24 The Tangled Web
A thousand streamers flaunted fair; Various in shape, device, and hue.
—Sir Walter Scott (1771–1832)1
The quotation that introduces this chapter is from Marmion, an epic poem by Sir Walter Scott about the Battle of Flodden in 1513, and it describes the army raised by King James IV of Scotland from the clans of Scotland. The poem includes the line: “O, what a tangled web we weave…”
I began this book with a reflection that although I, like you, am a human being, a Homo sapiens; we are not desiccated calculating machines,2 but creatures of passion. We share the same genes, with a few minor differences, but those genes did not solely determine who we are now, nor did the differences solely determine the ways in which we differ. The environment into which we were born, our early life experiences and our interactions with close kin, made a big and lasting difference. I asked: how much of our behavior is really governed by reason? How often are the reasons that we give merely self-serving narratives, justifying behaviors governed by things of which we are unaware or only dimly aware or which we prefer not to acknowledge?
I sketched out the “classical” view of the brain—which presents it as a computational structure whose power resides in its scale, in its complexity, in the ability of neurons to compute rapidly, and in the ability of neuronal networks to modify their connections in the light of experience—and asked, where in this edifice are our passions?
Behaviors important to who we are—love and hate, how much we eat and what we eat, how we respond to threat and to stress—are governed by the hypothalamus, but not by the map of how the neurons are connected, rather by where the receptors for peptide signals are found. The neurons comprise many subpopulations—clans—each with its characteristic phenotype, dictated by the “tartan” of genes that it wears. Clans talk to clans in many different ways with many types of signals on different spatial and temporal scales; they use not one language but many. I argued that these clans have their evolutionary origins in the neurons of “simple” organisms, which combined properties that we have thought of as separate properties of endocrine cells and neurons. They used a diversity of signaling mechanisms, made both peptides and neurotransmitters, and were endowed with a wide range of specialized senses. They had not a single role to which they were committed, but multiple behavioral and physiological functions. As these ancestor neurons proliferated in descendant species, populations differentiated not primarily by gaining properties but by losing properties. Paradoxically, we have become more intelligent as our neurons became less complex. Nevertheless, the neurons of the hypothalamus retain the multifunctionality of their distant ancestors, along with their multitude of sensory abilities.
The classical view of the brain is hierarchical—the “higher centers” integrate, refine, and evaluate information received from more primitive regions. It is a view that ascribes autonomy to neurons, ascribes salience to their electrical activity, and posits that the information carried by each neuron can be understood in isolation from that of all other neurons. Sometimes, this view is argued from evidence that neurons in areas like the cerebral cortex can apparently be extremely selective about the signals to which they respond. But I have argued that populations of neurons in the hypothalamus collectively encode information that is not present in any individual neuron. For example, individual vasopressin neurons fire extremely erratically, and can respond to the sensory signals that are important for vasopressin secretion over only a narrow dynamic range. But because the neurons are heterogeneous, the population as a whole can produce an orderly, sensitive, and proportionate response to physiological challenge; the erratic phasic firing of individual vasopressin cells allows the aggregate population to filter out irrelevant, transient perturbations. In short, the population can do things that the individuals within it cannot.
Peptide signals do not fit the classical view of the brain. Peptides released in the brain play many roles, but three of these I highlighted. First, they can serve as “autoregulators” of neuronal activity—in vasopressin neurons, dynorphin secreted from dendrites is essential for the generation of bursts of activity. Second, peptides can serve as paracrine factors, binding a population together. I gave two examples: oxytocin, released from the dendrites of oxytocin cells by suckling, propagates intense bursts of activity throughout the population of oxytocin cells; vasopressin, released from the dendrites of vasopressin cells, inhibits neighboring vasopressin cells and thereby distributes the burden of a physiological response equitably among them. Third, peptides are so potent and so persistent that intermittent pulsatile secretion from a population can generate a hormonal-like signal within the brain that has effects unconstrained by anatomical connectivity, governed only by the distribution of receptors. Such signals, through effects on gene expression, can alter the phenotype of target neurons, and by priming the release mechanisms of their targets they can produce profound and prolonged changes in functional connectivity.
In considering each of these roles, we can make sense of information transmission in the brain only by considering a population of neurons, the clan, as the unit of information processing. Whereas neurotransmitters are whispered secrets that pass from one neuron to another at a very specific time and place, peptides are public announcements, broadcast from one population of neurons to another.
Neurons of the hypothalamus are not all alike; individually, they are erratic, messy, quarrelsome, and unreliable. This heterogeneity is not by design but by accident. The patterns of gene expression in any neuron are not rigidly fixed by genetic nature, but are governed by the unique experience of each cell in its life from birth to adulthood. The innervation of each cell is not predetermined with precision. At any one release site, the release of peptide-containing vesicles is a very, very rare event. It is inconceivable that such rare events are rigidly determined by spike activity; there must be a noisy probabilistic relationship between spike activity and these events. It is only the large numbers involved that give an illusion of determinacy—the many endings in a cell, the many cells in the population.
Communication between our brain and the other organs of our bodies goes in both directions. Just as our brains control how our organs behave, so do our organs control how we behave, and they do so through the heart of the brain, the hypothalamus. In response to hormonal signals from the adrenal glands and the gonads, some neurons in the hypothalamus don’t merely change their electrical activity, they also change the messengers that they produce and who they talk to. The pattern in which they release their messengers not only affects the electrical and secretory activity of their targets, it can also affect the expression of genes in those targets. We are glimpsing a system that is not like a giant supercomputer executing some vast and sophisticated program, but like an ecology of many small analog computers that are constantly reprogramming themselves or being reprogrammed by external events.
Analog computers are now a mere footnote in the history of computing, a history dominated by the rise and rise of digital computers. But when I began as a PhD student in Birmingham, the digital computer available to me was a PDP-9, one of only 445 ever built. It had 32K of core memory, and it filled a small room. At the time I was studying the responses of auditory neurons, and I used that computer to control the frequency and intensity of auditory stimuli and to collect the resulting spike discharge. I could construct a “tuning curve” for an individual neuron that measured the thresholds of response across a range of frequencies in about two minutes—something that otherwise would have taken me about twenty minutes—longer than I could usually maintain a stable recording. But the range of things that could then be done with such primitive machines was limited. When Richard FitzHugh first began studying the Hodgkin-Huxley model of spike generation, to solve the equations on the digital computers then available took a week or more between programming and receiving a solution. So instead he used an analog computer for the work that led to the FitzHugh-Nagumo model, a model that became a standard tool in computational neuroscience,3 and he produced a version of the model that could easily be implemented on the small analog computers then available in student classrooms,4 including one to which I had access.
Digital computers represent quantities as binary data and process data serially in discrete time steps determined by the clock rate of the central processing unit. To program a digital computer involves writing an algorithm that enables a problem to be solved by a generally very long string of individually simple calculations, successively implemented. By contrast, to program an electronic analog computer involves first representing the problem to be solved by an electrical circuit, and then constructing that circuit—something that the small analog computers made simple by providing elements such as resistors, capacitors, and operational amplifiers to be easily connected in different configurations. In analog computers, quantities are represented as a continuous variable—voltage—and computation is intrinsically parallel, not serial. The results of the computation are instantly available and can be displayed on an oscilloscope in real time. At the time this was an ideal arrangement for studying the theoretical properties of neurons, whose membrane properties are naturally analogous to those of electronic components. Now, the vast increase in the processing power of digital computers has eliminated any advantage that analog computers once had. However, analog computers remain interesting because the way they compute feels much closer to the way the brain works: the brain has no central “clock,” all signals exchanged between neurons are continuous, not discrete, and all processing is parallel, not serial.
Reprogramming an analog computer involves physically changing the wiring of its components, an operation that can be seen as analogous to the functional rewiring that occurs in peptide-dependent priming and to the plasticity of gene expression in neurons, as well as to conventional synaptic plasticity. This potential for functional “rewiring”—for reprogramming target populations to fit different physiological states—inspires a different perspective on how the brain works. Computers are programmed once, by an external brain, and the best computers execute that program brilliantly and perfectly. Our brains organize themselves, and reprogram themselves constantly. They guess more than they calculate, they see analogies and jump to conclusions, and they are motivated not by a fixed program but by ephemeral passions and ever-changing needs.
The hormones of the hypothalamus—the heart of the brain—have a particular importance in this process: they are the signals of emotional salience, the links between our passions and our reason, the agents of our urges, our hopes, our fears—of the things that make us human.
Notes