Transcript Slide 1

Is it good to Share?
The Parallel between Transfer of Information via Memes
and Horizontal Gene Transfer
Paul Higgs
Dept of Physics and Astronomy
McMaster University
Thanks to:
Cedoljub Bundalovic-Torma
Jeffrey Wainberg
Supported by Canada Research
Chairs and NSERC.
Who am I ?
CRC in Biophysics, McMaster University, Hamilton, Ontario.
Polymer physics
RNA structure
prediction
Molecular Evolution and Phylogenetics
Mouse
Lemur
Tarsier
SakiMonkey
Marmoset
Baboon
Gibbon
Orangutan
Gorilla
PygmyChimp
Chimp
Human
:
:
:
:
:
:
:
:
:
:
:
:
*
20
*
40
*
60
*
CUCACCAUCUCUUGCUAAUUCAGCCUAUAUACCGCCAUCUUCAGCAAACCCUAAAAAGG-UAUUAAAGUAAGCAAAAGA
CUCACCACUUCUUGCUAAUUCAACUUAUAUACCGCCAUCCCCAGCAAACCCUAUUAAGGCCC-CAAAGUAAGCAAAAAC
CUUACCACCUCUUGCUAAUUCAGUCUAUAUACCGCCAUCUUCAGCAAACCCUAAUAAAGGUUUUAAAGUAAGCACAAGU
CUUACCACCUCUUGCC-AU-CAGCCUGUAUACCGCCAUCUUCAGCAAACUCUA-UAAUGACAGUAAAGUAAGCACAAGU
CUCACCACGUCUAGCC-AU-CAGCCUGUAUACCGCCAUCUUCAGCAAACUCCU-UAAUGAUUGUAAAGUAAGCAGAAGU
CCCACCCUCUCUUGCU----UAGUCUAUAUACCGCCAUCUUCAGCAAACCCUGAUGAAGGCUACGAAGUGAGCGCAAAU
CUCACCAUCUCUUGCU----CAGCCUAUAUACCGCCAUCUUCAGCAAACCCUGACAAAGGCUAUAAAGUAAGCACAAAC
CUCACCACCCCUUGCU----CAGCCUAUAUACCGCCAUCUUCAGCAAACCCUGAUGAAGGCCACGAAGUAAGCGCAAAC
CUCACCACCUCUUGCU----CAGCCUAUAUACCGCCAUCUUCAGCAAACCCUGACGAAGGCCACAAAGUAAGCACAAGU
CUCACCGCCUCUUGCU----CAGCCUAUAUACCGCCAUCUUCAGCAAACCCUGAUGAAGGUUACAAAGUAAGCGCAAGU
CUCACCGCCUCUUGCU----CAGCCUAUAUACCGCCAUCUUCAGCAAACCCUGAUGAAGGUUACAAAGUAAGCGCAAGU
CUCACCACCUCUUGCU----CAGCCUAUAUACCGCCAUCUUCAGCAAACCCUGAUGAAGGCUACAAAGUAAGCGCAAGU
CucACC cuCUuGCu
cAgccUaUAUACCGCCAUCuuCAGCAAACcCu
A G
aAAGUaAGC AA
:
:
:
:
:
:
:
:
:
:
:
:
78
78
79
76
76
75
75
75
75
75
75
75
Why am I interested in memes?
Started with Susan Blackmore’s book.
Big Brain argument:
My model confirms this – selection for large brain
size leads to sudden onset of culture – the mimetic
transition
(Higgs – Proc Roy Soc B. 2000: 267: 1355-1361).
In general – Evolution is fundamental in Biology.
Cultural Evolution is a fundamental part of human evolution that is poorly
understood.
Need for theory and models:
• Lack of generally agreed foundational principles (my perception)
• Verbal arguments tend to be ‘just so stories’
• Designing a model forces you to make assumptions very precise
• Outcome of simulations is not always as expected.
What is a meme ?
Anything transmitted culturally from person to person by imitation or
teaching/learning:
• Fashion / Story / Tune
• Idea / Belief / Scientific Theory
• Skill / Practical Knowledge
Do memes exist?
Of course they do!
Almost everything we know was learned from someone else.
A child reared in isolation could not function in society.
Memes affect our fitness –
influence behaviour, survival and reproduction.
Vertical and Horizontal Transmission
Vertical Transmission – from a parent
Natural Selection works by Vertical Transmission of Genes:
A gene increases fitness of an organism. The organism multiplies more rapidly.
The gene is passed to the offspring. The gene increases in frequency.
Conversely, bad genes would decrease in frequency.
Vertical Transmission of Memes is equivalent:
A meme increases the fitness of a person. The person has more children.
The meme is taught to children. The meme increases in frequency.
Conversely, bad memes would decrease in frequency.
Horizontal Transmission – from an unrelated individual
The donor individual does not necessarily have a high fitness.
Bad genes and memes can spread horizontally.
Viruses / Transposable elements
Memes as viruses of the mind…..
Mechanisms of Horizontal Gene Transfer (HGT)
Transformation – uptake of foreign DNA from surrounding medium and
incorporation into the genome
Conjugation – transfer of plasmid DNA directly between cells
Bacteriophages (viruses) – evolve to transfer their own genes. Can also
transfer a host gene from one cell to another.
E.Coli Conjugation (F Plasmid):
T4 Phage
Evidence for horizontal transfer –
Closely related bacterial genomes contain different genes.
Welch et al (2002).
Is there an evolutionary tree for bacteria?
HGT means
different genes
have different
trees.
Is there a tangle at the root?
Woese argues that HGT so frequent early on that
there are no separate lineages.
Lineages emerge later – “Darwinian Threshold”
Is it Good to Share Genes?
In a population of cells that differ in their ability to accept horizontally
transferred genes, does natural selection favour high or low rates of HGT?
Advantages:
New beneficial genes arise rarely.
Much quicker to acquire a new gene horizontally than to invent it for yourself.
Can gain new metabolic pathways (e.g. photosynthesis, antibiotic resistance).
Can replace lost or damaged genes (similar to the argument that sex beats Muller’s
ratchet)
Disadvantages:
New gene may be a duplicate or non functional – burden of junk DNA
New gene may disrupt existing gene
New gene may be a selfish replicator (transposable elements)
New gene may be a harmful parasite (bacteriophages)
Who Controls Gene Transfer?
The Recipient Cell: YES
Mechanisms of DNA uptake
Mechanisms recombination inside cell
Mechanisms of break-up of DNA fragments
Mechanisms of silencing inserted genes
Variation in cell wall thickness
Develop model to
describe evolution
of the recipient
The Donor Cell: NO
For transformation it is probably a fragment of DNA from a dead cell, so
there isn’t a donor.
When live cells transfer genes it is usually not controlled by the donor
genome.
The Genes Themselves: YES
Transposable elements
Plasmids
Viruses
Model for Gene Transfer
Population of N cells, each with a genome = list of genes, e.g. 3-4-17-1-9-6-1
Fitness
w
1  sf
1  bn
n = total number of genes (7 in example),
f = number of functioning types (6)
s = selective benefit for each new type of gene in genome
b = burden of DNA replication (replication time should be 1 + bn)
s > b so fitness increases for genomes with larger repertoires of genes
But duplicate genes are penalized.
Moran model – chose one individual to replicate with probability proportional to
its fitness and one random individual to die and be replaced.
One generation = N birth/death events.
PLEASE INTERRUPT
IF NECESSARY
Replication Process
Each gene in the parent is copied successfully to the offspring with probability 1-u
or lost/deleted with probability u
A new gene arises in the new cell with probability v
The new cell has the opportunity to acquire genes horizontally – mean number
acquired is h.
hk
exp(h)
Probability of acquiring k genes is P(k ) 
k!
Each acquired gene is a copy of a random gene from a random individual in the
population (assumed to be representative of DNA fragments available).
Donor does not lose the gene.
Genome size controlled by balance of v + h to u.
Selection s also tends to increase genome size if genes are not duplicated.
N = 500, s = 0.1, b = 0.01.
v = 0.002 (low), u = 0.01 (high)
When h = 0, genomes remain small, fitness is low
When h = 1, genomes become larger but many duplicate genes, fitness is higher.
Mean fitness in stationary state as function of h
u = 0.0001 – optimum h = 0
u = 0.001 – optimum h ~ 0.02 – 0.04
u = 0.01 – optimum h ~ 0.5 – 1.0
Conclusions so far:
When heredity is poor (large u) HGT is favourable because large genomes cannot be
maintained by vertical transmission alone.
HGT replaces lost genes, provides backup copies, allows rapid spread of new
beneficial genes and facilitates the emergence of complex genomes.
Early organisms needed HGT to get going.
When heredity is good (small u) HGT is unfavourable because it creates junk DNA.
Large genomes can be maintained successfully without HGT.
This model does not account for disruption of existing genes by inserted genes,
selfish replicators or viruses. All these things would be further disadvantages of HGT.
Modern Organisms should avoid HGT if possible.
But it may not be possible –
viruses and transposable elements are evolving for themselves
Occasionally a cell may benefit by HGT even so (resistance genes etc).
Evolution should move from a tangle to a tree as heredity gets better.
Is it Good to Share Memes?
What is the best balance of learning from parents and unrelated individuals?
Advantages of horizontal meme transfer:
Wider range of memes. Can learn more than parents knew.
Disadvantages:
Vertical memes have been tested by natural selection but horizontal memes have not.
Who Controls Meme Transfer?
The Recipient : YES
Who shall I talk to?
Who shall I learn from / copy?
How much effort shall I put into learning?
Model will focus on the
learning strategy of the
recipient in order to
compare with model for
gene transfer.
The Donor : YES
Who shall I talk to?
Who shall I share information with?
Shall I put effort into deliberate teaching?
The Memes Themselves: YES
Catchy memes should spread.
Fitness for a meme means ease of transfer, not fitness of the organism
Viruses of the mind….
“Strong memetics” view– we are just vehicles for our memes. Memes control us.
“This meme is important, tell everyone”,
“Don’t talk to strangers”,
“Educate your children”.
Model for Meme Transfer
Population of N people, each with list of known memes, e.g. 3-4-17-1-9-6
Cannot know same meme twice.
Fitness
w  1  sn
n = total number of memes known
s = selective benefit for each meme known (all memes good)
No equivalent of duplicate genes and burden b.
At each step one new individual is created. The individual has two parents chosen
with probabilities proportional to their fitnesses. One individual dies at random.
One generation = N birth/death events.
Learning Process
The new recipient individual has L learning opportunities.
For each opportunity a donor individual is selected.
h is the fraction of memes learned horizontally.
Prob h
Prob (1-h)/2
Prob (1-h)/2
-
random individual (equal probability)
parent 1
parent 2
One meme known by the donor is chosen randomly.
The recipient learns this meme.
If he already knows it, this is a wasted opportunity.
Cannot learn the same thing twice or get the benefit from it twice.
Assumes cultural fitness independent of biological fitness.
A new meme arises in the new individual with probability v
N = 500, s = 0.1,
v = 0.2 (high),
L = 100
mean fitness increases with h
All memes are good.
It doesn’t matter who
you get them from.
Therefore chose widest
range of meme donors.
Incorporate Variable Fitness Memes
Previously all memes were good – selection coefficient s
Now assume each meme has a different selection coefficient si
uniformly distributed between +0.1 and -0.1 w  1 
si

i
completely vertical
transmission is
optimal
random memes have no effect –
good and bad memes spread
equally well
Two parents / Single parent
any random individual
parent 1
parent 2
2 parents (as before)
h
(1-h)/2
(1-h)/2
Single parent
h
1-h
0
2 parents only
1 parent + random is similar to
2 parents + random
1 parent only
STUDENTS SHOULD GO
HOME AT THIS POINT
Cultural Fitness and Choice of Donor
Biological fitness = prob of being a parent
Cultural fitness = prob of being donor of a horizontally transferred meme
So far assumed that horizontal meme donors chosen with equal prob
(i.e. cultural fitness independent of biological fitness).
Now assume horizontal meme donors chosen in proportion to biological fitness
(i.e. cultural fitness = biological fitness).
2 parents only
STUDENTS PLEASE
COME BACK
Ignore parents but
choose donors in
proportion to
biological fitness
Conclusions # 2 :
Learning from two parents is a successful strategy because
(i) you can learn more than either of them
(ii) parents are selected by fitness
Learning from one parent only is poor because you can only learn more than them if
you invent it yourself.
Learning from unrelated individuals is good if you can judge the fitness of the
individuals but poor if you cannot.
Caveat – we have not considered strategies in competition.
In game theory, high fitness strategies can sometimes be invaded.
Emergence of cultural groups.
Cultures diverge because of slow spread of memes between groups.
Partly geographical, but partly because of distrust of foreign memes.
Group Structure gives rise to Group Selection
Total population N is fixed.
Composed of many groups with variable population.
A group splits randomly in two when its population reaches a
maximum size Ngroup. Other groups die out.
Parent 1 chosen from whole population with probability proportional to w.
Parent 2 chosen from same group as parent 1 with prob. prop. to w.
 Prob. of being born into a group is prop. to mean fitness of group.
Is it better to learn from your own group?
H = fraction of memes learned from a different group.
Prob. 1-H - learn from a member of own group (do not distinguish parents, i.e. h = 1)
Prob. H - learn from a member of a different group.
Case 1: donors within group chosen in prop. to w, donors outside chosen with equal prob.
copy the good guys from your own
group plus a small fraction of
random memes from outside
copy the good
guys from your
own group
copy random
memes
N = 500
Ngroup = 100
-0.1 < s < 0.1
Is it better to learn from your own group?
Case 2: donors within group and outside group chosen in prop. to w.
Case 3: donors within group and outside group chosen with equal probability.
Case 2: a small amount of learning from
outside is almost as good as H = 1
Case 2: if you can judge external
information, learn from a wide
range of donors.
Case 3: a small amount from
outside improves things.
Case 3:
Random copying
from your own
group works by
group selection.
Conclusions # 3 :
Learning from your own group is a successful strategy because the probability that you
are born in a given group is proportional to the mean fitness of the group. Groups with
higher fitness grow and divide.
- this works even if you copy random members of the group.
Therefore we would expect groups to emerge that favour memes from their own
group relative to foreign memes.
Origin of ethnic markers.
Cultures of different groups will diverge – like origin of species in genetic case
In all cases, a small amount of learning from outside is beneficial with respect to only
learning from your own group.
Completely isolated cultures get left behind….
A large amount of learning from outside is bad if you cannot judge the quality of the
foreign memes as well as the local memes. It is good only if judgement of quality of
foreign memes is as good as local ones.
The last point is probably not true, so we expect strong bias toward in-group learning
Issues related to the Learning Mechanism
Currently – random copying. Means you can waste time seeing the same thing twice.
i.e. both donor and recipient are stupid.
What if the donor tells you a meme chosen randomly from those you don’t know?
What if the donor tells you the highest fitness memes they know first?
Do individuals actually know the fitness of their memes?
If you learn a negative fitness meme, can you ignore it? Maybe fitness should depend
only on the positive memes known.
What if there is a cost to being a good teacher – i.e. parent spends time with child?
What if the donor tells you memes in the order in which they learned them – i.e. simple
important things first?
What if memes have prerequisites? Deep versus superficial learning.
The title “Is it Good to Share?” was inspired by
Bob Hoskins
Campaign for British Telecom in 1990s
“It’s Good to Talk”
Do memes drive ever-increasing communication rates?
How much time do we spend communicating?
Is this necessary or beneficial?
Do we really need the latest devices – emails, mobile phones and
blackberries?
Memes that promote increasing effort in communication will spread.
Where do we go from here?
Group memory - division of knowledge – trades
Cooperation within groups
Cooperation/conflict between groups – Which memes lead
to successful survival and multiplication of cultures?
Is this a good subject to work on?
Lots of interesting questions regarding memes and cultural evolution.
These questions have importance for understanding human history, the
development of civilization and the quality of life.
BUT where are the data/observations/experiments?
Without data this subject will remain on the fringes of respectability…..
The Mimetic Transition
Higgs (2000) Proc. Roy. Soc. B 267:1355-1361