Computer modeling

A computer simulation was performed using Python and 184,202 potential DNA positions (size of CFTR) to model locations at which a SNP might occur. Single nucleotide changes were simulated one at a time and placed randomly in the target sequence. After 969 SNPs had been modeled for each run (see Table 1), the simulation was checked for numbers of residues receiving more than a single point mutation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Polymorphisms identified in 4,786 F508del CFTR alleles.

https://doi.org/10.1371/journal.pone.0305832.t001

To assess SNP distribution ratios and types of mutations occurring in CFTR exons, every coding SNP generated by simulation was recorded and the amino acid produced determined. Computer-generated mutations were next categorized as to whether the modified position led to a transition or a transversion, and whether the resulting amino acid was synonymous, non-synonymous, or a nonsense variant. Mutation placement for each simulation used the human CFTR DNA sequence with each codon referenced against codon usage [9].

Statistical analysis

SNPs generated by random computer simulation were distinct from those experimentally observed for F508del CFTR alleles. A binomial distribution calculator was used to determine probability of observed SNP ratios in comparison to computer modeling data. p-value < 0.05 was taken as significant.

Transition versus transversion polymorphisms

Transition SNPs (T↔C, G↔A) have been measured in numerous previous studies of species comparison, DNM trios, tumor genomes, etc., and are typically present at far higher numbers than would be expected by chance alone. CpG methylation can be invoked to explain such findings, but accounts for only a fraction of transitions, based on experiments in yeast (where DNA methylation is very rare) and DNM analysis in humans (where transition SNPs are strongly enhanced at both CpG and non-CpG locations) [9, 27]. On the F508del haplotype, transition SNPs have accumulated in marked preference to transversions (585 intronic transition SNPs versus 344 transversions (63% transition); 24 exonic transition SNPs versus 16 transversions (60% transition)). The expected proportion, which favors transversions by a ratio of 2 to 1, has therefore nearly been reversed (Table 1).

What is the significance of transition SNP enhancement? Among other features, transitions confer synonymous polymorphism as programmed by the genetic code [1, 28, 29]. The tendency to elicit synonymous mutations is especially pronounced for transitions at the 3^rd nucleotide position of a codon (the preferred site for transition polymorphisms), where 94% of transitions lead to a synonymous amino acid change. Transition preference as shown in Table 1 cannot be accounted for by purifying selection of a functionless CFTR protein (note also that strong transition preference in coding DNA is very similar to non-coding elements, further arguing against recent selective pressure as an explanation). The increase of transition SNPs on an F508del CFTR background also agrees with a previous study of human DNMs among 283 trios, where ~65% of DNMs were transitions [27]. Such findings strongly contradict the notion of random SNP production. DNM bias of this type would serve evolutionarily to preserve protein coding sequences by favoring synonymous SNPs (Table 1) and limit genomic attrition (so called “meltdown”) [1, 2, 9].

Non-synonymous versus synonymous SNPs

Over the past 50 years (and continuing at present), very large numbers of published reports have drawn conclusions about human genomic evolution based on adjusted non-synonymous to synonymous SNP ratios (dN/dS) or similar quantitative methods. An unproven supposition for these studies has been that diminished numbers of non-synonymous SNPs reflect purifying selection that removes deleterious mutations from the gene pool.

Data from sequence analysis of 2,393 individuals with CF who are homozygous for F508del CFTR establishes a synonymous SNP incidence of ~38%. This result has meaning, since the percentage of synonymous DNMs genome-wide in human DNA has been reported previously to be of the same magnitude (approximately 30–44%) among normal controls and patients with autism [30–32]. Our findings therefore indicate that in contrast to many traditional applications, the primary determinant for non-synonymous to synonymous SNP ratios may simply reflect de novo mutation bias (as opposed to purifying selection or classic constraint). We note that in semblance to human DNM data (indicating up to 100-fold differences in mutational frequency at specific loci genome-wide), our findings further contradict use of non-synonymous SNP depletion as a routine means for evaluating DNA evolution. Note that for human genes reportedly subject to either rapid evolution or strong negative selection, values of non-synonymous SNPs may only be a few percentage points from “expected”. Data from gnomAD, for example, indicates typical synonymous SNP percentages of 29–40% across many loci (including both essential and non-essential genes)—a value used to indicate magnitude of constraint but very similar to the F508del synonymous proportion shown here—despite clear evidence that DNM formation bias (and not purifying selection) is responsible in the present study.

Random SNP accrual is further confounded by “parallel” evolution

Studies of congenic mice indicate a remarkably high level of parallel evolution (i.e., DNMs observed at the same codon in distinct animal lineages represent approximately 20% of all newly formed SNPs) [2]. Similar evidence of strong parallelism has been noted in human DNA. In one study, for example, 43% of de novo SNPs among 1756 trios were found to have occurred more than once when compared to the ExAc mutation dataset [33] (on a ‘random’ basis, <1% of DNMs would have been expected to show recurrence). Among 2,393 patients with CF studied here, 40 exonic SNPs were observed on the F508del background, and at least 5 of these occurred at the same position twice. For intronic DNA encoding 929 mutations on an F508del background, at least 28 overlapped in the same manner. Such findings contradict a notion of “random” DNM production required by dN/dS or pLOF-type analyses, and indicate that other features—such as biased SNP formation—contribute much more prominently than acknowledged previously. Note the stochastic likelihood of observing parallel evolution of the magnitude shown here is infinitesimal. For example, computer simulation of the process indicates that when 969 total SNPs were placed randomly across 184,202 DNA locations (the approximate size of CFTR), the number of positions receiving more than one SNP never exceeded 14 when the simulation was conducted 10 million times. Moreover, while two distinct point mutations (e.g., G→A or G→T) at the same codon occur with surprising frequency and can be readily quantified by the present analysis, repeat occurrence of the same base replacement at the same position is not easily distinguished from a single mutation simply inherited by multiple individuals (which would only be counted once). In other words, the present analysis is likely to underestimate both frequency and extent of repetitive SNP formation.

As part of the F508del DNA sequence analysis, we also observed pronounced clustering of “parallel” SNPs. For example, in CFTR intron 20 (total size 2804 base pairs (bp)), while 59 mutations (29 of these being SNPs) were observed, at least 43 (including 12 SNPs) occurred more than once at the same position (i.e., ≥ 41% of SNPs were found at the same location as a different SNP; Fig 2). Such findings strongly defy random SNP production, and do not result from recent positive selection involving intron 20.

Origin of the F508del variant

CFTR is ancient—over 450 million years old [34]—with thousands of polymorphisms noted in the wildtype protein across numerous ethnicities. Many of the polymorphisms in human CFTR may have markedly predated Homo sapiens evolution and remain evident because of founder-type effects. The occurrence of F508del on a single CFTR haplotype [23, 24, 35, 36] provides compelling evidence for one-time establishment of the variant.

Previous studies using distinct methodology have verified a mutation rate in Homo sapiens of approximately 100 SNPs per generation per genome [3, 27, 37–39]. This frequency corresponds to one new SNP per generation for every 30 million bp of human DNA. A total of 1,460 new mutations (~970 SNPs plus small indels, etc.) were noted in the present study when the 184,000-nucleotides of CFTR were evaluated across 4,786 alleles (total of 880,624,000 bp). Thirty-eight generations or ~800 years (at 20 years per generation) would be required to achieve the level of diversity in F508del shown here. A similar analysis based on exonic mutations indicates 57 generations (~1,100 years) of F508del SNP accumulation. Note that these timeframe projections may represent a modest underestimate due to consanguinity or other causes of F508del CFTR homozygosity, leading to loss of mutant alleles from the gene pool. Moreover, the analysis requires that CFTR conforms reasonably well to genome-wide estimates of SNP formation, whereas mutation rates are not yet known with certainty for any individual human gene. In either case, previous estimates place original occurrence of F508del between ~1,100 and 52,000 years ago [23, 36, 40, 41], and our findings support the shorter time line proposed by Serre [41] and Farrell [36]. In the same context, it should be noted that invoking small, epistatic or other recent fitness effects to explain transition, non-synonymous, or parallel mutation preferences on the F508del background over such a short time period would be naive, based on an extensive body of knowledge—and low likelihood that CFTR or its role in human physiology have changed substantially over the past few thousand years.

Evolutionary aspects of CFTR adaptation, including significance of the F508del defect

Earlier studies describing cystic fibrosis-related evolutionary biology have focused on CFTR as. a member of the ATP binding cassette (ABC) gene family. Other mammalian ABC proteins utilize ATP hydrolysis as a means to transport solutes across the plasma membrane [11]. CFTR has evolved specialized adaptations that include establishment of an ion channel pore, ligand binding interactions, and unique protein kinase A responsive regulatory elements that help govern overall ion channel function. The ways in which these features represent a response to evolutionary pressure have been modeled in part using DNA sequence comparisons with other ABC proteins, such as ABCC4 (in mammals) or YOR1 (in yeast) [42–44]. For example, specific functional divergence of the sixth transmembrane alpha helix at residue R352 has been implicated as contributing to evolutionary establishment of the CFTR permeation pathway. Ancient CFTRs such as those encoded by dogfish shark and sea lamprey (>450 million years old) have also facilitated evolutionary inference, including features of both ion channel structure and gating [34]. Some of this earlier work has relied on sequence conservation between species or features of stochastic SNP formation (across much longer time intervals than evaluated here for F508del CFTR). The interpretation of such findings might therefore be influenced by features of non-random SNP production as described by the present report.

As another example of previous CFTR evolutionary analysis, and in semblance to relationships between sickle cell disease trait and Plasmodium falciparum, a heterozygote advantage is believed to account for the high prevalence of F508del CFTR carriers in modern-day North America and Europe. It has been argued that a serious infectious disease (e.g., cholera, salmonella, tuberculosis, or plague), or heavy metal poisoning among European populations during the past several thousand years, may have been less lethal among heterozygotes for F508del CFTR. Although physiologic and epidemiologic evidence of heterozygote advantage remain inconclusive, the present report shows ways in which the high prevalence of F508del heterozygosity (appx. 1 in 30 White individuals in the US) provides a novel means to evaluate de novo SNP accumulation. Our study utilizes the largest compendium of fully sequenced F508del CFTR alleles to date–and furnishes a powerful resource for investigating SNP accumulation in human DNA [9, 45–49].

Limitations of the current approach

Care was taken in the present study to avoid sequencing artifact or incorrect SNP identification by applying a dataset based on state-of-the-art technology and well validated analytic methods (Fig 1 and [26]). SNP prevalence can be subject to ascertainment bias when specific populations (such as patient cohorts that include individuals with cystic fibrosis) are being evaluated, since strong enrichment for “loss of function” alleles is expected. That potential source of bias has been minimized in the current study by focusing on SNPs that occurred subsequent to a CFTR-null (F508del) mutational event. Another potential confounder involves the possibility of second site suppressors that partially rescue F508del CFTR. Known suppressors have been studied by our laboratory and others to characterize CFTR molecular abnormalities [21, 22, 50–54], and if present in cis with F508del could restore CFTR activity. Suppressors of F508del include R1070W, F1068M, F1074M, V510D/E/A, I539T, G550E, R553M/Q, and R555K [21, 22, 50–55]. Importantly, none of these SNPs were observed in the coding sequences obtained from 2,393 patients homozygous for F508del. Additionally, no synonymous SNP in CFTR has been reported to rescue the F508del defect. As a result, explaining synonymous SNP enrichment (Table 1) as being caused by numerous synonymous polymorphisms that rescue F508del would appear untenable.

Adaptive significance of SNP formation bias

It is worth noting that non-random SNP formation as delineated by the current report could provide an important evolutionary advantage by favoring transition, synonymous, and parallel exonic polymorphisms. This mechanism would permit DNA evolutionary adaptiveness and genomic diversity (essential for overcoming environmental pressures and selective “bottlenecks”), while also providing an element of protection for vital DNA protein coding elements and blunting the process of genomic attrition (so-called “mutational meltdown”). Over many hundreds of millions of years (the period of known CFTR evolution), an advantage of this type could be substantial [9].