Evidence level: B
皮纹能否在 PRS 之上再贡献一块独立预测增量,取决于”皮纹独有信息”与”PRS 遗传信息”的重叠度——而这一重叠度从未被任何队列直接量化过。
正文
现代遗传流行病学已经确立了多基因风险评分(PRS)在冠心病、精神分裂等复杂疾病中的预测角色,但一个反复被追问的问题始终悬而未决:在 PRS 之上,再叠一层皮纹这类体表发育标记,预测能力到底还能提升多少?
这个问题之所以值得问,是因为皮纹与 PRS 的”信息源”不同。PRS 直接测量个体的基因组遗传负荷,回答”你从父母那里继承了多少风险等位基因”;而皮纹在孕早期(约第 13-19 周)由表皮与神经系统共同的外胚层来源定型、出生后终生不变,同时携带遗传与宫内环境(如缺氧、营养)的双重印记。理论上,如果宫内环境的扰动”刻”进了皮纹却没有完全反映在 PRS 里,皮纹就应该能提供 PRS 之外的增量信息——这就是”两条出生即定的信息通道”假说的核心。
先看皮纹本身有没有预测力。宁夏一项 258 名男性的病例对照研究显示,冠心病患者弓纹频率、ATD 角与 A-B 脊线数显著高于对照(PMID 26816877),提示皮纹形态可作为冠心病早期筛查的候选标志。更早的 Barker 学派前瞻性研究则发现,出生时较瘦的婴儿成年后斗型纹更多,且”至少一个手指有斗型纹”者收缩压平均高出 8 mmHg(PMID 8374451)——这是皮纹-血压关联的经典证据,直接把皮纹定位为”胎儿编程”(fetal programming)的体表印记。精神分裂方向,2024-2025 年的家族研究进一步显示,DISC1 基因变异与皮纹特征(较低的指型强度指数 PII、更宽的 ATD 角)在患病后代中共同传递(PMID 41037940),且 CNR1 基因多态性会反向调节 PII 与精神分裂风险的关联(PMID 39457583)——说明皮纹与特定基因确实存在可测量的交互。
但”皮纹能预测疾病”和”皮纹能贡献 PRS 之外的增量”是两回事。后者需要回答一个更苛刻的问题:当 PRS 已经进入模型后,皮纹还能不能显著提升区分度(ΔAUC)和临床重分类(净重分类改善 NRI)?这在方法学上已有成熟范式可循。UK Biobank 35 万余人队列中,在标准风险方程之上加入冠心病 PRS,C 统计量从 0.76 升至 0.78(ΔC=0.02),整体 NRI 4.0%(PMID 32068818);而 ARIC/MESA 分析显示加入 PRS 后 C 统计量变化仅为 -0.001 至 +0.021,NRI 未达显著(PMID 32068817)。一篇 2022 年 JAMA Intern Med 综述系统总结了 5 项最大规模研究:PRS 单用 C 统计量仅 0.549-0.623,加入传统风险模型后增量介于 -0.001 到 +0.021,NRI 在 0.001-0.097——”从可忽略到温和”(PMID 35994254)。
这个”温和甚至可忽略”的基准,正是皮纹增量检验必须面对的现实:即便 PRS 本身,增量也是有限的。 精神分裂方向更直白:UK Biobank 738 例病例的机器学习预测,把 PRS、性别、教育程度等全部纳入后最优 AUROC 也仅 0.71,且所有 ML 模型的判别力均未显著超越逻辑回归(PMID 35779327)。换句话说,如果我们期待皮纹在 PRS 之上贡献 ΔAUC≥0.05 的”惊喜”,很可能要落空;真实可期待的是 ΔAUC 在 0.01-0.03、NRI 小幅为正的”渐进改善”——而这恰好是 PRS 自身增量的量级。
那么”皮纹+PRS”联合模型该如何设计?三个关键决策点:
第一,测量什么。皮纹特征应选与疾病有机制关联的定量指标——精神分裂用 PII、总 a-b 脊线数(TABRC)及其波动不对称(FA-ABRC)、ATD 角(PMID 41037940);冠心病/高血压用斗型纹频率、A-B 脊线数、ATD 角(PMID 26816877, 8374451)。所有测量遵循标准化协议并做操作者间信度检验。
第二,检验什么增量。核心是三模型对比:①单用 PRS;②单用皮纹;③PRS+皮纹。在交叉验证下用 ΔAUC(DeLong 检验)与 NRI 判定③相对①的增量,并附重分类表与校准曲线。预设阈值:ΔAUC≥0.03 且 NRI 置信区间不跨零,才认定”皮纹有独立增量”。同时报告③相对②的增量——它回答的是反问题:”皮纹能不能替代 PRS 作为廉价筛查通道”。
第三,控制什么混杂。皮纹随族裔基线不同,PRS 也有跨族裔可移植性问题,必须做族裔分层或纳入族裔协变量,并在独立队列外部验证,避免”皮纹增量”其实是族裔混杂的产物。此前两篇文章(皮纹 CNN 跨族裔 AUC 衰减、域适应补偿)已经标定了这道坎的规模。
为什么这值得做?如果增量成立,皮纹就提供了 PRS 做不到的东西:出生即固定、零成本、无需基因检测的发育印记通道,正好补上 PRS 测不到的那部分”早期环境×基因交互”信息。如果增量不成立,那也是重要的负面结果——它告诉我们皮纹与 PRS 的信息重叠度高,皮纹主要反映的还是遗传负荷本身,单用皮纹做筛查的时代该翻篇了。
证据与局限
已验证: 皮纹与冠心病、高血压、精神分裂存在稳定关联(PMID 26816877, 8374451, 41037940);PRS 对冠心病的增量预测已有大规模实证,但增量幅度从可忽略到温和不等(PMID 32068818, 32068817, 35994254);精神分裂 PRS 预测的判别力上限已被 UK Biobank ML 研究标定(PMID 35779327)。
部分验证: 皮纹与特定基因(DISC1、CNR1)的交互在精神分裂家族样本中被观察到(PMID 41037940, 39457583),提示皮纹携带 PRS 之外的基因×环境交互信息;但尚未在任何队列中直接检验”皮纹+PRS”联合模型的增量。
待验证: 多疾病队列(精神分裂/高血压/冠心病)中皮纹+PRS 联合模型相对单用 PRS 的 ΔAUC 与 NRI 增量检验;皮纹独有信息的方差分解(与 PRS 的遗传相关估计);低 PRS 但皮纹异常者的独立风险检验(对应旧文命题二);族裔分层下的增量稳健性。
相关阅读
- 皮纹+基因风险评分:两条”出生即定”的信息通道能否联合预测疾病
- 皮纹CNN跨族裔体质分类AUC衰减曲线
- 皮纹CNN跨族裔衰减的补偿:域适应能救回多少
- 皮纹GWAS多效性与共定位筛查
- DISC1-rs999710 与皮纹异常共传递检验
- 生物频率密码研究池
证据等级: B(皮纹-疾病关联与 PRS 增量方法学均有直接实证;”皮纹+PRS 联合增量”无直接研究,为实验设计方案)
参考文献: PMID 26816877, PMID 8374451, PMID 41037940, PMID 39457583, PMID 32068818, PMID 32068817, PMID 35994254, PMID 35779327
免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。
Skin Ridge Patterns plus Polygenic Risk Scores versus PRS alone: Delta AUC and Net Reclassification Improvement across multiple disease cohorts
Evidence level: B
Whether dermatoglyphics can contribute an independent predictive increment beyond PRS depends on the degree of overlap between “dermatoglyphic-unique information” and “PRS genetic information”—an overlap that has never been directly quantified in any cohort.
Main Text
Modern genetic epidemiology has established the predictive role of Polygenic Risk Scores (PRS) in complex diseases such as coronary heart disease and schizophrenia, yet a question repeatedly raised remains unresolved: To what extent can adding a layer of dermatoglyphics—a somatic developmental marker—further enhance predictive capability beyond PRS?
This question is warranted because the “information sources” of dermatoglyphics and PRS differ. PRS directly measures an individual’s inherited genetic load, answering “how many risk alleles you inherited from your parents”; conversely, dermatoglyphics are determined by the ectodermal origin shared by the epidermis and nervous system during early pregnancy (approximately weeks 13–19), remain invariant throughout life after birth, and carry dual imprints of genetic factors and intrauterine environments (e.g., hypoxia, nutrition). Theoretically, if intrauterine environmental perturbations are “etched” into dermatoglyphics but not fully reflected in PRS, dermatoglyphics should provide incremental information outside PRS—this constitutes the core of the “Two Birth-Determined Information Channels” hypothesis.
First, does dermatoglyphics itself possess predictive power? A case-control study of 258 males in Ningxia showed that coronary heart disease patients had significantly higher frequencies of whorls, ATD angles, and A-B ridge counts compared to controls (PMID 26816877), suggesting dermatoglyphic morphology could serve as a candidate biomarker for early screening. Earlier prospective studies by the Barker school found that infants born with lower birth weight had more whorls as adults, and those with “at least one whorl on a finger” had an average systolic blood pressure 8 mmHg higher (PMID 8374451)—classic evidence linking dermatoglyphics to hypertension and positioning them as somatic markers of “fetal programming.” In the schizophrenia direction, family studies from 2024–2025 further indicate that DISC1 gene variants and dermatoglyphic features (lower Peak Intensity Index PII, wider ATD angles) are co-transmitted in affected offspring (PMID 41037940), and CNR1 gene polymorphisms modulate the association between PII and schizophrenia risk (PMID 39457583)—demonstrating measurable interactions between dermatoglyphics and specific genes.
However, “dermatoglyphics can predict disease” and “dermatoglyphics can contribute incremental value beyond PRS” are distinct. The latter addresses a stricter question: When PRS is already included in the model, can dermatoglyphics still significantly improve discrimination (ΔAUC) and clinical reclassification (Net Reclassification Improvement, NRI)? A mature methodological paradigm exists for this. In the UK Biobank cohort of over 350,000 individuals, adding a coronary heart disease PRS to a standard risk equation increased the C-statistic from 0.76 to 0.78 (ΔC=0.02), with an overall NRI of 4.0% (PMID 32068818); conversely, ARIC/MESA analyses showed only a change of -0.001 to +0.021 in the C-statistic after adding PRS, with NRI not reaching significance (PMID 32068817). A 2022 JAMA Intern Med review systematically summarized five largest studies: PRS alone yielded C-statistics of only 0.549–0.623, incremental gains after adding traditional risk models ranged from -0.001 to +0.021, and NRIs were between 0.001–0.097—”from negligible to modest” (PMID 35994254).
This “modest or negligible” baseline is the reality dermatoglyphic incremental testing must face: Even PRS itself offers limited increments. In the schizophrenia direction, it is even more explicit: machine learning predictions in 738 UK Biobank cases incorporating PRS, sex, education, etc., achieved an optimal AUROC of only 0.71, and the discriminative power of all ML models did not significantly exceed logistic regression (PMID 35779327). In other words, if we expect dermatoglyphics to contribute a “surprise” with ΔAUC ≥ 0.05 above PRS, we may likely be disappointed; the realistic expectation is a “gradual improvement” with ΔAUC in the 0.01–0.03 range and a small positive NRI—precisely the magnitude of PRS’s own increment.
So how should a “dermatoglyphics + PRS” joint model be designed? Three critical decision points:
First, what to measure. Dermatoglyphic features should select quantitative metrics mechanistically linked to the disease—PII, total A-B ridge counts (TABRC), and their fluctuation asymmetry (FA-ABRC), and ATD angle for schizophrenia (PMID 41037940); whorl frequency, A-B ridge counts, and ATD angle for coronary heart disease/hypertension (PMID 26816877, 8374451). All measurements should follow standardized protocols and undergo inter-rater reliability testing.
Second, what incremental value to test. The core involves a three-model comparison: ① PRS alone; ② Dermatoglyphics alone; ③ PRS + Dermatoglyphics. Under cross-validation, use ΔAUC (DeLong test) and NRI to determine the increment of ③ relative to ①, accompanied by reclassification tables and calibration curves. Preset threshold: Only if ΔAUC ≥ 0.03 AND the NRI confidence interval does not cross zero is dermatoglyphic “independent increment” confirmed. Simultaneously report the increment of ③ relative to ②—it answers the inverse question: “Can dermatoglyphics replace PRS as a low-cost screening channel?”
Third, what confounders to control. Dermatoglyphics vary by ethnic baseline, and PRS faces cross-ethnic portability issues; therefore, stratification by ethnicity or inclusion of ethnic covariates is mandatory, with external validation in independent cohorts to avoid the “dermatoglyphic increment” being a product of ethnic confounding. The scale of this hurdle has already been flagged by two prior articles (dermatoglyphic CNN cross-ethnic AUC decay, domain adaptation compensation).
Why is this worth doing? If the increment holds, dermatoglyphics provides what PRS cannot: a developmental imprinting channel fixed at birth, zero cost, and no need for genetic testing, perfectly complementing the “early environment × gene interaction” information PRS misses. If no increment is found, it is also an important negative result—it tells us the information overlap between dermatoglyphics and PRS is high, meaning dermatoglyphics primarily reflect genetic load itself, signaling that the era of single-dermatoglyphic screening should be concluded.
Evidence and Limitations
Verified: Stable associations between dermatoglyphics and coronary heart disease, hypertension, and schizophrenia exist (PMID 26816877, 8374451, 41037940); PRS offers large-scale empirical evidence for incremental prediction in coronary heart disease, though increment magnitude varies from negligible to modest (PMID 32068818, 32068817, 35994254); the upper limit of discriminative power for schizophrenia PRS prediction has been established by UK Biobank ML studies (PMID 35779327).
Partially Verified: Interactions between dermatoglyphics and specific genes (DISC1, CNR1) have been observed in schizophrenia family samples (PMID 41037940, 39457583), suggesting dermatoglyphics carry gene×environment interaction information beyond PRS; however, no cohort has directly tested the incremental value of a “dermatoglyphics + PRS” joint model yet.
To be Verified: ΔAUC and NRI incremental testing of dermatoglyphics + PRS joint models versus PRS alone in multi-disease cohorts (schizophrenia/hypertension/coronary heart disease); variance decomposition of dermatoglyphic unique information (estimation of genetic correlation with PRS); independent risk testing for individuals with low PRS but abnormal dermatoglyphics (corresponding to Proposition II in the previous article); robustness of increments under ethnic stratification.
Related Reading
- Dermatoglyphics + Gene Risk Scores: Can Two “Birth-Determined” Information Channels Jointly Predict Disease
- Dermatoglyphic CNN Cross-Ethnic Constitution Classification AUC Decay Curve
- Compensation for Cross-Ethnic Decay in Dermatoglyphic CNN: How Much Can Domain Adaptation Recover
- Dermatoglyphic GWAS Pleiotropy and Colocalization Screening
- DISC1-rs999710 Co-transmission with Abnormal Dermatoglyphics Testing
- Biomolecular Frequency Code Research Pool
Evidence Level: B (Direct empirical evidence exists for dermatoglyphics-disease associations and PRS incremental methodology; “dermatoglyphics + PRS joint increment” lacks direct research and constitutes an experimental design proposal)
References: PMID 26816877, PMID 8374451, PMID 41037940, PMID 39457583, PMID 32068818, PMID 32068817, PMID 35994254, PMID 35779327
Disclaimer: This article compiles traditional knowledge and scientific literature for comparison and research-design discussion only. It is not medical advice and cannot replace professional diagnosis or treatment. Effect sizes, frequency ranges and exposure conclusions cited here are largely inferences at the research-report level, and individual circumstances vary widely. If you have any discomfort or health concerns, please consult a physician.
Leave a Reply