1. Introduction

It has seemed to many philosophers that chemistry was a bastion of monism—roughly, the view that there is just one good way of dividing things up. On the other hand, things get messy in the biological world. The thesis that species have microstructural essences has fallen on hard times. Pluralism now dominates in biology. So goes conventional wisdom. This neat domain division, however, is too simple. I propose to address a case somewhere in the middle. Should we treat biochemical kinds monistically or pluralistically? In particular, I shall focus on the status of proteins.

Biochemists sometimes tell us that the typical human body possesses 100,000 proteins—referring not to the number of individual molecules in the typical human body, but to the number of distinct kinds of proteins typically found in the human body. But how should we count? How do we individuate these biological beasts of burden to make the claim come out (approximately) true?Footnote ¹

Following the lead of natural kinds essentialists of old, one might suggest individuating proteins (and other biological macromolecules) on the basis of their chemical structure. At first glance, this stance affords a tempting monism about biochemical taxonomy. But it is a costly and ad hoc monism. For there are several plausible contenders for the concept of a protein's structure, none of which accommodates biological practice (see Section 2). While structure clearly plays a role in biochemical research, it often plays second fiddle to biochemical function when it comes to biochemical taxonomy (Section 4). However, counting proteins by functions introduces complications that strongly legislate for taxonomic pluralism (Section 5). I tentatively conclude that pluralism about proteins best accommodates scientific practice and our present philosophical resources.

2. An Embarrassment of Riches

Proteins, recall, are polymers of amino acids—“polypeptides”—their order “coded” by DNA. They serve as enzymes (catalysts for biochemical reactions) and structural features of our cells and bodies; they cause our hearts to pump and allow our blood to carry oxygen. They achieve these and countless other feats in large part because of their complex and multilevel structures.

To acquire its characteristic biochemical dispositions, a protein must usually be “folded.” From a protein's initial primary structure (its amino acid sequence), it usually acquires regular regions of secondary structure (e.g., α-helices and β-pleated sheets). Its tertiary structure consists in the subsequent three-dimensional folding of these structures. Quaternary structures form when multiple subunits with tertiary structure become complexed together. All this structure (at various levels of complexity) might seem a structure-obsessed monist's dream. It's really a nightmare. In the first place, the monist must apparently decide which of these various levels of structure is relevant to a protein's kind.Footnote ² But any choice involves serious compromise.

Consider primary structure. An amino acid sequence alone has no direct link with a protein's ultimate biochemical role. The same amino acid sequence can be folded into a variety of different tertiary structures with different biochemical dispositions and functions. Though one hears of “optimization studies” in which the three-dimensional structure (i.e., tertiary or quaternary structure) of a protein is surmised by computer simulations, these must take into account their different contexts and interactions. While many proteins fold more or less spontaneously into their active forms (as a low-energy conformation), many require a delicate dance of helper proteins, playfully referred to as ‘chaperonins’, to assist in folding. Different biochemical milieus also make a difference. Certain amino acids have hydrophobic residues and hence, in aqueous environments, tend together toward the center of the protein globule. They are held together—like antisocial members of an agoraphobic society—only by common aversion.

Even in the same biochemical context, the same sequence of amino acids may take on multiple conformations. So-called dual-topology proteins—for which the direction of insertion and orientation of the protein subunits depend only weakly on their charge characteristics—have recently come to light. It is hypothesized that the possibility of assuming multiple conformations within cells’ plasma membranes may in fact be crucial to their characteristic function (see Rapp et al. Reference Rapp, Granseth, Seppälä and Heijne2006). All these points apply equally to a polypeptide's secondary structure.

Thus, just as individuating chemical kinds by their ingredients significantly undercounts the variation introduced by different structures (hence ‘isomers’, from ‘same parts’), counting proteins by primary or secondary structure is thus apt to count too few. This stance presumes, of course, that the individuation of proteins will to a certain extent represent important biochemical facts about them. Tertiary and quaternary structure—the level at which the majority of a protein's enzymatic prowess is realized—might thus seem better candidates. For it is the particular three-dimensional folding of a polypeptide that confers a protein's enzymatic specificity. As the well-worn analogy has it, enzymes “recognize” their substrates rather as locks recognize their keys.

How should we understand tertiary structure? A natural suggestion would be to treat it as simply analogous to chemical structure. The examples are familiar. When asked to describe the structure of H₂O, we respond that an oxygen atom covalently bonds to two hydrogen atoms in a “bent” shape. We might go on to explain the tetrahedral orbital structure, the average bond angles that result, and so on. Such models can take physical form (think of Watson and Crick's model of the DNA double helix), but it is well to remember that they are models. The angles between the pieces of wood in a “ball and stick” model do not straightforwardly describe the essence of any protein. Individuating proteins—or, indeed, any other molecule—by such a condition casts too fine a net. At any given time, extraordinarily few if any water molecules have precisely the same bond structure—for molecules are not static entities.Footnote ³ The complexity of proteins only magnifies this problem.

More nuanced approaches to chemical structure are surely possible. Instead of describing various constituents united by bonds with certain precise internal angles, we might specify a range of bond angles through which a molecule of a certain kind might traverse. The essence of chemical kind could be represented as a volume in a certain kind of variably dimensional space depicting atomic connection, bond length, and angle ranges (among other potential parameters). Chemical kinds on this model would be clumps in a sort of abstract state-space.

Whatever the plausibility of this model is for “micromolecular” kinds—and it would take some spelling out—it seems far less plausible for macromolecular kinds. Molecular simulations reveal that the tertiary structure of many proteins undergoes dramatic changes. Remember: only the primary structure is held together by covalent bonds; tertiary structure accrues from hydrogen bonding and other weak intermolecular forces, perhaps at only a few points. Imagine a long hose fixed with magnets at certain points along its length which, when folded up, stuck together. Even if those magnets were strong, the structure of the folded hose might yet “flop around” a bit. This analogy suggests an alternative model. Perhaps we should understand a protein's tertiary structure as a series of nodal “connections” (if weak) at different spots on the polypeptide chain. But this too underestimates the extent of a polypeptide's flexibility, as I shall explain below.

It might be objected at this point that molecular biologists do speak of a protein's structure—they invest large sums of money and many computing clock-cycles trying to determine their structure, after all. Molecular biology texts and journals include scores of pictures and three-dimensional representations of these structures. The devil may be in the details, as usual, regarding how specifically to understand a protein's structure—but what are these representations of if not protein structure?

While molecular biologists do indeed speak this way, it's rarely (to my knowledge) in the context of describing a protein's essence. It is well to remember that the X-ray crystallographic techniques often used for elucidating a protein's structure require precipitating and crystallizing a quantity of the protein, stopping (or anyway significantly slowing) its native motion. These “snapshots” no more capture the essence of a protein than a single satellite photo does of a hurricane. More importantly, a protein's motion—the normal structural change it undergoes—is often crucial to its catalytic activity. Proteins are structures in motion.Footnote ⁴ Emil Fischer's long-standing static “lock-and-key” model of enzyme-substrate interaction—where target molecules fit precisely like molecular keys into the larger rigid protein lock where the reaction is catalyzed—has been replaced by Koshland's (Reference Koshland1958) “induced-fit” model. For Fischer's model makes sense of the target specificity of proteins but not their enzymatic activity. As Matthews and van Holde put it, “a lock does nothing to its key” (Reference Mathews and van Holde1996, 368). Consider hexokinase, the first enzyme involved in glycolysis (the digestion of glucose). When a glucose molecule binds to this enzyme's active site, it initiates a significant conformational change in hexokinase that initiates the phosphorylation of glucose to glucose-6-phosphate (i.e., adds a phosphate ion to the sixth carbon of the glucose molecule), leveraging the energy stored in an ATP molecule.

Other “protein machines” like the tetrameric sodium-potassium pump employ similar coupling reactions to power a repeated “airlockesque” motion that pushes ions across cell membranes against their concentration gradients. Another protein, ATP synthase, in turn harnesses the energy of these concentration gradients rather like a turbine to catalyze further reactions. Characterizing tertiary and quaternary structure by elaborations of the foregoing models begins to appear forlorn.Footnote ⁵

Perhaps we should take instruction from the machine analogy. A suitably competent and well-equipped engineer might, after all, construct a jet engine from a static plan or description. The analogy is not perfect. Jet engines are composed of materials and at scales at which quantum effects do not hold significant sway—but put that aside. Is it not the case that whatever a protein's dynamics, those dynamics are dictated by features that are somehow describable? A monist need cleave only to the in-principle possibility of such descriptions, not to their practical possibility. But the monist would also seem to have to answer a difficult further question: Why is this description (be it a range of precise descriptions, a volume in a state-space, or whatever) what it is to be a certain kind of protein? What about the world defines this (nonarbitrary) range of possible precise, static structures? Even if our abstract state-space features some actual clumping, this need not correspond to clumping across a wider range of possible worlds.

The monist encounters this embarrassment by taking the range of natural grouping to be properly part of the world. Withdrawing this claim by allowing that, strictly speaking, as the individual molecules of giant protein jiggle, they instantiate distinct “infimic kinds” (Ellis Reference Ellis2001, 3) avoids the embarrassment. It also forces the monist to bite a bullet: whereas we thought there were around 100,000 proteins in the human body, there may in fact be continuum-many. However, if monism is properly regarded as a thesis about what natural kinds there are, rather than about the systems of classification that characterize them, the monist can candy coat this bullet. So long as real kind distinctions are only ignored (rather than imputed), a plurality of pragmatically motivated systems of classification might sit atop a plenum of natural kinds. I believe that this kind of radical monism is harder to unseat than may initially appear.Footnote ⁶ So it goes with bit bullets. At best, we can push the radical into further radicality—for example, by pointing to the role of extrinsic properties in determining a protein's classification and thus inviting division of the infimic kinds on the basis of these as well. In such a case, there would be as many kinds as individuals (Ruphy [Reference Ruphyforthcoming] discusses this suggestion in the case of stars), a rather unpalatable position about which I’ll have nothing more to stay here. Can monism be maintained by less radical means?

3. Function's Attractions

The monist might attempt to better mirror the kind distinctions biochemists actually make by abandoning the presumption implicit above that individuating natural kinds of proteins requires cleaving to a single level of structure as relevant. A more opportunistic approach might thus adopt a “mixed stance”—“dappled structuralism,” we might call it—picking out primary sequences in some cases and (somehow) tertiary structures in others. However, one might fairly worry either that such decisions would be merely arbitrary, not reflecting anything about what natural kinds actually exist, or that any “mixed” approach to what level of structure was appropriate for a given object would already presuppose some conception of what that object's kind was.

Setting aside these problems, “dappled structuralism” coheres less well with scientific practice than it might initially seem. Consider the enzyme alcohol dehydrogenase (ADH), which catalyzes the conversion of ethanol to the hangover-inducing acetaldehyde. Insofar as the catalysis involves motion, primary structure would appear to be the indicated choice: there's just one kind of protein here, not continuum-many very slightly different kinds of proteins. But even in this case, counting by primary structure plausibly counts too many kinds of protein. For ADH comes in many “alleloforms”—very slightly different polypeptide sequences whose differences happen to be inconsequential to the protein's tertiary structure and biochemical function.Footnote ⁷ In most populations of the fruit fly Drosophila melanogaster, for example, ADH comes in two variants called “slow” (ADH^S) and “fast” (ADH^F), named after their speed through an electrophoresis gel. As Ridley puts it, “the enzyme called alcohol dehydrogenase is actually a class of two polypeptides with slightly different amino acid sequences” (2003, 83; my emphasis). Since normally such differences are inconsequential to the enzymatic activity of ADH, ADH^S and ADH^F are regarded as of a single kind of protein.Footnote ⁸

The appeal of this sort of dappled structuralism is an artifact of the importance of function for actual biochemical taxonomy. Suppose for the moment that we possessed a satisfyingly deep philosophical account of biochemical function that upheld ordinary scientific intuitions. How might we put it to work as the selection criterion properly sought by the monist? Individuating proteins by their primary structures failed to recognize the different tertiary structures they could attain and thus the function they might perform. The very same polypeptide sequence might get differently folded within a cell membrane to function in signal transduction or as a transcription regulator within different cells. How many kinds of proteins have we here? We hardly hesitate in answering “two,” counting by their distinct functions. A cheer for functional individuation! On the other hand, individuating proteins by tertiary structure either forecloses on grouping as a kind the various structures whose motion performs a certain function or leaves wide open the question of what “dynamic range” of structures correspond to a kind of protein. Shifting our focus to functional individuation allows us to put aside the complexities of structure—perhaps identifying occasional extensional overlap between structure and function ascriptions—and retain our monistic faith. So goes the optimistic thought. A second cheer!

4. Function's Perplexities

You know where this is going (optimism seldom pans out in philosophical essays!): the macromolecular monist now needs a monistic account of biochemical function. If we divide different kinds of proteins by function—and that's a big “if”Footnote ⁹—how should we divide functions? This is a complex and well-studied question about which I cannot hope to say anything too substantive. Fortunately, my conclusion tolerates some abstraction. However we understand biological function, ascription of function will likely be a holistic enterprise. This gives rise to a deep and theory-neutral pluralism. So long as functions cite different kinds of things (e.g., the function of adding a phosphate ion to the sixth carbon of the glucose molecule), some of which are functional kinds, there may be multiple different but equally acceptable ways of ascribing functions and thus of individuating functional kinds. Even if my transcendental approach fails to convince, I am reasonably confident that the details of various theories of function will bear out my pluralism (though I shall only be able to gesture at these lines of argument).

Broadly speaking, two approaches to function have dominated the literature: the etiological or selected-effects approach (championed by Wright [Reference Wright1973], Millikan [Reference Millikan1984, Reference Millikan1989], and Neander [Reference Neander1991]) and the causal role or systemic-capacity functions (developed by Cummins [Reference Cummins1975], Amundson and Lauder [Reference Amundson and Lauder1994], and Davies [Reference Davies2001]). According to the selected-effects approach, the function of a trait depends on whether performance of that function has figured in the selective history of that trait. Spelling out this intuitive thought has proved complex. How, specifically, must a particular capacity figure in a selective explanation? Must it explain the increase of the trait or just its maintenance? Cannot traits that are declining in frequency nevertheless possess functions? And, of course, settling these questions does not provide a way of gauging the contribution that a particular trait makes. Some traits are, in Gould and Vrba's (sometimes maligned Reference Gould and Vrba1982) term, exaptations: traits originally selected because of one kind of functionality but later co-opted for another. Even somehow restricting the contributions that trait has made to a lineage's “modern history” (see Godfrey-Smith Reference Godfrey-Smith1994) leaves open the question of how important possession of that trait has been to its possessors’ success.Footnote ¹⁰

Difficulties like these, while perhaps not insuperable, underlie much of the attraction to the more liberal systemic-capacity approach to function (upon which I shall focus). On this approach, we attempt to understand the function of a component in terms of its contribution to the capacities of the larger system. DNA polymerase, a key enzyme involved in DNA replication, has a “proofreading/error-correction” function that greatly reduces copying errors. High-fidelity DNA replication clearly contributes to the growth and development of organisms (not to mention their reproduction). It's not hard to appreciate how this systemic capacity likewise contributes to the fitness of organisms and serves as an explanation of its own persistence in the lineage.Footnote ¹¹

But we can divide up systemic capacities in various ways. When we ask, What is the function of DNA polymerase? a natural response is that it has several functions: DNA replication and error correction. But this may be an artifact of the fact that we understood its role in replication before we understood its role in error correction. Perhaps we ought to construe its function as high-fidelity replication. Which is correct? Is there a single function here or multiple functions? Is there a uniquely privileged way of dividing systemic-capacity functions?

The monist has an obvious response to this kind of pointed question (inspired by the radical monism above): analyze down—divide reality as far as it can be divided. Those divisions are the natural joints. Simply ignoring these fine distinctions—in, say, grouping different alleloforms that carry out the same function or even grouping different functional kinds into a higher-order kind—happens all the time and hardly impugns their reality. There is a privileged system of biochemical classification, even if it is not the one we commonly use. In the case of DNA polymerase, while it is natural to include a variety of structures under the functional umbrella of high-fidelity DNA replication, biochemists can distinguish between various types of DNA polymerase on the basis of precisely how they carry out this function: “When [a mismatched DNA complex] was supplied to E. coli DNA polymerase I, the incorrectly hydrogen-bonded base was removed by a 3’ → 5’ exonuclease activity. … In DNA polymerase III, this function resides in the Θ subunit of the core polymerase” (Lodish et al. Reference Lodish, Baltimore, Berk, Zipursky, Matsudaira and Darnell1995, 387). The committed monist could once again argue that while strictly speaking there are multiple functional biochemical kinds here, whether we group them as a single kind (owing to the sameness of a systemic capacity at a coarse level of analysis) or acknowledge the difference and divide them as different kinds is up to us.

This response may work well for some cases, but I do not think it will wash in general. For whether a proffered systemic capacity represents the “lowest level” of functional analysis will sometimes depend on how we divide into kinds other portions of the world—(crucially) including the very biochemical world which is at issue. I mentioned chaperonins above. These proteins fold other polypeptide sequences into their active tertiary structures. Now imagine two alleloforms of a certain protein that are shaped by a similarly heterogeneous collection of chaperonins. Do these shapings count as two different functions? That depends, presumably, on whether the two alleloforms are in fact the same kind of protein—whether they have the same lowest-level biochemical functionality. Suppose, for example, that one alleloform served to initiate cell division via a pathway that was interruptible by high concentrations of sodium or potassium and the other was only sensitive to sodium. As we “analyze down,” we would assign identifiably different (though readily groupable) functions to each alleloform that would potentially trickle down to the alleloforms of the chaperonins that helped form them. So far, so good. But now imagine, on the other hand, that the functionality of our two alleloforms is to dismantle the very chaperonins that built them! A wasteful pathway, to be sure, but certainly a possible one (even if it would not be favored by selection).Footnote ¹² Any appeal in this case to the objectivity of a lowest level of functional analysis simply goes round in a circle.Footnote ¹³

This suggests that there will sometimes be no uniquely privileged way of exhaustively carrying out a functional analysis on the systemic-capacity approach. Insofar as the individuation of biochemical kinds depends on just this sort of functional analysis (I believe the story is similar on the etiological approach), there is no uniquely privileged way of dividing up biochemical kinds. By hitching their wagons to function, macromolecular monists face the problems concomitant to functional monism. As with the structural route, the functional path has reached a dead end. The foregoing does not yet amount to a knock-down argument for macromolecular pluralism. Other routes may yet reach the monists’ destination. I am not hopeful, however.

5. Modes of Biochemical Classification

I began with a question that appears to have an answer: How many kinds of proteins are in the typical human body? Perhaps there is an answer in that it has many answers. Is there then any point to calling any of these answers correct? Some answers are clearly incorrect: there are considerably more than five or a hundred or even a thousand kinds of proteins. How can our pluralism be made “modest”? This is a difficult question that crops up anywhere pluralism tempts us. As yet, I think we have only the slimmest grasp on how to unite pluralism with some sense of realism. This is a project for another time, but one on which I believe consideration of proteins may shed some light.

Pluralism makes good sense of the actual classificatory activities of biochemists. Our classification schemes typically reflect our best inductive and explanatory strategies—and these are rarely neat and tidy. Coarse taxonomies develop in contexts where fine functional analysis serves no apparent utility. Clearly, these may be revised if circumstances demand it (say, if we cannot make sense of the effect of introducing a particular protein).

Biochemists can avail themselves of a plurality of legitimate classificatory systems simultaneously by invoking what I’d like to call different “modes” of classification. The example of the alleloforms of ADH highlights an important feature of how we classify the world. Notice how Ridley described this case: the same kind of enzyme is actually a class of two kinds of polypeptides. The role of these organizing terms cannot be ignored. Polypeptides are individuated by their sequence; I conjecture that enzymes (and proteins more generally) are individuated, at least in large measure, by their function. We needn’t maintain a one-to-one correspondence between different kinds of proteins and functions in order to appreciate this importance. If proteins are not purely functional kinds, though, the details of their (pluralistic) individuation seem to me an interesting and wide-open area for future research—one that would go a long way toward “de-radicalizing” the pluralism.

I believe the notion of different “modes” of dividing the world may help with this as well. While these different modes may cross-classify reality (or merely reveal different “layers” of structure), they need not be regarded as forcing hard choices. Perhaps the ambiguity of our nomenclature (as perhaps ‘protein’ is ambiguous between enzymatic function and the various levels of chemical structure) simply eases transitions between these modes, discourse that is triggered by other univocal organizing terms. Drosophila ADH is a member of what Benach et al. call “a broad and heterogeneous family of alcohol dehydrogenases, named short-chain dehydrogenases/reductases (SDR), and is the only member that utilizes small alcohols as substrates” (Reference Benach, Atrian, Fibla, Gonzàlez-Duarte and Ladenstein2000, 3613). But though heterogeneous, SDRs possess certain structural commonalities; they describe one ADH's membership in this class thusly: “the subunit shows the typical fold of the short-chain dehydrogenase/reductases enzyme cores thus exhibiting a three-dimensional structure similar to that reported for other SDR proteins whose three-dimensional structure has been solved to date. The subunit of [Drosophila ADH] has an α/β single domain structure with a characteristic Rossmann fold dinucleotide-binding motif” (3614).

Our rich system of different interleaved levels of functional and structural classification, I think, is best understood as the result of coming to terms with the genuine plurality of systems of functional and structural kinds. I suggest that we look to the role of organizing terms in our conceptualization of the world. There is no dividing the world into the natural kinds, perhaps—but that doesn’t mean that there aren’t natural kinds. There are natural kinds of enzymes, natural kinds of polypeptides, natural kinds of organisms, natural kinds of celestial objects, and so on—each dependent on norms of classification devolving from our best inductive and explanatory practices.Footnote ¹⁴