A null mutation in SERPINE1 protects against biological aging in humans

Khan SS et al. · Science Advances 3(11):eaao1617 · 2017 · DOI: 10.1126/sciadv.aao1617

TL;DR

In a cross-sectional study of 177 adults from the Berne Old Order Amish founder population, 43 heterozygous carriers of a private loss-of-function SERPINE1 mutation had approximately 50% lower circulating plasminogen activator inhibitor-1 (PAI-1), 10% longer mean leukocyte telomere length, 28% lower adjusted fasting insulin, and no observed type 2 diabetes cases, compared with 127 noncarriers. An extended-pedigree analysis also associated carrier status with older age at death. This is unusually informative human genetic evidence linking lifelong partial PAI-1 deficiency to aging-related phenotypes, but it remains a single-kindred observational study: the mutation was not randomized, telomere attrition was not measured longitudinally, diabetes was prevalent rather than incident, and the lifespan analysis inferred some genotypes from ancestry and excluded deaths before age 45.¹

Study design

Berne Amish recruitment and genotype groups

The investigators invited 450 Old Order Amish adults aged 18 years or older from the Berne, Indiana, founder population; 177 enrolled (39.3%) and were genotyped for the exon-4 SERPINE1 frameshift mutation c.699_700dupTA. The enrolled sample comprised 127 unaffected noncarriers, 43 heterozygous carriers, and 7 homozygotes, yielding a null-allele frequency of 16%. The seven homozygotes were excluded from primary analyses because they were few and young (18–34 years) but were included in supplemental secondary analyses; the paper does not state that those homozygote-inclusive analyses were prespecified.¹

The primary analytic sample was therefore 170 participants (127 noncarriers and 43 heterozygotes). The supplement locates 40 of these participants in pedigree generation V (born 1926–1950), 54 in generation VI (1951–1975), and 76 in generation VII (1976–1997). Mean age was 46 ± 20 years in noncarriers and 44 ± 17 years in heterozygotes; women comprised 55% and 65%, respectively.²

Prespecified endpoints

• Primary endpoint: leukocyte telomere length (LTL), quantified from peripheral-blood leukocyte DNA by quantitative polymerase chain reaction (PCR), with terminal restriction fragment length by Southern blot as an orthogonal measurement.
• Secondary endpoints: fasting insulin, prevalent type 2 diabetes mellitus, and lifespan. The paper states that these endpoints were selected prospectively, but it reports no public protocol or registration; preregistered: false is therefore retained.
• Exploratory endpoints: echocardiographic tissue-Doppler e′ velocity, brachial pulse pressure, carotid intima-media thickness, and three post-composed biological-aging scores.

Polygenic models adjusted the primary genotype associations for age, sex, familial relatedness, and the kindred structure using Sequential Oligogenic Linkage Analysis Routines (SOLAR). Laboratory technicians measuring telomere length and sonographers assessing cardiovascular structure/function were blinded to genotype or clinical data as applicable.¹

Extended-pedigree lifespan cohort

Lifespan was not analyzed only within the 177 living/enrolled participants. Genealogical fieldwork identified vital-status records for 221 enrolled people or direct ancestors; the supplement reports 113 deceased individuals. Among the deceased, SERPINE1 status was directly genotyped in 18 and inferred by high-confidence obligate pedigree ascertainment in another 38, for 56 deceased people with assigned genotype. The authors restricted lifespan analysis to deaths at age 45 or older to reduce deaths from infection, trauma, and childbirth. Figure 2’s number-at-risk table gives the analyzed genotype groups as 15 carriers and 41 unaffected relatives.²

CARDIA validation cohort

The mutation was not tested in the Coronary Artery Risk Development in Young Adults (CARDIA) cohort. Instead, CARDIA was used to assess whether the study’s newly constructed aging scores predicted five-year cardiovascular morbidity/mortality and whether plasma PAI-1 correlated with fasting insulin and those scores. Of the original 5,115 CARDIA participants, 2,793 year-25 attendees had data for scores 1 and 2; the Methods and table S4 body report 892 with score-3 data.²

Key results

Lifelong PAI-1 reduction

Heterozygotes had mean circulating PAI-1 antigen of 5.9 ± 6.6 ng/mL versus 12.7 ± 9.8 ng/mL in noncarriers—approximately 50% lower (P < 0.0001). PAI-1 antigen was undetectable in the seven homozygotes, consistent with complete loss of function.¹

Leukocyte telomere length

Heterozygous carriers had 10% longer mean LTL than noncarriers after adjustment for age, sex, and familial relatedness (prespecified primary endpoint; P = 0.007 by quantitative PCR). Southern-blot terminal restriction fragment length supported the association (P = 0.039). Adding the seven homozygotes to the carrier group produced a similar result (relative LTL 1.04 ± 0.25 versus 0.95 ± 0.25; P = 0.020).¹²

Among unaffected participants, LTL correlated inversely with age (R = −0.70, P < 0.0001) and was approximately 9% shorter per decade; each additional year corresponded to 0.0087 lower relative LTL and approximately 30 fewer terminal-restriction-fragment base pairs (both P < 0.0001). Crucially, the age-by-carrier interaction was not significant. The cross-sectional data therefore support a higher age-adjusted LTL level in carriers, not evidence that their individual telomeres shortened more slowly over time.¹

Fasting insulin and diabetes

Raw fasting-insulin medians were 4.0 μIU/mL (2.9–5.1) in heterozygotes and 4.9 μIU/mL (3.3–6.7) in noncarriers (table P = 0.04). After adjustment for age, sex, and familial relatedness, fasting insulin was 28% lower in heterozygotes (P = 0.035). Inclusion of the seven homozygotes preserved the association (P = 0.026), and excluding participants with diabetes did not alter it.¹²

No heterozygous carrier had type 2 diabetes (0/43), compared with 8/127 noncarriers (reported as 7%; Results-text P = 0.001, while Table 1 rounds this to P < 0.01). This was prevalent disease in a cross-sectional sample, not incident diabetes observed prospectively. Body mass index did not differ significantly between carriers and noncarriers.¹

PAI-1 and fasting insulin were positively correlated in both the Amish sample (R = 0.55, P < 0.0001) and CARDIA (R = 0.48, P < 0.05). These correlations are consistent with metabolic mediation but do not establish that the lower insulin values caused the telomere or lifespan findings.¹

Lifespan: 7-year mean difference versus 10-year median difference

The paper reports the same pedigree comparison in two distinct summaries that should not be conflated:

• Mean ± SD age at death: 82 ± 10 years in null-allele carriers versus 75 ± 12 years in unaffected relatives—a 7-year mean difference (Figure 2 caption; P = 0.037 by Wilcoxon rank-sum test).
• Median survival: 85 years (73–88) in carriers versus 75 years (70–83) in unaffected relatives—a 10-year difference between medians (Results text; P = 0.037).

Thus, “seven years longer” describes the arithmetic means, whereas “ten years longer” describes the medians. Neither should be presented as a randomized causal lifespan extension; the Figure 2 risk table indicates 15 carriers and 41 unaffected relatives in the age-45-and-older analysis.¹²

Cardiovascular measures and composite scores

The direct cardiovascular measurements were largely null: tissue-Doppler e′ velocity was 12.8 ± 4.0 versus 12.3 ± 4.1, brachial pulse pressure was 47 (42–53) versus 51 (44–57) mmHg, and carotid intima-media thickness was 0.60 (0.51–0.72) versus 0.61 (0.52–0.79) mm; none differed significantly by carrier status. The cardiovascular-only composite score was 0.6 units lower in carriers but also nonsignificant (P = 0.09).¹

The composite incorporating cardiovascular measures plus fasting insulin was 1.03 units lower in carriers (P = 0.03), and the further composite adding LTL was 0.53 units lower (P = 0.005). In CARDIA, each one-unit increase in scores 1, 2, and 3 was associated with five-year cardiovascular morbidity/mortality odds ratios of 1.46 (95% CI 1.20–1.78), 1.35 (1.14–1.60), and 1.61 (1.18–2.22), respectively. This validates prognostic association of the constructed scores, not the longevity effect of the Amish mutation.²

Interpretation

The private null allele creates a human genetic “natural experiment”: heterozygotes have lifelong, large-magnitude partial reduction of SERPINE1/PAI-1, and multiple aging-related measures move in a direction concordant with preclinical PAI-1 inhibition. That strengthens causal inference beyond an ordinary association between circulating PAI-1 and disease. It does not eliminate confounding by shared ancestry, linked alleles, family environment, selection into the study, or genotype inference in the lifespan cohort.¹

The study did not directly measure cellular senescence, broader senescence-associated secretory phenotype activity, healthspan, frailty, or cause-specific mortality. The title’s “protects against biological aging” therefore rests on LTL, insulin/diabetes, novel composite scores, and retrospective age-at-death—not a validated global biological-aging endpoint.

The authors proposed altered metabolism as a possible mediator, but mediation was not formally tested. Longer LTL may have been established early in life rather than preserved by slower attrition: the absence of a genotype-by-age interaction and the cross-sectional design cannot distinguish those possibilities.¹ gap/no-mechanism

Extrapolation and evidence weight

Dimension	Status	Notes
Pathway conserved in humans?	yes	The variant, PAI-1 antigen, insulin, telomeres, diabetes, and lifespan were measured in humans.
Phenotype conserved in humans?	yes, narrowly	The reported phenotypes occurred in one genetically and geographically isolated human kindred.
Replicated in humans?	no	CARDIA supported PAI-1/metabolic and score associations but did not replicate the mutation-longevity comparison. gap/needs-replication
Pharmacological equivalence established?	no	Lifelong heterozygous loss of function is not equivalent to starting a PAI-1 inhibitor later in life.

The human genetic evidence is notable but hypothesis-strengthening rather than definitive. Independent loss-of-function alleles, prospective follow-up within the kindred, and a pharmacological target-engagement trial would be needed to separate PAI-1-specific effects from founder haplotype and lifelong developmental effects.

Limitations

• Single founder kindred and common ancestor. All heterozygotes traced to one late-1800s ancestral couple. SOLAR adjusted for relatedness, but another linked genetic or epigenetic factor could cosegregate with the null allele.
• Participation and survivor selection. Only 177 of 450 invited adults enrolled. The lifespan analysis further conditioned on death at age 45 or older, deliberately excluding earlier mortality and potentially changing the estimand.
• Pedigree-derived genotypes. Of 56 deceased people assigned a genotype for lifespan analysis, 38 were inferred by obligate ancestry rather than directly genotyped; Figure 2’s number-at-risk table indicates 15 carriers and 41 unaffected relatives.
• Cross-sectional biomarkers. LTL and metabolic variables were measured once. The study cannot estimate within-person telomere attrition or incident diabetes.
• Young homozygotes. Only seven homozygotes were available, all 18–34 years old, preventing meaningful aging or lifespan inference for complete PAI-1 deficiency.
• Novel composite scores. The scores were constructed in the Amish data and then associated with limited five-year CARDIA events. They were not established aging clocks and need external replication with longer follow-up.
• Supplementary n inconsistency. The main Methods, supplementary Methods, and table S4 body report CARDIA score-3 N = 892, but the table S4 title and main-article supplementary-material list state n = 872. The analysis population should be treated as 892 provisionally because it is repeated in the Methods and table body, pending author clarification.
• Population generalizability. Even unaffected Amish participants had approximately 50% lower mean plasma PAI-1 than CARDIA and Framingham participants, and the community differed from urban U.S. cohorts in lifestyle and cardiometabolic profile.
• No independent replication. The genotype associations and lifespan comparison come from one extended family. gap/needs-replication

The work was supported by U.S. National Institutes of Health grants and CARDIA contracts; the authors declared no competing interests.¹

Cross-references

• PAI-1 (SERPINE1) — the encoded protein and target page
• sasp and cellular-senescence — mechanistic framing not directly assayed in this cohort
• telomere-attrition — prespecified primary endpoint
• deregulated-nutrient-sensing, insulin, and type-2-diabetes — metabolic secondary endpoints
• altered-intercellular-communication — possible relevance of a secreted senescence-associated factor
• homo-sapiens — human-evidence and founder-population context

Footnotes

Footnotes

1. khan-2017-serpine1-longevity · n=177 enrolled Amish adults (primary analysis: 127 noncarriers + 43 heterozygotes; 7 homozygotes analyzed secondarily) · observational · adjusted primary LTL P=0.007; lifespan P=0.037 · model: Berne Old Order Amish adults and extended pedigree · doi:10.1126/sciadv.aao1617 · PMID 29152572 · PMC5687852 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴

2. khan-2017-serpine1-longevity supplementary materials · observational · model: extended Berne Amish pedigree + CARDIA external score-validation cohort · ancestral records: 221 with vital status, 113 deceased, 56 with direct or obligate genotype ascertainment (15 carriers, 41 unaffected relatives in Figure 2); CARDIA n=2793 for scores 1–2 and Methods/table-body n=892 for score 3 · doi:10.1126/sciadv.aao1617 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷