ENIGMAs --- Innovations vs. Mutations

2 views
Skip to first unread message

z@z

unread,
Jul 11, 1999, 3:00:00 AM7/11/99
to
In Joe's ENIGMAs-mega-thread there has been a very interesting
discussion on the occurence of mutations in bacteria. The defenders
of neo-Darwinian orthodoxy could not provide convincing evidence
that adaptive mutations are totally random with respect to function.

It is generally acknowledged that mutation rates are very different
within a single genome. There are genes such as cytochrome-c and
ubiquitin with a mutation rate of almost zero. Also within genes,
regions with high mutation rates are confronted with highly
conserved regions.

As a general rule we can state that mutation rates are the higher,
the lower the functional constraints are. But whereas mutations
occur in the DNA, the functional constraints are on proteins. That's
really enigmatic!

Furthermore a number of experiments have shown an increased mutation
probability when cells are grown under stress (e.g., with low levels
of an antibiotic). It would be interesting to know how the very
different mutation rates of a single genome are affected by this
increased probability.

Around 70 percent of the mutations of the third position of a DNA
triplet are silent. According to Motoo Kimura (Scientific American,
11/1979) such positions seem to be especially variable. If they
actually have a higher mutation rate than the first two positions
of triples then neo-Darwinism has a big problem.

There are cases where bacterial resistance to an antibiotic depends
on one single point mutation. It is very suprising how easy it is for
an enzyme that is inhibited by an antibiotic to become an enzyme that
hydrolyzes that antibiotic.

One single point mutation and the enzymes defends itself by attacking
the aggressor!

One single point mutation and the enzyme carries out in addition to
its normal tasks the task of hydrolyzing its own aggressor!

One single point mutation resolving all biochemical problems involved
in adding such a complex function to a protein without spoiling its
normal functions!

To explain all this by a lucky conformational change resulting from
the substitution of one amino acid by another defies common sense,
logical reasoning and simple probability estimates.


Wolfgang


"DNA mutations do not occur everywhere with the same freqency.
Otherwise a fatal mutation in a vital protein would be much more
probable than a mutation e.g. having an effect on the length of the
neck. Moreover, evolutionarily older sequences (e.g. ubiquitin) are
less susceptible to mutations. However, according to reductionism
the probability of mutations should not depend on the evolutionary
age or on the effects of the mutations."
http://members.lol.li/twostone/E/psychon.html#a06

Stephen Knight

unread,
Jul 11, 1999, 3:00:00 AM7/11/99
to
On 11 Jul 1999 18:16:36 -0400, "z@z" <z...@z.lol.li> wrote:

>In Joe's ENIGMAs-mega-thread there has been a very interesting
>discussion on the occurence of mutations in bacteria. The defenders
>of neo-Darwinian orthodoxy could not provide convincing evidence
>that adaptive mutations are totally random with respect to function.
>
>It is generally acknowledged that mutation rates are very different
>within a single genome. There are genes such as cytochrome-c and
>ubiquitin with a mutation rate of almost zero. Also within genes,
>regions with high mutation rates are confronted with highly
>conserved regions.

A Laymen Point of View.

When life evolved, the Earth did not resemble it's present state.
Couldn't mutations be, during the "early" years, the example instead
of the rule we have today. Natural radiation bombarded our planet. We
did not have O3 to protect us (them/it) so couldn't mutations have
taken advantage of this effect? Taken selection to a point where
mutations could be beneficial due to harsh (my definition) conditions?

Just a wondering bricklayer.

Steve Knight #855
Knight of BAAWA


Michael Cooper

unread,
Jul 11, 1999, 3:00:00 AM7/11/99
to

z@z <z...@z.lol.li> wrote in message news:7mb5mm$j6p$1...@pollux.ip-plus.net...

> In Joe's ENIGMAs-mega-thread there has been a very interesting
> discussion on the occurence of mutations in bacteria. The defenders
> of neo-Darwinian orthodoxy could not provide convincing evidence
> that adaptive mutations are totally random with respect to function.

Come on this is basic genetics....

> It is generally acknowledged that mutation rates are very different
> within a single genome. There are genes such as cytochrome-c and
> ubiquitin with a mutation rate of almost zero. Also within genes,
> regions with high mutation rates are confronted with highly
> conserved regions.

> As a general rule we can state that mutation rates are the higher,

Fun facts about molecular biology....

There are 4 bases or "letters" in the DNA code.
Each codon or "word" in the DNA code is made up of 3 bases or "letters"
Thus the number of codons or "words" is 64 (4*4*4=64)
There are roughly 22 amino acids in use in cells. (I can't be bothered
checking this but I know I am in the right ballpark)
So that means that there are 42 spare codons or "words" doesn't it?
No it doesn't. For starters there is a "START" codon meaning the start of a
polypeptide (chain of amino acids) and several "STOP" codons meaning the end
of a polypeptide.
So what about the other 40 or so?
The gentic code has a wonderful property called redundency. This means that
there is more then one way to spell a particular "word" and have it mean the
same thing. Heres an example.

The amino acid arganine has 6 diffrent ways of "spelling" it.
(lets ignore what the letters mean for a minute...)
CGU
CGC
CGA
CGG
AGA
AGG

Lets say that the word CGA is used at a spot in the genetic code. If a
mutation occurs and the last letter is changed then as you can see it will
still be one of the same words and there will be no functional change. If
the second letter is changed then the meaning of the word has changed. If
the first letter changes then there is a chance that it will still have the
same meaning...

Ken Cox

unread,
Jul 12, 1999, 3:00:00 AM7/12/99
to
z@z wrote:
> In Joe's ENIGMAs-mega-thread there has been a very interesting
> discussion on the occurence of mutations in bacteria. The defenders
> of neo-Darwinian orthodoxy could not provide convincing evidence
> that adaptive mutations are totally random with respect to function.

Other than describing, in some detail, the experiments that show this.
Several people even provided references to the literature.

Or are you using "convincing" to mean "they couldn't get God Herself
to come down and tell me they're right, so I refuse to believe it"?

--
Ken Cox k...@research.bell-labs.com


Adam Noel Harris

unread,
Jul 12, 1999, 3:00:00 AM7/12/99
to
z@z <z...@z.lol.li> wrote:
:In Joe's ENIGMAs-mega-thread there has been a very interesting
:discussion on the occurence of mutations in bacteria. The defenders
:of neo-Darwinian orthodoxy could not provide convincing evidence
:that adaptive mutations are totally random with respect to function.

We cited experiments that show this is the case.

:It is generally acknowledged that mutation rates are very different


:within a single genome. There are genes such as cytochrome-c and
:ubiquitin with a mutation rate of almost zero. Also within genes,
:regions with high mutation rates are confronted with highly
:conserved regions.
:
:As a general rule we can state that mutation rates are the higher,
:the lower the functional constraints are. But whereas mutations
:occur in the DNA, the functional constraints are on proteins. That's
:really enigmatic!

Where do you get this "general rule?" Are you really talking about
mutation rates, or are you talking about substitution rates (where
mutations go to near-fixation in the population)?

:Furthermore a number of experiments have shown an increased mutation


:probability when cells are grown under stress (e.g., with low levels
:of an antibiotic). It would be interesting to know how the very
:different mutation rates of a single genome are affected by this
:increased probability.

Yeah, that sounds interesting. I would hazard to guess that the rates go
up somewhat proportionally to the initial rates, since the increased rates
seem to be a result of reduced proofreading.

:Around 70 percent of the mutations of the third position of a DNA


:triplet are silent. According to Motoo Kimura (Scientific American,
:11/1979) such positions seem to be especially variable. If they
:actually have a higher mutation rate than the first two positions
:of triples then neo-Darwinism has a big problem.

Oh good. So do they?

:There are cases where bacterial resistance to an antibiotic depends


:on one single point mutation. It is very suprising how easy it is for
:an enzyme that is inhibited by an antibiotic to become an enzyme that
:hydrolyzes that antibiotic.
:
:One single point mutation and the enzymes defends itself by attacking
:the aggressor!
:
:One single point mutation and the enzyme carries out in addition to
:its normal tasks the task of hydrolyzing its own aggressor!
:
:One single point mutation resolving all biochemical problems involved
:in adding such a complex function to a protein without spoiling its
:normal functions!
:
:To explain all this by a lucky conformational change resulting from
:the substitution of one amino acid by another defies common sense,
:logical reasoning and simple probability estimates.

Well, let's see some structure-function data first. Do you have any?

[sig]
:"DNA mutations do not occur everywhere with the same freqency.


:Otherwise a fatal mutation in a vital protein would be much more
:probable than a mutation e.g. having an effect on the length of the
:neck. Moreover, evolutionarily older sequences (e.g. ubiquitin) are
:less susceptible to mutations. However, according to reductionism
:the probability of mutations should not depend on the evolutionary
:age or on the effects of the mutations."
:http://members.lol.li/twostone/E/psychon.html#a06

What evidence do you have that evolutionarily older sequences are less
susceptible to mutations? I hope you aren't using the conservation of
sequences to suggest lower mutation rates. That would be a big blunder.
So what _are_ you using?

-Adam
--
Opinions expressed are not necessarily those of Stanford University.
PGP Fingerprint = C0 65 A2 BD 8A 67 B3 19 F9 8B C1 4C 8E F2 EA 0E


Ian Musgrave & Peta O'Donohue

unread,
Jul 12, 1999, 3:00:00 AM7/12/99
to
G'Day All
Address altered to avoid spam, delete RemoveInsert

Courtesy copy also posted to Wolfgang due to the late reply to this
(and various weirdness at my server and post propagation)

On 11 Jul 1999 18:16:36 -0400, "z@z" <z...@z.lol.li> wrote:

>In Joe's ENIGMAs-mega-thread there has been a very interesting
>discussion on the occurence of mutations in bacteria. The defenders
>of neo-Darwinian orthodoxy could not provide convincing evidence
>that adaptive mutations are totally random with respect to function.

Which of the cited experiments do you not find convincing?

>It is generally acknowledged that mutation rates are very different
>within a single genome.

What Wolfgang is confusing here is the the rate at which mutations
occur in the genome, and the proportion of those mutations that are
benifical/neutral vs deleterious and enter into the population after
_selection_ (the substitution rate).

>There are genes such as cytochrome-c and
>ubiquitin with a mutation rate of almost zero.

No, these are _substitution_ rates, rather than mutation rates
themselves. You are comparing the sequences of different existing
organisms (rather than measuring the mutation rate in any given DNA
copying event). This is the combined effect of mutation rate AND
selection pressure. The substitution rates of these genes must be
compared to the substition rate of pseudogenes (copies of genes which
are no longer expressed), where there is no selection constraint, and
the substitution rate is equivalent to the mutation rate. The
substitution rate for cytochrome C is far from zero, and is a bit
under half that of pseudogenes, which is compatible with the average
of 7 neutral substitutions per site determined from Yockeys
calculations.

The substition rate of ubiqutin is very low, but still not almost
zero.

>Also within genes,
>regions with high mutation rates are confronted with highly
>conserved regions.

Again, these are _substitution_ rates, which are the end result of
mutation AND selection (or neutral drift).

Not suprisingly, regions such as active sites have more functional
constraints than simple structural regions for example. The
hydrophobic core of Barnase, an RNAse enzyme, can be replaced by
scrambled peptides just so long as the core stays roughly the same
size.

So although the mutation rate is relatively constant, and the
mutational probability of any given nucleotide roughly equivalent to
any other in any sequence mutations that inactivate or reduce the
function of any given expresssed protein will be selected out, so that
the subsitution rate in functionally constrained regions will be
slower than those with no functional constraint.

>As a general rule we can state that mutation rates are the higher,
>the lower the functional constraints are.

That's because you are looking at sequence comparisons form existing
organisms, which shows the effects of selection (and neutral drift)
not the actual mutation rate.

>But whereas mutations
>occur in the DNA, the functional constraints are on proteins. That's
>really enigmatic!

Not at all, the protein is the phenotype that is acted on by
selection.

>Furthermore a number of experiments have shown an increased mutation
>probability when cells are grown under stress (e.g., with low levels
>of an antibiotic).

Actually the typical example is due to metabolic stress, eg
starvation. There are two different effects here. In some organisms,
stress turns off (or severely inhibits) the proof reading mechanisms
which normally reduces copying errors. In others, the genes are
"overtranscribed", (more copies are made than usual) and the coding
DNA has a higher probability of damage during this process.

>It would be interesting to know how the very
>different mutation rates of a single genome are affected by this
>increased probability.

Even in these examples, the probability of a given base mutating is
roughly the same as any other base.

>Around 70 percent of the mutations of the third position of a DNA
>triplet are silent. According to Motoo Kimura (Scientific American,
>11/1979) such positions seem to be especially variable. If they
>actually have a higher mutation rate than the first two positions
>of triples then neo-Darwinism has a big problem.
>

>There are cases where bacterial resistance to an antibiotic depends
>on one single point mutation. It is very suprising how easy it is for
>an enzyme that is inhibited by an antibiotic to become an enzyme that
>hydrolyzes that antibiotic.

This shows the perils of ignoring chemistry. This example seems to be
a confusion between the mutations that convert D-Ala-DAla-ligases into
beta-lactamases (penicillin resistance), and the single point mutation
than converts a D-Ala-D-Ala ligase to a D-Ala-D-Lactate ligase (VanA
Vancomycin resistance)[1].

But it is instructive to consider the actual biochemistry involved.

To start with the reaction catalysed by any enzyme is reversible
(although the equilibrium usually favors one direction), and the
D-Ala-D-Ala ligases will also cleave D-Ala-D-Ala sequences.

Kforward
D-Ala + D-Ala + Enz <--> D-Ala-D-Ala + Enz
Kback

So it is not too dificult to imagine that a single active site
mutation could reverse the equilibrium by favoring the back reaction.

You may imagine the enzyme active site in the above reaction as having
a few amino acids to act as a vice, and one to act as a "welder"[2]
(this is an over simplification, but not too far from the truth). It
takes very little imagination to realise that by substituting the
amino acid acting as the "welder" with another one of not too
different properites that acts as a "cutter" (serine is a common
"cutter amino acid in enzymes). (Alternatively, you might greatly
increase the affinity of the active site for the dipeptide vs the
amino acids, this would also force the reaction backwards)

In fact this is what happens in VanX (another vancomycin resistance
mutant) where a G237D mutation in the D-Ala-D-Ala ligase converts it
into a D-Ala-D-Ala peptidase.

Also, you can change the chemistry of the active site. Changing the
residue at site 261 from phenylalanine to tryosine changes the enzyme
from a D-Ala-D-Ala ligase to a D-Ala-D-Lactate ligase (Van A
vancomycin resistance) because of the chemical nature phe vs Try.
Similarly, changing leucine 282 to arginine changes a D-Ala-D-Ala
ligase to a D-Ala-D-Ser ligase (Van C vancomycin resistance).

In the case of penicllin resistance, a D-Ala-D-Ala ligase is converted
to a beta-lactamase. The penicillin molecule mimics D-Ala and is bound
by the active site, but the beta lactam ring prevents any further
reaction. It is not to hard to imagine a change analogous to the F261Y
mutation that will now attack the ring (In the case of beta
lactamases, it is serine 70 which has the primary role in breaking the
ring).

>One single point mutation and the enzymes defends itself by attacking
>the aggressor!

Single point mutations can radically change the chemistry of the
active site (for example by converting a nucleophilic attack to a
electrophillic attack) because of the chemical properties of the
subsituted amino acid. There are several examples of this, the best
deacribed being enzymes in the enolase superfamily, where single site
mutations can change the chemistry of the active site quite markedly.

>One single point mutation and the enzyme carries out in addition to
>its normal tasks the task of hydrolyzing its own aggressor!

In this case, no, in the case of beta-lactamases and VanA and VanX,
these mutations occur in _duplicates_ of the D-Ala-D-Ala ligase (gene
duplication is reasonably common). VanC has 500 fold less D Ala-D-Ala
ligase activity vs D Ala-D-Ser ligase activity.

>One single point mutation resolving all biochemical problems involved
>in adding such a complex function to a protein without spoiling its
>normal functions!

This is not relevant to the beta-lactamase/VanA example. An example of
this kind of mutation is Halls experimental evolution of
galactosidease activity, where existing metabolic functions of the
mutant enzyme were not comprimised.

>To explain all this by a lucky conformational change

It's not a "lucky" conformational change, but a predictable
consequence of the change in the chemistry of the active site by
changing the chemical entities in it.

>resulting from
>the substitution of one amino acid by another defies common sense,
>logical reasoning and simple probability estimates.

But it doesn't defy chemistry. Yes, simple common sense can be
confounded by the intricacies of chemistry, but that is to be
expected. The importance of single amino acid substitutions on the
fundamental chemistry of the active site is expected on theoretical
grounds, observed in experimetal mutations and seen in the wild.

>Wolfgang


>
>"DNA mutations do not occur everywhere with the same freqency.

Yes they (mostly[3]) do. This has been experimentally determined.

>Otherwise a fatal mutation in a vital protein would be much more
>probable than a mutation e.g. having an effect on the length of the
>neck. Moreover, evolutionarily older sequences (e.g. ubiquitin) are
>less susceptible to mutations.

No, they are _not_ less susceptible to mutation. I must emphasise
again that you have confused the substitution rate (derived from
comparison of existing ubiquitin sequences), with the mutation rate.
Non-viable/less viable mutants have been selected out in the
comparison of existing organisms.

>However, according to reductionism
>the probability of mutations should not depend on the evolutionary
>age or on the effects of the mutations."
>http://members.lol.li/twostone/E/psychon.html#a06

And they don't, this is bourne out by experimental evidence.

[1] Penicillin acts by binding to the D-Ala-D-Ala ligase active site,
preventing the formation of the peptide chains tha will form the
bacterial cell wall. Beta lactamases break down penicillin and prevent
it from binding to D-Ala-D-Ala ligases.

Vancomycin binds to D-Ala-D-Ala, preventing the peptide chain from
binding to the cell wall. Chains ending in D-Ala-D-Lactate of
D-Ala-D-Ser (produced by Van A and Van C) will not bind vancomycin and
insert into the cell wall. Van X will chop up any D-Ala-D-Ala chains
produced so that they can be recycled into D-Ala-D-Ser/Lactate chains.

[2] In most enzymes the "welder" amino acid swings in close proximity
to the two substrates to be joined, forms a reaction intermediate with
them and is released at the end of the reaction after the substrates
have bonded. A similar sequnce of reactions occurs with "cutter" amino
acids such as serine, but starts with the polymer forming a reaction
intermediate with the cutter.

[3] A-T conversions are more common than other point mutations,
because of the chemical stability of A and T. A flanked by long G
sequences tend have higher mutation rates than A in a normal sequence.
However, ther is no bias in wheter mutations occur in functional vs
no-functional sequences.

References:

Joseph Petrosino, Carlos Cantu, III and Timothy Palzkill.
b-Lactamases: protein evolution in real time. Trends Micro 6, 323-327.
1998

Hasson MS, Schlichting I, Moulai J, Taylor K, Barrett W, Kenyon GL,
Babbitt PC, Gerlt JA, Petsko GA, and Ringe D. (1998 Sep 1). Evolution
of an enzyme active site: the structure of a new crystal form of
muconate lactonizing enzyme compared with mandelate racemase and
enolase. Proc Natl Acad Sci U S A , 95, 10396-401.

Arthur M, Depardieu F, Cabanie L, Reynolds P, and Courvalin P. (1998
Nov). Requirement of the VanY and VanX D,D-peptidases for glycopeptide
resistance in enterococci. Mol Microbiol , 30, 819-30.

Healy VL, Park IS, and Walsh CT. (1998 Apr). Active-site mutants of
the VanC2 D-alanyl-D-serine ligase, characteristic of one
vancomycin-resistant bacterial phenotype, revert towards wild-type
D-alanyl-D-alanine ligases. Chem Biol , 5, 197-207.

Babbitt PC, and Gerlt JA. (1997 Dec 5). Understanding enzyme
superfamilies. Chemistry As the fundamental determinant in the
evolution of new catalytic activities. J Biol Chem , 272, 30591-4.

Park IS, and Walsh CT. (1997 Apr 4). D-Alanyl-D-lactate and
D-alanyl-D-alanine synthesis by D-alanyl-D- alanine ligase from
vancomycin-resistant Leuconostoc mesenteroides. Effects of a
phenylalanine 261 to tyrosine mutation. J Biol Chem , 272, 9210-4.

Park IS, Lin CH, and Walsh CT. (1996 Aug 13). Gain of
D-alanyl-D-lactate or D-lactyl-D-alanine synthetase activities in
three active-site mutants of the Escherichia coli D-alanyl-D-alanine
ligase B. Biochemistry , 35, 10464-71.

Hall BG. (1982 Jul-Aug). Evolution of a regulated operon in the
laboratory. Genetics , 101, 335-44.

Hall BG. (1978 Jul). Experimental evolution of a new enzymatic
function. II. Evolution of multiple functions for ebg enzyme in E.
coli. Genetics , 89, 453-65.

Cheers! Ian
=====================================================
Ian Musgrave Peta O'Donohue,Jack Francis and Michael James Musgrave
reyn...@werple.mira.net.au http://werple.mira.net.au/~reynella/
a collection of Dawkins inspired weasle programs http://www-personal.monash.edu.au/~ianm/whale.htm
Southern Sky Watch http://www.abc.net.au/science/space/default.htm


howard hershey

unread,
Jul 13, 1999, 3:00:00 AM7/13/99
to
Adam Noel Harris wrote:
>
> z@z <z...@z.lol.li> wrote:
> :In Joe's ENIGMAs-mega-thread there has been a very interesting
> :discussion on the occurence of mutations in bacteria. The defenders
> :of neo-Darwinian orthodoxy could not provide convincing evidence
> :that adaptive mutations are totally random with respect to function.
>
> We cited experiments that show this is the case.
>
> :It is generally acknowledged that mutation rates are very different
> :within a single genome. There are genes such as cytochrome-c and
> :ubiquitin with a mutation rate of almost zero. Also within genes,

> :regions with high mutation rates are confronted with highly
> :conserved regions.
> :
> :As a general rule we can state that mutation rates are the higher,
> :the lower the functional constraints are. But whereas mutations

> :occur in the DNA, the functional constraints are on proteins. That's
> :really enigmatic!
>
> Where do you get this "general rule?" Are you really talking about
> mutation rates, or are you talking about substitution rates (where
> mutations go to near-fixation in the population)?

Let me add an amen to that. You really are confusing substitution rates
with mutation rates thoughout.
>
> :Furthermore a number of experiments have shown an increased mutation


> :probability when cells are grown under stress (e.g., with low levels

> :of an antibiotic). It would be interesting to know how the very


> :different mutation rates of a single genome are affected by this
> :increased probability.
>

> Yeah, that sounds interesting. I would hazard to guess that the rates go
> up somewhat proportionally to the initial rates, since the increased rates
> seem to be a result of reduced proofreading.

And some kinds of mutation are more likely to occur in functionally
active genes (those that are actively transcribing) than in functionally
inactive genes.
>
> :Around 70 percent of the mutations of the third position of a DNA


> :triplet are silent. According to Motoo Kimura (Scientific American,
> :11/1979) such positions seem to be especially variable. If they
> :actually have a higher mutation rate than the first two positions
> :of triples then neo-Darwinism has a big problem.

Again, this is a confusion of substitution rates with mutation rates.
The third letter of a codon has the same mutation rate but a higher
substitution rate (except for methionine and tryptophan, of course).
Perhaps you can tell me why I made those exceptions? If you can, you
are well on the way to understanding the difference between the third
and first and second letters of a codon).


>
> Oh good. So do they?
>

> :There are cases where bacterial resistance to an antibiotic depends


> :on one single point mutation. It is very suprising how easy it is for
> :an enzyme that is inhibited by an antibiotic to become an enzyme that
> :hydrolyzes that antibiotic.

> :
> :One single point mutation and the enzymes defends itself by attacking
> :the aggressor!
> :
> :One single point mutation and the enzyme carries out in addition to


> :its normal tasks the task of hydrolyzing its own aggressor!

> :
> :One single point mutation resolving all biochemical problems involved


> :in adding such a complex function to a protein without spoiling its
> :normal functions!

> :
> :To explain all this by a lucky conformational change resulting from


> :the substitution of one amino acid by another defies common sense,
> :logical reasoning and simple probability estimates.

Why? What do you need to both hydrolyze a molecule and to be inhibited
by a molecule? Ans: You need to bind the molecule. A protein that can
do that is on its way (preadapted) to either function, is it not?
Hydrolysis involves putting extra strain on a particular chemical bond.


>
> Well, let's see some structure-function data first. Do you have any?
>
> [sig]

> :"DNA mutations do not occur everywhere with the same freqency.
> :Otherwise a fatal mutation in a vital protein would be much more


> :probable than a mutation e.g. having an effect on the length of the
> :neck. Moreover, evolutionarily older sequences (e.g. ubiquitin) are

> :less susceptible to mutations. However, according to reductionism


> :the probability of mutations should not depend on the evolutionary
> :age or on the effects of the mutations."
> :http://members.lol.li/twostone/E/psychon.html#a06

This is, again, a confusing of substitution rate and mutation rate.
What makes you think that ubiquitin is less susceptible to mutation than
any other gene?

z@z

unread,
Jul 13, 1999, 3:00:00 AM7/13/99
to
Hi Adam Noel Harris!

>> In Joe's ENIGMAs-mega-thread there has been a very interesting
>> discussion on the occurence of mutations in bacteria. The defenders
>> of neo-Darwinian orthodoxy could not provide convincing evidence
>> that adaptive mutations are totally random with respect to function.

> We cited experiments that show this is the case.

One cannot refute the possibility of directed mutations by a few
experiments whose results allow a random mutation interpretation.

Some of the cited experiments only deal with the question whether
mutations leading e.g. to phage resistance occur only after exposure
to phages or whether these mutations had occured before exposure.
But they do not deal with the question, whether the presence of
phages could make the appearance of such mutations more likely.

Only in the case of single point mutations a reasonable mathematical
model was provided. A simplified version (ignoring some details)
is the following:

o 10^7 different point mutations are possible in a bacterium
o The probability of the appearence of one point mutation
during replication is 0.1

Under these conditions, during 10^10 replications every possible
point mutation occurs around 100 times, and therefore one gets
around 100 resistent bacteria with the desired mutation.

10^10 replications is not a big quantity, 10^20 however is.

That all or most possible point mutations are regularly tried
out with more or less same probability seems to me merely an
unjustified assumption.

If in the above example two point mutations were necessary
for resistance of appear then already more than 10^16
cell divisions would be necessary to find it that way.

Apart from point mutations there are also insertions,
deletions and translocations which even can result in frame
shifts.

In a genome consisting of 10^7 base pairs there are each
10^7 possibilities to cut out one, two, three or more base
pairs. There are 4 x 10^7 possiblities to insert one base
pair, 4^2 x 10^7 to insert two and 4^n x 10^7 possibilities
to insert n base pairs. 20^10 x 10^7 = 10^20 possibilities
exist for the insertion of a sequence of ten amino-acids.

"Hall began working with a system that required two
mutations to occur before E. coli could use the sugar
salicin. One of the mutations, he documented in a 1988
Genetics article, is so rare in growing cells that he
failed to detect it. But plated on salicin, Sal+ revertants
appeared after a delay of about 12 days. Their frequency
was orders of magnitude greater than expected if the two
mutations were occurring independently and spontaneously.

Furthermore, the rarer spontaneous mutation happened first.
This mutation was excision of an insertion sequence. It
was followed relatively quickly by the second mutation,
which only then permitted transcription. When salicin was
absent, Sal+ revertants had no advantage, and Hall was
unable to detect excision mutants."
http://www.aaas.org/spp/dspp/dbsr/EVOLUT/goodman.htm (1992)

Another example from the same article:

"Shapiro's construct had an arabinose regulatory gene
followed by an arabinose structural gene into which was
inserted a Mu prophage. ... This arrangement, in which the
strain could grow on neither arabinose nor lactose, is
designated Lac(Ara)-. However, appropriate excision of the
Mu element fused the initial region of the ara gene to the
beginning of the lacZ gene, allowing the bacterium to grow
as long as lactose and arabinose are present.

Shapiro demonstrated that ara-lac fusions occurred only
after plating on a selective medium, with both arabinose and
lactose, and only after a long delay. Cairns confirmed these
results." ...

"Mittler and Lenski therefore designed an experiment to focus
on the effect of starvation on the rate of Mu excisions. ...
After controlling for cell death and growth, Mittler and
Lenski concluded that the increased rate of excision of Mu
elements could be attributed entirely to the stress of
starvation. Among starving cells excision mutations occurred
at comparable rates in the presence or absence of lactose
and arabinose; there was no need to invoke directed mutation
to explain the excisions."

But isn't it astonishing that stress leads just to the excision
of the 'non-native' Mu-sequence? How probable is an excision of
an Mu-element-long sequence? Has anybody shown that all the
other possible excisions of varions lengths and locations also
take place with a similar probability during starvation? If
not, Mittler and Lenski have provided rather evidence for
directed mutations (stress-induced restoration of a fully
functioning genome) than for random stress-induced mutations.

Hall's hypermutable-state model:

"He suggested that during prolonged starvation, some cells
in a colony enter a hypermutable state. In this condition,
both extensive random DNA damage and error-prone repair
occur. By chance, an error introduced into the DNA during
this process might solve the problem facing the cell at the
time. The cell then would replicate its DNA and stabilize
the advantageous mutation. The cell then would exit the
hypermutable state."

Model tested 1991 by Hall himself:

"He measured the reversion to trp+ of a strain that
was doubly mutant trpA- trpB-. He found double mutants
(revertants to tryptophan prototrophy) approximately 100
million times more frequently than expected if the
mutations were independent." ...

"Hall wondered whether such a high rate might apply only to
a small region of the genome around the trpA and trpB loci,
so he sequenced 700 base pairs around the two loci in 11
double revertants. The hypermutable-state model predicts that
other base substitutions should occur among those 700 base
pairs. He failed to find any, ..."

It is always possible to explain away the "apparent" increased
probability of advantageous mutations by some ad-hoc-hypothesis.

"Hall wrote that the problem 'is to explain how selective
conditions could increase the frequency of useful mutations
without increasing the frequency of mutations at other loci.'"

I would even say that the random mutation hypothesis is almost
unfalsifiable. A central principle of the psychon theory,
environment continuity, suggests a higher probability for
an adapive mutation (not representing an innovation) in a
corresponding environment. If e.g. the probability for a
mutation to antibiotic resistence is actually increased from
10^-10 to 10^-8 by the presence of a non-lethal amount of the
antibiotic, this increase will simply be explained by random
stress-induced errors.

The psychon theory also can explain why genes, even in the
case of large populations, do not drift apart into uncountable
alleles but rather converge to a few or even to one single
allele. It's not selection, it's MUTATION!


Cheers, Wolfgang

The psychon theory:
http://members.lol.li/twostone/E/psychon.html

Ken Cox

unread,
Jul 13, 1999, 3:00:00 AM7/13/99
to
z@z wrote:
> One cannot refute the possibility of directed mutations by a few
> experiments whose results allow a random mutation interpretation.

True. One also cannot refute the possibility that the planets are
pushed around by invisible and undetectable giant scarab beetles
that just happen to exactly follow the paths predicted by gravity.

However, since we have gravitational theory, which suffices to
explain all the observed motions of the planets, someone trying
to sell the beetle theory would have to come up with some positive
evidence for it, like a way to detect the beetles. For that matter,
it would help if that person could at least find an observation that
is better explained by the beetles than by current theory.

The directed mutation theory is in somewhat the same boat. The
theory that mutations are random explains the observations (such
as the probability that a bacterial culture develops antibiotic
resistance, for several variations on the culturing and exposure
protocols) to a high degree of precision. So just to even get
started, you need to find an an observation that directed mutation
explains better than does random mutation.

*Some* of the experiments that have been described *may* be such
observations that need explaining. However, they are still rather
preliminary; some have not even been replicated, and some others
are explainable simply as a result of the mutation rate increasing
when cells are under stress.

And again, such experiments are at best indications that *some*
modification of the random-mutation theory may be needed. To show
that directed modification is the *specific* modification that is
correct, you need to isolate the mechanism itself and show how it
works. That is, where is the little bit of protoplasm that analyzes
a new antibiotic, determines that it is harmful to the cell, designs
a protein to destroy the antibiotic, searches the genome for a site
that can be mutated to produce the protein, and causes the mutation?
And how does it manage to carry out these computationally-difficult
feats while remaining unobserved most of the time?


[snip the rest of the post, in which for some inexplicable reason
Wolfgang essentially says, "I don't like the actual measurements
that have been done on mutation rates. Let me make up some other
numbers that are better for my case."]

--
Ken Cox k...@research.bell-labs.com


Michael Konstantine Kalandros

unread,
Jul 13, 1999, 3:00:00 AM7/13/99
to
"z@z" wrote:

>
> I would even say that the random mutation hypothesis is almost
> unfalsifiable. A central principle of the psychon theory,
> environment continuity, suggests a higher probability for
> an adapive mutation (not representing an innovation) in a
> corresponding environment. If e.g. the probability for a
> mutation to antibiotic resistence is actually increased from
> 10^-10 to 10^-8 by the presence of a non-lethal amount of the
> antibiotic, this increase will simply be explained by random
> stress-induced errors.

Actually, that's not true, but the experiment is rather tough. Look
at the entire genome of each offspring. Record the mutations on each.
Under the psychon theory, each offspring should have the antibiotic
resistance mutation more often than not. Under the random mutation
hypothesis, only a relatively few will have that mutation. This must be

done in each generation (select a subsample to make it easier) since
those with the beneficial mutation will eventually dominate the
population, masking the difference between the two theories.

Mike


howard hershey

unread,
Jul 13, 1999, 3:00:00 AM7/13/99
to

Not completely. Many temperate phages of the lambda type regularly are
induced into lytic growth (following excision) by environmental stress.

> How probable is an excision of
> an Mu-element-long sequence?

Do you know how the mu virus works? Excision and integration are a
normal part of its life cycle.

> Has anybody shown that all the
> other possible excisions of varions lengths and locations also
> take place with a similar probability during starvation?

Like point mutations, there are excision 'hot-spots'. I work with a
plasmid in a teaching lab that we use to generate 'random' mutation in
the lac operon on the plasmid. Invariably these spontaneous lac
mutations essentially all have a deletion that has one end in a sequence
no bigger than 50 nucleotides. The other end varies, but there are some
deletions that are more common than others.

And you would be wrong.

> A central principle of the psychon theory,
> environment continuity, suggests a higher probability for
> an adapive mutation (not representing an innovation) in a
> corresponding environment. If e.g. the probability for a
> mutation to antibiotic resistence is actually increased from
> 10^-10 to 10^-8 by the presence of a non-lethal amount of the
> antibiotic, this increase will simply be explained by random
> stress-induced errors.
>
> The psychon theory also can explain why genes, even in the
> case of large populations, do not drift apart into uncountable
> alleles but rather converge to a few or even to one single
> allele. It's not selection, it's MUTATION!

The psychon theory is unnecessary. Drift is sufficient to explain why
most genese only exhibit a single or few neutral alleles.

z@z

unread,
Jul 14, 1999, 3:00:00 AM7/14/99
to
Hello Ken Cox!


| z@z wrote:
|
| > One cannot refute the possibility of directed mutations by a few
| > experiments whose results allow a random mutation interpretation.
|

| True. One also cannot refute the possibility that the planets are
| pushed around by invisible and undetectable giant scarab beetles
| that just happen to exactly follow the paths predicted by gravity.
|
| However, since we have gravitational theory, which suffices to
| explain all the observed motions of the planets, someone trying
| to sell the beetle theory would have to come up with some positive
| evidence for it, like a way to detect the beetles. For that matter,
| it would help if that person could at least find an observation that
| is better explained by the beetles than by current theory.

Some centuries ago you would have written (against Kepler):

One also cannot refute the possibility that the planets are

pushed around by invisible and undetectable mathematical
rules that just happen to exactly follow the paths predicted
by the epicycles.

However, since we know the epicycles, which suffices to explain


all the observed motions of the planets, someone trying to sell

the physical law theory would have to come up with some positive
evidence for it, like a way to concretely detect the physical
laws. For that matter, it would help if that person could at
least find an observation that is better explained by imaginary
laws than by concrete epicycles.

[snip]

| And again, such experiments are at best indications that *some*
| modification of the random-mutation theory may be needed.

There is never a limit on adding further epicycles!

[snip]

| [snip the rest of the post, in which for some inexplicable reason
| Wolfgang essentially says, "I don't like the actual measurements
| that have been done on mutation rates. Let me make up some other
| numbers that are better for my case."]

Do you really think that what you write here is a honest summary
of my post (http://www.deja.com/=dnc/getdoc.xp?AN=500641694)?


Cheers, Wolfgang

Joseph Potter

unread,
Jul 14, 1999, 3:00:00 AM7/14/99
to

z@z <z...@z.lol.li> wrote in message news:7mhqji$26a$1...@pollux.ip-plus.net...

> Hello Ken Cox!
>
>
> | z@z wrote:
> |
> | > One cannot refute the possibility of directed mutations by a few
> | > experiments whose results allow a random mutation interpretation.
> |
> | True. One also cannot refute the possibility that the planets are
> | pushed around by invisible and undetectable giant scarab beetles
> | that just happen to exactly follow the paths predicted by gravity.
> |
> | However, since we have gravitational theory, which suffices to
> | explain all the observed motions of the planets, someone trying
> | to sell the beetle theory would have to come up with some positive
> | evidence for it, like a way to detect the beetles. For that matter,
> | it would help if that person could at least find an observation that
> | is better explained by the beetles than by current theory.
>
> Some centuries ago you would have written (against Kepler):
>
> One also cannot refute the possibility that the planets are
> pushed around by invisible and undetectable mathematical
> rules that just happen to exactly follow the paths predicted
> by the epicycles.
>
> However, since we know the epicycles, which suffices to explain
> all the observed motions of the planets, someone trying to sell
> the physical law theory would have to come up with some positive
> evidence for it, like a way to concretely detect the physical
> laws. For that matter, it would help if that person could at
> least find an observation that is better explained by imaginary
> laws than by concrete epicycles.
>

One might also point out that working with the equations
of circles is a tad simpler than working with the equations
for the ellipse. Hence, several would call out the old Razor
to say we "don't need no stinking elliptical orbits."


Regards, Joe

Ken Cox

unread,
Jul 14, 1999, 3:00:00 AM7/14/99
to
Joseph Potter wrote:
> z@z <z...@z.lol.li> wrote:

> > Some centuries ago you would have written (against Kepler):
> > One also cannot refute the possibility that the planets are
> > pushed around by invisible and undetectable mathematical
> > rules that just happen to exactly follow the paths predicted
> > by the epicycles.

> One might also point out that working with the equations


> of circles is a tad simpler than working with the equations
> for the ellipse. Hence, several would call out the old Razor
> to say we "don't need no stinking elliptical orbits."

Zelda is perhaps correct in that I might have opposed Keplerian
ellipses. However, once it was clear that the orbits really are
ellipses, and once Newton showed how such orbits are generated
by a central force, I assure you I would have embraced them.

However, it is not true that the ellipses are harder than the
epicycles. What you must remember is that the epicycles are
really a series expansion of the ellipses -- you get *exactly
the same orbit* using a single ellipse as you do using a sum
of a hundred or so circles, each of different radii and period.
This makes the circles *much* harder to work on; the ellipses
requires only six parameters, while the circles require several
hundred (five parameters per circle).

Besides, the epicycles have the problem of the hubs. Where are
these points of attachment on which the circles rotate, and how
do they work? It's similar to the way the actual *mechanism*
whereby directed mutation -- the little chunk of protoplasm that
does the analysis -- remains missing no matter where we look.

--
Ken Cox k...@research.bell-labs.com


z@z

unread,
Jul 14, 1999, 3:00:00 AM7/14/99
to
Hi Ian!

[snip]

| >It is generally acknowledged that mutation rates are very different
| >within a single genome.
|
| What Wolfgang is confusing here is the the rate at which mutations
| occur in the genome, and the proportion of those mutations that are
| benifical/neutral vs deleterious and enter into the population after
| _selection_ (the substitution rate).

Howard Hershey, 7-Jul-99:

"Point mutations, for example, occur at a rate in the range
of 10^-4 to 10^-12 (modal about 10^-9) depending upon the
particular site (some sites are highly mutable (hot spots)
compared to others. For example, the point mutation that
causes achondroplasia has perhaps the highest known mutation
rate in the human genome (about 10^-6)."

If all point mutations were as likely as this one, humans
would have around 6'000 point mutations (or even more, if
achondroplasia depends on the substitution of the particular
site by a particular amino acid.)

But according to Richard Harter, "the average human being has
about 50-100 mutations, of which about 3 matter, i.e., they
actually change a protein."
http://www.tiac.net/users/cri/mutate.html


| >There are genes such as cytochrome-c and
| >ubiquitin with a mutation rate of almost zero.

Maybe, I should have been more precise:

There are genes such as cytochrome-c and ubiquitin with

a rate of non-silent mutations of almost zero.

But the problem of silent mutations has been specifically
addressed only three paragraphs later.

| No, these are _substitution_ rates, rather than mutation rates
| themselves. You are comparing the sequences of different existing
| organisms (rather than measuring the mutation rate in any given DNA
| copying event). This is the combined effect of mutation rate AND
| selection pressure.

When I write mutation rate then I mean mutation rate. We have
already had a discussion on the related problems. The most
convincing case is ubiquitin.
See: http://members.lol.li/twostone/E/deja2.html

Because of the huge number of animals it is impossible to explain
by genetic drift the fact that there is only one single allele of
ubiquitin in all animals. Drift only works for little populations,
but certainly not for all animals considered as one species.

The alternative that every mutation in animal ubiquitin is lethal
or harmful enough to disappear by selection seems much more
unreasonable to me than the assumption that (non-silent) mutations
are almost inexistent in ubiquitin. Also the existence of 'hot
spots' could suggest the existence of 'cold spots'.

Think about the evolution of mankind during the last million of
years. The mean age at death was certainly lower than today. So
we can assume that the average number of children per couple was
ten (the less, the better for my argument). We can further assume
that children had on average three mutations actually influencing
the phenotype and that fifty percent of the children died
independently of their genetic constitution.

If the probability that a mutation is harmful is 80 percent,
then it is very improbable that mankind could have further
evolved, because genetic selection could only act on five
children two of which had to survive.

Ian, if one takes into account all facts and tries to be
consistent, then one concludes that the many highly conserved
regions of genomes cannot be the result of continuous selection,
but must be the result of low rates of mutations.

| The substitution rates of these genes must be compared
| to the substition rate of pseudogenes (copies of genes which are
| no longer expressed), where there is no selection constraint, and
| the substitution rate is equivalent to the mutation rate. The
| substitution rate for cytochrome C is far from zero, and is a bit
| under half that of pseudogenes, which is compatible with the average
| of 7 neutral substitutions per site determined from Yockeys
| calculations.

According to my premises one cannot conclude from mutation or
substitution rates of pseudogenes to the rates of the actual
genes because functional constraints are very different.

| The substition rate of ubiqutin is very low, but still not almost
| zero.

It would be certainly very interesting and instructive to
examine in detail the question of different DNA-alleles of
ubiquitin and cytochrome-c.

[snip]

Cheers, Wolfgang

Matt Silberstein

unread,
Jul 14, 1999, 3:00:00 AM7/14/99
to
In talk.origins I read this message from "Joseph Potter"
<joe.p...@worldnet.att.net>:

[snip]


>
>One might also point out that working with the equations
>of circles is a tad simpler than working with the equations
>for the ellipse. Hence, several would call out the old Razor
>to say we "don't need no stinking elliptical orbits."
>

I agree that it is easier to work with circles than ellipses, but I
have never seen anything in the Razor that speaks of making things
easy to work. Nor do I see how circles are somehow simpler than
ellipses, in fact, I could make an argument that ellipses are simpler.
But this is irrelevant. The circle model does not predict
observations, the ellipse model does. Models that don't make accurate
predictions do not get submitted to the Razor, they get eliminated
before that.

Matt Silberstein
-----------------------------------------------------------------------
She used her body just like a bandage, she use my body just like a wound
I'll probably never know where she disappeared
But I can see her rising up from the back seat now just like an angel
rising up from a tomb

But it was long ago and it was far away, oh God it seems so very far
And if life is just a highway, then the soul is just a car

And objects in the rear view mirror may appear closer than they are.

J.S.


Ian Musgrave & Peta O'Donohue

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to
G'Day All
Address altered to avoid spam, delete RemoveInsert

On 14 Jul 1999 13:30:26 -0400, "z@z" <z...@z.lol.li> wrote:

>Hi Ian!
>
>[snip]
>
[snip mutation rate stuff to be dealt with in another post.]

>| >There are genes such as cytochrome-c and
>| >ubiquitin with a mutation rate of almost zero.
>

>Maybe, I should have been more precise:
>

> There are genes such as cytochrome-c and ubiquitin with

> a rate of non-silent mutations of almost zero.

Where does this figure come from, given that the substitution rate of
cytochrome-C is around 0.28 substitutions per site per 10^9 years (cf
the substitution rates of pesudogenes, not under selection pressure,
which are 1.3 substitutions/ site per 10^9 years [*])

This means that yeast should differ from humans in around 50 amino
acids out of 101, as it happens they differ in 46 amino acids (that is
there have been 46 amino acid _substitutions_). Again, this is NOT the
mutation rate, the substitution we see are the combined result of
mutation AND selection/neutral drift.

>But the problem of silent mutations has been specifically
>addressed only three paragraphs later.

Silent mutations are not the same as substitutions (another reality
check on mutation rates is to express the results as substitutions per
silent replacement)

>| No, these are _substitution_ rates, rather than mutation rates
>| themselves. You are comparing the sequences of different existing
>| organisms (rather than measuring the mutation rate in any given DNA
>| copying event). This is the combined effect of mutation rate AND
>| selection pressure.
>

>When I write mutation rate then I mean mutation rate. We have
>already had a discussion on the related problems. The most
>convincing case is ubiquitin.
>See: http://members.lol.li/twostone/E/deja2.html

But here you arn't quoting mutation rates, you are comparing
substition rates. (ie yeast ubiquitin differs from human ubiquitin in
3 amino acid positions, this gives a _substitution_ rate of around
0.005 substitutions per site per 10^9 years, yet again this is NOT the
mutation rate)

see http://x37.deja.com/=dnc/getdoc.xp?AN=480456524

for more detailed analysis of this case.

>Because of the huge number of animals it is impossible to explain
>by genetic drift the fact that there is only one single allele of
>ubiquitin in all animals. Drift only works for little populations,
>but certainly not for all animals considered as one species.

Despite you claim, drift works in arbitarily large populations, not
just small ones, try out this population simulator with large
populations and see what happens.
http://http.bsd.uchicago.edu/hgd-sad/HWSimulator/sim.cgi
to see why neutral drift will eliminate multiple allelles in any
arbitaryly sized population.

>The alternative that every mutation in animal ubiquitin is lethal
>or harmful enough to disappear by selection seems much more
>unreasonable to me than the assumption that (non-silent) mutations
>are almost inexistent in ubiquitin.

Why, have you done structure function analyses on ubiquitin and its
mutants?

>Also the existence of 'hot
>spots' could suggest the existence of 'cold spots'.

See Watson et al, "Molecular biology of the gene" 4th edition volume I
page 340-342 for the actual frequency distributions of mutation, no
cold spots.

>Think about the evolution of mankind during the last million of
>years. The mean age at death was certainly lower than today. So
>we can assume that the average number of children per couple was
>ten (the less, the better for my argument). We can further assume
>that children had on average three mutations actually influencing
>the phenotype and that fifty percent of the children died
>independently of their genetic constitution.

You could of course actually check this out with real demography.

>If the probability that a mutation is harmful is 80 percent,
>then it is very improbable that mankind could have further
>evolved, because genetic selection could only act on five
>children two of which had to survive.

Kimura suggests that it is more like 50%. You have also ignore the
60-80% rate of preimplantation failures in typical pregnancies, which
seem to be due to mutations.

>Ian, if one takes into account all facts and tries to be
>consistent, then one concludes that the many highly conserved
>regions of genomes cannot be the result of continuous selection,
>but must be the result of low rates of mutations.

Or you could do the calculations correctly.

>| The substitution rates of these genes must be compared
>| to the substition rate of pseudogenes (copies of genes which are
>| no longer expressed), where there is no selection constraint, and
>| the substitution rate is equivalent to the mutation rate. The
>| substitution rate for cytochrome C is far from zero, and is a bit
>| under half that of pseudogenes, which is compatible with the average
>| of 7 neutral substitutions per site determined from Yockeys
>| calculations.
>

>According to my premises one cannot conclude from mutation or
>substitution rates of pseudogenes to the rates of the actual
>genes because functional constraints are very different.

That's because pseudogenes reflect _only_ the mutation rate, where as
working genes represent _selection_ on top of mutation.

>| The substition rate of ubiqutin is very low, but still not almost
>| zero.
>

>It would be certainly very interesting and instructive to
>examine in detail the question of different DNA-alleles of
>ubiquitin and cytochrome-c.

I have the begining of such an analysis for ubiquitin at
http://x37.deja.com/=dnc/getdoc.xp?AN=480456524

Wesley Elsberry has a general one for cytochrome C.
http://inia.cls.org/~welsberr/evobio/evc/argresp/sequence.html

Joseph Potter

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to

Ken Cox <k...@lucent.com> wrote in message news:378C9F...@research.bell-labs.com...

> Joseph Potter wrote:
> > z@z <z...@z.lol.li> wrote:
>
> > > Some centuries ago you would have written (against Kepler):
> > > One also cannot refute the possibility that the planets are
> > > pushed around by invisible and undetectable mathematical
> > > rules that just happen to exactly follow the paths predicted
> > > by the epicycles.
>
> > One might also point out that working with the equations
> > of circles is a tad simpler than working with the equations
> > for the ellipse. Hence, several would call out the old Razor
> > to say we "don't need no stinking elliptical orbits."
>
> Zelda is perhaps correct in that I might have opposed Keplerian
> ellipses. However, once it was clear that the orbits really are
> ellipses, and once Newton showed how such orbits are generated
> by a central force, I assure you I would have embraced them.
>
> However, it is not true that the ellipses are harder than the
> epicycles. What you must remember is that the epicycles are
> really a series expansion of the ellipses -- you get *exactly
> the same orbit* using a single ellipse as you do using a sum
> of a hundred or so circles, each of different radii and period.
> This makes the circles *much* harder to work on; the ellipses
> requires only six parameters, while the circles require several
> hundred (five parameters per circle).
>

I once read an essay by Kune (sp?) where he talked of this very
issue. He said that in those days the precision necessary did
not take many circles and that the methods were well understood.
Hence, at least for a time, the ellipse was harder to work with.


Regards, Joe

Joseph Potter

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to

Matt Silberstein <mat...@ix.netcom.com> wrote in message
news:37925584...@nntp.ix.netcom.com...

> In talk.origins I read this message from "Joseph Potter"
> <joe.p...@worldnet.att.net>:
>
> [snip]
> >
> >One might also point out that working with the equations
> >of circles is a tad simpler than working with the equations
> >for the ellipse. Hence, several would call out the old Razor
> >to say we "don't need no stinking elliptical orbits."
> >
> I agree that it is easier to work with circles than ellipses, but I
> have never seen anything in the Razor that speaks of making things
> easy to work. Nor do I see how circles are somehow simpler than
> ellipses, in fact, I could make an argument that ellipses are simpler.
> But this is irrelevant. ...
>

Ever try to find out how much area is swept out in a unit
of time as a particle travels in the orbit elliptically verses
one that travels in a circular orbit?

I seem to recall that a fellow named Newton had to deal with this
problem and had to come up with something called Calculus.

(of course a fellow in France claimed to have beat
him to it, but that's life)


>
>The circle model does not predict

> observations, the ellipse model does. ...
>

The point is that scientists at that time thought the model
worked well. It did allow for predictions that "wowed
the King," and kept everyone employed.

>
> ... Models that don't make accurate


> predictions do not get submitted to the Razor, they get eliminated
> before that.
>

I can not recall exactly, but I think that the circular orbit
bit was in place for quite some time. I also recall that
there was some resistance to going to the ellipse as
a model.

I am glad we did go to the ellipse, of course. Otherwise
I think NASA might have had even more trouble than
they did in the early days.


Regards, Joe

Matt Silberstein

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to
In talk.origins I read this message from "Joseph Potter"
<joe.p...@worldnet.att.net>:

>
>Matt Silberstein <mat...@ix.netcom.com> wrote in message
>news:37925584...@nntp.ix.netcom.com...
>> In talk.origins I read this message from "Joseph Potter"
>> <joe.p...@worldnet.att.net>:
>>
>> [snip]
>> >
>> >One might also point out that working with the equations
>> >of circles is a tad simpler than working with the equations
>> >for the ellipse. Hence, several would call out the old Razor
>> >to say we "don't need no stinking elliptical orbits."
>> >
>> I agree that it is easier to work with circles than ellipses, but I
>> have never seen anything in the Razor that speaks of making things
>> easy to work. Nor do I see how circles are somehow simpler than
>> ellipses, in fact, I could make an argument that ellipses are simpler.
>> But this is irrelevant. ...
>>
>
>Ever try to find out how much area is swept out in a unit
>of time as a particle travels in the orbit elliptically verses
>one that travels in a circular orbit?
>
>I seem to recall that a fellow named Newton had to deal with this
>problem and had to come up with something called Calculus.
>
>(of course a fellow in France claimed to have beat
>him to it, but that's life)
>

I fail to see the relevance of this. Newton did not develop the
Calculus to figure out the area swept, he developed it (partially) to
figure out the result of an attractive force and to show that a sphere
would act as a point source.

But that says nothing about whether an ellipse is a simpler or more
complicated model than a circle.

>
>>
>>The circle model does not predict
>> observations, the ellipse model does. ...
>>
>
>The point is that scientists at that time thought the model
>worked well. It did allow for predictions that "wowed
>the King," and kept everyone employed.
>

Which time do you mean? Copernicus? Kepler? Newton? And do you really
claim that the scientists made these claims just to wow the King and
keep employed? I had thought your point had something to do with the
Razor, not with the lack of ethics in scientific behavior. Yes,
Kepler's model worked well, it worked much better and "simpler" than
the previous models.

>> ... Models that don't make accurate
>> predictions do not get submitted to the Razor, they get eliminated
>> before that.
>
>I can not recall exactly, but I think that the circular orbit
>bit was in place for quite some time. I also recall that
>there was some resistance to going to the ellipse as
>a model.
>

When you can recall make your point.

>I am glad we did go to the ellipse, of course. Otherwise
>I think NASA might have had even more trouble than
>they did in the early days.
>

Or maybe they would just try to wow the Congress/President and keep
employed and would ignore the data and models. After all, why should
they act different from other scientists?

Joseph Potter

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to

Matt Silberstein <mat...@ix.netcom.com> wrote in message
news:378ed962...@nntp.ix.netcom.com...

> In talk.origins I read this message from "Joseph Potter"
> <joe.p...@worldnet.att.net>:
>
...

> >
> >
> >The point is that scientists at that time thought the model
> >worked well. It did allow for predictions that "wowed
> >the King," and kept everyone employed.
> >
> Which time do you mean? Copernicus? Kepler? Newton? And do you really
> claim that the scientists made these claims just to wow the King and
> keep employed? I had thought your point had something to do with the
> Razor, not with the lack of ethics in scientific behavior. Yes,
> Kepler's model worked well, it worked much better and "simpler" than
> the previous models.
>
...


What lack of ethics? All experts are in the "wow" business,
and all professions are generally resistant to change if
what they are doing "works" fairly well.

Regards, Joe

Ken Cox

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to
Joseph Potter wrote:
> Ever try to find out how much area is swept out in a unit
> of time as a particle travels in the orbit elliptically verses
> one that travels in a circular orbit?

It is nasty. Unfortunately, the planets *actually do travel in
ellipses*. This means to get an exact answer with circles will
take an infinite number of circles, which is a bit less tractable
than working with the ellipses.

--
Ken Cox k...@research.bell-labs.com


Joseph Potter

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to

Ken Cox <k...@lucent.com> wrote in message news:378DF7...@research.bell-labs.com...

At the point in time in question (I always loved that
expression, see Watergate), I do not believe they
were getting *exact* answers.

In fact, I seem to recall we do not get *exact* answers even
today because the planets do not really travel in perfect
little ellipses.

Exact answers are not always the best answers for the
task at hand.

Regards, Joe


z@z

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to
Joseph Potter wrote:
| Matt Silberstein wrote:

| > I agree that it is easier to work with circles than ellipses, but I
| > have never seen anything in the Razor that speaks of making things
| > easy to work. Nor do I see how circles are somehow simpler than
| > ellipses, in fact, I could make an argument that ellipses are simpler.
| > But this is irrelevant. ...
| >
|

| Ever try to find out how much area is swept out in a unit
| of time as a particle travels in the orbit elliptically verses
| one that travels in a circular orbit?
|

| I seem to recall that a fellow named Newton had to deal with this
| problem and had to come up with something called Calculus.

Joe, do you suggest here that Kepler had found his three laws
without being able to do the necessary calculations?

| (of course a fellow in France claimed to have beat
| him to it, but that's life)

Leibniz's work on infinitesimals had the same common ancestor
as Newton's Calculus: Kepler's works on infinitesimals.

Cheers, Wolfgang

Joseph Potter

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to

z@z <z...@z.lol.li> wrote in message news:7mku6h$gt6$1...@pollux.ip-plus.net...

> Joseph Potter wrote:
> | Matt Silberstein wrote:
>
> | > I agree that it is easier to work with circles than ellipses, but I
> | > have never seen anything in the Razor that speaks of making things
> | > easy to work. Nor do I see how circles are somehow simpler than
> | > ellipses, in fact, I could make an argument that ellipses are simpler.
> | > But this is irrelevant. ...
> | >
> |
> | Ever try to find out how much area is swept out in a unit
> | of time as a particle travels in the orbit elliptically verses
> | one that travels in a circular orbit?
> |
> | I seem to recall that a fellow named Newton had to deal with this
> | problem and had to come up with something called Calculus.
>
> Joe, do you suggest here that Kepler had found his three laws
> without being able to do the necessary calculations?
>

Not at all. I just point out that the calculations would be
much easier if the planets did go round in nice little
circles. Unfortunately they refuse to cooperate.

>
> | (of course a fellow in France claimed to have beat
> | him to it, but that's life)
>
> Leibniz's work on infinitesimals had the same common ancestor
> as Newton's Calculus: Kepler's works on infinitesimals.
>

Newton gave credit to the giants that came before him, as
all scientists do if they are honest about it.

I seem to recall from math history that it took another
hundred years or so to firm up the "divide by zero"
problem and put the Calculus on a firm foundation.

Regards, Joe


z@z

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to
Hello Ian!

[snip]

| >Maybe, I should have been more precise:
| >
| > There are genes such as cytochrome-c and ubiquitin with
| > a rate of non-silent mutations of almost zero.
|
| Where does this figure come from, given that the substitution rate of
| cytochrome-C is around 0.28 substitutions per site per 10^9 years (cf
| the substitution rates of pesudogenes, not under selection pressure,
| which are 1.3 substitutions/ site per 10^9 years [*])

You probably agree with me that less than 4 billion years ago, no
genetic sequences did exist. So for every existing sequence we
necessarily have at least 0.25 substitutions (insertions) per site
per 10^9 years.

| This means that yeast should differ from humans in around 50 amino
| acids out of 101, as it happens they differ in 46 amino acids (that is
| there have been 46 amino acid _substitutions_).

According to Wesley's page [1] the human cytochrome-c does not
differ from the one of chimps and according to my biochemistry
encyclopedia pigs differ neither from cows nor from sheep.

Your six-substitution-rule would result in around 100^6 or 10^80
possible selectively neutral alleles of cytochrome-c. You even
claim that "for the vast majority of proteins, such as cytochrome C,
trypsin, haemoglobin, barnase and lactate dehydrogenase it is an
_underestimate_" [2].

In any case, the fact that pigs, cows and sheep all have the same
cytochrome-c is a huge ENIGMA within neo-Darwinism, I even dare to
say that it's a further knockout blow to pure materialism. Think
about the odds!

[snip]

| >When I write mutation rate then I mean mutation rate. We have
| >already had a discussion on the related problems. The most
| >convincing case is ubiquitin.
| >See: http://members.lol.li/twostone/E/deja2.html
|
| But here you arn't quoting mutation rates, you are comparing
| substition rates. (ie yeast ubiquitin differs from human ubiquitin in
| 3 amino acid positions, this gives a _substitution_ rate of around
| 0.005 substitutions per site per 10^9 years, yet again this is NOT
| the mutation rate)

1444 different alleles can emerge from animal ubiquitin by a single
point mutation. The fact that none of these alleles actually can be
found (with a reasonable frequency?) in animals must be explained
somehow.

[snip]

| >Because of the huge number of animals it is impossible to explain
| >by genetic drift the fact that there is only one single allele of
| >ubiquitin in all animals. Drift only works for little populations,
| >but certainly not for all animals considered as one species.
|
| Despite you claim, drift works in arbitarily large populations, not
| just small ones, try out this population simulator with large
| populations and see what happens.
| http://http.bsd.uchicago.edu/hgd-sad/HWSimulator/sim.cgi
| to see why neutral drift will eliminate multiple allelles in any
| arbitaryly sized population.

It is at least the third time you post this simulator site. Did you
try it out? Probably not. If you had you would know that it is
completely irrelevant to our discussion, because it works only for
unrealistically small populations (where drift actually works). Do
you know how many animals exist and of how many individuals animal
species normally consist?

Your claim that "drift works in arbitarily large populations" is
simply wrong. That genes have converged to only very few or even
one single allele during evolution is an empirical fact. Its
explanation by genetic drift, however, is rather a typical case
of the "ordinary unintentional scientific dishonesty". And it is
exactly this ordinary dishonesty only consisting in exaggerations
and withholding contradicting evidence which has led to such a
catastrophe as the HIV-AIDS-thesis, resulting in tens of thousands
of persons killed (most of them reborn in the meanwhile :-) by
pharmaceutical drugs.

[snip]

[1] http://inia.cls.org/~welsberr/evobio/evc/argresp/sequence.html
[2] http://x37.deja.com/=dnc/getdoc.xp?AN=480456524

Cheers, Wolfgang

Thread history backwards:
http://www.deja.com/=dnc/getdoc.xp?AN=501234777 (Ian)
http://www.deja.com/=dnc/getdoc.xp?AN=501004397 (z@z)
http://www.deja.com/=dnc/getdoc.xp?AN=500345463 (Ian)
http://www.deja.com/=dnc/getdoc.xp?AN=499894064 (z@z)

John Wilkins

unread,
Jul 15, 1999, 3:00:00 AM7/15/99
to
In article <7mki48$9e7$1...@bgtnsc02.worldnet.att.net>, "Joseph Potter"
<joe.p...@worldnet.att.net> wrote:

|Ken Cox <k...@lucent.com> wrote in message

news:378C9F...@research.bell-labs.com...
|> Joseph Potter wrote:
|> > z@z <z...@z.lol.li> wrote:
|>
|> > > Some centuries ago you would have written (against Kepler):
|> > > One also cannot refute the possibility that the planets are
|> > > pushed around by invisible and undetectable mathematical
|> > > rules that just happen to exactly follow the paths predicted
|> > > by the epicycles.
|>

|> > One might also point out that working with the equations
|> > of circles is a tad simpler than working with the equations
|> > for the ellipse. Hence, several would call out the old Razor
|> > to say we "don't need no stinking elliptical orbits."
|>

|> Zelda is perhaps correct in that I might have opposed Keplerian
|> ellipses. However, once it was clear that the orbits really are
|> ellipses, and once Newton showed how such orbits are generated
|> by a central force, I assure you I would have embraced them.
|>
|> However, it is not true that the ellipses are harder than the
|> epicycles. What you must remember is that the epicycles are
|> really a series expansion of the ellipses -- you get *exactly
|> the same orbit* using a single ellipse as you do using a sum
|> of a hundred or so circles, each of different radii and period.
|> This makes the circles *much* harder to work on; the ellipses
|> requires only six parameters, while the circles require several
|> hundred (five parameters per circle).
|>
|
|I once read an essay by Kune (sp?) where he talked of this very
|issue. He said that in those days the precision necessary did
|not take many circles and that the methods were well understood.
|Hence, at least for a time, the ellipse was harder to work with.

That would be Thomas Kuhn of paradigm fame, in his book _The Copernican
Revolution_. In the initial stages, epicycles were more or less the same
general degree of precision. But the observations of Kepler made them much
less workable, and ellipses became easier both mathematically and in terms
of parsimony, as the number of required epicycles went up.

Kuhn, Thomas S. The Copernican revolution: planetary astronomy in the
development of Western thought. New York: Vintage Books, 1959.

see also

Kuhn, Thomas S. The essential tension: selected studies in scientific
tradition and change. Chicago: University of Chicago Press, 1977.

--
John Wilkins, Head, Graphic Production
The Walter and Eliza Hall Institute of Medical Research, Melbourne,
Australia <mailto:wil...@WEHI.EDU.AU><http://www.wehi.edu.au/~wilkins>
Homo homini aut deus aut lupus - Erasmus of Rotterdam


Joseph Potter

unread,
Jul 16, 1999, 3:00:00 AM7/16/99
to

I remember now, it was _The Essential Tension_ --- thanks for the
note. I think it is time to read it once again. I recall that it was a good read.

Regards, Joe
--------------
"Current utility may not be equated with
historical origin, or, when you demonstrate
that something works well, you have not
solved the problem of how, when, or why
it arose." --- Stephen Jay Gould


maff91

unread,
Jul 16, 1999, 3:00:00 AM7/16/99
to
On 15 Jul 1999 19:40:52 -0400, wil...@wehi.edu.au (John Wilkins)
wrote:

[snip]

>That would be Thomas Kuhn of paradigm fame, in his book _The Copernican
>Revolution_. In the initial stages, epicycles were more or less the same

http://www.amazon.com/exec/obidos/ASIN/1567312179/


>general degree of precision. But the observations of Kepler made them much
>less workable, and ellipses became easier both mathematically and in terms
>of parsimony, as the number of required epicycles went up.
>
>Kuhn, Thomas S. The Copernican revolution: planetary astronomy in the
>development of Western thought. New York: Vintage Books, 1959.
>
>see also
>
>Kuhn, Thomas S. The essential tension: selected studies in scientific
>tradition and change. Chicago: University of Chicago Press, 1977.

http://www.amazon.com/exec/obidos/ASIN/0226458067/
http://www.amazon.com/exec/obidos/Author=Kuhn%2C%20Thomas/

--
L.P.#0000000001


howard hershey

unread,
Jul 16, 1999, 3:00:00 AM7/16/99
to
z@z wrote:
>
> Hi Ian!
>
> [snip]

>
> | >It is generally acknowledged that mutation rates are very different
> | >within a single genome.
> |
> | What Wolfgang is confusing here is the the rate at which mutations
> | occur in the genome, and the proportion of those mutations that are
> | benifical/neutral vs deleterious and enter into the population after
> | _selection_ (the substitution rate).
>
> Howard Hershey, 7-Jul-99:
>
> "Point mutations, for example, occur at a rate in the range
> of 10^-4 to 10^-12 (modal about 10^-9) depending upon the
> particular site (some sites are highly mutable (hot spots)
> compared to others. For example, the point mutation that
> causes achondroplasia has perhaps the highest known mutation
> rate in the human genome (about 10^-6)."
>
> If all point mutations were as likely as this one,

Perhaps you did not read what I just wrote above. This is decidedly NOT
the average mutation. It is one with a very high rate of spontaneous
occurance. Most dominant mutations that occur frequently do so because
of dosage effects (knock out one copy by any mutation that eliminates
its activity). That is not the case with achrondroplasia, which (with a
couple of exceptions, always involves a particular nucleotide change and
amino acid change).

> humans
> would have around 6'000 point mutations (or even more, if
> achondroplasia depends on the substitution of the particular
> site by a particular amino acid.)

And if pigs had wings, we would all need umbrellas.


>
> But according to Richard Harter, "the average human being has
> about 50-100 mutations, of which about 3 matter, i.e., they
> actually change a protein."
> http://www.tiac.net/users/cri/mutate.html
>

> | >There are genes such as cytochrome-c and
> | >ubiquitin with a mutation rate of almost zero.
>

> Maybe, I should have been more precise:
>

> There are genes such as cytochrome-c and ubiquitin with

> a rate of non-silent mutations of almost zero.
>

> But the problem of silent mutations has been specifically
> addressed only three paragraphs later.
>

> | No, these are _substitution_ rates, rather than mutation rates
> | themselves. You are comparing the sequences of different existing
> | organisms (rather than measuring the mutation rate in any given DNA
> | copying event). This is the combined effect of mutation rate AND
> | selection pressure.
>

> When I write mutation rate then I mean mutation rate.

Except that when you are talking about the differences between
ubiquitins of different species, you are talking about substitution
rates and calling it mutation rates.

> We have
> already had a discussion on the related problems. The most
> convincing case is ubiquitin.
> See: http://members.lol.li/twostone/E/deja2.html
>

> Because of the huge number of animals it is impossible to explain
> by genetic drift the fact that there is only one single allele of
> ubiquitin in all animals. Drift only works for little populations,
> but certainly not for all animals considered as one species.

Drift has a larger percentage effect in small populations each
generation. But drift occurs in both large and small populations
(spcifically a species population).


>
> The alternative that every mutation in animal ubiquitin is lethal
> or harmful enough to disappear by selection seems much more
> unreasonable to me than the assumption that (non-silent) mutations

> are almost inexistent in ubiquitin. Also the existence of 'hot


> spots' could suggest the existence of 'cold spots'.

Ubiquitin has a low level of selectively neutral sites. That is why
neutral changes are rare.

You may be erroneously thinking that most genes should have a 'build-up'
of neutral or deleterious alleles over time. But actually, the fate of
most neutral alleles (and certainly most deleterious dominant ones) is
loss. Imagine a population with nearly all of its alleles being allele
B. A mutation introduces a new allele which is essentially equivalent,
B'. What is the likely fate of B'? The probability that it will
'drift' and increase by drift until *it* becomes the most common allele
is 1/N where N is the total number of alleles in the population. That
means that the chance that allele B' (actually progeny that include
copies of this allele) will be lost purely by chance is (1-N)/N, which
is pretty close to 1 for reasonably large populations.

Let's assume that, instead of starting with nearly all B and having B'
enter by mutation, that we start with 50% B and 50% B' (the two are
functionally equivalent, remember). Now drift will occur, and the next
generation has 51% B and 49% B' just by chance. At this point, it is
important to remember the key rule that gamblers forget: CHANCE HAS NO
MEMORY. Thus, what happens the next generation will be chance variation
from 51% and 49% and not a tendancy to return to the original 50% and
50%. Because chance has no memory, let's say that drift leads to 90% B
and 10% B'. At this point, it is easier to see that drift could lead to
the elimination of B'. Drift could, with equal probability, have gone
the other direction; indeed if you had started several populations with
50/50 B/B', half the time it would drift so as to eliminate the B and
half the time to eliminate the B'. But what you will not get is
retention of the 50/50 ratio you started with. THAT is how chance
really works.
>
[snip]

> | The substitution rates of these genes must be compared
> | to the substition rate of pseudogenes (copies of genes which are
> | no longer expressed), where there is no selection constraint, and
> | the substitution rate is equivalent to the mutation rate. The
> | substitution rate for cytochrome C is far from zero, and is a bit
> | under half that of pseudogenes, which is compatible with the average
> | of 7 neutral substitutions per site determined from Yockeys
> | calculations.
>

> According to my premises one cannot conclude from mutation or
> substitution rates of pseudogenes to the rates of the actual
> genes because functional constraints are very different.

Yes, the substitution rates are quite different when comparing
functional and pseudogenes. The mutation rates, however, are not
significantly different between pseudogenes and functional genes.


>
> | The substition rate of ubiqutin is very low, but still not almost
> | zero.
>

> It would be certainly very interesting and instructive to
> examine in detail the question of different DNA-alleles of
> ubiquitin and cytochrome-c.

Within or between species populations. Research tends to focus on the
frequency of alternate alleles that have functional sequellae (like
genetic disease) rather than those that have no phenotypic effect.
Easier to find, for one thing. But there are plenty of alternative
neutral alleles found when one actually searches for them (there are
something like 300 variants of hemoglobin alone). Most genes have some
variants within the species, if you are willing to put in the resources
to search for them (randomly seuencing DNA is too much work; most
searches only pick up electrophoretic variants of some sort). It all
depends on how many toads you are willing to kiss to find the prince.
>
> [snip]
>
> Cheers, Wolfgang


z@z

unread,
Jul 18, 1999, 3:00:00 AM7/18/99
to
Hello Howard!


| > Howard Hershey, 7-Jul-99:
| >
| > "Point mutations, for example, occur at a rate in the range
| > of 10^-4 to 10^-12 (modal about 10^-9) depending upon the
| > particular site (some sites are highly mutable (hot spots)
| > compared to others. For example, the point mutation that
| > causes achondroplasia has perhaps the highest known mutation
| > rate in the human genome (about 10^-6)."
| >
| > If all point mutations were as likely as this one,
|
| Perhaps you did not read what I just wrote above. This is decidedly NOT
| the average mutation. It is one with a very high rate of spontaneous
| occurance. Most dominant mutations that occur frequently do so because
| of dosage effects (knock out one copy by any mutation that eliminates
| its activity). That is not the case with achrondroplasia, which (with a
| couple of exceptions, always involves a particular nucleotide change and
| amino acid change).

That the achrondroplasia mutation occurs at a rate of about 10^-6
and "with a couple of exceptions, always involves a particular
nucleotide change and amino acid change" is certainly not
consistent with the principle of random mutations. In a similar
way also mutations leading to cancer are far from being totally
random!

The psychon theory explains substantially increased mutation
probabilities by the existence of psychons corresponding to
alternative alleles. In such a way, alleles can survive even
if they lead to death before reproduction.

What I write here may seem strange to someone who believes in
orthodox materialism, but the psychon thesis leads to lots of
testable predictions (and even to completely new medical methods).
For instance it predicts that artificially induced mutations
will mutate back to the original sequence with a substantially
increased probability.

[snip]

| > The alternative that every mutation in animal ubiquitin is lethal
| > or harmful enough to disappear by selection seems much more
| > unreasonable to me than the assumption that (non-silent) mutations
| > are almost inexistent in ubiquitin. Also the existence of 'hot
| > spots' could suggest the existence of 'cold spots'.
|
| Ubiquitin has a low level of selectively neutral sites. That is why
| neutral changes are rare.
|
| You may be erroneously thinking that most genes should have a 'build-up'
| of neutral or deleterious alleles over time. But actually, the fate of
| most neutral alleles (and certainly most deleterious dominant ones) is
| loss. Imagine a population with nearly all of its alleles being allele
| B. A mutation introduces a new allele which is essentially equivalent,
| B'. What is the likely fate of B'? The probability that it will
| 'drift' and increase by drift until *it* becomes the most common allele
| is 1/N where N is the total number of alleles in the population. That
| means that the chance that allele B' (actually progeny that include
| copies of this allele) will be lost purely by chance is (1-N)/N, which
| is pretty close to 1 for reasonably large populations.

I do not understand the expression (1-N)/N.

Assume that the probability for a selectively neutral random
mutation in a certain protein is 10^6. If we assume a constant
population of 10^9 with one single allele at the respective
locus in the first generation, we get 10^3 alleles differing
from the original form in the second generation. After a million
generations a huge number of different selectively neutral
allels have appeared and the frequency of the orginal allele
is reduced to around 37 percent.

That's simplest probability theory and cannot be denied in a
reasonable way.

| Let's assume that, instead of starting with nearly all B and having B'
| enter by mutation, that we start with 50% B and 50% B' (the two are
| functionally equivalent, remember). Now drift will occur, and the next
| generation has 51% B and 49% B' just by chance.

If the population consists of 1000 individuals then 510 B versus
490 B' is a reasonable assumption, but if the population consists
of only a million individuals then 510'000 versus 490'000 is very
improbable. Many (maybe even most) species have populations which
are much larger than 10^6.

| At this point, it is important to remember the key rule that gamblers
| forget: CHANCE HAS NO MEMORY.

For drift to work in reality, a MEMORY would be necessary. Drift
only works if unrealistically small populations are assumed.

[snip]


Cheers, Wolfgang


P.S. Have you seen the post "Molecular Sequence Proof of Common
Descent" of Zeus Thibault? In trying to prove common descent
Zeus inadvertently has shot down neo-Darwinism!

See: http://www.deja.com/=dnc/getdoc.xp?AN=502140449

WRowe0521

unread,
Jul 18, 1999, 3:00:00 AM7/18/99
to
>"z@z" <z...@z.lol.li> wrote:
>he psychon thesis leads to lots of
>testable predictions (and even to completely new medical methods).
>For instance it predicts that artificially induced mutations
>will mutate back to the original sequence with a substantially
>increased probability.

--> quote
Antibiotic resistance: Road of no return
http://www.weeds.iastate.edu/weednews/antibiotics.htm
Bob Hartzler Iowa State University Extension

October 15, 1998 -

The article describes findings reported by Bruce Levin, a population geneticist
at Emory University in Atlanta at a meeting of the European Society for
Evolutionary Biology. Researchers had hoped that bacteria that have become
resistant to antibiotics would 'evolve backward', losing there resistance when
doctors stopped prescribing the antibiotic. The driving force for backward
evolution was thought to be that the resistance mechanism would make the
resistant biotype less fit, therefore in the absence of the antibiotic
susceptible biotypes would reestablish themselves as the dominant biotype in
the population.
The results suggest that resistance in E. coli may be much more persistent than
previously thought. The researchers found that 25% of E. coli populations
isolated from baby's diapers at a day care center were still resistant to
streptomycin, an antibiotic that has been used rarely during the last 30 years.
--> end quote

Looks like your theory has failed its first test.


Bill Rowe


howard hershey

unread,
Jul 19, 1999, 3:00:00 AM7/19/99
to
z@z wrote:
>
> Hello Howard!

>
> | > Howard Hershey, 7-Jul-99:
> | >
> | > "Point mutations, for example, occur at a rate in the range
> | > of 10^-4 to 10^-12 (modal about 10^-9) depending upon the
> | > particular site (some sites are highly mutable (hot spots)
> | > compared to others. For example, the point mutation that
> | > causes achondroplasia has perhaps the highest known mutation
> | > rate in the human genome (about 10^-6)."
> | >
> | > If all point mutations were as likely as this one,
> |
> | Perhaps you did not read what I just wrote above. This is decidedly NOT
> | the average mutation. It is one with a very high rate of spontaneous
> | occurance. Most dominant mutations that occur frequently do so because
> | of dosage effects (knock out one copy by any mutation that eliminates
> | its activity). That is not the case with achrondroplasia, which (with a
> | couple of exceptions, always involves a particular nucleotide change and
> | amino acid change).
>
> That the achrondroplasia mutation occurs at a rate of about 10^-6
> and "with a couple of exceptions, always involves a particular
> nucleotide change and amino acid change" is certainly not
> consistent with the principle of random mutations. In a similar
> way also mutations leading to cancer are far from being totally
> random!

Well, in the sense that not *every* mutation causes cancer and not
*every* mutation causes achondroplastic dwarfism, you are quite right.
But I never would have claimed that *any* old random mutation would
cause these phenotypes. Only specific mutations cause particular
phenotypes. But these mutations are occuring entirely without respect
to need (at random in the sense I use it). What, by the way, do you
think the "need" for mutations that produce achondroplasia or the
somatic mutations that cause cancer, is?


>
> The psychon theory explains substantially increased mutation
> probabilities by the existence of psychons corresponding to
> alternative alleles. In such a way, alleles can survive even
> if they lead to death before reproduction.
>
> What I write here may seem strange to someone who believes in

> orthodox materialism, but the psychon thesis leads to lots of


> testable predictions (and even to completely new medical methods).
> For instance it predicts that artificially induced mutations
> will mutate back to the original sequence with a substantially
> increased probability.

That would be a shock to all those studies of reversion, which generally
find that reversion is less frequent than forward mutation. But do
explain your psychon theory and its testable predictions. Sounds
interesting, in a totally ridiculous way.
>
> [snip]


>
> | > The alternative that every mutation in animal ubiquitin is lethal
> | > or harmful enough to disappear by selection seems much more
> | > unreasonable to me than the assumption that (non-silent) mutations
> | > are almost inexistent in ubiquitin. Also the existence of 'hot
> | > spots' could suggest the existence of 'cold spots'.
> |
> | Ubiquitin has a low level of selectively neutral sites. That is why
> | neutral changes are rare.
> |
> | You may be erroneously thinking that most genes should have a 'build-up'
> | of neutral or deleterious alleles over time. But actually, the fate of
> | most neutral alleles (and certainly most deleterious dominant ones) is
> | loss. Imagine a population with nearly all of its alleles being allele
> | B. A mutation introduces a new allele which is essentially equivalent,
> | B'. What is the likely fate of B'? The probability that it will
> | 'drift' and increase by drift until *it* becomes the most common allele
> | is 1/N where N is the total number of alleles in the population. That
> | means that the chance that allele B' (actually progeny that include
> | copies of this allele) will be lost purely by chance is (1-N)/N, which
> | is pretty close to 1 for reasonably large populations.
>

> I do not understand the expression (1-N)/N.

Whoops. That should be (N-1). Must be my dyslexia acting up.


>
> Assume that the probability for a selectively neutral random
> mutation in a certain protein is 10^6.

This, remember, is a *high* frequency for point mutation at the DNA
level and not an average frequency. [You may be confusing the rate of
mutation at a particular nucleotide with the rate of phenotypic mutation
in particular proteins, where all changes that cause null protein
synthesis are counted.] More 'typical' (for point mutation) is mutation
frequencies around 10^-9.

> If we assume a constant
> population of 10^9 with one single allele at the respective
> locus in the first generation, we get 10^3 alleles differing
> from the original form in the second generation. After a million
> generations a huge number of different selectively neutral
> allels have appeared and the frequency of the orginal allele
> is reduced to around 37 percent.
>
> That's simplest probability theory and cannot be denied in a
> reasonable way.

Sure it can. Because it won't happen that way in real populations.
Your assumption seems to be that only the 'wild-type' allele can get
lost by neutral drift. But that allele starts out in a predominant
position. Let's say that

>
> | Let's assume that, instead of starting with nearly all B and having B'
> | enter by mutation, that we start with 50% B and 50% B' (the two are
> | functionally equivalent, remember). Now drift will occur, and the next
> | generation has 51% B and 49% B' just by chance.
>

> If the population consists of 1000 individuals then 510 B versus
> 490 B' is a reasonable assumption, but if the population consists
> of only a million individuals then 510'000 versus 490'000 is very
> improbable. Many (maybe even most) species have populations which
> are much larger than 10^6.

Yes. I agree. Large populations dampen the % change that occurs due to
chance alone. That is why flipping a coin 10 times and getting 7 heads
is not as remarkable as getting 700 heads when you flip 1000 times.
Chance changes are relatively *more* important in small populations.
They do not, however, disappear in large populations until you start
hypothesizing infinite populations or start excluding chance deviations
entirely (as you do in your model).


>
> | At this point, it is important to remember the key rule that gamblers
> | forget: CHANCE HAS NO MEMORY.
>

> For drift to work in reality, a MEMORY would be necessary. Drift
> only works if unrealistically small populations are assumed.

No. It is here you are wrong. Drift works precisely *because* chance
has no memory. If the *existing* generation has 51% B and 49% B',
random fluctuation in the next will be around *that* central tendancy
and not around the 50% B and 50% B' that existed in the grandparents
generation. That is, it might, say, go either up (and again approach
50%/50%). But it is equally likely to go down, to say, 52%/48%. *If*
it goes down, the next generations offspring will fluctuate around the
52%/48% central tendancy, with no memory whatsoever of its original
50%/50% ratio. Essentially, what you will get, by the action of chance,
is a random walk up (and down) the percentages. We call this random
walk the process of neutral drift, BUT, and it is a big BUT, one cannot
go beyond 0% B' (or 0% B). Once that particular 'fixation' point is
reached, the only way to get new B' (or B) is by new mutation. Drift is
*slower* in a large population, but it does continue.

OTOH, the rate of neutral substitution is, surprisingly I am sure to
you, roughly the same in small and large populations. In small
populations, the chance of any particular mutation becoming fixed by
neutral drift (which is equal to 1/N, where N = total population of
alleles, since every new allele starts as a single mutation). If the
mutation rate is u, that means that each generation there will be Nu of
any particular mutation. The larger the population, the more frequent
any particular mutation will be. Thus, the substitution rate for any
particular mutation is roughly Nu * (1/N) = u regardless of population
size. But this is population genetics 101 stuff.

Perhaps a model you can understand is that of incompatible plasmids in
bacteria. If you have two plasmids that use the same replicative
machinery and differ by, let us say, a single nucleotide, what happens
when the bacteria divides? For the sake of simplicity, let us say that
we start with a population of bacteria that, each, at the time of cell
division into two, has eight plasmids, four of type A and four of type
B, which are distributed randomly and equally (4/daughter cell). That
is, after cell division, there is a 1/16 chance (for each of the 4
plasmids, there is a 1/2 probability that it will be of the B type) that
a given daughter cell will have only the B plasmid (and an equal
probability of a cell having only the A plasmid). There will be a 3/16
probability of a given daughter having a 3:1 B:A ratio (and an equal
probability of a cell with a 3:1 A:B ratio). And more than half (9/16)
will have the 2:2 ratio of the parents). The next generation (after the
plasmids in each cell has duplicated), however, the 1/16 that have only
the A plasmid (and the equivalent number having only B) can only "breed
true". The ones that have the 3:1 (6:2 after duplication) ratio have a
3/16 probability of producing a cell only having one plasmid type. In
each generation, those remaining bacteria with mixed plasmid populations
have a certain probability of generating non-mixed daughters. But the
reverse, a non-mixed daughter producing a mixed plasmid is extremely
rare (it requires mutation). Very shortly, in terms of generations, you
will have almost half of the population of bacteria that will *only*
have plasmid A and an equal population that *only* has plasmid B. There
will remain a tiny and decreasing population of cells with mixed
plasmids. This segregation is accomplished solely by the action of
chance because there is a "wall" of 100% A (or B) beyond which one
cannot go and from which one cannot retreat. In drift in populations,
the two "walls" are 'extinction' of the allele and 'fixation' of the
allele. Neither wall is absolute; both are fuzzy. 'Extinction' of an
allele *can* be countered by new mutation. 'Fixation' can also be
affected by new mutation (no allele probably ever reaches 100%).

>
> [snip]
>
> Cheers, Wolfgang
>
> P.S. Have you seen the post "Molecular Sequence Proof of Common
> Descent" of Zeus Thibault? In trying to prove common descent
> Zeus inadvertently has shot down neo-Darwinism!

I think I have read some of that thread. It, too, is nonsense.
>
> See: http://www.deja.com/=dnc/getdoc.xp?AN=502140449


z@z

unread,
Jul 21, 1999, 3:00:00 AM7/21/99
to
Hi Howard Hershey!

[snip]

| That would be a shock to all those studies of reversion, which generally
| find that reversion is less frequent than forward mutation.

What kind of reversion studies? And what about domesticated
species becoming wild. After a few generations individuals with
reduced domestication traits can appear. How can this happen in
a few generations if the necessary mutations do not occur? Also
in the case of extremely inbred rats, having undergone prolonged
gene purgation, genetic variability can reemerge rather quickly.

[snip]

| > Assume that the probability for a selectively neutral random

| > mutation in a certain protein is 10^-6.


|
| This, remember, is a *high* frequency for point mutation at the DNA
| level and not an average frequency. [You may be confusing the rate of
| mutation at a particular nucleotide with the rate of phenotypic mutation
| in particular proteins, where all changes that cause null protein
| synthesis are counted.] More 'typical' (for point mutation) is mutation
| frequencies around 10^-9.

I do not understand your comment in parentheses.

Take a protein consisting of around 1000 amino acids. If the
probability of a point mutation is 10^-9 then the probability of
a mutation in any of the 3000 base pairs is 3 x 10^-6. If we
further assume that one third of these mutations is neutral, the
probability of a selectively neutral random mutation is 10^-6.

| > If we assume a constant
| > population of 10^9 with one single allele at the respective
| > locus in the first generation, we get 10^3 alleles differing
| > from the original form in the second generation. After a million
| > generations a huge number of different selectively neutral
| > allels have appeared and the frequency of the orginal allele
| > is reduced to around 37 percent.
| >
| > That's simplest probability theory and cannot be denied in a
| > reasonable way.
|
| Sure it can. Because it won't happen that way in real populations.

That it does not happen is an empirical fact. Nevertheless it should
happen according to the premises of neo-Darwinism. So neo-Darwinism
is refuted.

[snip]

| > For drift to work in reality, a MEMORY would be necessary. Drift
| > only works if unrealistically small populations are assumed.
|
| No. It is here you are wrong. Drift works precisely *because* chance
| has no memory. If the *existing* generation has 51% B and 49% B',
| random fluctuation in the next will be around *that* central tendancy
| and not around the 50% B and 50% B' that existed in the grandparents
| generation. That is, it might, say, go either up (and again approach
| 50%/50%). But it is equally likely to go down, to say, 52%/48%. *If*
| it goes down, the next generations offspring will fluctuate around the
| 52%/48% central tendancy, with no memory whatsoever of its original
| 50%/50% ratio.

Repeat your reasoning and change the 'displacement rate' per
generation from 1% to more realistic 0.1%, 0.01% or 0.001%.

| Essentially, what you will get, by the action of chance,
| is a random walk up (and down) the percentages. We call this random
| walk the process of neutral drift, BUT, and it is a big BUT, one cannot
| go beyond 0% B' (or 0% B). Once that particular 'fixation' point is
| reached, the only way to get new B' (or B) is by new mutation. Drift is
| *slower* in a large population, but it does continue.

Because of the similarity to Brownian motion I would guess that
the mean (square?) number of generations needed in order to reach a
particular fixation point is inversely proportional to the square
of the 'displacement rate'. Furthermore, the larger the population
the lower the displacement rate. (It would be helpful if a
specialist of probability theory could provide an analysis).

In my (homozygous) example above, in the second generation 1000
new alleles have appeared. Every of these alleles would have a
theoretical probability of 10^-9 (inverse of the population size)
to reach 'fixation' by drift after a huge amount of generations,
only if no more mutations occured. But in every new generation
around 1000 alleles differing from the most frequent allele do
appear.

| OTOH, the rate of neutral substitution is, surprisingly I am sure to
| you, roughly the same in small and large populations. In small
| populations, the chance of any particular mutation becoming fixed by
| neutral drift (which is equal to 1/N, where N = total population of
| alleles, since every new allele starts as a single mutation). If the
| mutation rate is u, that means that each generation there will be Nu of
| any particular mutation. The larger the population, the more frequent
| any particular mutation will be. Thus, the substitution rate for any
| particular mutation is roughly Nu * (1/N) = u regardless of population
| size. But this is population genetics 101 stuff.

Interesting, I have been accused of confusing neutral mutation
rates with neutral substitution rates, and now you write that
they are identical! I agree with you that within neo-Darwinism
they should be identical. In reality however, they are quite
different.

| Perhaps a model you can understand is that of incompatible plasmids in
| bacteria. If you have two plasmids that use the same replicative
| machinery and differ by, let us say, a single nucleotide, what happens

| when the bacteria divides? [snip] Very shortly, in terms of generations,


| you will have almost half of the population of bacteria that will *only*
| have plasmid A and an equal population that *only* has plasmid B.

This is not relevant to our discussion as neither A nor B do disappear.

[snip]

| > P.S. Have you seen the post "Molecular Sequence Proof of Common
| > Descent" of Zeus Thibault? In trying to prove common descent
| > Zeus inadvertently has shot down neo-Darwinism!
|
| I think I have read some of that thread. It, too, is nonsense.
|
| > See: http://www.deja.com/=dnc/getdoc.xp?AN=502140449

What do you think is nonsense? Is it Zeus' conclusion?

"As mentioned above, the cytochrome c proteins in chimps and humans are
exactly the same. The clincher is that the two DNA sequences that code
for cytochrome c in humans and chimps differ by only one codon, even
though there are over 10^49 different sequences that could code for
these two proteins."

Or is it my comment to Zeus' conclusion?

"Because neither selection nor genetic drift can explain that
human and chimp DNA-sequences did not drift apart by random
mutations, neo-Darwinism is definitively dead !!!"

How do you explain the many very obvious differences between
humans and chimps, if mutations are so rare?


Cheers, Wolfgang


Thread backwards:
http://www.deja.com/=dnc/getdoc.xp?AN=502836392 (Howard)
http://www.deja.com/=dnc/getdoc.xp?AN=502417358 (me)
http://www.deja.com/=dnc/getdoc.xp?AN=501873642 (Howard)
http://www.deja.com/=dnc/getdoc.xp?AN=501004397 (me)

Reply all
Reply to author
Forward
0 new messages