biology is technology

Rob Carlson :: Learning to Fly, Ch 3

Here are a few pages of Chapter 3:

Ch 3.  Learning to Fly (or Geese, Yeast, and 747s).

There is an old saw in physics that geese can’t fly.  It is to this day a mystery how geese manage to carry all that mass around on long flights.  Our understanding of the physics of flight suggests that geese should not be graceful and efficient, despite clear evidence to the contrary.  Estimates of how much power is required for a bird to maintain a certain mass aloft have always been somewhat confused, as illustrated by the difference between previous theoretical predictions and experimental measurements published recently in the journal Nature [1] .  It turns out that heavy birds are considerably more metabolically efficient than we have given them credit for.  The lesson is that we still have a long way to go in explaining seemingly simple biological phenomena that are plentiful just outside the window.  This may seem a bit incongruous, given the fact that we build aircraft that fly much further and faster than the average honking goose.

Another paper in the same issue of Nature made some progress in resolving the ongoing debate about whether birds fly in formation for physical or social reasons.  It turns out that unlike humans, who tend to fly in formation around members of the opposite sex, a bird flying in formation can ride updrafts created by the bird just ahead, allowing more gliding and fewer power strokes [2] .  A theoretical prediction that flying in formation is energetically beneficial was made in 1914, but it took another 85 years to make the measurement.  Even so, the new paper only claims qualitative, rather than quantitative, agreement with theory, meaning that experiments can now be done with sufficient precision to confirm that flying in formation is generally a good idea for birds.  But the experiments are not yet good enough to confirm detailed numerical predictions of the theory.  It may still turn out that the theory gets the details wrong and must be revised.

This caveat is characteristic of the sorts of stories we tell about the way biology works: It is much easier to write down a quantitative description of a biological system than it is to test that description.  The primary division between the study of biological systems and physical systems more amenable to traditional engineering can be found in storytelling.  Most models and experimental predictions in biology are natural language stories.   Models of protein function often have the structure of “Protein X binds to Protein Y”, or “Protein X recognizes and cleaves a certain DNA sequence”.  For the most part, there aren’t any numbers included.  These sorts of models lack quantitative predictive power, whereas engineering generally requires a framework of quantitative models based on quantitative experiments.  Quantitative experiments have certainly been done in biology, and models constructed to describe the results, but they have generally been aimed at describing the behavior of populations rather than individuals.

Indeed, there is a long and fruitful history of experiments and models dealing with the behavior of a system of molecules or a system of organisms.  It is, in many cases, relatively straightforward to create statistical descriptions at the level of populations.  This could be a population of molecules, where quantitative tools in statistical mechanics and biochemistry have been very successful, particularly in describing the many molecules that contribute to the electrical behavior of neural cells, or to the way bacteria and other single celled organisms respond to chemical gradients, known as chemotaxis.  Both of these examples will be explored in a later chapter.  Alternatively, statistical descriptions can be created for a population of rabbits (prey) and foxes (predators).  Ecological modeling derived from MacArthur and Wilson’s seminal work The Theory of Island Biogeography [3] is both quantitative and predictive.  The existence of quantitative population models in these fields provides a basis for biochemical engineering and the beginnings of ecological engineering, which both deal with the statistical behavior of large numbers of individuals.  It does not, however, help predict the behavior of individual members of the population.

This difference is particularly important when considering the transition of a field of study from “pure” science to technology.  For example, building things requires the ability to predict how much product will result from a particular process, or how much weight a bridge can support.  We are extremely interested, for example, in how much water a particular dam can hold back, rather than relying on a qualitative story about how much water can be restrained by a pile of concrete, or even how much water dams hold back on average.  Thus our limited understanding of biological aviation betrays something profound about our ability to implement real biological technology.  While technological progress based on natural language stories is certainly possible, quantitatively predictive models, in contrast, rapidly become design tools and enable true engineering.

The history of human attempts at heavier-than-air flight demonstrates this point very well.  Early investigations into flight were disparate endeavors guided as much by ego and opinion as observation.  In the absence of a theoretical underpinning, it took many years of effort to lay stable foundations for human powered flight.  But soon after controlled flight was finally demonstrated aviation became a discipline in its own right and soon changed the world.

The Frenchman Louis-Pierre Mouillard, in his 1881 book L’Empire de l’Air [4] , was the first to suggest that powered flight would only be possible after learning how to control airplanes via gliding.  He distinguished “pilots”, who would really know how to fly, from “chauffeurs”, who were focused on powering their way into the air without any knowledge or ability to control their craft.  Mouillard was truly ahead of his time.  In a letter to Octave Chanute in 1891, he identified aluminum as “the metal for aviation” [5] *.  Despite Mouillard’s insight into the need for flight controls and new materials, he managed only short flights with small model airplanes and never produced a truly functional passenger carrying glider [5] .

The distinction between the approaches to aviation elucidated by Mouillard is a useful metaphor for the future of Biological Technologies.  It is only through learning how the parts work, and how they work together, that we will truly be able to produce engineered biological systems.  Just as we learned to fly aircraft, so must we learn to fly biology.

Aviation, a term that seems to have been born in correspondence between Mouillard and Octave Chanute in the early 1890’s [5] , emerged in a distributed fashion under the influence of only the most embryonic notions about general principles of flight.  Early attempts at building successful artificial wings were in some cases based on philosophy as much as science.  Otto Lilienthal made his first glider flight in Germany in 1891, and later spent several years making vaguely controlled jumps off an artificial hill near Berlin.  He was among the first early experimenters to recognize that cambered, rather than flat, wings were required for flight.  He also was among the first to implement Mouillard’s notions of control, flying his craft by shifting his weight around the center of lift, foreshadowing today’s hang gliders.  But Lilienthal was cursed by an ancient Greek preference for geometric forms, which led him to favor airfoils that were sections of perfect circles [6] .  At first glance, the curve of a bird’s wing may indeed look like a section of a circle, but evolution spent many years discovering that parabolic arcs are the better shape for flying.

While Lilienthal struggled with an airfoil shape that was inefficient and thus difficult to control, Octave Chanute spent the better part of the 1890’s compiling data that suggested a parabolic arc was indeed better.  On a visit to San Diego he learned of the shape from John Joseph Montgomery, who is attributed with the first use of parabolic wing sections on a glider in 1883 [7] .  Near the shores of Lake Michigan in Indiana, Chanute’s assistants and collaborators flew gliders with parabolic cross-section wings several hundred times in the late 1890’s [8] .

Incidentally, Chanute practiced a form of Open Source Aviation.  Through his correspondence with all the major experimenters of the day he freely gave away all his findings, including extensive data tables on airfoil efficiency, providing a core of airfoil design information to the community [8] .  The Wright Brothers were recipients of this generosity, and benefited from Chanute’s enthusiasm, encouragement, and his design of the now familiar braced-box (also called a Pratt truss) bi-plane construction.  They put the latter information to good use, combining it with their own extensive observations and experimental results, first taking wing in a glider in 1900.  Experience in flying this 1900 glider gave the Wrights enough confidence to later affix an engine to a craft of very similar design, making the first powered flight by a human in December of 1903*.

Although Chanute and the Wrights were frequent correspondents in the 1890’s and early 1900’s, Chanute temporarily fell out with the Wrights during their efforts to patent aviation technology.  He felt the information and inventions should be a public commodity, that the Wrights were impeding broader development, and that he deserved some credit for their specific inventions.  The parties had begun to reconcile at the time of Chanute’s death in 1910 [8] .

The program of careful observation and methodical experimentation begun by Chanute and the Wrights was slowly augmented by a growing body of empirical evidence concerning the flow of air over and around various shapes.  But during its origins, the early practitioners of aviation relied on shared, though differing, stories, including those data tables describing the efficiencies of different airfoil shapes.  Only with time were conclusions based on this data justified with physics.  In particular, Lilienthal’s story, which included the use of circular section airfoils, wasn’t as good as Chanute’s and contributed significantly to control problems that led to a serious crash in 1896.  Lilienthal died from his injuries.  The Wrights diagnosed the cause of his crash several years after it occurred, and this effort led in part to the design of the novel control features in their own glider [8] .

            Throughout this early age of discovery and invention, and in the decades that followed, the beginnings of a quantitative theory of flight encompassing airfoils, control systems, and power systems slowly accumulated in the minds of mathematicians, physicists, and budding aeronautical engineers.  Thus the early qualitative stories about airfoil shape and wing construction evolved into a body of knowledge that superseded individuals and egos, becoming codified in more mathematical terms.  Theory and experiment advanced in turn.  Hypotheses were confirmed or overturned by experiment, and experiments were explained and guided by new hypotheses.

At the end of this process – if it is the end – aeronautical engineering has advanced to the point where theory is so accurate that airplanes are designed on computers, tested on computers, and built predominantly without wind tunnel testing.  The first airframe of a new model is often flown as soon as it is constructed.

The Boeing 777 was the first airplane built in this way.  It was designed based on expertise from building previous aircraft and 50 years of intense academic and industrial effort compiling experience, theory, and shared technologies.

It is remarkable that all this technological development has recently returned aviation technology to where it started: the garage.  The X-Prize was intended to inspire the same ferment in spacecraft as existed in the early days of powered flight, and it succeeded admirably.  As demonstrated recently by Burt Rutan and his crew at Scaled Composites, garage spacecraft are now within the range of relatively modest investment.  True, not every garage inventor has twenty-five million dollars to spend on this sort of project, and SpaceShipOne was not capable of reaching orbit during its demonstration flights in late 2003, but until recently this achievement was beyond the reach of anyone but governments and large corporations.  SpaceShipOne was designed and tested primarily on inexpensive desktop workstations, running commercial computational fluid dynamics software.  When physical testing was necessary, the builders strapped the ship to the back of a pick-up truck and drove through the desert at high speed.  The success of the project owed as much to the expertise and experience that produced the 777 as it did Rutan’s evident brilliance as a designer.  Improvements in physical, conceptual, and computational tools are what made this possible.

            The history of aviation, with its progression from qualitative stories to computer based design, very likely demonstrates the future course of biological technologies.  But before getting overly excited about the possibility of rational biological design, it is instructive to compare the state of our knowledge of biology with our knowledge of aviation technology.  In particular, we can examine how well we know an organism we have relied on for thousands of years.

Yeast are single-celled organisms that provide us beer, bread, cheese, and wine, and may well serve as an initial manufacturing platform for other biologically constructed items.  How does our knowledge of the bits and pieces that make up yeast compare to our knowledge of the parts in workhorse aircraft like the Boeing 747? 

            The original design and construction of the 747 represents roughly the midpoint of the journey from the Wrights to the 777.  The 747 is built from approximately 50,000 kinds of parts (there are 6 million total components in a 747, including fasteners), and at its heart is 50-year-old technology.  Each part, whether wing strut, flap, or turbine blade, is described by its own quantitative model – its device physics – that predicts its behavior under the conditions expected during flight.  The notion of device physics is the cartoon at the heart of Composability (introduced in Chapter 2).  It is a highly functional abstraction that allows design using relevant features of a component as a whole rather than, for example, worrying about what all the atoms in a turbopump are doing during takeoff.

All of the structural shapes in a 747 are the product of long experience optimizing weight and performance, and the materials chosen for each part are thoroughly tested for strength and longevity.  The description of each piece of a 747 fits within a larger model that originally resided partially in blueprints and partially in the heads of Boeing engineers.  Each part was tested to ensure that it behaved as specified by the overall model.  Each part met those design specifications.

            Beginning with only the components described by their genome (proteins and RNAs), yeast have approximately 6300 kinds of moving parts and are built with 3 billion year old technology.  We have models for less than 50% of those parts, and those models are built at very low resolution.  That is, we have given each of those parts a name, we know generally what sort of molecule it is, and we may have some knowledge of its function.  We don’t know most of the design specs of the parts, and we know the device physics for only a handful.   Including the molecules in yeast not directly described by the genome would swell the parts list by many more thousands, only some of which are known in name and in function.  These include the lipid and carbohydrate components of the cell membrane, and the various components of metabolism.  This list is continually growing and we have only a rough idea of how many of each part are present in the cell, which other parts they interact with, and under what conditions.

            One important difference between designing biological systems and designing airplanes is that with airplanes we choose to work within performance envelopes that are relatively easy to model and understand.  Clearly the range of possible flying things exceeds those built by humans.  It is just as clear that we do not understand the physics of most of the things in the air, as demonstrated by the difficulty in explaining how geese stay aloft.  But by restricting ourselves to modes of propulsion and shapes that are within our modeling skills, we can base design tools on those models and thereby provide an infrastructure for building airplanes like the 777.

A considerable technological advance from the 747, the 777 is built from approximately 130,000 kinds of parts (with about 4 million components in each airplane) and was designed from scratch entirely using computer aided design (CAD).  The engines add another approximately 50,000 parts, some of which move at supersonic speeds.  The device physics and the design specifications for each part resided in a large computer model that was repeatedly simulated to understand the likely operation of the completed aircraft.  Though the individual parts, and assembled subsystems such as doors, were thoroughly tested, there were no engineering mockups used in construction of the aircraft [9] .  The first 777 airframe served as a flight test bed.  Boeing put the airplane together from its constituent parts, and then flew it straight away.  It was a remarkable engineering feat.

We trust our lives to this process every day.

In contrast, we have very few quantitative models of biology, and we have no design tools for biology.  Despite the paucity of such tools, we forge ahead (with mixed success) building rudimentary synthetic biological systems, introducing genetic modifications into the environment.

We trust our lives to this process every day.

...