Connectome (49 page)

[>]
   
render parts:
The thicker object is a short segment of a dendrite, with spines protruding. The thinner objects are parts of axons.

[>]
   
a million person-years:
Helmstaedter, Briggman, and Denk 2008.

[>]
   
machine learning:
How can you help a computer learn? First, devise an algorithm that performs the task, but put a lot of adjustable parameters in it. Depending on the parameter settings, the algorithm performs the task differently. Second, devise a quantitative measure of disagreement between the computer and the humans on the database of examples. This measure is a function of the adjustable parameters in the computer program. It is known as a cost function, or objective function for learning. We would like to minimize this function with respect to the adjustable parameters. To do this, we carry out the third and final step of writing a program that searches for the optimal setting of the parameters. Often this is done in an iterative fashion. The program finds a small change of the parameters that lowers the cost function. It does this repeatedly, in an attempt to find the lowest possible value.

[>]
   
Viren Jain and Srini Turaga:
Jain, Seung, and Turaga 2010.

[>]
   
Old-time computer hackers:
Kelly 1994.

[>]
   
Intelligence Amplification: Engelbart credited the term to Ross Ashby, a pioneer in the field of cybernetics.
Engelbart credited the term to Ross Ashby, a pioneer in the field of cybernetics.

[>]
   
possible to “crowdsource”:
I haven't mentioned that humans also make errors when tracing neurites, though at a lower rate than computers. Helmstaedter, Briggman, and Denk 2011 shows how to combine the efforts of multiple humans to improve accuracy, an example of the “wisdom of crowds.”

[>]
   
cost per letter:
Shendure et al. 2004.

 

10. Carving

 

[>]
   
brain forest:
Cajal may have originated the metaphor, describing the brain as “a jungle, in whose impenetrable thickets many explorers had lost their way” (Ramón y Cajal 1989).

[>]
   
Huntington's disease:
Utter and Basso 2008.

[>]
   
it is very important:
In this book I've been guilty of cortical chauvinism. For the sake of simplicity, I've spoken of localizing mental functions within cortical areas, but this is admittedly naïve. Every other brain region has its partisans, who can explain why the region is so important, even if smaller than the cortex. Fans of the basal ganglia have mapped its connections with the cortex and the thalamus to understand how these regions cooperate to carry out mental functions (Middleton and Strick 2000).

[>]
   
Each strip:
Masland 2001. The figure presents a classification of neurons valid for a generic mammalian retina. Some larger types of neurons are omitted. I've used the terms
class
and
type
to denote two levels of taxonomy, but my usage is by no means standard in neuroscience. To classify plants and animals, biologists use the official terms
species, genus, family, order,
and so on. A similar scheme is needed for neurons.

[>]
   
make out six layers:
According to the standard convention, most of the cortex has six layers, and is called
neocortex
or
isocortex.
“Neo-” refers to the evolutionary theory that six-layered cortex is newest. Those who don't believe this theory prefer the prefix “iso-”, which emphasizes that all six-layered cortex has a similar appearance. Other parts of the cortex have fewer (or more) than six layers and are known as
allocortex.
A famous example is the hippocampus.

[>]
   
layering was uniform:
The arrangement of cell bodies into layers is known as
cytoarchitecture,
as “cyto-” means “cell.”

[>]
   
Oskar and Cécile Vogt:
Zilles and Amunts 2010. Their stain marked a substance called myelin, a fatty material that sheaths many axons. This revealed “myeloarchitecture” rather than the “cytoarchitecture” used by Brodmann.

[>]
   
Sir Grafton Smith:
Smith was an interesting character who straddled the fields of neuroanatomy and archaeology. He investigated and x-rayed the brains of Egyptian mummies.

[>]
   
Percival Bailey and Gerhardt von Bonin:
Bailey and von Bonin used a “double-blind” method to see whether cortical areas could be reliably distinguished by cytoarchitecture, and mostly found negative results (Bailey and von Bonin 1951).

[>]
   
hundreds of neuron types: Stevens 1998.
Stevens 1998.

[>]
   
Neuroscientists continue to argue:
Nelson, Sugino, and Hempel 2006.

[>]
   
A great many wires:
I should qualify my statement. It might make sense to regard the grooves (sulci) in the cortex as its “joints.” Neurons on opposite sides of a groove are connected by longer axons than neurons within the same convolution (gyrus). By the principle of wiring economy, there should be fewer wires connecting opposite sides of a groove, and cutting along the grooves is analogous to carving at joints. This justifies the practice of MRI researchers, who locate cortical areas relative to grooves, because they can't see the layering that Brodmann relied upon.

[>]
   
Unlike poultry:
If we want to preserve Socrates' metaphor, we can think of classification as happening by cuts in a high-dimensional feature space rather than three-dimensional space.

[>]
   
over one hundred types:
White et al. 1986.

[>]
   
Collapse all neurons of one type:
Ibid.

[>]
   
Connections are directly related:
As described by Nelson, Sugino, and Hempel 2006, it's also important to define neuron types by molecular criteria such as the expression of a particular gene or genes. A beautiful example in the retina is provided by Kim et al. 2008. The molecular definition is useful for controlling neuron types and for understanding how they emerge during development. As I mentioned earlier, neurons of one type should also share similar functions, as revealed by measurements of spiking. Therefore, I anticipate three definitions of neuron types based on molecules, connectivity, and activity, which will ideally coincide with each other. These three definitions parallel three meanings of the term
neuron
,
which were delineated by Golgi in his Nobel lecture. He pointed out that the neuron is supposed to be an embryological, anatomical, and functional unit, before he proceeded to question its existence.

[>]
   
Layer 4 of area 17:
The axons reaching layer 4 of area 17 come from neurons in the LGN, which in turn receives axons from the retina. The LGN is a subdivision of the thalamus devoted to vision. As a general rule, sensory pathways reach the neocortex through thalamic axons terminating in layer 4. The text focuses on connections between areas, but differences in layering also reflect connections between neurons in the same cortical area, because of rules of connection that are based on layers. For example, excitatory neurons in layer 4 make synapses on pyramidal neurons in layers 2 and 3, which in turn make synapses on pyramidal neurons in layer 5. Therefore, when the thickness and density of layers change, connectivity is probably changing too.

[>]
   
area 17 has a thicker layer 4:
Furthermore, there is potentially much more information in connectivity than in layering. Brodmann and his contemporaries disagreed over their cortical maps precisely because differences in layering are so subtle. Cortical layers are not very distinct in the first place, as we saw earlier, and variations in them are even less distinct. I predict that differences in connectivity will be much more marked.

[>]
   
regional or neuron type connectome:
You may find it confusing that by now I've defined three kinds of connectome. According to Lederberg and McCray 2001, the term
genome
also has multiple meanings. When first coined in 1920, it referred to the totality of chromosomes in an organism. (Your DNA is divided into twenty-three pairs of molecules known as chromosomes, which are like volumes of an encyclopedia.) Later it came to refer to the totality of genes, and today it means all the letters in the DNA sequence. Similarly, I expect that the most common meaning of
connectome
will shift over time toward the neuronal one, which has the highest resolution.

[>]
   
Wernicke called it:
Eling 1994.

[>]
   
different flavor from the neural:
Catani and ffytche 2005; Mesulam 1998; Geschwind, 1965a, 1965b.

[>]
   
Olaf Sporns and his colleagues:
Sporns, Tononi, and Kotter 2005. Around the same time, Patric Hagmann independently coined the term in his Ph.D. thesis.

[>]
   
lesions that spare:
Mohr 1976.

[>]
   
less localized than previously:
Lieberman 2002; Poeppel and Hickok 2004; Rilling 2008.

[>]
   
deny that the arcuate fasciculus: Bernal and Altman 2010.
Bernal and Altman 2010.

[>]
   
other pathways that do:
Friederici 2009.

[>]
   
the Broca–Wernicke model:
Hickok and Poeppel 2007.

[>]
   
formation of cortical areas:
Fukuchi-Shimogori and Grove 2001.

 

11. Codebreaking

 

[>]
   
Michael Ventris and John Chadwick:
Chadwick 1960 recounts the story of their collaboration. Kahn 1967 tells a shorter version, along with providing other examples of codebreaking throughout history.

[>]
   
a number of lost languages:
Robinson 2002.

[>]
   
stick in your computer:
A similar scenario is the basis of Anthony Doerr's short story “Memory Wall.”

[>]
   
name the president:
Corkin 2002.

[>]
   
MTL seemed essential:
In technical terms, H.M.'s condition is described as severe anterograde amnesia. “Anterograde” means that his amnesia only applied to events after his operation. His memory of events from before his surgery was mainly intact, though it was worse for events just prior to the surgery than from events long before. Therefore he had a mild degree of retrograde amnesia, which was temporally graded.

[>]
   
role in memory recall:
Gelbard-Sagiv et al. 2008.

[>]
   
groups of CA
3
neurons:
This idea is due to David Marr, who first theorized about cell assemblies in CA3. Neurons in other parts of the hippocampus make synapses on neurons in other brain regions, rather than their neighbors.

[>]
   
the human CA
3
:
Furthermore, it's not clear whether memories and cell assemblies are really confined to CA3. This might be the case for new memories, if they are initially stored in the hippocampus and later transferred to the neocortex, as some theorists believe. Alternatively, cell assemblies might be distributed across both the hippocampus and the neocortex from the very beginning. They might start out with more neurons in the hippocampus but end up with more neurons in the neocortex as memories are consolidated.

[>]
   
It includes autobiographical:
These are called “episodic” and “semantic” types of memory, respectively. H.M.'s semantic memory was not as impaired as his episodic memory.

[>]
   
declarative memory in animals:
Declarative memory might seem to depend on language, as the term implies that recall occurs through “declaring.” But Eichenbaum 2000 argues that the term should nevertheless be extended to animals, because they have mnemonic capabilities that correspond to those included in human declarative memory and depend on analogous brain regions. Also, parrots and other animals might be able to “declare” memories through vocalization or other communication skills.

[>]
   
don't nurse their young:
You might think that egg-laying distinguishes birds from mammals, but a few mammalian species like the platypus also lay eggs.

[>]
   
Males of other species also sing:
Not all sounds from birds are considered song. Less complex sounds are known as “calls.”

[>]
   
Mozart kept a pet starling:
West and King 1990.

[>]
   
starts to “babble”:
Doupe and Kuhl 1999.

[>]
   
learns to copy his father's:
If young zebra finches do not hear an adult male's song, they will still sing when they grow up, but abnormally. However, Fehér et al. 2009 showed that if such isolated birds are bred for several generations, so that each generation learns from the previous one, the song eventually sounds more normal again. This suggests that there is some innate preference for certain song properties, in addition to the preference learned from experience.

[>]
   
muscles around the syrinx:
Also involved are the muscles for respiration, which control the rate of air flow through the syrinx.

[>]
   
converted into sounds:
It's admittedly simplistic to say that RA and nXII merely relay or amplify signals. For a more accurate account, you can consult the scientific literature. Also, you might question whether a straight pathway is a good model. Since the bird hears its own song, maybe there should be an additional step from the syrinx back into the brain, which would turn the pathway into a circular loop. In this view, each note of the song would serve as a stimulus that drives the bird to produce the next note. Such a loop was proposed as a model for sequence generation in the nineteenth century, by people like the American psychologist William James. It does not appear to be a good model for birdsong, because adult zebra finches can still sing even if they are deaf.