Combine mechanisms and evidence without mistaking a plausible model for an established history.
By the end: Assess a resistance case and distinguish demonstrated change, causal mechanism, historical inference, and remaining uncertainty.
Combine independent kinds of evidence
Fossil succession, geographic distributions, inherited anatomical patterns, and molecular similarities constrain evolutionary explanations in different ways. Fossils document past forms and contexts but incompletely sample past life. Shared derived traits and sequence patterns help reconstruct relationships; converging evidence is stronger than one isolated resemblance.
For example, a shared DNA insertion at corresponding genomic positions can support common ancestry when inheritance is a better explanation than repeated independent insertion. Comparisons across additional loci test whether that pattern is consistent; linked sites are not all independent confirmations. Scientists also observe evolution in contemporary populations and controlled studies. Such observations can establish a frequency change or test a mechanism under particular conditions. They do not automatically reconstruct every earlier event. Distinguish the broad evidence for descent with modification from the evidence for a particular adaptation, divergence date, or causal variant.
Match the evidence to the historical claim
An evolutionary explanation can contain several claims with different evidence requirements. That a population changed is a measurement claim. That selection contributed is a causal claim about unequal inherited reproductive contributions. That two lineages inherited a feature from a shared ancestor is a historical claim. That the feature originally spread because of its present function is a further historical claim about selection. Evidence can establish one without establishing the others. Write the exact claim rather than treating every observation as a test of evolution in general.
Consider a constructed comparison of three lineages, A, B, and C. They share corresponding DNA regions, but only A and B contain a particular insertion between the same neighboring sequences. If those neighboring sequences identify the same inherited genomic position, one insertion in the ancestral lineage of A and B can explain both copies. An alternative requires separate insertions or an earlier insertion followed by loss in C. Comparing explanations requires assumptions about how often insertion, loss, and transfer occur. Event counts alone do not establish the most probable history.
Corresponding positions support a history
Constructed sequence and ancestry schematic, not measured genomes. Blue regions mark corresponding flanks; the gold segment I is present in A and B. One inherited insertion explains this pattern, but loss, repeated insertion, and transfer are alternatives to evaluate. Lengths do not encode base counts or time.
Now suppose ten nearby sequence differences agree with the insertion pattern. That is more information, but it need not represent ten independent histories. Nearby sites may have traveled together through inheritance because recombination seldom separated them in the relevant history. Evidence from another genomic region can test whether the same grouping extends beyond that inherited segment. Disagreement does not automatically invalidate ancestry. Ancestral variation, genetic exchange, and reconstruction errors can produce different gene histories. The question becomes which processes explain the agreement and disagreement across regions.
Fossils provide a different kind of constraint. A securely identified fossil in a dated context establishes that the represented lineage or feature existed by that time, within dating and identification uncertainty. It does not establish that the lineage first arose at that exact moment. Earlier organisms may have left no recoverable fossils, or suitable rocks may remain unsampled. In a hypothetical record with a feature at twelve million years ago and no older observation, twelve million years is evidence of existence, not a complete estimate of the feature's origin.
Absence becomes informative only relative to an opportunity to detect. Imagine ten independent suitable deposits, each with a one-half chance of preserving and revealing a feature if it was present. Missing it in all ten would have probability one-half multiplied by itself ten times, about 0.098%. If the chance per deposit were only one-tenth, missing all ten would instead have probability 0.9 to the tenth power, about 34.9%. These constructed probabilities illustrate why preservation and sampling assumptions matter. Real deposits are not necessarily independent or equally informative.
A missing observation depends on detection
Case if the feature is present
Miss per opportunity
Miss all opportunities
10 suitable deposits; detect with probability 0.5
0.5
0.5¹⁰ ≈ 0.098%
10 suitable deposits; detect with probability 0.1
0.9
0.9¹⁰ ≈ 34.87%
100 cells; R frequency 0.01
0.99
0.99¹⁰⁰ ≈ 36.60%
Constructed probabilities, not estimates of fossil preservation or microbial sampling. All rows assume independent opportunities, a constant probability per opportunity, and perfect classification when detected.
Compare the sequence and fossil examples before combining them. The insertion comparison tests inheritance relationships under a model of sequence change. The fossil observation constrains when a feature already existed. Neither alone measures the reproductive advantage of that feature in its ancestral environment. A present benefit therefore cannot fill the missing historical steps. Converging evidence means independently constrained observations fit one history, not several descriptions of the same sequence pattern. State which part of the proposed history each observation actually restricts.
Follow a resistance case through the life cycle
Suppose an inherited resistance variant is initially rare and becomes common under antimicrobial exposure. Possible contributors include selection on existing variants, new mutation, transfer of resistance DNA, migration of resistant lineages, and drift. Resistance is a property of microbes in this discussion, not the host person becoming resistant to a medicine.
Ask whether susceptible and resistant types differ in reproductive contribution under exposure, whether measurements count living descendants or only DNA, and whether population composition changed. More resistance can coexist with fewer total microbes. If resistance has a cost without exposure, selection may favor susceptibility there; costs, compensatory evolution, and continued exchange can prevent a simple reversal.
Measure resistance without confusing the quantities
For a resistance audit, define both the inherited variant and the measured response. Resistance describes a microbial property under specified exposure conditions. A DNA sequence classified as a candidate resistance allele is not automatically a measurement of that response. Other genes or environmental conditions may alter its effect. Conversely, distinct genetic changes can produce similar responses. An audit should say whether the observations concern sequence copies, organisms carrying a sequence, or organisms expressing a resistant phenotype. Those categories can coincide but are not interchangeable.
Use a constructed haploid population with one tracked copy per cell. Initially it contains ten thousand cells, of which one hundred carry R and express resistance in the defined environment. Later it contains one thousand cells, of which eighty carry R and express resistance. R frequency rises from one hundred divided by ten thousand, or 1%, to eighty divided by one thousand, or 8%. Yet the number of R cells falls by twenty. Total abundance falls by nine thousand. A larger share can result because the alternative type declined still more strongly.
The haploid assumptions made the copy and carrier fractions equal. In a separate diploid example, one hundred individuals include ten RR, twenty Rr, and seventy rr. There are forty R copies among two hundred total copies, so R frequency is 20%. Thirty individuals carry R, giving a 30% carrier fraction. If resistance requires RR in this constructed example, the resistant phenotype fraction is only 10%. The measurements count different things. The phenotype rule requires evidence rather than an allele name.
Copies, carriers, phenotype, and abundance
Population and quantity
Numerator / denominator
Result
Haploid, earlier R frequency
100 R copies / 10,000 copies
1%
Haploid, later R frequency
80 R copies / 1,000 copies
8%
Haploid, change in R cell count
80 cells − 100 cells
−20 cells
Diploid, R frequency
(2 × 10 + 20) copies / 200 copies
20%
Diploid, R carriers
(10 + 20) individuals / 100 individuals
30%
Diploid, resistant phenotype
10 RR individuals / 100 individuals
10%
Constructed examples with explicit assumptions. The diploid example assumes RR alone expresses resistance. The haploid example assumes one tracked copy per cell and matching R genotype and phenotype. These are separate populations.
Before interpreting a frequency shift, check what entered the samples. A sample enriched for one habitat or life stage can differ from an earlier sample even if neither local breeding population changed. DNA measurements may count material that does not represent living reproductive contributors. Moving organisms may change a site census without contributing genes to its descendants. Genuine gene flow requires inherited genetic contribution to the recipient population, whether through reproduction or another specified transfer route. Sampling design and population boundaries belong in the explanation.
An earlier sample with no R does not establish that R originated later. If the true frequency is 1% and one hundred cells are independently sampled, each draw has probability 0.99 of missing R. Missing it every time has probability 0.99 to the hundredth power, about 36.6%. A negative sample is therefore quite plausible despite pre-existing variation. This calculation assumes representative independent draws and perfect classification. Clustering, incomplete detection, and changes in the sampled population require a different model, so the number is not a universal detection limit.
Distinguish mutation timing from subsequent multiplication. An early mutation can leave many descendants even before it has any reproductive advantage. The secondary history of the Luria and Delbruck experiment describes variable resistant counts consistent with such timing differences, against a simpler exposure-induced model. That historical experiment concerned virus resistance in bacteria, not antibiotic treatment. Here the original paper's identity was checked, but its full text was not reviewed. Its lesson for this audit is to compare predictions about lineages and distributions, rather than infer a mutation's origin from its eventual frequency.
One mutation event can leave different counts
Invented binary lineages, not historical experimental observations. Time runs downward; each circle represents one cell. R labels mark inherited mutant copies. Both histories have one origin event, equal division, and no selection or reversion. Endpoint mutant counts therefore do not count mutation events.
Work through the mechanism
Observation
Possible inference
Important alternative
Variant frequency rises
Differential reproduction may favor it
Migration or sampling change
Resistant group expands
Earlier origin may have more descendants
Different fitness or transferred DNA
Variant and phenotype co-occur
Candidate causal relationship
Linkage or genetic background
Each inference requires a design that can distinguish the listed alternatives.
Combine mechanisms in an explicit life cycle
These are paper calculations, not additional lesson controls. Begin with haploid R frequency 0.1. Suppose R has twice the expected reproductive contribution of the alternative S under the specified exposure. Weighting gives 0.1 times two, or 0.2, for R and 0.9 times one, or 0.9, for S. Their total contribution is 1.1. The predicted offspring frequency is therefore 0.2 divided by 1.1, or about 18.18%. Relative contributions do not predict total offspring numbers.
This change does not require any new mutation during the interval. Selection can change the share of an inherited variant already present. Nor does the calculation prove that selection caused an observed increase to 18.18%. A mixture with another population might produce the same endpoint. Compare explanations that include genetic exchange and uncertainty in reproductive contributions. Reproducing a final number with one model establishes consistency, not unique identification of its cause.
Life-cycle order can separate predictions. Suppose an external source has R frequency 0.6 and supplies 20% of the recipient gene pool. Mixing first gives 0.8 times 0.1 plus 0.2 times 0.6, or 0.20. Applying the same twofold reproductive difference afterward gives R contribution 0.40 and S contribution 0.80. The result is one-third, about 33.33% R. Here every source copy enters before the reproductive comparison and experiences the same stated environment. No further exchange, mutation, or random sampling occurs in this deterministic step.
Reverse the order while retaining the specified genetic-contribution weight at mixing. Selection first produces 2/11 R. Mixing afterward gives 0.8 times 2/11 plus 0.2 times 0.6, or 73/275, about 26.55%. The incoming contribution bypasses that interval of local selection, so the result differs from one-third. This comparison is not a claim that migrant arrival counts remain identical between histories. The weight is defined as a share of the pool at the mixing stage. Drawing the stages exposes which biological quantity a parameter must represent.
The same mechanisms in different orders
Life-cycle order
Intermediate R frequency
Final R frequency
Mix, then select
0.8 × 0.1 + 0.2 × 0.6 = 0.20
0.40 / (0.40 + 0.80) = 1/3 ≈ 33.33%
Select, then mix
0.20 / (0.20 + 0.90) = 2/11
0.8 × 2/11 + 0.2 × 0.6 = 73/275 ≈ 26.55%
Constructed deterministic haploid arithmetic. Initial R frequency is 0.1; source frequency is 0.6; incoming genetic contribution is 20% at the mixing stage. R contribution is twice S in the selection stage. No new mutation or drift is included.
Finite reproduction adds uncertainty even after the weights are correct. Take one R among ten haploid copies and retain the twofold advantage in the relevant environment. Its weighted share is two out of eleven. If ten offspring copies are sampled independently from that pool, the chance that every draw is S is nine-elevenths to the tenth power, about 13.44%. The expected R count is ten times two-elevenths, about 1.82, but the realized count can be zero. Selection changes sampling probabilities; it does not remove genetic drift.
Try changing one assumption at a time on paper. Set the reproductive contributions equal and the expected R frequency stays at 0.1, while ten neutral offspring draws lose R with probability 0.9 to the tenth power, about 34.87%. Set the incoming source frequency equal to the resident frequency and mixing alone changes nothing. Finally remove exposure and suppose R contributes only 0.9 relative to S at one. Starting at 0.1 gives 0.09 divided by 0.99, about 9.09%. That decline is conditional on a new fitness assumption, not an automatic consequence of ending exposure.
Expected contribution does not guarantee persistence
Reproductive model
R share per draw
Expected R count
Chance of zero R
Equal contributions
1/10
10 × 1/10 = 1
(9/10)¹⁰ ≈ 34.87%
R contributes twice as much as S
2/11
10 × 2/11 ≈ 1.82
(9/11)¹⁰ ≈ 13.44%
Constructed model: one R among ten haploid copies, followed by ten independent offspring draws from the weighted pool. Fitness values apply only to the stated environment. The table compares one-generation loss, not eventual fixation.
Specify what would change your conclusion
A useful study compares hypotheses rather than merely observing that a story is possible. Relevant evidence may include before-and-after samples, independent lineages, genetic sequence histories, fitness comparisons, and controls for environmental or sampling differences. A sequence association alone does not prove that a variant caused the phenotype; linked variants and background effects can matter.
The models in this course isolate processes so their logic can be checked. Real inference needs measured parameters, uncertainty, and appropriate comparisons. Do not convert a toy selection coefficient into a treatment recommendation. The course provides no instructions for culturing resistant organisms or choosing antimicrobial treatment; clinical decisions require appropriate professional guidance.
Separate causal evidence from a plausible narrative
Suppose a sequence variant and resistance appear together in a set of sampled lineages. That association makes the variant a candidate explanation, but shared ancestry can confound the comparison. Related lineages may inherit both the candidate and another causal difference. If every carrier also carries that second difference, the observations cannot distinguish their effects. Repeated observations within one inherited group do not create independent tests of the candidate. An audit must ask what comparison separates the proposed cause from the other inherited differences accompanying it.
Consider a constructed dataset with two genetic backgrounds, X and Y, each represented by one hundred organisms. In X, eighty carry the candidate and twenty do not; all express resistance. In Y, twenty carry the candidate and eighty do not; none express resistance. Across backgrounds, eighty of one hundred carriers are resistant, compared with twenty of one hundred noncarriers. That looks like a strong association. Within either background, however, carrying the candidate makes no difference to the observed response. Background explains the pattern without a candidate effect.
An association can come from genetic background
Sample group
Candidate carriers: resistant / total
Noncarriers: resistant / total
Background X
80 / 80 = 100%
20 / 20 = 100%
Background Y
0 / 20 = 0%
0 / 80 = 0%
Pooled
80 / 100 = 80%
20 / 100 = 20%
Constructed counts of organisms, not empirical data or allele-copy frequencies. Within background X everyone is resistant, and within Y nobody is. Pooling backgrounds produces an association that cannot distinguish a candidate effect from background differences.
This example does not establish that real candidate variants are harmless markers. It establishes what the constructed observations fail to distinguish. More informative evidence compares the candidate across relevant backgrounds and environments and asks whether its proposed molecular effect explains the phenotype. Repeated natural separations of linked variants can also help distinguish their histories. The table cannot replace evidence about inheritance and mechanism. Report the candidate relationship and unresolved alternative rather than calling a correlation an established cause.
A historical adaptation claim asks a further question even after a causal effect is established. A variant might improve resistance today but have arisen in another environment, spread by drift, or been maintained for a different effect before the current exposure. It might also have arrived through transferred DNA after evolving elsewhere. Demonstrating its present biochemical effect and reproductive benefit supports a contemporary mechanism. Establishing why it originally spread needs evidence about ancestral environments and lineage history. Current usefulness does not by itself identify the historical selective cause.
As a guided audit, write two competing explanations for a rise in R frequency. One uses selection on standing variation; the other uses incoming R lineages with equal local reproductive contributions. Both can predict an increase, so another before-and-after count alone may not distinguish them. Ask instead whether local descendants show unequal contribution and whether genetic histories support incoming ancestry. Then consider transfer of only the relevant DNA segment. That alternative need not bring the rest of the donor lineage's genome, so evidence about whole-organism ancestry and segment ancestry answers different questions.
Finish by naming a result that would revise each claim. Equal reproductive contributions in the measured environment would weaken the proposed selection explanation for that interval. Evidence of an earlier R lineage would contradict a claim that R first originated afterward, without disproving later selection. Failure to reproduce a candidate effect across relevant backgrounds would restrict its causal scope. These revisions target particular claims, not the broad fact of inherited population change. Report what is measured, what the model assumes, and which historical alternatives remain. None of these teaching calculations determines a treatment choice.
A result revises a specific claim
Proposed claim
Result that challenges it
What can still be true
Selection favored R in this interval
Equal inherited contributions under the measured conditions
R frequency changed through another process
R first originated after exposure
An authenticated earlier lineage carries R
Exposure later favored R
Candidate alone explains resistance
Response differs across relevant genetic backgrounds
Candidate has a conditional effect
R originated for its current function
Earlier history supports another function or neutral spread
R is useful in the current environment
Hypothetical evidence comparisons, not a study protocol. Each result addresses a narrower question than whether populations evolve. The strength of a revision depends on representative samples and the stated environmental scope.
Exercises
Resistance rises while burden falls
Before exposure, a haploid population has 1,000 cells, 10 resistant. Afterward it has 100 cells, 20 resistant. Calculate frequency and abundance changes, then state what mechanism is established.
Hint
Do not equate a change in composition with a uniquely identified cause.
Worked answer
Measurement
Before
After
Total cells
1,000
100
Resistant cells
10
20
Resistance frequency
1%
20%
Population size and composition must both be reported.
Resistance frequency rose from 1% to 20%; resistant abundance rose from ten to twenty while total abundance fell by 90%. These observations establish those measured changes. They are consistent with selection but do not by themselves exclude mutation, transfer, migration, or sampling differences. Demonstrating a mechanism requires more evidence.
Will stopping exposure erase resistance?
A constant-fitness model predicts resistance rising under exposure. Someone concludes that removing exposure must return it to its original frequency. Is that justified?
Hint
The model says nothing about relative fitness after the environment changes unless you specify it.
Worked answer
After exposure ends
Possible consequence
Resistance neutral
No deterministic decline from selection alone
Resistance costly
Decline possible if alternatives persist
Compensation or gene flow
Persistence can occur
Alternative allele absent
Cannot restore it by selection alone
Predicting reversal needs a new environmental and genetic model.
No. Resistance could be neutral, costly, or beneficial for another reason without exposure. Compensatory changes or continued gene flow can preserve it, and if susceptibility has been lost it needs a source to reappear. Even a cost does not imply a fixed return time or the original frequency. The toy model cannot determine a clinical strategy.
Sources and further reading
The explanations and exercises are complete without these optional links. The references cover inheritance, population mechanisms, and evidence for ancestry. Source annotations distinguish textbook explanations, secondary history, and a primary paper whose bibliographic identity alone was checked. The examples are constructed teaching models, not empirical estimates or professional decision advice.
Public-health context and the distinction between microbial resistance and host response. No clinical treatment strategy or current burden estimate is inferred from our models.