SARS-CoV-2

"Our estimates imply that every single base mutation is being generated de novo and transmitted to a new SARS-CoV-2 host, somewhere in the world, every day." — Ron Sender and colleagues, Proceedings of the National Academy of Sciences, 2021¹

"For things like viruses, which have very high population sizes and very high mutation rates, if a specific change is possible, it should also be expected to occur." — Robert Carter, Creation Ministries International, 2021²

Summary

No organism has ever been watched evolving as closely as SARS-CoV-2.  More than nine million of its genomes are in the public databases and the changes that spread were described within weeks.  This article counts how much mutational search happened between December 2019 and September 2026 and lists what the search produced.  Each change is scored as a gain, a loss, a modification or a transfer of function using the definitions in Measuring Biological Information.  Influenza, which had eighty years rather than six and can swap whole gene segments, is covered separately in Influenza.

The main findings:

1. The record is real and very large. NCBI held 9,218,691 SARS-CoV-2 nucleotide records on 27 September 2026, of which 9,165,053 were near-complete genomes;³ the Nextstrain open dataset held 9,437,121 sequences the same day.³ Reported cases, 779,395,900, badly undercount infections.⁴ The Institute for Health Metrics and Evaluation put total infections and reinfections at 3.80 billion by 14 November 2021, before Omicron.⁵
2. Between 4 × 10²¹ and 1 × 10²³ virions have been made, carrying between 1.5 × 10²⁰ and 4.4 × 10²¹ new mutations. The inputs are 3.80 billion infections, 10¹² to 3 × 10¹³ virions produced per infection,¹ a genome of 29,903 bases⁶ and a measured rate of 1.3 × 10⁻⁶ mutations per base per replication cycle.⁷ The arithmetic is in the note.⁸
3. Every one-letter change arose about 10¹⁵ to 10¹⁶ times, every two-letter pair about 10⁹ to 10¹⁰ times, and every three-letter combination a few hundred to a few thousand times. Four specific changes at once is where the supply runs out: fewer than one in two hundred of the possible four-change sets was ever assembled in a single replication.⁹
4. Robert Carter, writing for Creation Ministries International, concedes all of that. "Essentially, every possible simple mutation has already occurred," he wrote in July 2021.² The creationist case disputes not the size of the search but what it bought.
5. What it bought, scored: no gain of function in Michael Behe's 2010 sense. Receptor affinity up, antibody binding down, one host protease used less and another more, two drug targets altered at a fitness cost, and three accessory genes knocked out in lineages that spread. The table is in Part 4.
6. The one change that looks like a coordinated pair is not one. N501Y and Q498R together raise receptor affinity 387-fold, and Q498R alone slightly lowers it.¹⁰ But N501Y raises affinity 15-fold on its own and appeared first, in Alpha. The pair was climbed one useful step at a time.
7. The one change with Behe's shape is drug resistance that did not spread. Mpro E166V gives about 100-fold resistance to nirmatrelvir and costs replicative fitness, which L50F or T21I restores.¹¹ It was assembled in laboratory passage. In patients, 0.5% of 13.4 million sequences carried any resistance-associated change, with no rise after the drug came into use.¹²
8. Nothing new was built. No new gene, protein fold or enzyme activity has been reported since December 2019. The extra reading frames sometimes offered as counter-examples, ORF3c and ORF9c, are older than the pandemic or are not genes at all.¹³
9. The strongest point for the young-earth side is the transmission bottleneck. One to eight virions found each new infection,¹⁴ so nearly all of that diversity is thrown away before selection between hosts can see it.
10. The strongest point against it is that the ledger is a lower bound. Only changes somebody looked for are on the list, and the fitness effect of most Omicron changes has never been measured one at a time.
11. The genetic-entropy reading is half right. Omicron really is intrinsically milder, in rodents with no immunity¹⁵ and in unvaccinated people.¹⁶ But the cause is a shift in which tissue the virus prefers, which came with better transmission, and the C-to-U mutation bias the argument rests on is opposed by selection rather than running unchecked.¹⁷

Part 1: The record

Three numbers are needed and all three have to be read fresh, because two of them change daily.

Genomes.  On 27 September 2026 the NCBI Nucleotide database held 9,218,691 records for taxon 2697049, the SARS-CoV-2 taxon, of which 9,165,053 were between 29,000 and 30,500 bases long and so were near-complete genomes.³  The Nextstrain open dataset, which is built from the same public deposits, reported 9,437,121 sequences.³  GISAID holds more, because many countries deposit there and not in GenBank, but GISAID publishes no SARS-CoV-2-only total outside its login.  Its front page on the same day gave 22,536,977 submissions across all five of its databases, influenza included, which is only an upper bound.³  The best published GISAID figure found for this article is Jonathan Ip and colleagues' count of 13,446,588 sequences as of 14 December 2022.¹²

Infections.  Reported cases are not infections.  The World Health Organization's data file, whose last dated row when it was downloaded on 27 September 2026 was 6 September 2026, sums to 779,395,900 confirmed cases.⁴  The gap between that and the truth was measured.  Isabel Bergeri and colleagues, pooling standardised seroprevalence surveys for the World Health Organization, found that the ratio of seroprevalence to cumulative reported cases in the third quarter of 2021 "ranged from around 2:1 in the Americas and Europe HICs to over 100:1 in Africa (LMICs)."¹⁸

The infection estimate this article uses comes from the COVID-19 Cumulative Infection Collaborators, who combined reported cases, reported deaths, excess deaths, hospital admissions and seroprevalence surveys for 190 countries.  Through 14 November 2021 they estimated "3·80 billion (95% uncertainty interval 3·44–4·08) total SARS-CoV-2 infections and reinfections combined, and an estimated 3·39 billion (3·08–3·63) individuals, or 43·9% (39·9–46·9) of the global population, had been infected one or more times."⁵

That count stops one week before Omicron was named.  Omicron and its descendants then infected a large further share of the world, much of it a second or third time, and no comparable whole-pandemic estimate was found.  So 3.80 billion is used below as a lower bound on infections, not as the total.  Every figure derived from it is therefore also a lower bound.

Genome length.  29,903 nucleotides, from the first published genome.⁶  Three possible substitutions at each position gives 89,709 possible single-nucleotide changes, which is the number the rest of this article divides by.

Part 2: The ledger

How many virions

Ron Sender and colleagues at the Weizmann Institute converted measured viral RNA concentrations in rhesus macaque tissue into whole-body counts for a 70 kg person.  The lungs dominate:  10⁶ to 10⁸ RNA copies per gram times 1,000 grams of lung gives "10⁹ to 10¹¹ virions" at the peak of infection.¹  Over a whole infection the total produced is three to thirty times the peak, "3 × 10⁹ to 3 × 10¹² virions."¹  Because viral loads are roughly log-normal, the right figure to multiply by a count of people is the arithmetic mean rather than the median, and that is "10¹² to 3 × 10¹³ viral particles."¹

Multiplying by 3.80 billion infections gives 3.8 × 10²¹ to 1.1 × 10²³ virions produced in the pandemic through November 2021.⁸  For scale, Sender's own independent estimate of the standing crop at any one moment during the pandemic was 10¹⁷ to 10¹⁹ virions.¹

How many mutations

Two different rates are quoted for SARS-CoV-2 and they are not the same quantity.  Mixing them is the commonest error in this kind of arithmetic.

Massimo Amicone and Isabel Gordo's group measured the first figure directly, by passaging two SARS-CoV-2 lineages fifteen times in Vero cells and sequencing the populations.  Their paper states the problem plainly:  estimates of the per-year rate existed, but "we still lack a direct quantification of the amount of mutations that can be generated in a single infection cycle."⁷  Before their work, published ledgers used rates from related viruses;  Sender's paper used 3 × 10⁻⁶ from mouse hepatitis virus.¹  Amicone's directly measured figure is 2.3 times lower, so this article uses it and its mutation supply is correspondingly smaller than Sender's.

The second figure, 1.1 × 10⁻³ per site per year, comes from Sebastián Duchene and colleagues fitting a molecular clock to the first 122 genomes.¹⁹  Across 29,903 bases that is about 33 substitutions per genome per year along a transmission chain.  The per-cycle rate across the same genome is 0.039 mutations per copy.  The two differ by nearly three orders of magnitude because the second counts only what survived hundreds of replications and a bottleneck at every transmission.

How thoroughly the genome was searched

Divide the mutational events among the possible changes.⁹

The cliff sits between three and four.  Any change that needs three specific mutations in one copying event has been offered to the virus hundreds of times.  Any change that needs four has almost certainly never been offered that way at all:  between 3.5 × 10¹⁴ and 1.0 × 10¹⁶ four-change sets were assembled, out of 2.7 × 10¹⁸ possible, which is 0.01% to 0.4% of them.

Sender's group reached the same conclusion for single changes.  Taking the E484K change as an example, they estimated that "such a specific mutation will be observed in one out of every ∼200,000 infections."¹  With 0.3 million to 3 million new cases a day, the mutations generated daily "likely exceeds the total number of possible single nucleotide substitutions to the SARS-CoV-2 genome," so "every single base mutation is being generated de novo and transmitted to a new SARS-CoV-2 host, somewhere in the world, every day."¹  Their abstract puts it in one clause:  "the genetic diversity of virions in an infected host covers all possible single nucleotide substitutions."¹

What the bottleneck does to all of that

The numbers above count mutations made.  They do not count mutations that get anywhere.

Katrina Lythgoe and colleagues deep-sequenced 1,313 samples from the first UK wave, fixed a reproducibility threshold using replicate RNA samples, and then estimated how many virions actually start a new infection.  "Comparing iSNVs among 14 household contact pairs, we estimated transmission bottleneck sizes of one to eight viruses."¹⁴  Their conclusion was "a consistent pattern of low within-host diversity, purifying selection, and narrow transmission bottlenecks," and the consequence they drew is the one that matters here:  "if mutations do arise, they will be prone to loss at the point of transmission."^{14 14}

So the effective search is much smaller than 10²².  A new mutation has to arise, rise to a large share of the virions in one person's airway, win a lottery with one to eight tickets at transmission, and do that again in the next host.  Lythgoe's own summary is that "within-host emergence of vaccine and therapeutic escape mutations is likely to be relatively rare, at least during early infection, when viral loads are high."¹⁴

This is the strongest quantitative point in the young-earth column.  The virion count sets what is chemically available.  The infection count, 10⁹ to 10¹⁰, sets what selection between hosts can pick from.  The two differ by twelve orders of magnitude.

Part 3: The furin cleavage site, and what this article will not settle

SARS-CoV-2's spike protein is cut in two places.  One cut, at the S1 to S2 boundary, is made by the host enzyme furin, and the site it acts on was already there in the first genome sequenced in December 2019.  Kristian Andersen and colleagues describe it as "a functional polybasic (furin) cleavage site at the S1–S2 boundary through the insertion of 12 nucleotides."²⁰  That insertion is not present in the virus's closest known bat relatives, and how it got there is disputed.  Andersen's group concluded that "SARS-CoV-2 is not a laboratory construct or a purposefully manipulated virus."²⁰  Rossana Segreto and Yuri Deigin argued the opposite in BioEssays, that "the genetic structure of SARS-CoV-2 does not rule out a laboratory origin."²¹

Nobody watched the site arise, so it is outside this article's subject.  This series counts what was observed happening after December 2019.  The origin question is left to the two sources above and to those who work on it.

What the site did afterwards is inside the subject, and it matters, because the cleavage site controls which way into the cell the virus takes.  Bo Meng and colleagues state the dependence:  "The plasma membrane route of entry, and indeed transmissibility and disease severity in animal models, is critically dependent on the polybasic cleavage site (PBCS) between S1 and S2 … and cleavage of spike before virion release from producer cells;  this contrasts with the endosomal entry route, which does not require spike cleavage in producer cells."²²

Two changes at position 681, just before the cut, spread in different lineages.  Akatsuki Saito and colleagues engineered Delta's P681R into a virus and found that it "facilitates cleavage of the spike protein and enhances viral fusogenicity," and that the P681R virus was more pathogenic in hamsters than its parent.²³  Alpha's P681H is a different substitution at the same position and did not behave the same way.  Bailey Lubinski and colleagues tested it with fluorogenic peptides and "observed no significant difference in furin cleavage."²⁴  Two changes at one position, one with a measured effect on the enzyme and one without.  Omicron carries P681H along with two more changes in the same region, H655Y and N679K.²²

Part 4: What the search produced

Spike D614G

The first change to sweep the world was a single nucleotide, A23403G, turning aspartate 614 of spike into glycine.  Bette Korber's group at Los Alamos was watching GISAID for exactly this and reported in Cell that "a SARS-CoV-2 variant with Spike G614 has replaced D614 as the dominant pandemic form."²⁵  The rise was fast:  the haplotype carrying it "was found in 10% of 997 global sequences" before 1 March 2020, "67% of 14,951 sequences" in March, and "78% of 12,194 sequences" between 1 April and 18 May.²⁵

One detail is usually dropped and should not be.  D614G did not sweep alone.  It swept as one of four linked changes, together with C241T in the untranslated region, a silent change at position 3,037, and C14408T, which alters the polymerase.²⁵  Any price put on D614G is really a price on the haplotype.

Jessica Plante and colleagues then engineered the change into a cloned virus so the comparison was isogenic.  "D614G enhances replication on human lung epithelial cells and primary human airway tissues through an improved infectivity of virions," and the G614 virus produced "modest 1.2-, 2.4-, and 1.9-fold more infectious virus than the D614 virus at 24, 36, and 48 hpi."^{26 26}  In hamsters it reached higher titres in the nose and trachea but not the lungs.²⁶  The effect is real, measured, and about twofold.

Score:  modification of function, up.  One nucleotide, in a haplotype of four.

Receptor binding, and the prediction that came true

Tyler Starr and Jesse Bloom measured the effect of every single amino-acid change in the receptor-binding domain on both folding and receptor affinity, before any variant of concern existed.  "Most mutations are deleterious for RBD expression and ACE2 binding," they reported, but a minority were tolerated or better, and in mid-2020 they added that "we find no evidence that these ACE2-affinity-enhancing mutations have been selected in current SARS-CoV-2 pandemic isolates."^{27 27}  Their isogenic checks confirmed affinity gains at positions 501 and 498.²⁷

Four months later Alpha appeared carrying N501Y, and Beta and Gamma appeared carrying N501Y with K417N or K417T and E484K.  Starr's later paper records the sequence of events without any hedging:  "we identified the N501Y mutation as enhancing ACE2-binding affinity prior to the emergence of this consequential mutation in the Alpha variant."¹⁰  A laboratory scan found the change before selection did.  That is a strong result for the predictive power of the mainstream account and should be conceded plainly.

Score for N501Y and its companions:  modification of function, up.  One nucleotide each.

Omicron

Omicron is the hard case and the interesting one.  Raquel Viana and colleagues, who described it from southern Africa, call it "exceptional for carrying over 30 mutations in the spike glycoprotein."²⁸  Fifteen of those sit in the receptor-binding domain.  Five of the fifteen, G339D, N440K, S477N, T478K and N501Y, had already been shown individually to improve receptor binding, and seven were expected to hurt at least three of the four main classes of neutralising antibody.²⁸  The branch leading to Omicron carries the strongest selection signal in the tree, with nine positively selected sites in spike alone.²⁸

The epistasis.  A set of fifteen changes to one small domain ought to break it.  Starr and colleagues measured why it does not.  "Although Q498R alone weakly reduces ACE2 affinity in the Wuhan-Hu-1 RBD, it confers a 25-fold enhancement in affinity when present in conjunction with N501Y (which itself improves binding 15-fold in Wuhan-Hu-1), such that the double mutant has a 387-fold increased binding affinity."¹⁰  That extra affinity is what pays for the rest:  Omicron's RBD changes, summed as single-mutant effects in the original background, should strongly impair binding, but summed in a background that already has N501Y they come to about zero.¹⁰

This looks at first like a coordinated pair of the kind Behe's argument is about, and it is not one.  N501Y helps by itself, and helps a great deal.  It appeared first, in Alpha, and spread on its own merits.  Q498R only became useful afterwards.  The order was one useful step, then a second step made useful by the first.  That is a ratchet, not a coincidence, and the ratchet is the mainstream account's mechanism working exactly as advertised.  One honest qualification cuts the other way:  Starr's group notes that the same pair "was first discovered in directed evolution studies," so the laboratory found it before the virus did.¹⁰

The entry route.  Meng and colleagues report that Omicron's spike binds the receptor more tightly than Delta's, escapes antibodies far better, and shows "a shift in tropism away from TMPRSS2-expressing cells and compromised ability to generate syncytia."²²  This is the most interesting single result in the whole record, because it is a change in mechanism rather than a change in degree.  It is still not a new mechanism.  Both routes into the cell, cutting at the surface with TMPRSS2 and cutting inside the endosome with cathepsins, were available to the virus from the start.²²  What changed is which one it leans on.

Where Omicron came from.  Omicron's branch carries far more changes than the clock predicts for the time available, and no intermediate was ever sampled.  The leading explanation is that it evolved inside one person whose immune system could not clear it.  Sheri Harari and colleagues state the hypothesis and its basis:  "Some substitutions found in chronic infections are lineage-defining mutations in variants of concern (VOCs), which has led to the hypothesis that VOCs emerged from chronic infections."²⁹  Every chronic case on record has been in a severely immunosuppressed patient.²⁹  Lawrence Corey and colleagues had described the mechanism in 2021 as "a form of rapid, multistage evolutionary jumps (saltational evolution), which could preferentially occur in the milieu of partial immune control. | note: Superscript reference numbers follow "(Covid-19)" in the source and are omitted here."³⁰

That explanation is an inference, not an observation.  No sample of Omicron's ancestor was ever taken.  What was observed is that other chronic infections do produce variant-defining changes, which makes the story plausible without confirming it.  A rival account has Omicron passing through an animal population, and Viana's recombination tests found only weak signals for a recombinant origin between the BA lineages.²⁸

Score for Omicron:  a large set of modifications, most of them unmeasured individually.  The shift in entry route is a modification of function, shifted.  Nothing in it is a new coded element.

Antibody escape

Yunlong Cao's group at Peking University mapped escape across the Omicron sublineages and found "BA.2.12.1, BA.4 and BA.5 escape antibodies elicited by Omicron infection," which is to say that infection with one Omicron did not protect against the next.³¹  The escape has a price.  BA.2's loss of G496S regained a hydrogen bond with the receptor and raised binding, while "the F486V mutation in BA.4/BA.5 spike decreases hACE2 binding activity."³¹  Later sublineages are not straightforwardly fitter;  they are differently fitted, trading receptor grip for invisibility.

Deletions did some of the same work.  Kevin McCarthy and colleagues found that because "coronaviruses acquire substitutions more slowly than other RNA viruses," recurrent deletions in spike "overcome this slow substitution rate," fall in a few recurrent deletion regions that map to antibody epitopes, and "confer resistance to neutralizing antibodies."³²

Score:  modification of function.  Antibody escape is a reduction in the binding of a host protein to a viral one.  It is a real benefit to the virus and it is not a new capability.

Drug resistance

Nirmatrelvir, the protease inhibitor in Paxlovid, was the drug most likely to be defeated, and in the laboratory it was.  Sho Iketani and colleagues passaged SARS-CoV-2 under increasing drug pressure and then rebuilt thirteen recombinant clones to test each candidate change.  "Construction and analysis of 13 recombinant SARS-CoV-2 clones showed that these mutations mediated only low-level resistance, whereas greater resistance required accumulation of additional mutations.  E166V mutation conferred the strongest resistance (around 100-fold), but this mutation resulted in a loss of viral replicative fitness that was restored by compensatory changes such as L50F and T21I."¹¹  The changes appeared "in a stepwise manner, mirroring the increases in drug resistance."¹¹

This is the shape Behe's argument predicts:  the big win is costly alone and needs a second, compensating change.  And the prediction that goes with it held.  Jonathan Ip and colleagues screened 13,446,588 GISAID sequences a year after the drug was approved.  "Overall, 0.5% (67,095/13,446,588) of the sequences contained at least one mutation that was shown to affect the inhibitory activity of nirmatrelvir or ensitrelvir … We did not observe any increasing trend of resistance after the widespread clinical use of nirmatrelvir-ritonavir."¹²  Most of that 0.5% is G15S and T21I, which raise resistance only slightly.  Two reasons for the absence are worth naming:  the drug is given for five days to people who are about to recover anyway, so there is little time for selection to work inside one host, and the costly changes are selected against everywhere else.

Remdesivir resistance has been seen once, clearly, in a patient.  Shiv Gandhi and colleagues sequenced the virus of an immunocompromised woman before and during treatment and found a change that was absent beforehand:  "a mutation, E802D, in the nsp12 RNA-dependent RNA polymerase … the mutation conferred a ~6-fold increase in remdesivir IC50 but resulted in a fitness cost in the absence of remdesivir."³³  Sixfold, with a cost.  Laboratory selection reaches other polymerase changes by more than one mechanism.³⁴

Score for both:  modification of function, activity down, with a measured fitness cost.  One nucleotide each, plus one more for the compensating change.

Losses

Several lineages that spread carry knocked-out genes, and this part of the record is unusually clear.

Yvonne Su and colleagues found a deletion in Singapore in January 2020:  "a 382-nucleotide (nt) deletion in SARS-CoV-2 that truncates open reading frame 7b (ORF7b) and ORF8, removing the ORF8 transcription regulatory sequence (TRS) and eliminating ORF8 transcription."  It reached "23.6% (45/191) of SARS-CoV-2 samples screened."³⁵  Barnaby Young and colleagues then followed the patients.  Nobody infected with the deletion variant needed supplemental oxygen, against 28% of those with the intact virus, an adjusted odds ratio of 0.07.³⁶  Alpha carried a different ORF8 knockout, a stop codon early in the gene.³⁷

Jesse Bloom and Richard Neher then priced these losses across the whole phylogeny by counting how often each mutation appears against how often mutation should put it there.  "Stop-codon and amino acid mutations to ORF7a and ORF8 are not more deleterious than synonymous mutations," and "the lack of deleterious mutations to ORF8 is consistent with the fact that viruses with deletions in this gene have spread in humans … and that major variants had stop codons early in ORF8."³⁷  More generally, "most accessory proteins are under little constraint," and the one exception is ORF3a, which is under strong selection against stop codons.^{37 37}

One case complicates the picture and should not be left out.  Artem Nemudryi and colleagues found that ORF7a truncations arise often, isolated one from a patient, and showed it had a real cost:  the truncation "negates anti-immune activities of the protein, which results in elevated type I interferon response to the viral infection," and the isolate had a growth defect.³⁸  So "accessory genes are free to lose" is too strong.  Some are.

Score:  loss of function.  Negative nucleotide count.

The ledger in one table

Fourteen entries can be scored.  None of them is a gain of function.  Twelve are modifications, two are losses, and one row cannot be scored at all.  No new gene, no new protein, no new enzyme activity, and no new binding between two proteins is among them.

Part 5: What did not appear

The claim that gains did not appear has to be tested rather than asserted, so here is what a virus under this much selection might have gained.

A new gene.  None reported, and this is the point easiest to get wrong, because papers do report extra reading frames in SARS-CoV-2 and those are sometimes offered as new genes.  Irwin Jungreis, Rachel Sealfon and Manolis Kellis settled it by comparative genomics across 44 sarbecovirus genomes.  "We find strong protein-coding signatures for ORFs 3a, 6, 7a, 7b, 8, 9b, and a novel alternate-frame gene, ORF3c, whereas ORFs 2b, 3d/3d-2, 3b, 9c, and 10 lack protein-coding signatures or convincing experimental evidence of protein-coding function.  Furthermore, we show no other conserved protein-coding genes remain to be discovered."¹³

ORF3c is a real gene, and it was new to the annotation, not new to the virus:  the evidence for it is conservation across the whole sarbecovirus group, so it existed long before December 2019.  ORF9c, ORF3d and ORF3b are the ones sometimes called new, and they are the ones the analysis rejects:  "None of these are conserved in the species Severe acute respiratory syndrome-related coronavirus."¹³  Either way, neither is a gain during the observed period.

A new protein fold.  None reported.

A new enzyme activity.  None reported.  Every change above alters an existing activity:  binding tighter, being cut more, escaping an antibody, resisting an inhibitor.

A new interaction between two proteins.  None reported.

What could count against this.  Two things.  The altered entry route is one.  Omicron leans on cathepsins in the endosome rather than TMPRSS2 at the surface, which is a change in mechanism rather than degree.  Under the scoring rule of Measuring Biological Information it is still a modification, because both routes existed in the virus already and both were in use.  A reader who thinks a shift between two existing pathways should count as more has a case worth arguing;  it is still not a new coded element.

The other is the number of unmeasured changes.  Omicron's thirty-odd spike changes have never been assayed one at a time, and the deep mutational scanning that exists covers folding, receptor affinity and antibody binding, not every function the protein has.  If one of them turned out to create something, this section would be wrong, and nobody has looked hard enough to rule it out.

Part 6: What cannot be priced, and the unit of account

Gains that cannot be priced

Three whole categories of adaptation are invisible to the ledger above.

Everything that happened inside a host and was thrown away.  Most of the search happened within single infections, where selection acts on replication speed and on evading a partly-primed immune system.  Almost all of it is discarded at the bottleneck and was never sequenced.

Everything nobody looked for.  Drug resistance gets found because patients stop improving, and antibody escape because escape assays are cheap and were run continuously.  An adaptation to a different cell type, or to a change in human behaviour, would leave no such signal.  The accessory-gene knockouts were found because they were large;  smaller regulatory changes would not have been.

The fitness effects nobody has decomposed.  Bloom and Neher's method is honest about its own floor:  their estimates "are only sensitive to fitness costs greater than a few percent."³⁷  A change worth 1% per generation to a virus with a doubling time of five days is strong in the real world and invisible to that method.

So the ledger is a lower bound on what the virus produced, and the true count of adaptive changes is higher than fourteen by an unknown amount.  That weakens the argument this series makes, and the data to remove the weakness do not exist.

The unit of account

This series measures unique functional sequence:  nucleotides that had to be originated rather than copied, per organism produced.  For SARS-CoV-2 the count is small and the range is wide.

The lower bound counts only positions whose functional effect was measured in a controlled comparison:  D614G, N501Y, Q498R, G339D, N440K, S477N, T478K, F486V, P681R, Mpro E166V, Mpro L50F and nsp12 E802D.  That is twelve positions, twelve nucleotides.  The upper bound takes Omicron's whole spike set as though every one of its thirty-plus changes mattered, adds the alternative forms in Alpha, Beta, Gamma and Delta, the escape deletions and the two drug targets, and lands under about 100 nucleotides.  Deleted genes add nothing, and negative counts are not subtracted here.

Divided by the virions produced, that is 1 × 10⁻²² to 3 × 10⁻²⁰ new functional nucleotides per virion.³⁹  Divided by infections, which is the unit selection between hosts actually works on, it is 3 × 10⁻⁹ to 3 × 10⁻⁸ per infection.  For comparison, the Plasmodium falciparum ledger gives 2 × 10⁻²¹ to 7 × 10⁻²⁰ nucleotides per parasite, and the HIV figures give about 8 × 10⁻²⁰ per virion.  Three organisms with nothing in common but heavy sampling land within two orders of magnitude of each other.

Two cautions belong with the number.  It is a lower bound, for the three reasons above.  And it counts sequence, which is not the same as function:  a single nucleotide bought a 15-fold change in receptor affinity, and no arithmetic on nucleotide counts captures that.

Part 7: The genetic-entropy reading

What the creationist side actually claims

The young-earth argument about SARS-CoV-2 has to be stated from its own source before it is weighed.  Robert Carter wrote it for Creation Ministries International in July 2021, and the article is called "Is Covid-19 evolving?  No, but it is changing rapidly."

Carter does not dispute that the virus changes, and he does not dispute the size of the search.  He states it himself, taking Behe's malaria arithmetic as the model:  one in 10²⁰ parasites will carry the mutations needed for chloroquine resistance, and there are about 10²¹ parasites in a billion infected people.²  From that he draws the rule:  "For things like viruses, which have very high population sizes and very high mutation rates, if a specific change is possible, it should also be expected to occur."  And more directly:  "if a simple mutation will make a virus more infectious to humans, we should be looking for it.  It will happen."^{2 2}  On supply he anticipates Part 2 of this article:  "the number of viruses that have existed over this timeframe is truly vast.  Essentially, every possible simple mutation has already occurred."²

His two claims are about what that buys.  The first is that single-step change is not evolution in the sense that matters, because it makes no new information.  The second is the genetic-entropy prediction:  "over time viral infectivity should wane.  As each viral strain in circulation picks up more and more mutations, it should become less and less robust.  This is the essence of genetic entropy and we saw it happen within the human H1N1 virus.  Yet, the process took decades in H1N1 and SARS-CoV-2 has only been around for a short time.  Will genetic entropy degrade the virus?  Yes.  Will it happen this year?  No."²

No Sanford, Carter or Journal of Creation publication specific to SARS-CoV-2 mutation accumulation was found for this article, and the routes tried are listed in the research notes.  Carter's web article is the primary that exists, and the H1N1 work it rests on belongs to Influenza.

Testing the prediction

Did severity fall?  Yes.  Tommy Nyberg and colleagues compared 1.07 million Omicron infections with Delta infections in England.  "The adjusted hazard ratio (HR) of hospital attendance … with omicron compared with delta was 0·56 (95% CI 0·54–0·58);  for hospital admission and death, HR estimates were 0·41 (0·39–0·43) and 0·31 (0·26–0·37), respectively."¹⁶

Was it the virus or the population's immunity?  Mostly the virus.  Nyberg's group separated the two:  "Underlying the observed risks is a larger reduction in intrinsic severity (in unvaccinated individuals) counterbalanced by a reduction in vaccine effectiveness."¹⁶  Peter Halfmann and colleagues settled it independently in animals that have no prior immunity of any kind:  "B.1.1.529 replicates less efficiently in the lungs of Syrian hamsters, which results in less severe pneumonia compared to that resulting from the B.1.617.2 variant."¹⁵  So Carter's expectation of attenuation was met, and faster than he expected.

Was it degradation?  This is where the argument fails, and it fails on the mechanism.  Omicron is milder because it replicates less well in the lungs and better in the upper airway, and that shift came together with a large rise in transmissibility.  A virus that spreads faster and kills less is not a degraded virus;  it is a differently adapted one.  Yao Liu's review states the attenuation case about as strongly as the evidence allows, arguing that "selection for antibody-escaping variants and accumulation of near-neutral mutations have led to suboptimal codon usage and reduced replicative capacity, as demonstrated in non-respiratory cell lines," while also listing "reduced infectivity in lung tissues and increased tropism for the upper airway, resulting in shorter incubation periods, milder diseases, and more efficient transmission between people" as beneficial adaptation.⁴⁰  That review is a single-author narrative synthesis in Cureus, not a measurement, and should be read as the best statement of the case rather than as evidence for it.

Is the mutation bias unchecked?  No, and the paper Carter cites for the bias says so.  Alan Rice and colleagues measured the mutational profile at fourfold degenerate sites and found it "heavily GC→AU biased," implying that if mutation alone ruled, the genome would settle at 17.13% GC with 65.67% U at those sites.¹⁷  The genome is nowhere near that.  Their conclusion is that "analysis of mutational profiles indicates a strong mutation bias toward U.  In turn, analysis of observed synonymous site composition implicates selection against U … observed UU content being a quarter of that expected under neutrality."¹⁷

That is the crux.  The C-to-U pressure that the genetic-entropy reading points at is real, and selection is visibly holding it back.  A ratchet that is being opposed is not a ratchet.  Whether it will win over decades is an open question for a virus that has been circulating for under seven years, and Carter says as much himself.

Objections and responses

A virus already fitted to its job is under stabilising selection, so of course nothing new appeared.  This partly succeeds, and it is the strongest objection here.  Two things blunt it.  The virus was new to the human host and by any account not yet optimal in it, which is why D614G, P681R and the Omicron entry shift were available at all.  And the objection concedes the point the series is making:  if stabilising selection is why a 10²²-virion search produced only modifications, then stabilising selection is doing most of the work in nature, which is what the young-earth model says.

The target was transmission among people whose immunity keeps changing, and surface changes meet that target.  This succeeds.  Antigenic change is cheap, one amino acid at a time on an exposed loop, and it was what paid.  The honest conclusion is that this record cannot show what the virus would have done under a pressure that required a new function.  It shows what it did under the pressure it faced.

Five and a half years is short.  This partly succeeds, and Carter makes the same point for the other side.  But the argument here is about tickets, not years.  Counted per site, 10²² virions and 10²¹ mutations across a 29,903-base genome is a larger search than every mammal that has ever lived has run across a three-billion-base one, on the mammal figures derived in Plasmodium falciparum, and it happened in six years.  A lottery is won by buying tickets, not by playing for a long time.

A 30 kb genome under proofreading has little slack, so there was nowhere to put a new gene.  This partly succeeds and it is a real constraint.  Coronaviruses carry a proofreading exonuclease, which is why their substitution rate is low for an RNA virus;  McCarthy's group says so directly:  "coronaviruses acquire substitutions more slowly than other RNA viruses."³²  But it cuts both ways.  It weakens the young-earth argument as a general claim, because a constrained genome is a poor test of what an unconstrained one could do.  And it weakens the evolutionary reply, because it concedes that genome architecture, not mutation supply, sets the limit on what can be found.

The ledger only counts what was measured, so it is rigged.  This succeeds and Part 6 concedes it.  The right response is to call the count a floor, not to widen it with guesses.

Antibody escape is obviously a benefit, so calling it a modification is unfair.  The score describes the molecule and the benefit is reported beside it in the table, as the scoring rule requires.  A change that lowers the binding of a host antibody to a viral protein has not created a coded element.

The Omicron entry shift is a new function.  Possibly, on a looser definition.  Not on the definition this series uses, because both entry routes were already available and in use.  A reader who wants to call it a gain should notice that it cost something at the same time:  Omicron makes syncytia poorly and replicates worse in lung tissue.

If every pair of mutations arose 10⁹ times, why has no two-step gain been found?  Because arising is not spreading.  A pair that arises inside one host still has to pass a bottleneck of one to eight virions, twice, before selection between hosts can favour it.  The nirmatrelvir case is the clean test:  the two-step path exists, it was built in the laboratory, and in 13.4 million patient sequences it did not take off.

Issues and unanswered questions

1. No whole-pandemic infection estimate was found. The 3.80 billion figure stops on 14 November 2021, one week before Omicron was named, and Omicron infected a very large further share of the world. Every total in this article is therefore low by an unknown factor, probably between two and four. A post-Omicron cumulative-infection estimate would fix it.
2. GISAID's own SARS-CoV-2 count could not be read live. Four open routes failed. The published figure of 13,446,588 as of December 2022 is used instead, and the live counts are NCBI's and Nextstrain's.
3. The mutation rate rests on one experiment. Amicone's 1.3 × 10⁻⁶ per base per cycle is the only direct measurement found, and the two lineages in that experiment gave synonymous-site estimates of 3.8 × 10⁻⁶ and 1.2 × 10⁻⁶, three-fold apart.⁷ A second independent measurement would tighten every number in Part 2.
4. Most Omicron changes have never been assayed individually. The upper bound of 100 nucleotides in Part 6 is a ceiling, not a count.
5. No creationist publication on SARS-CoV-2 attenuation exists, as far as this article could determine. The genetic-entropy case for this virus has to be reconstructed from Carter's 2021 web article and from the H1N1 work behind it.
6. The bottleneck estimate rests on 14 household pairs and Lythgoe's own paper notes that it is "considerably lower than estimates in a recent Austrian study."¹⁴ The disagreement is unresolved and the effective search size depends on it.
7. Whether selection will keep holding the C-to-U bias back over decades is not known, and it is the question on which the genetic-entropy reading will eventually be settled.

Verdict

SARS-CoV-2 ran the largest and best-documented mutational search any organism has ever been watched running.  Between 3.8 × 10²¹ and 1.1 × 10²³ virions carried between 1.5 × 10²⁰ and 4.4 × 10²¹ new mutations, which offered every one-letter change about 10¹⁶ times, every two-letter pair about 10⁹ times, and every three-letter combination a few hundred times.  Robert Carter's article for Creation Ministries International concedes exactly this, and the argument therefore turns on what the search bought.

It bought fourteen scoreable changes:  twelve modifications and two losses.  Receptor affinity up as much as 387-fold from two amino acids, wide antibody escape, more spike cleavage in one lineage and none in another at the same position, a shift between two entry routes the virus already had, two drug targets altered at a measured cost, and three accessory genes knocked out in lineages that spread.  No gain of function by Behe's 2010 test, and no new gene, protein fold, enzyme activity or protein-protein interaction.  The reported new reading frames are older than the pandemic or are not genes.

Three things weaken that verdict.  The ledger counts only what somebody measured, so it is a floor.  The pressure was almost entirely antigenic, so the record cannot show what this virus would do under a pressure demanding a new function.  And a 30 kb proofreading genome with overlapping frames has little room to build anything, which makes it a poor test case in both directions.

Two things strengthen it.  The bottleneck of one to eight virions means the population selection acts on is twelve orders of magnitude smaller than the virion count, which is the honest reason a 10²² search behaves like a 10¹⁰ one.  And the one change with the shape the design argument predicts, nirmatrelvir resistance needing a costly step plus a repair, was reachable in a laboratory and has not spread in 13.4 million patient sequences.

On genetic entropy the split is clean.  The prediction of attenuation was met, and met intrinsically rather than through population immunity, which is a real success for that side.  The proposed mechanism was not:  Omicron is milder because it moved up the airway and spread faster, and the C-to-U bias the argument rests on is being visibly opposed by selection.

Sources