Mapping and manipulating connections in small animals
2016-11-05T19:50:47+05:30
“THE most obvious differences between different animals are differences of size, but for some reason the zoologists have paid singularly little attention to them. In a large textbook of zoology before me I find no indication that the eagle is larger than the sparrow, or the hippopotamus bigger than the hare, though some grudging admissions are made in the case of the mouse and the whale. But yet it is easy to show that a hare could not be as large as a hippopotamus, or a while as small as a herring. For every type of animal there is a most convenient size, and a large change in size inevitably carries with it a change of form.” (J.B.S. Haldane)
State of play
After Ptolemy there was a dark age for world map making, partly because of the immense effort such an enterprise entails. Similarly, after the 14-year C. elegans feat (White et al. 1986), the field went quiet. It has now returned, this time with the aid of modern computing. At the time of writing, only two species’ full, published connectomes exist. The C. elegans’ connectome of the adult hermaphrodite has 302 (adult male 382) neurons, 6,393 chemical synapses and 890 gap junctions ultimately controlling 95 muscles at 1,410 neuromuscular junctions (Cook et al. 2019; Jarrell et al. 2012; Varshney et al. 2011; White et al. 1986). The connectome of the tadpole larva of the ascidian Ciona intestinalis (Ryan et al. 2016), the closest invertebrate relatives of vertebrates (Delsuc et al. 2006), weighs in at 177 neurons. Additionally, multiple D. melanogaster connectomes are on the horizon, ranging from ~15,000-150,000 neurons (Ohyama et al. 2015; Scheffer and Meinertzhagen 2019).
Electron microscopy can reveal all connections
Electron microscopy (EM) is the only means by which to resolve fine neurites (<200 nm, as small as 15 nm (Ian A. Meinertzhagen 2016)), synaptic vesicles (~40 nm) and synaptic clefts (~20 nm) in densely stained tissue, and so see all synaptic connections between neurons in a sample. It provides data that can be mined for both a parts-list and an adjacency matrix of connections between those parts. Other techniques, such as expansion microscopy (Wassie, Zhao, and Boyden 2019), super resolution microscopy (Igarashi et al. 2018) and molecular barcode sequencing (Kebschull and Zador 2018) can provide sparse information on synaptic connectivity or map single neuron’s projections at scale. While they can provide orthogonal useful information, e.g. on gene expression, they lack the morphological and synaptic connectivity ‘completeness’ EM can provide. There is promise that X-ray tomography may provide another avenue to dense circuit reconstruction and allow researchers to use larger tissue volumes than can typically be easily handled with EM pipelines (Kuan et al. 2020).
However, due to the high resolution required to resolve fine neurites and synapses, the EM connectomics produces terabyte-scale datasets for even tissue samples <1 mm3. These datasets are small relative to the size of the nervous systems that neurobiologists typically study, because it is difficult, time-consuming and expensive to section, image, store and extract neuroanatomical data for larger pieces. The field is therefore heavily size restricted. Typically these datasets are only in the millions of cubic microns, and frequently only either a smaller sub-volume has been densely reconstructed (Helmstaedter et al. 2013), or small numbers of neurons have been sparsely reconstructed in a larger volume (Dolan et al. 2019). The image data itself can be acquired by various modern automated methodologies.
Here’s a little summary of modern EM methods and the datasets that have been acquired using them:
What did the C. elegans connectome do for its neuroscience
Some commentators have criticised the modern shift in neuroscientific research back towards descriptive endeavours such as connectomics (Krakauer et al. 2017), however, the reality is that the C. elegans connectome was both immediately useful and continues to be useful today. To consider what a from-connectome approach looks like we can examine the case of the escape response in C. elegans (Pirri and Alkema 2012). Early investigators used the connectome to guide laser microsurgery (Chalfie et al. 1985) and genetic perturbations (Y. Zheng et al. 1999), dissociating two motor pathways. Critically they were able to evaluate the interneurons within them, finding four command-like interneurons. The authors could hypothesise which neurons were involved in the forward or backward crawl in response to the activation of touch-sensors, which was later further validated using optogenetics (Leifer et al. 2011) and calcium imaging (Chronis, Zimmer, and Bargmann 2007). They could also predict and test that backward crawl suppressed head-casting movements via a specific set of connections (Pirri et al. 2009). Without the connectome as a guide, investigators would have been less able to make the precise interventions necessary to characterise and implicate these ~13/118 cell types in this circuit.
Once the circuit was established, investigators were able to ask after its behavioural significance. It is known that C. elegans’ escape response is required to survive ‘ring’ traps laid by predacious fungi (Pirri and Alkema 2012). Investigators found that touch-sensitive reversal proves key to escaping ring traps (Maguire et al. 2011). Moreover, reversal with head-casting was more likely to trigger these traps than a smooth exit; these mutants with this phenotype were easily outcompeted by the wild-type. By contrast, insect-associating Pleiorhabditis nematodes do not exhibit this suppression, which might be the result of different selection pressures (Pirri and Alkema 2012). An interesting next question will be whether, and if so in what way, do these species’ escape circuits differ; cell type specific synaptogenesis, receptor expression profiles and neurotransmitter use are all likely, evolutionarily labile substrates.
Why flies are now the frontier
Big Ben and your wristwatch are fundamentally similar in their operation and while there are important differences that allow the former to exist at a grander scale, a careful examination of the latter still reveals how timekeeping can be accomplished. In terms of components and complexity, it is not yet clear that the mouse has vastly more neuronal cell types than the fly. A compressed fly connectome could be a graph of ~10-30k cell types, or even more. We do know that mammalian cell types are arranged very differently, that they have many more neurons per type, and they are bigger and probably electrically different from flies’. So it feels like a case of building the same device from different but analogous infrastructure and design principles.
But why not then persist with the C. elegans connectome (e.g. for my PhD for example), where there exist significant foundations but not yet a convincing understanding of how different dynamic states in its nervous system result in behaviour? The fly is clearly both complex and efficient. It possesses a highly optimised, innately structured brain for the tasks it is likely to face in its ~32-day life. Similar brains have spread from sub-Saharan Africa to the Arctic, diversifying into ~80,000 species that comprise ~8.5% of the Earth’s ten quintillion insects (McAlister 2017). They are a testament to how evolutionary elaboration on grossly similar neuroanatomy can help an order master a large range of environments.
It is also commonly thought that the behavioural repertoire of the roundworm, and its capacity for more ‘cognitive’ tasks such as learning, is much lesser than the insect (Olsen and Wilson 2008). Unlike roundworms, flies can recognise familiar but displaced objects (Liu et al. 2006; Tang et al. 2004), recall where a displaced object once was (Neuser et al. 2008), respond to small odour concentration changes dependent on the greater sensory context (Faucher et al. 2006) and co-compute olfactory and visual information to direct behaviour on foot (Frye and Dickinson 2004; Guo and Guo 2005) or in-flight (Duistermars and Frye 2008). They can also learn to associate certain stimuli with outcomes, even learning to adapt mating strategies in the face of rejection (Ejima et al. 2007). They can discriminate between pheromones and mating songs of similar Drosophilids (Ritchie, Halsey, and Gleason 1999) and fight over resources (Yurkovic et al. 2006).
One of the largest pieces of nervous tissue collected and imaged by EM is a full female adult fly brain (Z. Zheng et al. 2018) known as FAFB. It is 40 teravoxels, 21 million serial section transmission EM camera images, within a 995 x 537 x 283 μm volume at a 4 x 4 x 40 nanometer resolution, and takes up 106 terabytes on disk. Sparse connectomic approaches in this dataset are helping to verify and build a range of behaviorally relevant hypotheses (Z. Zheng et al. 2018; Dolan et al. 2018; Felsenberg et al. 2018; Frechter et al. 2019; Huoviala et al. 2018; Sayin et al. 2018; Bates et al. 2020).
EM methods will soon generate a synaptic-resolution ‘parts lists’ for a small business of flies (perhaps 1-2 larval, 2-3 adult brains and ~2 VNSs in < 5 years), which can be compared to extant, annotated light-level morphologies (Schlegel, Costa, and Jefferis 2017) and leveraged to improve our understanding of circuit computations (e.g. (Takemura et al. 2013)). EM morphologies can also be matched to those revealed by genetic labelling (Davis et al. 2018). Genetic driver lines, especially the intersectional split-GAL4 system, give stable genetic access to small constellations of neuronal types, each of a shared morphology (Dionne et al. 2018; Luan et al. 2006; Pfeiffer et al. 2010). In the best cases, they can selectively label just one morphological neuronal type. I therefore expect strong similarities between subdividing neurons by morphology and by gene expression.
Strong stereotypy in insect nervous systems mean that cell types - which exist in only about 3-5 copies of cell per hemisphere per brain, can be found across different animals in a reproducible manner. Therefore, you can conduct an experiment in, say, many live flies and if you know which cell type you manipulated/recorded from, you can link it to a static connectome from a single, separate dead fly. Like so:
Cell type specific genetic access means that investigators can knock out or down gene expression, gain optogenetic, chemogenetic and thermogenetic temporal control, try to map connections using genetic tricks [Datta et al. (2008); Huang2017-ct; Talay2017-uh], label single neurons (Nern, Pfeiffer, and Rubin 2015), perform functional imaging with genetically encoded fluorescent calcium sensors (Griesbeck 2004; Miesenböck and Kevrekidis 2005) or voltage sensors [Abdelfattah et al. (2018); Grimm2017-yj] and repeatedly target the same cells with whole-cell patch-clamp (e.g. (Frechter et al. 2019)). Behaviour may be assessed during the manipulation of individual cell types; modern cameras and machine learning pipelines extract a wealth of information about a freely moving animal’s responses to presented stimuli in minute detail, in a high-throughput manner (e.g. (Branson et al. 2009)).
However, one of the great limitations of the fly is also its size. Insect neurites can be thinner than 50 nanometers (I. A. Meinertzhagen and O’Neil 1991), which is about the diameter of a synaptic vesicle. This strains the segmentation of these neurites in ssTEM data, which has a typical Z-resolution of 40-50 nanometers. Section loss through machine and human error can make the z resolution dramatically worse for some neuroanatomical loci (Ohyama et al. 2015). If we consider the input resistance to a neurite of this diameter, it is easy to imagine that the activation of even a single ion channel could result in a membrane voltage fluctuation that rivals that of an action potential. Our understanding of the biophysics of insect neurons is poor, and as the field progresses this will become an increasing burden.
Concluding thoughts
While connectomes do not provide mechanistic answers to neurobiological questions alone, the connectome is a neuroinformatics scaffold (Swanson and Lichtman 2016) onto which other information can be pinned, such as transcriptomes, cell type annotations, response properties of neurons, etc. However, today’s connectomes are single snapshots of single animals. Even if they are representative of a species, we know that synaptic plasticity is rampant across all studied nervous systems. What is important is whether we can extract connectivity motifs that help explain the potency of innate structure in nervous systems.