MLB7052.LT: Modern Methods in Molecular Ecology
TalTech / UT Institute of Technology
Loeng 11 — molekulaarne evolutsioon (molecular evolution).
Every marker in this course is a piece of inherited text. This lecture is about how that text changes over time, why most of the change means nothing, and how a tree is reconstructed from it anyway.
Two questions sit under everything that follows:
Neither is answered by looking at an organism. Both are answered by reading molecules as documents — with the understanding that the document has been copied, imperfectly, for about four billion years.
Extinction rate across the Phanerozoic. The “Big Five” are the red points — the deep-time backdrop against which molecular lineages survive or end.
LUCA (last universal common ancestor) is the inferred organism from which every living thing on Earth descends. It is not the first cell; it is the last one whose descendants are all still with us.
Everything reconstructed about LUCA is reconstructed backwards from living sequences — the same logic you will use on a ten-sample dataset, applied to the whole tree.
Märkus
LUCA is a node, not a fossil. Nobody has a specimen. What we have is the set of features so widely shared that the most economical explanation is common inheritance rather than repeated invention: the genetic code, ribosomes, ATP, membrane chemistry.
Primordial / prebiotic soup, roughly 3.7–4.0 billion years ago.
What remains hard is not making monomers. It is getting a polymer that both carries information and acts as a catalyst.
Evolutionary change is not a separate compartment from ecology. It feeds populations, communities and ecosystems, and is fed back by them.
The chicken-and-egg problem: DNA stores but does not catalyse, protein catalyses but is not copied directly. Something had to do both.
The competing systems-chemistry model says the prebiotic soup never contained only one polymer: RNA and DNA building blocks coexisted, formed RNA–DNA chimeras, and the clean division of labour came later.
TNA — threose nucleic acid — is a genetic polymer in which RNA’s natural five-carbon ribose is replaced by an unnatural four-carbon sugar. It base-pairs with RNA and DNA, which makes it a candidate pre-RNA genetic system.
Tähtis
The question is not which molecule came first, but how a genetic takeover happens — one polymer handing its information to another without losing it.
Molecular evolution is the evolution of biological macromolecules — DNA, RNA, proteins. Three things at once:
It sits between fields you already know, and borrows from all of them:
| Field | What it contributes |
|---|---|
| Genetics, molecular biology | what a mutation is, mechanistically |
| Population genetics | how a variant’s frequency behaves in a population |
| Genomics | the scale — whole genomes, not single loci |
| Mathematics, bioinformatics | the models and the algorithms that fit them |
Märkus
Reconstructing the evolution of genes and of groups of organisms is molecular phylogenetics (molekulaarne fülogeneetika) — the applied core of this lecture.
Fülogeneetika (phylogenetics) — the branch of biology that studies the origin and descent of groups of organisms: populations, species, families. From Greek phylē (tribe) + genetikos (relating to origin).
Fülogenees (phylogeny) — the descent history itself. phylē + genesis (birth).
The aims are concrete:
A tree is a hypothesis with parts: branch points are inferred common ancestors, horizontal lines are lineages, and extinct tips are as much a part of the topology as living ones.
Why molecules beat morphology for this job:
The neutral theory holds that most evolutionary change at the molecular level, and most of the variation within and between species, is due to random genetic drift of selectively neutral mutants — not to selection.
Read the qualifications carefully, because they are what make it testable:
Tähtis
A neutral mutation is one that does not affect the organism’s ability to survive and reproduce. Because only a fraction of gametes is sampled each generation, such an allele can reach fixation by chance alone.
The mathematical machinery came from population genetics — Haldane, Fisher, Wright.
Haldane’s dilemma was the motivation. Haldane estimated that a beneficial mutation takes ~300 generations to fix in a mammalian lineage. The substitution rate observed between humans and chimpanzees (~1.5 per year) is far too high to be explained by beneficial mutations alone. Something else must be doing most of the work.
Drift in action: identical starting frequency, 20 replicate populations, and by generation 20 the allele is fixed in some and lost in others. No selection anywhere in this picture.
The prediction: as functional constraint diminishes, the probability that a mutation is neutral rises — and so should the rate of sequence divergence.
The evidence, and it is the reason the theory was taken seriously:
Märkus
This is the logic behind choosing a marker. A gene region is useful at a given taxonomic depth because of how constrained it is. Constraint is not a nuisance — it is the tuning knob.
Kimura’s infinite sites model (ISM). With diploid individuals and a neutral mutation rate per gene copy, the number of new mutants per generation is , and each has probability of eventually fixing:
The population size cancels. The rate at which fixed differences accumulate between diverging lineages equals the per-individual mutation rate, independent of how large the populations are.
That is the theoretical basis of the molecular clock — and note that the clock was an empirical observation before the neutral theory explained it.
Three clocks in the same lineage. Each is linear, each ticks at its own rate, and the rate is set by constraint.
Tomoko Ohta added the mutations that are neither neutral nor clearly deleterious: slightly deleterious ones.
Their population dynamics differ from strictly neutral mutations only when the selection coefficient exceeds . Below that threshold, drift overwhelms selection and a slightly harmful allele behaves as if it were neutral.
The consequence is important, and it is why this belongs next to the population-genetics lecture:
The effective population size decides how much of the genome is effectively neutral. In a small population, more mutations behave neutrally; in a large one, selection sees them.
Small populations lose genetic diversity and gain drift. Ohta’s threshold says the same populations also lose the ability to purge mildly harmful variants.
Selection moves a distribution in a direction, splits it, or narrows it. Each mode leaves a different footprint in allele frequencies — and none of them looks like drift.
A bottleneck removes alleles wholesale. The greater prairie chicken went from millions to fewer than 50 birds, and the allelic diversity did not come back with the population.
The neutral theory is the null. Both panels are departures from it — but only one of them is adaptation, and telling them apart is the everyday work of molecular ecology.
A character is homologous (homoloogne) when it is inherited from a common ancestor.
A character is homoplasious (homoplaasne) when it arose independently in two lineages — convergence, parallelism, or reversal.
Phylogenies may be reconstructed from homologous characters only. Homoplasy is not a minor nuisance: it actively supports the wrong tree, because it makes unrelated lineages look similar.
The trouble is that you cannot tell them apart by inspection. Similarity is the observation; homology is the inference.
Dark coats on dark lava, light coats on light sand — and the dark phenotype evolved separately in different lava flows. Group the mice by coat colour and you recover geology, not genealogy.
Not all homology is the kind you want. Three flavours, and only one of them tracks speciation:
Tähtis
Build a species tree out of paralogs by accident and you will get a confident, well-supported, wrong answer. This is the single most common way a beginner’s phylogeny fails.
Speciation as a gradient: a barrier to gene flow, then genetic divergence, then reproductive isolation. Orthologs date from the branch point; the barrier is what made it one.
Before anything can be compared, positions have to be made comparable. Alignment (joondamine) is the step that decides which nucleotide in sequence A corresponds to which nucleotide in sequence B.
Every number downstream — distance, tree, support value, divergence date — inherits whatever the alignment decided.
Märkus
An alignment is a hypothesis of positional homology. It is not data preparation; it is the first inference in the pipeline, and the one that is most often accepted without inspection.
A dot is placed wherever the two sequences carry the same nucleotide at that pair of positions.
A diagonal run of dots is an alignment. Where the diagonal shifts, an insertion or deletion is being proposed. Off-diagonal dots are chance matches — with four bases, a quarter of all cells hit by accident.
Two different paths through the same plot are two different alignments of the same pair of sequences. Neither is given by the data; you have to choose.
The cost of an alignment:
The weight sets the exchange rate between “call it a substitution” and “call it a gap”:
| Alignment with gaps | Alignment with substitutions | |
|---|---|---|
At the gapped alignment wins. At the substitution-based one does. Same sequences, same algorithm, different answer.
Tähtis
The gap penalty is a parameter you choose, not a property of the data. If you have never changed it, you have never tested whether your alignment depends on it.
In practice you will not set by hand — the aligner has a default. That is precisely the problem: the default was chosen for someone else’s data.
Aligning three or more sequences is not the pairwise problem repeated. The cost depends on which sequences are compared against which:
The catch is circular and worth stating out loud: the guide tree used to build the alignment is built from a quick-and-dirty distance estimate, and the final tree is then built from that alignment.
Phylogenetic trees are constructed only from homologous sequences. Methods differ, but they share a skeleton:
The shape of a sequence-based tree is its topology (topoloogia).
The more sequences you analyse, the more possible topologies exist — the number of distinct trees grows factorially, which is why this is a search problem and not a calculation.
Juurteta puu (unrooted tree) — shows relationships, but not the direction of time.
Juurega puu (rooted tree) — adds a root, and therefore an ancestor and a direction from past to present. The root normally comes from an outgroup, not from the data itself.
Märkus
An unrooted tree is often all the data support. Drawing it with a root anyway is a decision that should be defended, not a formatting choice.
The principle of parsimony: the best tree is the one requiring the fewest mutational events to explain the observed sequences from a common ancestor.
The classic worked example — the hominoid tree from 896 bp of mitochondrial DNA:
| Topology | Mutations required |
|---|---|
| Variant 1 — (chimpanzee, gorilla), human | 145 |
| Variant 2 — (human, chimpanzee), gorilla | 147 |
| Variant 3 — (human, gorilla), chimpanzee | 148 |
By parsimony, variant 1 wins. On current data, variant 2 is the correct one — human and chimpanzee are sister taxa.
Tähtis
Three mutations separate the “best” tree from the right one. Parsimony is a criterion, not an oracle: it is a statement about this alignment under this criterion, not a fact about the organisms.
Count the positions at which two sequences differ and you get the p-distance. The dashed line is what you would like it to be; the curve is what it is.
The gap opens because of multiple hits: a site can be substituted twice and look unchanged, or change back. As divergence grows, the observed difference saturates near 0.75 — the value expected for two random sequences.
Correction models (Jukes–Cantor, Kimura 2-parameter, and on upward) estimate the true number of substitutions from the observed one. Choosing a model is choosing which kinds of change you believe are happening.
The ancestor of modern humans lived in Africa 100 000–200 000 years ago. Both mtDNA and Y-chromosome sequencing put the out-of-Africa expansion at roughly 100 000 years ago.
The route, as read from the molecules:
From mtDNA — the mt-MRCA, popularly “mitochondrial Eve”. From the Y chromosome — the Y-MRCA, “Y-chromosomal Adam”.
Tähtis
“Eve” was not the only woman alive, and she and “Adam” need not have been contemporaries. An MRCA is the coalescence point of one locus — mtDNA is maternally inherited and the Y is paternal, so they trace two different lineages through the same population.
Every other locus in the genome has its own MRCA, at its own date. The species has no single ancestor.
A latitudinal cline in Adh allele frequency in Drosophila — and the cline has shifted about 4° of latitude in twenty years as the climate warmed.
An insecticide-resistance allele arises once and is then carried between populations. Gene flow, not repeated mutation.
Thirty years of trophy hunting: mean body weight and horn length both decline. Selection with a date and a cause.
These are the timescales molecular ecology actually works on. The neutral theory is the null model against which each of them is a departure.
Because identity has no scale attached. Without knowing the substitution rate for that region, the constraint acting on it, and the saturation level, 98% could mean a recent split or an ancient one under strong purifying selection. Identity is a measurement; divergence time is a model-based estimate from it.
Because the number of new neutral mutants per generation scales with population size as , while the fixation probability of each scales inversely as . The two cancel, leaving . This holds only for strictly neutral mutations — under Ohta’s nearly neutral theory, re-enters through the threshold.
Whether the shared character states are homologous or homoplasious, and whether the genes compared are orthologs. Convergence and accidental paralog comparison both produce well-supported wrong trees. Also check whether the sequences are saturated — at high divergence, chance similarity alone approaches 75%.
That the topology is not supported by the data independently of the alignment — the signal is in a region whose positional homology is ambiguous. The honest responses are to report the sensitivity, to exclude the ambiguously aligned regions, or to use a method that integrates over alignment uncertainty. Silently reporting whichever tree you preferred is the one unacceptable option.
Because if most substitutions are neutral, they accumulate at a rate set by the mutation rate rather than by the environment, giving an approximately clock-like accumulation of differences. Where selection dominates, or where swings widely enough to move Ohta’s threshold, the clock becomes unreliable — which is why dates need calibration and error bars.
How long species last in the fossil record — the timescale a molecular clock is being asked to measure.
Tähtis
peeter.laas@taltech.ee · peeter.laas@ut.ee
Worth your time:
Figures from Bowman & Hacker, Ecology, 6th ed., used for teaching.
Fall 2026