`23 Redundancy and Degeneracy`

If an organised body is not in the situation and circumstances best adapted to its sustenance and propagation, then, in conceiving an indefinite variety among the individuals of that species, we must be assured, that, on the one hand, those which depart most from the best adapted constitution, will be the most liable to perish, while, on the other hand, those organised bodies, which most approach to the best constitution for the present circumstances, will be best adapted to continue, in preserving themselves and multiplying the individuals of their race.

—James Hutton (1726–1797)¹

How did the complex brains of vertebrates evolve? C. elegans has only about 300 neurons, each with a specific and unique function. To execute its function, each must exchange signals with other neurons and with the other cells whose function they regulate; they communicate using both electrical and chemical signals, including both neurotransmitters and hormone-like messengers. They receive signals from the animal’s external and internal environment, and many have specialized sensory properties, enabling them to respond directly to stimuli such as light, osmotic pressure, electrolytes, nutrients, and a variety of chemicals. These neurons are multifunctional, combining properties of sensory cells, neurons, and endocrine cells.

In bigger brains, no single neuron is particularly important: each is part of a clan of hundreds or thousands of similar neurons that collectively execute a particular function. The zebrafish has a handful of vasotocin neurons; the rat many thousands of vasopressin neurons in several clans. The clans differ in many ways—in the shape and size of their neurons; in where they project to and where they receive inputs from; in what other proteins and peptides they make. Through the evolution of increasingly large and complex brains, the different neurons of simple brains have proliferated and diversified into distinct clans, each retaining some features of an ancestor cell while acquiring new features through mutations.

But there are important differences between a nervous system in which a function is enacted by a single cell and a nervous system in which the analogous function is executed by a clan of cells. The death of any one neuron in a simple nervous system has major consequences, but a single C. elegans lives for just two or three weeks, has a generation time of four days, and may produce a thousand progeny in each generation. We live for seventy years or more and raise just a few children; our lives are, in this sense, more precious. Our brains must be robust against the insults that our lives deliver and the random neuronal death that ensues with these insults, and the brains of our children must be robust against the vagaries of random gene mutation and a process of development that is, to a large extent, stochastic. That robustness is in part enacted by redundancy: when a function is enacted by more neurons than are needed, it can withstand a degree of attrition by random loss of neurons and tolerate inefficiency introduced by minor gene mutations or developmental imprecision.

Redundancy brings problems. A signal implemented by more neurons than are needed may be a signal that is too large, and a signal that must be precisely timed may be “smeared” if implemented by a large population of autonomous neurons. The clan response must be orchestrated; the members of the clan need to know what the other members are doing. The signals that pass among cells of a clan may have to be different from the signals that pass from a clan to its target cells: to efficiently inhibit a target the cells in a clan may have to excite each other, and we see examples of both mutual excitation and mutual inhibition in the hypothalamic clans.

There are other ways of implementing robustness. Redundancy, in the strict sense, is the overprovision of agents to fulfill a given task: this provides robustness against the death or inactivation of some of those agents but not against the inactivation of all of them. Robustness against that eventuality is provided when the same task is achieved by different means. The ability of structurally different elements to perform the same function is called degeneracy, and it is a characteristic of many biological systems. Degeneracy is both necessary for, and an inevitable outcome of, natural selection.² If a system is not degenerate then a mutation is likely to be lethal, but if it is degenerate then a mutation can open the door to the evolution of a novel function. The genetic code itself is degenerate: different nucleotide sequences can encode the same amino acid. Many genes are degenerate: transcription can start and stop at different sites and still yield the same protein product. At the highest level, our language is degenerate: we can express the same meaning in many different ways. In our language the existence of synonyms allows words to evolve in meaning over time: “viral” does not have the same meaning now that it did when I was young.

Degeneracy arises in biological systems at many levels and in many ways. Natural selection can only work in a population where there is heritable variability, and that variability comes from the accumulation of neutral gene mutations. A mutation may be neutral because it has no functional consequences, or because it degenerately or redundantly duplicates an existing function. Such mutations accumulate because there is no strong selection pressure to eliminate them. In the process of genetic recombination that accompanies sexual reproduction, new combinations of neutral mutations might display novel functionality, and, if adaptive, these will be selected for.

Our evolvability—our ability as a species to adapt to environmental change—thus depends on genetic variability and degeneracy. On a different timescale, our ability as individuals to adapt to life challenges also depends on degeneracy, in the form of our ability to solve problems by another way when one way is blocked. This poses problems for neuroscientists. We typically infer the function of a neuronal population from three sorts of experiments: by observing it in different physiological circumstances; by showing what effect its activation has on physiology or behavior; and by showing the physiological consequences of removing or disabling it. When these three sorts of experiments converge on a conclusion, we are likely to have reasonable confidence in that conclusion. But there are pitfalls.

If a neuronal population responds in some way to a particular physiological challenge, how do we know that that response has any physiological significance? Some responses may be too small or brief to be functionally relevant, while others may not be what they seem. At the end of pregnancy, oxytocin neurons respond to many stimuli less strongly than they do at other times. But at the end of pregnancy there is a massive store of oxytocin in the pituitary—the store has been upregulated in anticipation of the demands of parturition and lactation. Because the stores are greater, spikes release more oxytocin—so the weaker spiking activity seen in response to many stimuli evokes a normal amount of secretion. The attenuated spiking response is an adaptation that preserves a normal function.³

When we give a peptide into the brain, we typically inject a large amount to see a large effect, but if we give too much the peptide might act at other receptors. Oxytocin can act at vasopressin receptors and vasopressin at oxytocin receptors at concentrations only modestly higher than those at which the native ligand acts. If giving a peptide into a part of the brain changes some behavior, we need to know that in the behaving animal the endogenous peptide is actually released in relevant circumstances and reaches that site in sufficient amounts—because the receptors, as a result of a neutral mutation, might be expressed at sites that never see the endogenous ligand.

If we remove a peptide from an animal, we may see no effect because we have intervened in a degenerate system. This is not a hypothetical pitfall; any true and full account of how our understanding has emerged would be replete with such tales.

In the arcuate nucleus, one clan of neurons makes two peptides, NPY and AgRP. Both are potent orexigens: when injected into the brain in tiny amounts they stimulate eating. These neurons are primary drivers of appetite: they are activated when the empty stomach secretes ghrelin, which acts directly on the NPY/AgRP neurons, and are inhibited by leptin when our energy stores are in excess. The expression of mRNA for NPY and AgRP is increased after fasting, antagonists to either one of these peptides will inhibit feeding, and activation of these neurons by optogenetics will activate feeding. AgRP acts at the MC4 receptors through which the satiety signal α-MSH acts, but with the opposite effect of α-MSH—it is an “inverse agonist.”

Yet transgenic mice that lack NPY have a normal body weight. Similarly, transgenic mice that lack AgRP have a normal body weight. So is the absence of either compensated for by the other? Transgenic mice that lack both NPY and AgRP also have a normal body weight. So did we get it wrong, are these neurons unimportant? To test this, two groups of scientists independently and at about the same time found ways to selectively kill the NPY/AgRP neurons. The outcomes were clear: when most of these neurons were killed, the mice stopped eating and were kept alive only by gastric feeding.⁴^,⁵

It seemed that the neurons could still drive feeding despite the loss of both NPY and AgRP, and they could do so because they remained able to release the neurotransmitter GABA. Remarkably, delivering a GABA agonist to mice, using a tiny implanted pump to deliver it into the peritoneal cavity, completely prevented the anorexic effects of destroying the NPY/AgRP neurons—and the mice continued to eat normally even after stopping the infusion a few days later.⁶ Thus, given just a few days, the mice could compensate fully for the loss of these neurons, at least as judged by their ability to eat a normal amount of food each day.

Some peptides are essential for the things we believe them to be important for: mice without GnRH or kisspeptin are infertile; mice without oxytocin cannot feed their young; rats without vasopressin cannot concentrate their urine; animals without growth hormone or its releasing hormone remain very small. Many, though, are not. Ghrelin, known as the “hunger hormone” for its potent effects on appetite and adiposity, was named not for this function but for its potency at stimulating growth hormone secretion. But ghrelin-deficient mice grow normally, eat normally, and have a normal body weight.⁷

Oxytocin is named from the Greek for “quick birth,” and is best known for its use in obstetrics to facilitate delivery. Oxytocin-deficient mice deliver their young normally and at about the normal time,⁸ but from this we cannot conclude that oxytocin is unimportant in parturition. As John Russell and I wrote in 1998, “The neurohypophysial system regulates the delivery of progeny in all vertebrates, and, during its long evolutionary history, other mechanisms may have evolved convergent roles simply by a process of exclusion. When everything that opposes the actions of oxytocin in parturition is excluded, the things that remain are neutral, assist oxytocin or, in dogging the footsteps of oxytocin, can substitute for it.”⁹

Thousands of gene products are involved in the processes by which a few cells in the early embryo proliferate, diversify, migrate, and form the networks that will comprise the mammalian brain. That brain must be robust both to environmental challenges and to intrinsic sources of variation, including the noise in gene expression that comes from the probabilistic nature of gene transcription. Our brains can be robust to even extreme insults in early development, as shown by cases of people with large parts of the brain missing who go on to lead normal lives with little apparent impairment.¹⁰^,¹¹ Many mechanisms support that robustness. As mentioned earlier, neurons that receive more or less of a particular chemical signal typically respond by downregulating or upregulating expression of receptors for that chemical, while neurons that receive a stronger or weaker excitatory drive often alter their expression of ion channels to restore a “normal” level of excitability. The distributed nature of information representation means that, even in the developed brains of adults, focal brain injuries tend to degrade performance in a broad domain rather than ablate specific elements of it, and because connections are plastic, considerable functional recovery can occur. Natural selection is likely to favor system architectures that are robust in this way, and systems that have evolved to be robust to extrinsic variation are likely also to be robust to mutations.

Most if not all neurons in the brain produce and secrete one or more peptides as well as conventional neurotransmitters, and express receptors for many peptides. We cannot safely infer that all of these play an important role in brain function. Often, they might be evidence less of the ubiquity of peptide signaling than of the “messiness” of neurons, and of widespread redundancy and degeneracy. I am not arguing that what is true in the heart of the brain, where peptides are certainly vitally important, is necessarily true of the whole brain. But in the next and final chapter I will try to pull together the disparate elements of a hypothalamocentric vision of the brain, and contrast it with the classical view.

Notes