox, a mismatch between the contraction of an entire crustacean muscle, and the activity of the individual fast and slow motoneurons innervating it, was shown to incorporate a misidentification of all the fibres constituting that muscle, and the failure to identify a small class of additional fibres that produced muscle tension but were not recorded from electrophysiologically. Parenthetically, I might add that I am probably the only Canadian to have read this paper by Harold long before I read anything written by his younger sister. At any rate, the paper led me to a number of false starts and cul-de-sacs in neurobiology, to more fertile ground in work on insect brains, and in particular their wiring accuracy (Horridge & Meinertzhagen, 1970), a theme to which I returned much later at a synaptic level (Takemura et al., 2015). The theme of accuracy in nervous systems will in fact serve as a unifying theme for this short chapter and appreciation, and while I cannot claim that my own switch in 1968 from crustacean muscle to work on flies was any more than coincidental, it did in fact anticipate the same switch also made by Harold in starting to work on flies, after a sabbatical spent with this Journal’s Editor, Chun-Fang Wu (Atwood, Govind, & Wu, 1993).

Introduction

In an important dichotomy Seymour Benzer pointed out long ago, in a typical one-liner, that the component neurons of insect brains are constituted on a digital basis, by comparison with those of vertebrate brains, which have analogue characteristics. As with all generalizations there are of course many exceptions. Even so, neurons in insects, as well as other arthropods and invertebrates (Bullock & Horridge, 1965), are in many cases individually recognizable so that we can assign an index number or descriptor to each specific neuron, in a manner only rarely possible (Bullock, 1978) in a vertebrate brain. Why is the ability to distinguish many different types of insect neuron, each anatomically distinct, so satisfying? Not just for reasons of the concussive cellular book-keeping it invites. Instead this principle enables precision in the documentation of neurons, such that in favourable cases we can return to the same neuron repeatedly, at a level of thoroughness, precision and with an analytical depth which is generally not possible in the brains of any vertebrate. In the case of the fruit fly Drosophila melanogaster this exclusivity has demonstrated genetic basis, and each cell type, of which 16,000 single types of neurons have been genetically identified in the brain (Chiang et al., 2011), can be identified using the GAL4/UAS system (Brand & Perrimon, 1993) and intersectional variants such as the split Gal4 system (Luan, Peabody, Vinson, & White, 2006), as reviewed (del Valle Rodríguez, Didiano, & Desplan, 2011; Nern, Pfeiffer, & Rubin, 2015), or a split GAL4 repressor system (Dolan et al., 2017), or stochastic recombinase-mediated MultiColor FlpOut labelling (Nern et al., 2015), all of which can be used to target expression of a reporter gene, such as that for green fluorescent protein (GFP). Alternatively flip-out single-cell mosaic methods to generate single-cell clones of labeled neurons (Lee & Luo, 1999; Wong, Wang, & Axel, 2002) collectively provide means to confirm the many cell shapes previously seen by Golgi impregnation and other earlier methods. These may not dwarf the planet’s 60,000 known species of trees (Beech, Rivers, Oldfield, & Smith, 2017), but are clearly not trivial numbers either.

These genetic possibilities in Drosophila underpin the concept of identified neurons that has been promulgated extensively in many previous reports on invertebrate brains, starting from long ago (Bullock, 2000; Comer & Robertson, 2001; Hoyle, 1977; Kandel, 1976; Wiersma, 1967) and implicit in all work on arthropod brains since, and that has enabled the discovery or refinement of many key concepts in neurobiology (Clarac & Pearlstein, 2007). As a result, we know many neurons in insect brains in enormous and reproducible detail, especially those in Drosophila in which many instrumental genetic manipulations are additionally possible.

Cell types and identified neurons

The development of rigorous ways to classify neurons into biologically meaningful types has been pioneered by studies on visual systems, but in vertebrates (Masland, 2004; Seung & Sümbül, 2014; Sanes & Masland, 2015). Various bases for such categorization include the spatial distribution of the neuron’s soma and its dendrites (e.g. Masland, 2001, 2004; Cook, 2003), cluster analyses of functional and morphological parameters (Badea & Nathans, 2004; Kong, Fish, Rockhill, & Masland, 2005; Sümbül, Zlateski, Vishwanathan, Masland, & Seung, 2014, and others) and the molecular signatures of their genome (e.g. Hobert, Carrera, & Stefanakis, 2010) or transcriptome (e.g. Henry, Davis, Picard, & Eddy, 2012; Shekhar, Lapan, Whitney, Tran, Macosko, & Kowalczyk, 2016). Indeed, the recognition of cells and cell classes may eventually be based on differential expression profiling data, with the prospect that these will reveal the evolutionary diversification of the different cell types, and possibly the recognition of the same cell type in different species (Arendt, 2008). Collectively these approaches provide comprehensive criteria for the recognition of cell types, but on the other hand these are also born of the necessity of doing so in vertebrate brains, which lack the cellular stereotypy of insects.

Given their discreteness in invertebrates of many types, on the other hand, neuron classes fall out far more clearly, as we see perhaps most clearly in the fly’s visual system (below), which shares deep organizational similarities to cells and strata of the vertebrate retina (Sanes & Zipursky, 2010) but is built with almost crystalline precision (Ready et al., 1976). Given our own native ability in everyday life to recognize visually the branching patterns of trees of all species and types, the most powerful character is generally the three-dimensional structure of neuron arbours.

A cell type can be considered a population of cells with similar molecular, anatomical, and physiological properties (Seung & Sümbül, 2014) and, for many invertebrate neurons the population comprises one, or more usually one of a pair of neurons, each unique (White, Southgate, Thomson, & Brenner, 1986; Ryan et al., 2016, Ryan, Lu, & Meinertzhagen, 2018). For most, however, the definition of its type is solely morphological. Insofar as the dendritic arbour ensures the pattern of the cell’s input and output synapses, we may assume that a neuron’s shape is also an influential reporter for its physiology and especially its connectivity, merely one that is the easiest to record and classify.

Morphological cell types

Morphology presents us with an exquisite phenotype, one that is visually intuitive. For those who study them in any detail, especially with respect to their morphology, neurons become a little like people, each an individual with identifying structural features. Thus the transmedulla visual interneuron Tm2 is a major input neuron to the dendrites of T5 cells (Shinomiya et al., 2014) as part of the (ON-channel) pathway sensing luminance increments in motion detection, but known more informally as the Johnny Walker cell from the shape of its two descending dendrites (Figure 1) (Meinertzhagen et al., 2009).

Figure 1. Comparison between the neuron shape for transmedulla cell Tm2 reported from Golgi impregnation (Fischbach & Dittrich, 1989), and its counterpart from a genetic reporter, Tm2- Gal4;UAS-GFP (right), showing the twin descending “walking” legs, reproduced from Meinertzhagen et al. (2009) in the Journal of Neurogenetics.

The range of different morphological classes of neurons is enormous, even in invertebrate species with relatively few neurons. In particular, the nematode C. elegans (White et al., 1986); the fruit fly Drosophila (Fischbach & Dittrich, 1989; Takemura et al., 2013, 2017a; Wolff, Iyer, & Rubin, 2015); many other fly species (e.g. Strausfeld, 1970, 1976; 2012); the larval marine polychaete Platynereis (Randel, Asadulina, Bezares-Calderón, Verasztó, Williams, Conzelmann, Shahidi & Jékely, 2014, 2015); and the tadpole larva of the ascidian Ciona intestinalis (Ryan, Lu, & Meinertzhagen, 2016); all provide a richly populated range of comprehensively documented neuron classes.

The different morphological classes of a nervous system were of course an early focus of studies that pioneered the method of Golgi impregnation and supported the collection of illustrated libraries of cell types. Because of its repeating modular composition as parallel pathways, many studies have been undertaken in the visual system, based early on especially by Cajal (Cajal & Sánchez, 1915) and later by Strausfeld (e.g. Strausfeld, 1971), and finally in Drosophila (Fischbach & Dittrich, 1989; Hanesch, Fischbach, & Heisenberg, 1989) when this species attained eminence as a genetic model in neurobiology. Golgi-labelled profiles (Fischbach & Dittrich, 1989) have since served as an important basis for recent EM studies in the visual system of Drosophila (Takemura et al., 2013, 2015, 2017a). With recent advances in connectomic approaches in Drosophila, work on the adult visual system has again leapt into prominence (Meinertzhagen, 2018).

Reports derived from Golgi impregnation and other selective methods are all partial, however, suffering from the failure to saturate an entire cell population and thus to reveal all examples. Later studies, for example from genetic reporters in Drosophila, e.g. for the antennal lobe (Tanaka, Endo, & Ito, 2012), gustatory centre (Miyazaki & Ito, 2010), or central complex (Wolff et al., 2015), to choose three stations at random, are no less susceptible to this uncertainty but offer additional advantages, especially for identifying wide-field neurons and in establishing a common database and nomenclature (Shinomiya, Matsuda, Oishi, Otsuna, & Ito, 2011).

Saturation of the sample population is less of a problem for studies based on dense reconstruction using various electron microscopy (EM) methods, in which no cell can hide in the forest of others. Initial approaches involving the analysis of serial ultrathin sections (ssEM) (Takemura, Lu, & Meinertzhagen, 2008, Takemura et al., 2013) are now supplanted in Drosophila by the focussed ion beam milling and scanning electron microscopy (FIB-SEM) imaging method (Knott, Rosset, & Cantoni, 2011; Xu et al., 2017). The latter is well suited to the analysis of tiny brain volumes and yields image stacks with equal resolution in all three axes, x, y and z, thus avoiding the poor z-axis resolution inherent in ssEM (Meinertzhagen, 2016). But for all they are comprehensive, existing EM imaging and reconstruction technologies can reconstruct only tiny volumes which, although they may be comprehensive, fail to capture fully especially those cells with wide arbours that extend out of the image stack. An alternative, to examine species with miniaturized brains, such as certain parasitic Hymenoptera, is ideal for FIB-SEM, but attained at least partly by a reduction in neurite calibre as well as by neuron loss (e.g. Fischer et al., 2018), and thus requires increased voxel resolution that offsets the imaging advantages of a smaller brain volume.

An additional imaging limitation to early methods using conventional light microscopy is that they report profiles only from a single view. This problem is now offset by modern image capture using light microscopy, but resolution in the z axis is still typically poorer than in x,y.

Numerical dimensions of cell classes

The range of cell types in the insect visual system has been reported accurately (Fischbach & Dittrich, 1989), with major omissions primarily of wide-field neurons having neurites that spread out of the section or field of view. Even though the problem of saturation in Drosophila (above) is less for libraries compiled from genetic reporters, or based on dense reconstruction using FIB-SEM methods, it is entirely possible that some fly neurons will escape attention, and especially those in busy thoroughfares such as the optic lobes, central complex (Wolff et al., 2015) or mushroom body (Takemura et al., 2017b), are destined to suffer the fate of all minorities everywhere and be forever unrecognized and unidentified. Some brain regions, such as that underlying the eye’s dorsal rim region (Weir et al., 2016), may be of sufficient specific behavioural importance to warrant the special search for limited classes of medulla cell. Other brain regions may never be deemed worthy of such study.

The number of classes is still considerable even for these small nervous systems. Looking overall, 118 classes are reported among the 302 neurons in C. elegans (White et al., 1986); at least 25 different classes and 52 subtypes occur among 177 neurons reported in a tadpole larva of Ciona intestinalis (Ryan et al., 2016); with the 84 different types (Takemura, personal communication) now identified from FIB-SEM among the estimated 27,000 cells of the medulla cortex (Meinertzhagen & Sorra, 2001) distributed among about 750 columns, the number corresponding to the overlying array of ommatidia (Ready et al., 1976). There are innumerable qualifications and disqualifications to append to these numbers, which offer no easy comparison, neither to each other nor to the brains of other species, but they suggest that on average there may be as few as two to four cells for each morphological class. A major difference is of course the morphological complexity of neurons in different nervous systems. Those of Drosophila, in particular, are exquisitely branched, their dendritic arbour providng an anatomical fingerprint that rivals the complexity of dendritic arbours seen in vertebrate neurons except in having a soma that is separated from its arbour, and providing us with many characters. Those of C. elegans and Ciona are inclined to be simple, tubular and unbranched with, in Ciona, at most a single dendrite (Ryan et al., 2016).

Genetic and morphological relationships between cell types suggest that some neurons at least, such as motion sensing T4 and T5 cells, may descend from a common cell type, and thus be evolutionary siblings (Shinomiya et al., 2015). This interesting possibility is likely to prevail even more extensively in other cell types, and we can expect this to be shown as further evidence appears, in particular – as Arendt (2008) predicts – from differential expression profiling data, with the prospect that these will reveal the evolutionary diversification of cell types.

The fly’s visual system

In insect visual systems the cell classes are more intuitive than in vertebrates, and cell libraries help to generate clearer classes than is true for example in the vertebrate retina. Among the clearest classes of neurons reported in the fly visual system, for example, are the lobula plate tangential cells (LPTCs) of the fly’s third visual neuropil, the lobula plate, that signal optic flow (Hausen, 1982; Hengstenberg, Hausen, & Hengstenberg, 1982). The dendrites of these integrate local motion information over an array of columnar elements and each cell has a planar array of these which is closely similar across individual cells, and relatively invariant in their anatomy (Cuntz, Forstner, Haag, & Borst, 2008). Emphasizing the biological basis of their structural similarities, such cell clusters survive scaling so as to include fly species of different sizes (Cuntz et al., 2013), and thus extend to accommodate phyletic diversity, probably because these cells are homologous between species and were inherited from an ancient common ancestor. Revealing the commonality of the representative isomorphic forms of another neuron class, the cell T1, a similar case has also been presented for representative cells of the crayfish and Drosophila (Fischbach & Dittrich, 1989).

Photoreceptor neurons of the fly’s compound eye and their interneurons provide some of the best-studied identified neurons. As is widely known, the photoreceptors are one of six classes, R1-R6, that innervate the first optic neuropile, or lamina, each ommatidial bundle diverging to innervate a small group of six underlying lamina columns, one per column (Braitenberg, 1967), and they provide model examples of the accuracy with which neurons can target their physiologically appropriate postsynaptic partner neurons (Horridge & Meinertzhagen, 1970; Schwabe, Neuert, & Clandinin, 2013). In regions away from the peripheral margins of the compound eye and from the equator between its dorsal and ventral halves, for example, Schwabe et al. (2013) report only nine targeting errors in nearly 4,000 lamina columns, corresponding to an average error rate of only 1 in 2,630 photoreceptor terminals, thus with 99.7% conforming to the predicted functional pattern (Schwabe et al., 2013). Somewhat higher rates of targeting errors were reported long ago in the blowfly Calliphora, when these were associated with a pattern dislocation in the equator (Horridge & Meinertzhagen, 1970).

Each photoreceptor of the six is a member of a single cell class and has a representative from each of the eye’s approximately 750 ommatidia (Ready et al., 1976). Along with other classes of Drosophila sensilla, photoreceptors are probably among the largest classes of identifiable neuron in the fly’s brain, relative for example to at least 1100, the approximate number of olfactory receptor neurons (Vosshall & Stocker, 2007). Such wiring precision is hardly likely to be the sole preserve of the photoreceptor neurons, however, but should also exist in their downstream target neurons, given that photoreceptor inputs are characterized by the precise spatiotemporal signals they provide to the rest of the visual system. Indeed, interneurons in the medulla to which the fly’s photoreceptors provide indirect input, establish synaptic connections with a comparable and corresponding precision. Thus, synapses of the columnar neurons, those that occupy all columns, one in each, vary little in number so that all synapses have a > 99% chance of belonging to a consistent connection. The few connections not reproduced in seven neighbouring columns are considered inconsistent connections which, it has been concluded (Takemura et al., 2015), are wiring errors. It is not clear whether these persist or are merely transient morphogenetic errors. An additional class of synaptic singularity are autapses, in which the same medulla neuron is presynaptic upon itself. This occurs at a relatively elevated rate (≥∼2%) but in only two medulla cells (Mi1 and Dm9), at least 20-fold more frequently in these than for other cell types (Takemura et al., 2015). A simple interpretation would be that the reduced rate of autapses for all other cell types reflects the baseline error in the fly’s ability to programme synaptogenesis, whereas in Mi1 and Dm9, the increased rate is functionally significant. But other interpretations are possible.

No doubt other features of their wiring precision will be revealed by these and the other cell types of the medulla cortex. Collectively the medulla cells number at least 80 different types reconstructed from FIB-SEM, each morphologically discriminable, including 12 lamina input cell types (Dr. S-Y. Takemura pers. comm.). In confirmation, most have a morphologically corresponding split-GAL4 driver line (Luan et al., 2006) as part of a larger collection expressed in all medulla cells (Dr. A. Nern, pers. comm.).

The foregoing examples serve to illustrate the digital precision of at least some neuron populations in the fly’s visual system, and the insights that these have provided, especially with respect to the accuracy of connections between identified neurons. We have no easy way to know how widely their exemplary precision is matched by other neurons and neuron populations, neither in the fly nor in other model species, but we may predict that they may provide a precedent for the wiring precision of any and perhaps all nervous systems. Uncovering the extent of that precision has illustrated the value of invertebrate systems and the precision of the digital neurons these harbour.

Disclosure statement

No potential conflict of interest was reported by the author.