Cherniak (1991) has explored the optimal wire saving morphologies for neurons. Bipolar neurons, for example, should obviously be linear while tripolar neurons should be planar with processes separated by 120 degrees. Diagrams of efficient bipolar and tripolar neurons are shown below:
It is suprising, therefore, that in C. elegans, the species that Cherniak (1995) uses to exemplify efficient wiring, and in which "optimization of placement is so sensitive that it fine-tunes even the positioning of individual somata" (Cherniak 1995 p 525), many neurons' morphology is strikingly unlike those of the "wire saving" examples above. In fact, over 30% of the bipolar neurons drawn by White et al. (1986) differ very markedly from the wire saving ideal. In many of these cases, both of the neurons' projections pass in the same direction from the cell body. The single indentified neurons redrawn below from White et al (1986) are clearly not 'micro-optimized' for saving wire. They illustrate the fact that other constraints determine the organization of nervous systems, such as, perhaps, the constraint that a nervous system develops using processes that are simple to implement biologically.
Non-wire-saving neurons in C.elegans
1) The Recurrent Laryngeal Nerve and the Ansa Subclavia
The fibres that become the recurrent laryngeal nerve travel down through the neck with the vagus nerve. They remain with the vagus as they pass by their final target (the larynx) and make a long detour into the thorax. On the right hand side, the recurrent laryngeal nerve splits from the vagus in front of the subclavian artery, winds below and behind the vessel, and then ascends back into the neck to the side of the trachea to enter the larynx. On the left side, the nerve arises from the vagus on the left of the arch of the aorta, and winds below the arch before ascending along the trachea to the larynx (Williams and Warwick 1975).
Another nerve that makes a strikingly wire wasting detour around a major blood vessel is The Ansa Subclavia which links the middle and upper cervical ganglia via a loop down and around the right subclavian artery. These two examples are plainly contrary to the "Save Wire" directive, and illustrate how evolutionary and developmental factors can so greatly outweigh any wire length minimization directive as to make it invisible.
2) Thalamo-cortical connectivity
Each side of the thalamus is composed of 40 or so nuclei, which lie together near the center of each cerebral hemisphere. The nuclei have prolific connections with specific groups of cortical areas. In the cat visual system, for example, the lateral geniculate nucleus sends around 1 million fibres to, and receives around 10 million fibres from, the primary visual cortex (REF). Higher visual nuclei such as the the pulvinar and lateral posterior complex also connect extensively with the visual cortical areas. Analogous patterns occur in the somato-motor and auditory systems.
We have collated data on thalamo-cortical connectivity in the cat. Within each cerebral hemisphere, there are about 800 connections (each containing between 100,000 and 10,000,000 fibres) linking the 40 thalamic nuclei with the 60 cortical areas, which in turn are linked by over 1000 area to area connections. We have investigated the organization of this complex network using non-metric multidimensional scaling (NMDS); a procedure which produces a spatial configuration that optimally reflects the connectional proximities of a set of structures (e.g. Young 1992; Scannell and Young 1993; Young et al 1995).
If CPO determined the layout of the thalamo-cortical system (and indeed any system), the configuration produced by NMDS in 3 dimensions should resemble the anatomical configuration, since the positions of the anatomical structures would be 'micro-optimized' to bring connected structures as close as possible together. However, despite being anatomical neighbours, the thalamic nuclei do not connect with each other. NMDS analysis of the network yields configurations in which the thalamus 'explodes' (Scannell 1995). The non-interconnected thalamic nuclei lie as a 'shell' around the interconnnected cortical areas. According to CPO, therefore, the mammalian thalamo-cortical system is inside-out! We suppose that development and evolution constrain the thalamic nuclei to lie together. In any case it is clear that wiring cost is not the major factor in the organization of this system.
3) Exuberant connections, pruning and cell death
CPO claims that 'Save Wire' is the overiding directive. It is a puzzle, then, that mammalian brain development is characterised by an initial exuberance of neurons and connections, most of which are subsequently lost by cell death and 'pruning'. The adult human brain has been estimated to possess as little as 30% of the number of neurons found earlier in development (Rabinowicz et al 1996). The number of connections between neurons is also massively reduced in the adult compared to the neonate (Innocenti 1995; Payne et al. 1988). In the cat, for example, there is a huge decrease in the number of callosal axons during neonatal life, and a 90% reduction in the number of synapses and branches of the axonal arbours of the remaining fibres (Innocenti 1995; Payne et al. 1988). The disposable nature of so much neural machinery shows that building neurons, axons, arbours and synapses - neural wiring - is not the critical developmental cost. Indeed, neural wiring and machinery are so cheap biologically that the majority made during development can be thrown away. Again, it seems that the difficulty of wiring up a brain using a small number of mechanisms is much the more critical factor.
The neonatal brain is smaller but has more neurons and connections than the adult brain. In the adult, therefore, wiring volume is very unlikely to be anywhere near the biological maximum. In principle, the adult brain could support many more neurons, or else it could save wire by being much smaller. One must suppose that adult brains have sparser neuronal machinery for some reason other than CPO: perhaps a trade-off between conduction velocity, neuron diameter and brain volume, to have more reliable neurons, or else to allow plasticity during development (Innocenti 1995).
4) The mammalian retina
The retina, notoriously, is back to front. All the axons have to go the wrong, wiring-costly, way in order to pass into the optic nerve. This is in clear contradiction of CPO!