G'Day All
Address altered to avoid spam, delete RemoveInsert
</delurks>
On Tue, 11 Jun 2002 21:15:35 +0000 (UTC), drear...@aol.com (Drearash)
wrote:

To amplify this, I will repost (with some slight amendments) a post I
made on this subject some time ago.

Spetner uses two separate metrics of information, one is an
"expectation" measure whereby an ensemble of different strings has
less information that an ensemble of identical strings, this will
surprise people familiar with standard Shannon Weaver information or
Algorithmic information, but is a valid formulation under particular
circumstances.

He also uses an "addressing" measure of information. This is fairly
simple to understand.
"Brisbane" has less information than
"Cannon Hill, Brisbane" which has less information than
"Richmond Road, Cannon Hill, Brisbane" which has less information than
"666, Richmond Road, Cannon Hill, Brisbane"
(there is a formal way to do this using binary addresses for matrix
elements in n x n (or n x n x n) matrices which I can't be bothered to
go into now. It's a bit of a stretch mapping this onto substrate
binding, but there is nothing inherently more loony about this than
mapping telegraphy transmission to DNA replication).

He then claims that enzyme/hormone receptor specificity is analogous
to addressing specificity, and that enzymes with many substrates have
less information than enzymes with one substrate. On the face of it
this seems like a reasonable argument. Enzyme-substrate (or
hormone-homone receptor) binding is likened to a "lock and key"
mechanism, where the substrate is the key and the enzyme is the lock.
(important caveat, both the lock and key are "floppy").

Note that there is a subtle shift in the argument. One can easily see
that the address for "666, Richmond Road, Cannon Hill, Brisbane" takes
more information to specify than "Brisbane", but the claim that
enzymes have more information because they bind only one substrate is
similar to the claim that "666, Richmond Road, Cannon Hill, Brisbane"
has more information than "Brisbane" because fewer letters are
delivered there than there are to "Brisbane". This is a somewhat
different claim, that the information of the physical object at an
address has the information required to specify that address.

But accepting Spetner's claim, it would seem intuitively obvious that
a
more specific lock accepting only one key has more information a less
specific lock which accepts many keys (this is actually debatable),
and that analogously, an enzyme is more specific because it has more
binding points than an enzyme that is less specific (and there is also
the unstated assumption that single substrate enzymes are more
important than multi substrate enzymes. They are not).

For example we might imagine an active site that uses 4 binding points
has more info than one with two binding points

X--O     O--Y        X--0     O--Y
    \   /                \   /
    DRUG                 DRUG
    /   \
Z--N     N--Y

Where X, Y and Z are amino acids in the enzymes active site.

Unfortunately, biology doesn't work that way. The number of binding
points are important, to a degree. But other physical properties of
the enzyme are important too. For example, take the betalactamases
that break down cephalosporins, one mutant variant can breakdown
extended cephalosporins, where as the normal enzyme can't. The mutant
variant doesn't have fewer binding points than the normal enzyme, but
has a more flexible hinge so that the catalytic group can reach the
lactam ring easier.

Even within a binding site things are tricky. In adrenergic receptors,
the drug propranolol binds to 3 sites in the binding cleft, yet has
far more powerful binding to the enzyme than the natural hormones
noradrenaline and adrenaline, which bind to five sites in the cleft.

(see http://home.mira.net/~reynella/chime/adr_tute.htm and go to the
adrenoceptor tutorial to see what I mean, requires the chime plug-in)

Worse still, you can have substrates which bind at exactly the same
number of points, but which bind more firmly because you have
substituted a chlorine atom for an oxygen atom, and changed the
distribution of electrons on the drug molecule.

Binding specificity is not something simply analogous to "addressing",
it involves physical shape (which is amenable to an addressing
analysis) and physicochemical properties (which is not simply amenable
to an addressing analysis, especially when the properties are those of
the _substrate_) and structural properties of the enzyme, like hinge
flexibility or helix distortion which are also not amenable to an
addressing analysis (does a helix which flexes to the right have more
information than a helix which flexes to the left?)

Simply put, any analysis of the "information" of a protein that
addresses the substrate binding specificity, although superficially
plausible, is naive in the extreme, because substrate specificity is
not amenable to the proposed analysis (especially when some of that
information is in the _substrate_). Also, his analysis deals solely
with single substrate enzymes, how he deals with bi-bi-random two
substrate two product enzymes I don't know, he seems to feel that all
enzymes should be like (some of) the enzymes of the Krebs citric acids
cycle.

Worse still, in his analysis of the ribulose to xylitol mutation he
doesn't consider binding specificity, but a biochemical measure called
specificity. This is a ratio between catalytic efficiency and binding
specificity. Okay, you say now, this is pretty trivial, we can work
with this. Well, no. Catalytic efficiency is something else again. It
is definitely not amenable to an "addressing" measure, and is based on
a range of physicochemical properties such as the charge distribution
in the _substrate_, the free rotation of the catalytic side group
(does a catalytic group that rotates 10 angstroms have more or less
information than one that rotates 15 angstroms?) and so on.
Critically, two substrates can have the same _binding_ specificity,
but different catalytic efficacy which confounds the information
analysis.

Worse yet again. Spetner's argument is related to the number of
substrates acted on.  Yet in the ribulose dehydrogenase mutating to
xylitol dehydrogenase, in both cases the enzyme bound 3 substrates, by
his own definition no information change has occurred. The _rate_ of
dehydrogenation has changed, but
a) There is no simple way relative magnitudes can be incorporated into
his calculation. Remember "666, Richmond Road, Cannon Hill, Brisbane"
has more information that "Brisbane" by simple bit length measures.
How are we to assign a information measure to the fact that "666,
Richmond Road, Cannon Hill, Brisbane" receives twice as many election
newsletters form the Labour candidate as the Liberal candidate, while
"664, Richmond Road, Cannon Hill, Brisbane" receives the same number
(this situation is a closer analog to enzyme activity)

b) The relative magnitude change may be entirely due to changes in
catalytic efficiency, which has no information content in an
addressing scheme (nor can I see any meaningful way to calculate the
"information" in this with any existing information metric)

It gets worse still. Spetner compared the "specificities" of 3
substrates, because they were the only substrates measured in the
experiment. In reality ribulose dehydrogenase (and xylitol
dehydrogenase) binds a lot more substrates than just those 3 (although
the catalytic efficiency is very low for the majority of substrates
bound. Without assessing the full panel of substrates, any claim about
specificity is meaningless. (But which substrates? if we restrict
ourselves to natural substrates, we exclude xylitol, a synthetic sugar
not found in the natural environment, but developing xylitol
dehydrogenase activity was the whole point of the exercise) if we
include synthetic substrates, we have a potentially infinite number of
substrates to test.

It continues to get worse. Spetner assumed that the point reached in
the experiment (where Ribulose and Xylitol were being broken down at
roughly similar rates by the mutant enzyme), was as good as it gets,
no further improvement was possible. In the experiment he looked at,
the main point was to see if xylitol activity could be developed, not
getting the optimum activity. In fact, in other experiments mutant
enzymes were produced that broke down xylitol 20 times faster than
ribulose (which kind of destroys his thesis see Hartley, B.S. (1984),
Experimental evolution of ribitol dehydrogenase. In R.P. Mortlock
(ed.), "Microorganisms as Model Systems for Studying Evolution" (pp.
23 - 54) Plenum, New York). Furthermore, we have literally hundreds of
enzymes where random mutation results in high substrate specificity
(there has been enormous amounts of work on developing novel and
specific enzyme activities from generalist alpha-beta barrel proteins
eg see Matsumura I, Ellington AD. In vitro evolution of
beta-glucuronidase into a beta-galactosidase proceeds through
non-specific intermediates. J Mol Biol. 2001 Jan 12;305(2):331-9).

And it _still_ get worse. His other example is the case of the
streptomycin resistance mutation of the _rspL_ gene (which codes for
the S12 subunit of the 30S ribosomal particle. The S12 subunit with
the 16S RNA forms part of the proof reading center of the tRNA
acceptor binding site). The actual protein is
_more_ specific, it now only binds tRNA rather than tRNA and
streptomycin. Now, streptomycin isn't a substrate, so you might object
to using binding specificity in this case, but it does have an effect
on the substrate binding accuracy of the ribosome, and streptomycin
binding ribosomes turn out garbage proteins because streptomycin
messes up the proof-reading centre (which is how streptomycin kills
bacteria). The mutant version which doesn't bind streptomycin is
actually MORE accurate, ie more SPECIFIC, than the wild type (the
wild-type proof reading centre makes a few mistakes even in the
absence of streptomycin).

Either way, this  is a _clear_ increase in Spetner's binding
specificity (the mutant gene product doesn't bind streptomycin at all
and binds peptidyl tRNA more accurately). Spetner _now_ swaps to the
expectation measure and claims (without evidence) that since there
must be more S12 sequences that don't bind streptomycin than those
that do, information must have decreased (why didn't he do this
analysis on the ribulose
enzyme?).

However, he is dead wrong. The ensemble of all _rsPL_ genes that
produce streptomycin binding S12 is around 10^60. The ensemble of all
_rsPL_ genes that _don't_ bind streptomycin is around 10^60 too (from
an analysis of neutral mutations ala Yockey). So the information
difference is so small as to be non existent. If we only take amino
acid changes in the streptomycin binding site, then there are roughly
the same number of substitutions that will allow binding as those that
don't (10 resistant versus 6 normal see Tolvonen JM et al., Mol.
Micro. 1999, 1735-1746). Also, there is a particular mutation, AAA43
-> AGA43 which doesn't bind streptomycin AND has wild type accuracy
and translation rates. This is the only possible (single) mutation
that does this, so it's ensemble is smaller than the wild type
ensemble

So to summarize, Spetner swaps metrics when one shows inconvenient
changes and he is wrong in the details, these invalidate his analyses.
Although his arguments are superficially plausible, a deeper look with
some knowledge of biochemistry shows massive flaws.

Cheers! Ian
</relurks>
=====================================================
Ian Musgrave Peta O'Donohue,Jack Francis and Michael James Musgrave
reyne...@werple.mira.net.au http://werple.mira.net.au/~reynella/
Southern Sky Watch http://www.abc.net.au/science/space/default.htm