属丰度 vs 功能通路:跨体质标志物的「更稳层」到底是什么?

Written by

in

Language

Evidence level: B

在痰湿、阳虚、湿热三体质里,菌属水平(如 Faecalibacterium)方向不一致——但切到「功能层」(粪便丁酸 + 丁酸合成通路基因)是否更稳?这是一个工程问题,不是哲学问题。

正文

上一篇文章(#700)我们给出了一个反直觉但清楚的发现:同样叫”短链脂肪酸产菌明星”的 Faecalibacterium,在痰湿质里被耗竭(30 例 vs 30 例 vs 30 例肥胖对照;PMID 35719350),在阳虚质健康个体里却又被富集出来(”双向升高”,LDA>4,P=0.0484;PMID 35294128)。换句话说,属水平的方向在跨体质时会反向——属这层是错位的。但我们想要的应答标志物必须能跨体质套用,因为它是要给 RCT 入组与亚组分层用的。这就逼出一个工程问题:粒度再粗一点或再细一点能不能把这毛病修掉?

本文就给这个问题一个工程裁决方案。要修掉”方向翻转”,候选方案有三:(1) 切到种(F. prausnitzii vs F. duncaniae vs F. longum,#910 已展开);(2) 切到丁酸合成通路基因(but/buk/bcd);(3) 切到代谢终产物(粪便丁酸绝对定量)。后两个本质上都是”功能层”。本文重点放在后两层,也就是”功能层 vs 菌属层,跨体质方向稳定性头对头”的裁决上。

一、为什么”功能层”理论上应该更稳?

逻辑很简单:产丁酸是一个功能性 guild,不是一个属。Louis 与 Flint 团队 2010 年用 butyryl-CoA:acetate CoA-transferase 基因(but)在 10 名健康人粪便里识别出 12 个 OTU,4 个来自 Eubacterium rectale、Roseburia faecis、Eubacterium hallii 和未命名 SS2/1,其余 12% 序列属于 20 个未培养 OTU(PMID 19807780)。换句话说,”会产丁酸”的并不只有 Faecalibacterium——它是个功能”行会”,只要 but 或 buk 终端基因 + 上游 thl/hbd/cro/bcd 在,加上纤维底物,就都能把纤维变成丁酸。Vital 等 2014 年在 3,184 个基因组里筛出 225 个潜在丁酸产菌,跨 10 个门(Firmicutes、Actinobacteria、Bacteroidetes、Fusobacteria 等),Lachnospiraceae 与 Ruminococcaceae 是主力,但还有大量不归这两家管的(PMID 24757212)。所以”功能层”是去物种偏倚的等价物——它绕开了”谁来做”这件事,直接看”做了没有”。

更进一步,Tierney 等 2021 年用 13 个队列、>100 万微生物特征做跨疾病 meta,发现 T2D 这种病的”物种级签名”和”基因级签名”会给出不一致的答案——属级和基因级的张力是结构性事实(PMID 34006865)。他们专门强调:”at the species-level… not consistently identified across models—however, our gene-level analysis unveils a group of robust, strain-specific gene associations”——也就是说,属级很可能漏掉真信号,基因级反而能捞回来。

二、四个实证锚点

实证一:功能层比属层在跨疾病预测上更好。Pasolli 等 2016 年用 2,424 个 metagenomic 样本做 8 个研究的 RF/SVM,对比”物种级丰度”vs”菌株特异性标记”两种特征——结论是”in general improved by feature selection and use of strain-specific markers instead of species-level taxonomic abundance”(PMID 27400279)。同样的 AUC 在物种级上不去,切到菌株特异性标记就上得去。这是”功能层/菌株级 > 物种级”的方法学锚点。

实证二:丁酸代谢基因与产丁酸菌丰度在疾病里脱钩可以反过来。Stø 等 2022 年做了一个反直觉发现:严重颈动脉粥样硬化(n=43)患者粪便丁酸水平反而升高(P=0.024),伴随产丁酸通路的功能潜能升高(P=0.031),但同时肠屏障损伤标志物 CCL25 与 IFABP 也升高(PMID 36572703)。解释是:屏障吸收少了,丁酸”留在”肠腔里。粪便丁酸高 ≠ 体内丁酸高——这就是功能层和代谢终产物之间还可以再断一层。

实证三:属层在湿热质与痰湿质上都有 SCFA 产菌缺失。湿热多中心 249 例(DHC vs BC,Barnesiella/Coprobacter/Butyricimonas 显著下降,丙酸代谢通路抑制,PMID 39957168);痰湿质 PCOS 队列 54 例(痰湿 vs 非痰湿 vs 健康,Faecalibacterium 耗竭 + 丁酸与丙酸显著下降 P<0.01,PMID 41214659;痰湿肥胖队列 Faecalibacterium 耗竭 + 70 种功能改变,PMID 35719350)。两个偏颇质都缺 SCFA 产菌——但缺的是不同的菌。

实证四:痰湿质功能通路被抑制。痰湿肥胖队列 PICRUSt 分析显示 70 种功能差异,主要是 Arachidonic acid metabolism、LPS biosynthesis 升高,Arginine/proline metabolism、Flavone/flavonol biosynthesis、Glycolysis/Gluconeogenesis、primary bile acid biosynthesis 降低(PMID 35719350)。即菌属层与功能层在痰湿质这里是同步变化,但”哪种 SCFA 产菌减少”和”哪条 SCFA 通路减少”并不一一对应——属层和功能层的联系要靠分类/KEGG 通路重建(如 PICRUSt)才能连起来,且这层重建又会引入新偏倚。

三、裁决设计:四头对头实验

基于实证一、二、四的方法学建议与实证三的”双偏颇都缺 SCFA 产菌”现象,给以下裁决设计:

  1. 同队列、同步上”三类层”:①属水平(Faecalibacterium/Roseburia/F. prausnitzii)qPCR 绝对定量;②基因水平(粪便 DNA,宏基因组 KEGG map00650/butyyrate metabolism:but/buk/thl/hbd/cro/bcd 相对丰度);③代谢终产物(粪便 GC-MS 绝对定量三种 SCFA)。
  2. 三体质亚组:痰湿 vs 阳虚 vs 湿热,每亚组 ≥60 例;再叠平和质作为基线对照。终点:CCMQ 评分改善(应答者/非应答者按 MCID 分层)+ 同种菌群干预 RCT 前后三时点采样。
  3. 指标:每个体每层(属/基因/代谢)独立建模对”应答”的预测 ΔAUC + 跨体质方向一致性检验(同种斜率+bootstrap CI)+ 对 Z 标化后的属丰度-基因-代谢做 spearman 相关矩阵。
  4. 判读表(参照 #700 四象限扩展):
    – A 真稳定:三亚组斜率同号+CI 重叠,属/基因/代谢三层一致 → 功能层更优成立
    – B 仅代谢稳:属层方向翻转,但粪便丁酸/基因层方向一致 → 走代谢层(”功能层更优”局部成立)
    – C 仅基因稳:丁酸代谢产物不稳定(如吸收差的混淆)但基因层稳定 → 走基因层
    – D 全反转:三层都不稳 → 撤回”一菌一标志”假设,方向是诊断指纹而非应答预测器
  5. 关键警戒:参考实证二的反直觉——粪便 SCFA 必须同时测血清/血浆 SCFA(UPLC-MS/MS)与肠屏障(CCL25/IFABP/内毒素),否则”粪便丁酸升高”会被误读为”体内丁酸升高”,与”功能层更稳”的结论相反。这是本研究的方法学硬约束。

四、结论

属层不是金标准——它甚至可能不如基因层稳定(实证一),而基因层又不必然反映”体内稳态”(实证二)。但我们不知道功能层是否真比菌属层更稳直到实测——同一队列四个体的功能/菌属/代谢三层头对头裁决,是一个工程问题,不是哲学问题。建议:所有体质×菌群干预 RCT 在主分析之外,强制加跑功能层与代谢终产物作为敏感性分析。否则单凭菌属层 ΔAUC 给”应答预测器”,跨体质迁移时大概率会结论翻转(实证三、四的双锚点已经显示:属层即便在同一偏颇质内也只是”同步”现象,不是”真正原因”)。最后再次强调:功能层不必然替代菌属层——它是一个补充层;真正的处方必须三层都看,但本文重点是给”功能层在跨体质条件下是否更稳”这个命题一个可裁决、可复制的工程答案。

证据与局限

已验证:
– Louis 2010 实证 butyryl-CoA:acetate CoA-transferase 在健康人粪便 OTU 涵盖非 Faecalibacterium 类(F. prausnitzii 在菊粉后显著升高),功能层不依赖于单一属(PMID 19807780,A 级直接实证)。
– Vital 2014 实证 225 个潜在产丁酸菌跨 10 个门,”会产丁酸”是 functional guild 不是分类群(PMID 24757212,A 级直接实证)。
– Pasolli 2016 实证 2,424 例机器学习 meta,菌株标记/strain-specific markers 优于物种级丰度(PMID 27400279,A 级方法学锚点)。
– Tierney 2021 实证属级 vs 基因级会出现不一致/张力(PMID 34006865,B 级张力的结构性事实)。

部分验证:
– 痰湿质(PMID 35719350 + PMID 41214659)与湿热质(PMID 39957168)的 SCFA 产菌缺失方向不同——功能层是否也存在类似”双偏颇都缺”但缺的是不同的通路,需同队列叠加实测(实证缺口)。
– Stø 2022 实证粪便丁酸升高 = 屏障吸收差而非真正健康(PMID 36572703)——这是”功能层陷阱”:代谢终产物受吸收干扰,单独看粪便值会被误导(中等直接证据)。

待验证:
– 三体质同队列同步测属/基因/代谢三层,跨体质方向一致性的 ΔAUC 与 Jaccard 方向重叠率(本文主张的核心实证缺口)。
– 粪便/血浆 SCFA 比值作”吸收效率代理”在中医体质亚组里能否校正实证二的陷阱。

相关阅读

证据等级: B(PMID 19807780、24757212、27400279、34006865、35719350、35294128、36572703、39957168、41214659 为 A/B 级直接实证;本文主张的核心裁决”功能层跨体质更稳”尚无三体质同队列 + 三层同步实测,属于 B 级待验证据)
参考文献: PMID 19807780, PMID 24757212, PMID 27400279, PMID 34006865, PMID 35719350, PMID 35294128, PMID 36572703, PMID 39957168, PMID 41214659, PMID 30532706(背景方法学)


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。

Allele Frequency vs Functional Pathways: What is the \”More Stable Layer\” of Cross-Tissue Biomarkers?

Evidence level: B

In damp-phlegm, yang-deficiency, and damp-heat constitutions, genus-level abundance (e.g., Faecalibacterium) shows inconsistent directions— but does cutting to the “functional layer” (butyrate + butyrate synthesis pathway genes) offer greater stability? This is an engineering problem, not a philosophical one.

Main Text

In the previous article (#700), we presented an intuitive yet clear finding: Faecalibacterium, labeled a “short-chain fatty acid-producing star,” is depleted in damp-phlegm constitution (30 cases vs. 30 cases vs. 30 obese controls; PMID 35719350) but enriched in healthy yang-deficiency individuals (“bidirectional increase,” LDA>4, P=0.0484; PMID 35294128). In other words, the direction at the genus level reverses across constitutions—the genus layer is misaligned. However, our desired responder marker must be applicable across constitutions as it is used for RCT enrollment and subgroup stratification. This forces an engineering question: Can increasing or decreasing the granularity fix this issue?

This article provides an engineering adjudication solution. To correct the “direction flip,” there are three candidate strategies: (1) cut to the species level (F. prausnitzii vs. F. duncaniae vs. F. longum, #910 already expanded); (2) cut to the butyrate synthesis pathway genes (but/buk/bcd); (3) cut to the metabolic end product (absolute quantification of fecal butyrate). The latter two are essentially the “functional layer.” This article focuses on the latter two layers, specifically adjudicating “functional layer vs. genus layer, head-to-head cross-constitution directional stability.”

I. Why the “Functional Layer” Should Theoretically Be More Stable

The logic is simple: butyrate production is a functional guild, not a genus. The Louis and Flint team identified 12 OTUs in the feces of 10 healthy individuals using the butyryl-CoA:acetate CoA-transferase gene (but) in 2010, with 4 originating from Eubacterium rectale, Roseburia faecis, Eubacterium hallii, and unnamed SS2/1, while the remaining 12% of sequences belonged to 20 uncultured OTUs (PMID 19807780). In other words, not only Faecalibacterium produces butyrate—it is a functional “guild.” As long as the terminal but or buk genes plus upstream thl/hbd/cro/bcd are present, along with fiber substrates, any organism can convert fiber into butyrate. Vital et al. (2014) screened 3,184 genomes and identified 225 potential butyrate producers across 10 phyla (Firmicutes, Actinobacteria, Bacteroidetes, Fusobacteria, etc.), with Lachnospiraceae and Ruminococcaceae being dominant, yet many others fall outside these two families (PMID 24757212). Thus, the “functional layer” is a bias-free equivalent—it bypasses the question of “who does it” and directly observes “whether it is done.”

Furthermore, Tierney et al. (2021) conducted a cross-disease meta-analysis using 13 cohorts and >1 million microbial features, finding that “species-level signatures” and “gene-level signatures” for diseases like T2D yield inconsistent answers—the tension between genus-level and gene-level is a structural fact (PMID 34006865). They specifically emphasized: “at the species-level… not consistently identified across models—however, our gene-level analysis unveils a group of robust, strain-specific gene associations”—meaning the genus level likely misses true signals, while the gene level can recover them.

II. Four Empirical Anchors

Empirical Finding 1: The functional layer outperforms the genus layer in cross-disease prediction. Pasolli et al. (2016) used 2,424 metagenomic samples across 8 studies with RF/SVM, comparing “species-level abundance” vs. “strain-specific markers”—concluding that performance “in general improved by feature selection and use of strain-specific markers instead of species-level taxonomic abundance” (PMID 27400279). The same AUC fails at the species level but rises with strain-specific markers. This is a methodological anchor for “functional/strain layer > species level.”

Empirical Finding 2: Decoupling butyrate metabolism genes from butyrate-producing bacteria abundance in disease can reverse the trend. Stø et al. (2022) made an intuitive finding: patients with severe carotid atherosclerosis (n=43) showed elevated fecal butyrate levels (P=0.024), accompanied by elevated functional potential of the butyrate pathway (P=0.031), while intestinal barrier damage markers CCL25 and IFABP also increased (PMID 36572703). The explanation is reduced barrier absorption, leaving butyrate “trapped” in the intestinal lumen. High fecal butyrate ≠ high systemic butyrate—this represents another layer of distinction between the functional layer and metabolic end products.

Empirical Finding 3: The genus layer shows SCFA-producing bacteria depletion in both damp-heat and damp-phlegm constitutions. Damp-heat multi-center study with 249 cases (DHC vs. BC, significant decrease in Barnesiella/Coprobacter/Butyricimonas, inhibition of propionate metabolism pathway, PMID 39957168); Damp-phlegm PCOS cohort with 54 cases (damp-phlegm vs. non-damp-phlegm vs. healthy, Faecalibacterium depletion + significant decrease in butyrate and propionate P<0.01, PMID 41214659; damp-phlegm obese cohort Faecalibacterium depletion + 70 functional changes, PMID 35719350). Both biased constitutions lack SCFA producers—but they lack different bacteria.

Empirical Finding 4: The damp-phlegm constitution shows suppressed functional pathways. PICRUSt analysis of the damp-phlegm obese cohort revealed 70 functional differences, primarily increased Arachidonic acid metabolism, LPS biosynthesis, Arginine/proline metabolism, and decreased Flavone/flavonol biosynthesis, Glycolysis/Gluconeogenesis, and primary bile acid biosynthesis (PMID 35719350). That is, the genus layer and functional layer change synchronously in the damp-phlegm constitution, but “which SCFA producer decreases” and “which SCFA pathway decreases” do not correspond one-to-one—the link between genus and functional layers relies on reconstruction via classification/KEGG pathways (e.g., PICRUSt), which introduces new biases.

III. Adjudication Design: Four Head-to-Head Experiments

Based on methodological suggestions from Empirical Findings 1, 2, and 4, and the “double bias lacking SCFA producers” phenomenon in Empirical Finding 3, we propose the following adjudication design:

  1. Same cohort, simultaneous measurement of “three layers”: ① Genus level (Faecalibacterium/Roseburia/F. prausnitzii) qPCR absolute quantification; ② Gene level (fecal DNA, metagenome KEGG map00650/butyrate metabolism: relative abundance of but/buk/thl/hbd/cro/bcd); ③ Metabolic end product (fecal GC-MS absolute quantification of three SCFA).
  2. Three constitution subgroups: Damp-phlegm vs. Yang-deficiency vs. Damp-heat, with ≥60 cases per subgroup; plus normal and constitution baseline controls. Endpoints: CCMQ score improvement (responder/non-responder stratified by MCID) + pre/post-intervention sampling at three time points for the same microbial intervention RCT.
  3. Metrics: For each constitution and layer (genus/gene/metabolism), independent modeling for prediction of “response” ΔAUC + cross-constitution directional consistency test (same slope + bootstrap CI) + Spearman correlation matrix for Z-standardized genus abundance vs. gene vs. metabolism.
  4. Interpretation Table (referencing #700 four-quadrant expansion):
    – A True Stability: Same-sign slopes across three subgroups with overlapping CIs, consistency across genus/gene/metabolism layers → “Functional layer is superior” holds
    – B Only Metabolism Stable: Genus layer direction flips, but fecal butyrate/gene layer directions consistent → Proceed with metabolism layer (“Functional layer is superior” locally holds)
    – C Only Gene Stable: Butyrate metabolites unstable (e.g., absorption confounding) but gene layer stable → Proceed with gene layer
    – D Full Reversal: None of the three layers stable → Withdraw “one-bacterium-one-marker” assumption; direction is a diagnostic fingerprint rather than a responder predictor
  5. Critical Warning: Referencing the intuitive finding in Empirical Finding 2—fecal SCFA must be measured alongside serum/plasma SCFA (UPLC-MS/MS) and intestinal barrier markers (CCL25/IFABP/endotoxin), otherwise “elevated fecal butyrate” may be misread as “elevated systemic butyrate,” contradicting the conclusion that “the functional layer is more stable.” This is a methodological hard constraint for this study.

IV. Conclusion

The genus layer is not the gold standard—it may even be less stable than the gene layer (Empirical Finding 1), and the gene layer does not necessarily reflect “systemic homeostasis” (Empirical Finding 2). However, we do not know if the functional layer is truly more stable than the genus layer until measured—a head-to-head adjudication of the functional/genus/metabolism three layers within the same cohort is an engineering problem, not a philosophical one. Recommendation: All constitution × microbial intervention RCTs should run functional layers and metabolic end products as sensitivity analyses alongside primary analyses. Otherwise, relying solely on genus layer ΔAUC for “responder predictors” will likely lead to conclusion flips during cross-constitution migration (Empirical Findings 3 and 4 already show that even within the same biased constitution, the genus layer is merely a “synchronous” phenomenon, not the “true cause”). Finally, reiterate: The functional layer does not necessarily replace the genus layer—it is a supplementary layer. The true prescription must examine all three layers, but this article focuses on providing a adjudicable, replicable engineering answer to the proposition: “Is the functional layer more stable under cross-constitution conditions?”

Evidence and Limitations

Verified:
– Louis 2010 empirical evidence shows butyryl-CoA:acetate CoA-transferase covers non-Faecalibacterium OTUs in healthy feces (F. prausnitzii significantly increases after inulin), demonstrating functional layer independence from a single genus (PMID 19807780, Level A direct empirical evidence).
– Vital 2014 empirical evidence shows 225 potential butyrate producers across 10 phyla, confirming “ability to produce butyrate” is a functional guild, not a taxonomic group (PMID 24757212, Level A direct empirical evidence).
– Pasolli 2016 empirical evidence from a machine learning meta-analysis of 2,424 cases shows strain-specific markers outperform species-level abundance (PMID 27400279, Level A methodological anchor).
– Tierney 2021 empirical evidence shows inconsistency/tension between genus-level and gene-level (PMID 34006865, Level B structural fact of tension).

Partially Verified:
– Different directions of SCFA-producing bacteria depletion in damp-phlegm constitution (PMID 35719350 + PMID 41214659) and damp-heat constitution (PMID 39957168)—whether the functional layer exhibits a similar “double bias lacking different pathways” requires same-cohort combined measurement (empirical gap).
– Stø 2022 empirical evidence shows elevated fecal butyrate = poor barrier absorption rather than true health (PMID 36572703)—this is a “functional layer trap”: metabolic end products are affected by absorption; viewing fecal values alone can be misleading (moderate direct evidence).

To Be Verified:
– Cross-constitution directional consistency ΔAUC and Jaccard directional overlap rate from same-cohort simultaneous measurement of genus/gene/metabolism three layers across three constitutions (core empirical gap advocated in this article).
– Whether fecal/plasma SCFA ratio as an “absorption efficiency proxy” can correct the trap in Empirical Finding 2 within TCM constitution subgroups.

Related Reading

  • [Previous Article: Cross-Constitution Directional Consistency Adjudication] – tcm-constitution-responder-marker-cross-constitution-direction-consistency (#700 parent article, genus level vs. cross-constitution direction anchor)
  • [Parent Article: Cross-Constitution Stability Adjudication] – tcm-constitution-responder-marker-cross-constitution-stability (#700 parent series, this article is sub-topic #701)
  • [Deep Dive #910: Phylogroup I/II Cross-Constitution Directional Stability] – Species-level phylogroup (Layer 1 alternative advocated here; Layer 2 and 3 advocated here)
  • [Research Pool ④: Environment-Life Frequency Interface] (/research/pools/pool-4-environment-life-frequency-interface)
  • [Research Pool Homepage] (/topics/research-pools)

Evidence Level: B (PMIDs 19807780, 24757212, 27400279, 34006865, 35719350, 35294128, 36572703, 39957168, 41214659 are Level A/B direct empirical evidence; the core adjudication advocated here—”functional layer is more stable across constitutions”—lacks same-cohort three-constitution + three-layer simultaneous measurement, belonging to Level B pending evidence).
References: PMID 19807780, PMID 24757212, PMID 27400279, PMID 34006865, PMID 35719350, PMID 35294128, PMID 36572703, PMID 39957168, PMID 41214659, PMID 30532706 (background methodology)


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.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *