Lactase Persistence

"multiple independent variants have allowed various human populations to quickly modify LCT expression and have been strongly adaptive in adult milk-consuming populations" — Sarah Tishkoff and colleagues, 2007¹

"an existing promoter of the compensating gene could have its sequence altered, binding a transcription factor more strongly and thus leading to overexpression, which would be classed as a modification-of-function mutation" — Michael Behe, 2010²

Summary

Most mammals, and most people, stop making the milk-digesting enzyme lactase after weaning.  Lactase persistence is the trait of making it for life.  It is the best-documented case of recent natural selection in humans, and it is a small change.  This article scores it with the rules in Measuring Biological Information and prices it in the currency the Benchmarking Evolution series uses:  people, generations and mutations.

1. The change is one letter in a switch, not a new gene. In Europeans it is a C-to-T change 13,910 letters upstream of the lactase gene, inside an intron of a neighbouring gene called MCM6.^{3 4} The stretch of DNA around it is an enhancer, a docking site for proteins that turn a gene up. Both versions of the enhancer work in cultured intestinal cells. The persistence version works harder, by about 1.3 to 4 times depending on the laboratory.^{5 6} In adult intestines the ordinary version gets chemically silenced with age and the persistence version does not.⁷ No protein changed. The gene stays on.
2. It happened at least five separate times. Five different one-letter changes, in Europe, East Africa and Arabia, each on its own chromosome background, now have laboratory evidence that they raise expression. Two more are candidates.^{8 8} This repetition is the strongest evidence in the case for the mainstream reading.
3. Selection was strong, but nobody measured it directly. Published selection coefficients run from 0.014 to 0.19. They differ by method, by the dates assumed and by the way the number is defined. Ancient DNA shows the European allele was rare until about 2000 BC and then rose from 2 of 26 chromosomes sampled at a Bronze Age battlefield in northern Germany to 144 of 192 in modern Europeans, over about 130 generations.⁹
4. It was not watched. The allele counts in dated skeletons are the nearest thing to an observation in human evolution. Their time axis rests on radiocarbon dates of bone. The selection coefficients are model output. Part 4 separates the three.
5. By Behe's own rules the main variants are modifications, not gains and not losses. Behe's 2010 paper scores this exact situation: a control element changed so that it binds a transcription factor more strongly is "a modification-of-function mutation, as an existing FCT was altered, not gained or lost".² One East African variant adds a new protein-binding complex and is a candidate gain. The common creationist description, a broken off-switch and a loss of information, does not match the molecular data. The weaker creationist point, that no new gene or protein was made, is correct.
6. One-letter changes like this are cheap. Each specific change arises about once in every 160 to 530 million gene copies passed on. A population of a million people makes a new copy of the European variant about every 81 generations, and about one copy in ten escapes loss by chance. The five independent finds show that a single-letter tweak to an existing switch is easy to find, and nothing about changes that need several letters before any helps.
7. The timing suits a short chronology better than most cases, but not entirely. Most of the rise falls in the last 3,200 years, where young-earth and conventional dates roughly agree. The earliest carriers are radiocarbon-dated to about 4700 to 4600 BC, before the Flood on either the Masoretic or the Septuagint chronology. A young-earth reading has to compress those dates.

Verdict.  Lactase persistence is real, recent, strong selection on an existing gene.  It adds one working letter at a time to a switch that already worked, and it did so at least five times.  It is a modification that helped, not a new function and not a breakdown.  It shows that single-letter regulatory improvements are within easy reach of human populations.  It does not show how switches or genes are built.

Part 1: What the change is

The discovery

In 2002 Nabil Enattah, Irma Järvelä and colleagues in Finland traced the trait in nine extended families.  Their abstract describes the ancestral state first:  "Adult-type hypolactasia, also known as lactase non-persistence (lactose intolerance), is a common autosomal recessive condition resulting from the physiological decline in activity of the lactase-phlorizin hydrolase (LPH) in intestinal cells after weaning."³  They narrowed the region to 47,000 letters on chromosome 2, sequenced it, and found "a DNA variant, C/T-13910, roughly 14 kb upstream from the LCT locus, completely associates with biochemically verified lactase non-persistence in Finnish families and a sample set of 236 individuals from four different populations."³

The variant sits "in intron 13 of the neighbouring Minichromosome Maintenance 6 (MCM6) gene", so it is not in the lactase gene at all.⁴  It is a single letter in a region that does not code for protein.

Two points follow for the rest of the article.  First, persistence is dominant:  one copy is enough.  Second, the switch-off is the normal mammalian pattern.  Joachim Burger and colleagues open their 2007 paper with it:  "Most mammals lose the ability to digest the milk sugar lactose after weaning because of an irreversible reduction in expression of the intestinal enzyme lactase".¹⁰  Non-persistence is not a defect.  It is the ancestral state, and the creationist sources agree.¹¹

What was measured in cells

Several laboratories tested the European variant in cells.  The usual method is to cut out the stretch of DNA around position -13,910, join it to a lactase promoter driving a light-producing reporter gene, and put it into Caco-2 cells, a human intestinal cancer line.  More light means more transcription.

• Lynne Olds and Eric Sibley, 2003. "A 200 bp region surrounding the C(13910) variant, associated with lactase non-persistence, results in a 2.2-fold increase in lactase promoter activity. The T(13910) variant, associated with lactase persistence, results in an even greater 2.8-fold increase."⁵ Their promoter came from the rat.
• Jesper Troelsen and colleagues, 2003. "The -13910 regions from both lactase persistent (-13910T variant) and lactase nonpersistent (-13910C variant) have enhancer activity. However, the -13910T variant enhances the LPH promoter approximately 4 times more than the -13910C variant when analyzed in differentiated Caco-2 cells."⁶
• Rikke Lewinsky and colleagues, 2005. A nuclear protein "binds more strongly to the T-13,910 variant associated with lactase persistence than the C-13,910 variant" linked to non-persistence. Supershift tests identified it: "Oct-1 binds directly to the T-13,910 variant". Oct-1 raised expression from both versions, and only when a second factor, HNF1-alpha, was present. Sites for four more intestinal factors lie in the same stretch, and three are needed for the enhancer to work.¹²
• Tine Jensen and colleagues, 2011, summarising: the variant "falls in an Oct-1 binding site and shows greater Oct-1 binding than the ancestral variant and increases enhancer activity."¹³

The sizes differ.  Anke Liebert's group later put the effect at "about twofold".⁴  Viviane Labrie's group called it "modest (T-13910 variant shows a ~1.5-fold higher activity compared to ancestral variant)."⁷  What every study agrees on is the direction and the kind of change.  The ancestral sequence is already an enhancer, and it already binds Oct-1.  The persistence letter makes the binding stronger and the enhancer more active.

What is inferred for people

A reporter gene in a cancer cell line is not a person.  Caco-2 cells make little lactase, and Liebert's group warns that such assays may underestimate the effect in the body.⁴  The cell work also cannot explain the timing.  An enhancer that is always a little stronger does not by itself explain why one person's lactase falls sharply after childhood and another's does not.

Labrie and colleagues supplied the missing piece in 2016 by measuring DNA methylation, a chemical tag that silences DNA, in intestinal cells from 115 adults.  In people with two ancestral copies, the region around -13,910 was 82.1% methylated.  In people with two persistence copies it was 19.3%.⁷  In sperm all three genotypes were more than 94% methylated there, so the difference arises during life.  Their reading:

"Our findings suggest that the lactase non-persistence haplotypes containing the C-13910 allele accumulate modified cytosines that silence the regulatory elements in MCM6 and LCT, while the lactase persistence haplotype containing the T-13910 allele displays age-related modification changes to maintain LCT activity."⁷

Liebert's group joined the two findings:  "It seems likely that changes in transcription factor binding, might be responsible for preventing this methylation, and thus allowing continued adult expression of LCT in carriers of this, and the other functional variants."⁴  That link is an inference.  Nobody has watched stronger Oct-1 binding hold off methylation in a living intestine.

So the measured facts are these.  The variant is associated with the trait in people.  It strengthens an existing enhancer in cells.  The enhancer escapes age-related silencing in persistent adults.  The causal chain between them is inferred, and it is a reasonable inference.

Part 2: At least five separate origins

The African variants

The European letter is almost absent in Africa, yet many African herders digest milk as adults.  In 2007 Sarah Tishkoff and colleagues gave lactose tolerance tests to 470 people from 43 ethnic groups in Tanzania, Kenya and Sudan.  They "identified three SNPs (G/C-14010, T/G-13915 and C/G-13907) that are associated with lactase persistence and that have derived alleles that significantly enhance transcription from the LCT promoter in vitro.  These SNPs originated on different haplotype backgrounds from the European C/T-13910 SNP and from each other."¹  A haplotype background is the set of neighbouring letters a variant sits among.  Different backgrounds mean the same letter change did not travel from one population to another.  It arose again.

The effect in cells was smaller than for the European variant.  "The C-14010–, G-13915– and G-13907–derived haplotypes consistently drove higher expression (from ~18%-30%) than the haplotypes with the ancestral alleles."¹  C-14010 is the common one.  It is carried on 46% of chromosomes in the Afro-Asiatic speakers of Tanzania they sampled, 39% in the Nilo-Saharan speakers of Tanzania and 32% in those of Kenya.¹

Their conclusion was hedged:  "at least two, and probably four or more, distinct causal variants associated with lactase persistence (T-13910 in Europeans and C-14010, G-13907 and G-13915 in Africans) have evolved independently".¹

Arabia, and a disagreement

Enattah's group found the Arabian form in 2008.  In Saudis, persistence goes with G-13915 together with a second change 10,000 letters away, and "both SNPs of the compound allele, located 10 kb apart, are required for the enhancer effect" in their cell tests.¹⁴  The same paper disagreed with Tishkoff on one point:  "The European T−13910 and the earlier identified East African G−13907 LP allele share the same ancestral background".¹⁴  That does not merge the two.  They are different letters three positions apart, so they are two mutations whatever the background.  It does weaken the claim that every origin is on a separate lineage.

The count now

By 2017 Liebert and colleagues wrote that persistence "is associated with at least five independent functional single nucleotide variants in a regulatory region about 14 kb upstream of the lactase gene [−13910T (rs4988235), −13907G (rs41525747), −13915G (rs41380347), −14009G (rs869051967) and −14010*C (rs145946881)]."⁸  Two more, -14011T and -13779C, have what the same authors call "more limited evidence of function".⁸  The five lie in one stretch of 104 letters.

The "kind of change" column was worked out for this article from the DNA sequence around each variant, because it decides which mutation rate applies in Part 6.¹⁶

Liebert's group summed up the three neighbouring East African variants in one line:  "they have rather different effects on protein binding, which nevertheless leads to the same transcriptional outcome" as -14010C.⁴  Several different single letters, in one small stretch, each make the same switch stay on.

A caution about the haplotype evidence

The same group found that an older chromosome type, which carries none of the five variants, also shows the long unbroken stretches normally read as a sign of selection.  Recombination is suppressed in that region, and "such suppression may have exaggerated haplotype-based measures of past selection."⁸  That bears on the selection coefficients in Part 3 that come from haplotype length.  It does not touch the ancient DNA.

Part 3: How strong, and how fast

The selection coefficients

A selection coefficient, s, is the fitness edge of one genotype over another, per generation.  The numbers below are not comparable as printed.  Some measure the penalty on people with no persistence copy, some the gain per copy, some the gain only among dairy farmers.  They also rest on different assumed dates and population sizes.  They are listed with their methods and are not averaged.

The site's population genetics article explains why the convention matters.  Bersaglieri's model, "w11 = w12 = 1 and w22 = 1 − s", puts the whole number on the people who lack the allele.¹⁷  Read as a bonus to carriers, the top of the range would be about a quarter higher.

Iain Mathieson and Jonathan Terhorst showed why single numbers disagree.  When s is allowed to change, "selection for the persistence allele was strongest (s ≈ 0.08) from 150 to 100 generations before the present (roughly 4500–3000 BP) before decreasing to around 0.02 in the past 50 generations.  This large change in strength of selection might explain the wide range of estimates from models that assume a constant value".¹⁹  A constant s over several thousand years is an average of something that changed.

Mathieson and Terhorst also set the floor.  With ancient data their method can detect selection coefficients "around 0.02" and "cannot reliably distinguish smaller selection coefficients from zero."¹⁹  Most published values for lactase sit well above that floor.  The lowest, Bersaglieri's 0.014 and Evershed's constant 0.022, sit at it.  The haplotype values also carry the extra doubt raised in Part 2.

When the allele rose

Ancient DNA has turned the timing from inference into something close to a count.

• Early farmers lacked it. Nine Neolithic and Mesolithic skeletons from Germany, Hungary, Lithuania and Poland all carried two ancestral copies. Burger and colleagues concluded that persistence "was rare in early European farmers."¹⁰ Nine skeletons can show rarity. They cannot show absence.
• The first carriers. Laure Ségurel's group found the European allele first "in Ukraine 5,960 before present (BP)," about 4000 BC.²¹ Richard Evershed's group, with a larger set, put the earliest carrier at "c.4,700-4,600 BCE".²⁰ These are first detections, not the date of the mutation.
• Still rare in the Bronze Age. At the Tollense valley battlefield in northern Germany, about 3,200 years ago, 14 warriors gave a frequency of 7.1%. At Mokrin in Serbia, 4,100 to 3,700 years ago, 4.6%.⁹ Iain Mathieson's 2015 study found "an appreciable frequency of lactase persistence in Europe only dates to the last four thousand years."²²
• The rise. "Within 3,000 years, sample counts for these alleles increased from 2/26 to 144/192" for the persistence allele.⁹ That is from 8% to 75%. Ségurel's group found the same break: "After 3,000 BP, the mutation strongly increased in frequency in Europe, but not in Asia."²¹

Only the European allele has been followed through ancient DNA.  The African dates, about 3,000 to 7,000 years, are model output from haplotype length alone.¹

What drove it is disputed

The textbook story is that dairy farmers who could digest fresh milk out-bred those who could not.  Two recent studies undercut it.

Evershed's group mapped milk use from about 7,000 fat residues in prehistoric pottery.  "European milk use was widespread from the Neolithic period onwards but varied spatially and temporally in intensity.  Surprisingly, comparison of model likelihoods indicates that LP selection varying with levels of prehistoric milk exploitation provides no better explanation of LP allele frequency trajectories than uniform selection since the Neolithic."²⁰  Among about half a million people in the UK Biobank, "around 92% (95% CI 91.5% - 92.2%) of genetically LNP participants consume fresh cows’ milk".²⁰  They proposed that non-persistent people drank milk anyway, and that it hurt them only in famine and epidemic.²⁰

Ségurel's group looked at Central Asia.  Mongol herders get 35% of their summer calories from dairy, yet only 12% to 30% of Mongol and Kazakh herders are persistent, and in ancient Central Asian samples the allele stayed near 5% from the Iron Age on.²¹  Had it been selected at 4% from 3,000 years ago, starting at 5%, "the mutation should have reached a frequency of 51% (meaning a phenotypic prevalence of LP of 76%) in modern populations."²¹  They asked plainly:  "why has LP been selected for at all?"²¹

This dispute is about the cause.  Nobody in it doubts that the allele rose fast.  For a young-earth reader the dispute changes nothing about the timeline:  every side puts the selection inside the last several thousand years.

Part 4: Was it watched?

The Observed Evolution series admits a case only if something was actually watched happening.  Lactase persistence was not watched.  Nobody followed a population and counted who digested milk, generation by generation.  But the evidence comes in three layers, and they differ in how much they depend on assumptions.

The counted layer is as close to observation as human evolution gets.  The frequency of a single letter went from rare to common in a time series of real people.  Burger's generation count is an assumption stated in the paper:  "We assumed that 130 generations have passed between the Tollense battle and the modern CEU reference population, implying a generation time of roughly 25 years."⁹

The modelled layer is where the dates of haplotype studies come from, and those numbers are soft.  Tishkoff's interval for the East African age runs from 1,200 to 23,200 years.¹  Enattah's discovery paper judged from its spread across populations that the European variant was "very old."³  Ancient DNA later showed it was not common until the Bronze Age.

To make the case observed, someone would have had to type the variant in a living population over enough generations to see the frequency move, with the fitness of each genotype recorded.  The ancient-DNA series is the substitute, and it is a good one.

What a young-earth timeline does with these dates

On the Masoretic text the Flood falls about 2500 to 2350 BC, and on the Septuagint about 3300 BC, as the site's Old Testament text and chronology article works out.  Creation Ministries International places it "around 4,500 years ago".¹¹

Most of the European rise needs no reinterpretation.  Tollense, about 1250 BC, falls centuries after the Flood on either chronology, and the rise from 8% to 75% happened after it.  Mathieson's Bell Beaker carrier, about 2300 to 2200 BC, sits near the Masoretic Flood date.²²  This part of the record fits a short chronology as well as a long one.  A recent, fast, human-driven sweep is what a young-earth model expects, and it is the kind of adaptation Creation Ministries International already accepts.

The earliest carriers do not fit as dated.  Evershed's 4700 to 4600 BC and Ségurel's 5,960 BP are older than the Flood on both chronologies, and Evershed's date is older than the Masoretic creation.  A young-earth reading must hold that radiocarbon ages of about 6,000 to 7,000 years are too old by one to three thousand years.  The usual proposal is that the atmosphere held less carbon-14 in the centuries after the Flood.  The site's radiocarbon article weighs that model and reports the problem it has not answered:  tree-ring calibration runs continuously past 10,000 BC.  So the timing of the rise suits a short chronology.  The timing of the first carriers does not, unless those radiocarbon dates are wrong by a large and specific amount.

Part 5: How to score it

The definitions

This series scores with Michael Behe's 2010 definitions, as Measuring Biological Information explains.  His unit is the functional coded element, or FCT:  "a discrete but not necessarily contiguous region of a gene that, by means of its nucleotide sequence, influences the production, processing, or biological activity of a particular nucleic acid or protein, or its specific binding to another molecule.  Examples of FCTs are: promoters; enhancers; insulators;"²  An enhancer is on his list.

A gain "produces a specific, new, functional coded element", and his first example is a new promoter built by mutation.²  A loss removes the function of "a specific, pre-existing, coded element," and his examples include the loss of the ability "of an altered promoter to bind a transcription factor".²  Modification is everything else, and it "includes point mutations as well as other mutations that have a quantitative effect on a preexisting FCT, increasing or decreasing its strength".²

Behe scored this case in advance

Behe does not mention lactase persistence in the 2010 paper, in The Edge of Evolution or in Darwin Devolves.  Every mention of lactose in them concerns bacteria.²³  But on page 426 of the 2010 paper he worked through the ways a gene could come to be expressed more, and he scored each:

"1) a repressor element of the compensating gene could be deleted, rendering the gene constitutive, which would be a loss-of-FCT mutation; 2) an existing promoter of the compensating gene could have its sequence altered, binding a transcription factor more strongly and thus leading to overexpression, which would be classed as a modification-of-function mutation, as an existing FCT was altered, not gained or lost; 3) a new promoter could be constructed by mutation to increase the transcription rate of the gene, which would be a gain-of-FCT mutation"²

His case 2 is the European variant, with "enhancer" in place of "promoter".  The ancestral sequence is already an enhancer that binds Oct-1.  The persistence letter makes it bind "more strongly", and expression rises.  By Behe's own rule that is a modification.

The score, variant by variant

The four questions of the series' procedure give the same answer, and they separate the variants that differ.

1. Transfer? No. Each variant arose in the populations that carry it. None came from another species.
2. Loss? Only if a working element stopped working. The popular description says a shut-off switch broke. No study read found a repressor site destroyed. The ancestral enhancer still binds its factors and still works in cells. The shut-off in adults is DNA methylation spreading over the region with age, and the persistence version resists it.⁷ Something that normally happens to the enhancer stops happening, but no coded element was lost. One variant is unsettled: early reports said -13915G abolished Oct-1 binding, and a 2011 study found it binds.¹⁵
3. Gain? Only if a binding site or control relationship now works that did not exist. For -13910T and -14010C the Oct-1 site already existed and now binds more strongly. For -14009G the answer may be yes. Its probe formed "a prominent additional DNA–protein complex", most likely with a factor of the Ets family, which the ancestral probe did not form.⁴ That is a new binding site by Behe's definition, if it is confirmed in intestinal tissue.
4. Modification. -13910T and -14010C are modifications, marked up. -13907G, whose mechanism is least studied, goes here provisionally.

"New nucleotides" follows the conversion rule in the sibling article:  a modification up counts the substitutions it needed, here one each.

This agrees with the score already printed in Measuring Biological Information, which put the lactase variants under "modification, up, hard case" and left the mechanism open.  The functional papers and Behe's page 426 settle it for the main variants.  Two refinements should carry back to that table:  the count is at least five, not four, and -14009G should be scored apart from the others.

What the class does not say matters as much.  The adult can now digest milk, which is a new ability for the person.  In the molecule, one existing switch was turned up.  No protein changed, no gene was built, and no new control relationship was wired in, except possibly at -14009.

Part 6: The ledger

The inputs

The human mutation rate has been measured directly, by sequencing parents and children.  Augustine Kong and colleagues sequenced 78 Icelandic families and found an average of 1.20 × 10⁻⁸ new mutations per nucleotide per generation.²⁴  Their Table 2 splits the rate by kind of change.  A transition at a site that is not a CpG occurs at 6.18 × 10⁻⁹ per base per generation.  A transversion at a non-CpG site occurs at 3.76 × 10⁻⁹, shared between two possible transversions.  A transversion at a CpG occurs at 9.59 × 10⁻⁹, also shared between two.²⁴

So the rate of each specific persistence letter is:

• −13910 C>T, a transition not at a CpG: 6.18 × 10⁻⁹ per generation.
• −13907 C>G, −13915 T>G and −14009 T>G, each one of two possible transversions at a non-CpG site: 1.88 × 10⁻⁹.
• −14010 G>C, one of two possible transversions at a CpG: 4.80 × 10⁻⁹.²⁵

These are averages over every site of each kind, and the rate at a particular site can differ with its neighbours.

A new mutation also has to survive.  Most new copies of even a helpful allele are lost by chance while they are rare.  J. B. S. Haldane showed in 1927 that a new allele whose single copy gives an advantage s survives with probability about 2s.²⁶  Persistence is dominant, so the single copy gets the full advantage.  For scale this article uses s = 0.05, inside every range in Part 3, which gives a survival chance of about one in ten.

Waiting times

Each person passes on two copies of the region, so a population of N people makes 2N × rate new copies of a given letter each generation.  The table gives the waiting time for one new copy, and for one that survives drift.²⁷

The population sizes are inputs chosen for scale.  No source read measures the number of people in prehistoric Europe or East Africa.  The 8,000,000 row is the final population Burger's group used in its simulations, "approximately one tenth of the Census population of Germany".⁹  The 10,000 row is the effective population size Tishkoff's simulations assumed.¹  An effective size is not a head count.  It is the size of an ideal population that would drift at the same rate, and for humans it is much smaller than the census.

What the numbers say about the European variant

From the earliest detected carrier to today is about 6,650 years, or 238 generations at Evershed's 28 years each.²⁸  Hold a population constant for 240 generations and count how often the -13910 letter would arise:

• 10,000 people: 0.03 times, with 0.003 surviving.
• 100,000 people: 0.3 times, with 0.03 surviving.
• 1,000,000 people: 3 times, with 0.3 surviving.

So in a population the size of the models' effective size, the European variant would probably never have arisen in the farming era.  In a population of a million or more it arises a few times and has a fair chance of taking hold.  Either the mutation arose in a large census population, or it arose earlier and waited at low frequency until selection began.  Evershed's group notes that "the dates of origin, the start of selection, the earliest observation and the reaching of appreciable frequencies of this allele are all distinct ‘events’, each possibly separated by thousands of years."²⁰  The ledger cannot choose between these.

What the five origins imply

The five established variants sit in a stretch of 104 letters.  That stretch allows 312 different single-letter changes, and at least 5 of them raise expression enough to be selected.  At least one change in 63 in this stretch helps.²⁹  It is a lower bound, because a helpful change that never happened to arise, or arose and was lost, leaves no trace.

A target that large, hit at the rates above, in populations of millions, will be hit repeatedly within a few thousand years.  The repetition is what the arithmetic predicts.  This is the strongest point on the mainstream side, and it should be granted plainly:  a one-letter improvement to an existing switch is easy for human populations to find.

It does not follow that everything is easy to find.  Each lactase variant helps on its own, from the first copy.  Where a gain needs two specific letters and neither helps alone, both must arrive before selection can act.  The chance that one child receives both specific changes at once is the product of the two rates.  For the -13910 transition paired with one specific transversion, a population of 10 million would wait about 4 billion generations.³⁰  That figure is a floor on the difficulty's scale, not a model of any real case, because a harmless first letter can drift through the population and shorten the wait by many orders of magnitude.  The site's Haldane's Dilemma article takes up those longer paths.  The only lactase allele that may have needed two changes is the Arabian one, and its paper does not say which letter came first.¹⁴

Unique functional sequence per person-generation

The series ends each ledger on unique functional sequence created per organism-generation.  The numerator here is one modified nucleotide per variant.  The denominator is the number of people who lived in the populations concerned while the variants arose and spread.  No source read measures it, so the cell is left empty rather than filled with a guess.

Part 7: What each side says

Creationist sources

Creation Ministries International gives the fullest creationist treatment, by David Catchpoole in 2009.  It gets the ancestral state right:  "the gene for lactase normally switches off as children are weaned.  And a genetic mutation that results in lactase production not being switched off accounts for the ability of certain people to drink milk into adulthood."¹¹  It accepts that the change helps.  Then it scores it:  "these genetic changes are not “evolution” in the uphill molecules-to-milkman sense, as the changes are downhill, i.e., information has been lost (viz., the normal switching-off mechanism of lactase production following weaning)."¹¹  The footnote to that sentence cites a general essay by Carl Wieland, not a study of the enhancer.

Answers in Genesis, in an unsigned 2006 news item headed "Milk-Digestion Gene Disabled, Evolution Proved?", said the mutations "merely cause a malfunction in the automatic shutdown of lactose digestion in children."³¹  Jean Lightner, writing for the same ministry in 2009, took a different line:  "Originally humans were designed to produce lactase, the enzyme necessary to digest milk, only during early childhood.  This excellent design prevents wasteful production of the enzyme during adulthood.  However, the lactase gene was designed to allow for change."³²

Robert Carter has not written on lactase in any source found, but he has rejected the blanket claim the others rely on:  "The phrase, “Mutations cannot create new information” is almost a mantra among some creationists, yet I do not agree."³³  No page on lactase persistence by the Institute for Creation Research or by Nathaniel Jeanson was found.³⁴

Weighed against the molecular papers, the creationist sources are right on three points.  Non-persistence is the normal state.  No new gene or protein was made.  The change is recent.  They are wrong on the mechanism.  No gene was disabled, and no study read shows a shut-off element destroyed.  An existing enhancer was made to bind its activator more strongly, and in one East African variant a new binding complex may have appeared.  By Behe's rules that is a modification, and possibly a small gain.  It is not a loss.  Lightner's "designed to allow for change" is compatible with the data, but as stated it predicts nothing a test could check.

Mainstream sources

The mainstream papers say what the data show.  Todd Bersaglieri's group found "the signals of selection we observe are among the strongest yet seen for any gene in the genome."¹⁷  Evershed's group calls persistence "the most strongly selected monogenic trait to have evolved over the last 10,000 years."²⁰  Tishkoff's group drew the lesson about regulation:  multiple independent variants "have been strongly adaptive in adult milk-consuming populations, emphasizing the importance of regulatory mutations in recent human evolution."¹

None of these papers claims the case shows how genes or switches are built.  Some mainstream authors question their own field's numbers.  Liebert and colleagues call the published coefficients "extraordinarily high in view of the fact that the cultural adaptation of fermentation of milk products, which reduces the lactose concentration, allows milk to be used as a source of calories in the diet of lactase non-persistent people".⁸

Objections and responses

"It is just a broken switch.  Evolution only breaks things."  The molecular data say otherwise.  Both versions of the enhancer work, and the persistence versions bind their activator more strongly.^{6 13}  Behe's own rule scores that as a modification.  Calling it a loss misstates the evidence and hands a critic an easy correction.

"Adults can now digest milk.  That is a new function."  For the person, yes.  For the molecule, an existing gene stays on longer, and the series scores the molecule, as Behe does.²

"Five independent origins prove that evolution can find anything."  They prove that it can find one-letter improvements to a switch that already works, when each letter helps at once.  Part 6 shows why that is easy.  They say nothing about changes that need several letters before any help.

"The dates are radiocarbon, so the whole timeline is worthless."  The counts do not depend on the dates:  the allele went from rare in Bronze Age skeletons to 75% today.  The rise after about 1250 BC fits a short chronology as well as a long one.  Only the earliest carriers, before about 2500 BC, need their radiocarbon ages compressed.

"The selection coefficients are just model output."  They are, and the article treats them so.  But a rise from 2 of 26 chromosomes to 144 of 192 in about 130 generations, in populations of millions, is far too fast for chance.  Both of Burger's methods flagged the persistence allele as under "strong positive selection".⁹  The size of s is uncertain.  Its existence is not.

"The mutations were directed, not random."  Nothing in the data points that way.  Each variant is an ordinary transition or transversion, at a rate consistent with the genome-wide rate, in a stretch where several different letters work.  That cannot rule out design.  It gives design no evidential support either.

"The cell effects are tiny, 1.2 to 1.5-fold."  In cell lines, yes.  In adult intestines the methylation difference is 82% against 19% at the variant's own region.⁷  The cell assays may understate the effect in the body, as their authors warn.⁴

Issues and unanswered questions

• The bodies of four functional papers were not read. Enattah 2002, Olds and Sibley 2003, Troelsen 2003, Lewinsky 2005 and Jensen 2011 are paywalled and not in PubMed Central. Their abstracts carry every claim used here. Lewinsky's body would say whether the ancestral C binds Oct-1 weakly or not at all.
• Two mechanisms are unsettled. Whether -13915G loses or keeps Oct-1 binding is disputed. The new complex at -14009G has not been confirmed in intestinal tissue, and its factor is only probably an Ets protein.
• The Arabian allele needed two changes in one laboratory's tests. No replication was found.
• The origin of the European mutation is not dated. First detection is about 4700 BC. Whether it arose then or waited earlier at low frequency is open, and the ledger cannot decide it.
• The ledger's denominator needs census estimates for prehistoric Europe and East Africa, which no source read supplies.
• A test for the design-for-change claim. Mutate every letter of the 104-letter stretch in turn and measure expression in intestinal cells. If many of the 312 single changes raise expression, the five natural variants are what chance predicts. If only the natural five do, the stretch looks poised, and Lightner's claim gains content.
• The earliest carriers' radiocarbon dates sit before the Flood on both chronologies. No creationist source has addressed these dates specifically.

Verdict

Lactase persistence is the clearest case of recent strong selection in humans, and ancient DNA has counted most of its rise.  It is also a small change.  At least five times, one letter in one enhancer was changed so that an existing switch bound its activator more strongly and escaped silencing in adults.  By Behe's 2010 rules the main variants are modifications of an existing element, with one candidate gain.  They are not losses, and the creationist sources that call them losses should change that description.

The ledger favours the mainstream reading on one point and limits it on another.  One-letter improvements with an immediate benefit are cheap in populations of millions, and the repeated origins show it.  Nothing in the case measures what it costs to build a new switch or a new gene.  The timing of the rise fits a short chronology.  The first carriers do not, unless their radiocarbon dates are too old by one to three thousand years.

Sources