Falsification-aware triangulation of macrophage–stromal signaling in tendon repair prioritizes SPP1 across murine and human evidence layers

Abstract Public single-cell data can nominate macrophage–stromal signals in tendon repair but may overstate temporal replication and causality. We asked which ligand candidates remained supported after replicated murine temporal testing, matched-null network perturbation, within-patient human analysis, and bounded external-human assessment. Methods We used a versioned plan frozen before candidate-restricted downstream testing. Four pooled murine tendon-injury single-cell libraries from GSE288443 (d0, d1, d7, and d28; one library per time point) were analyzed separately for state reconstruction and within-library CellChat–NicheNet prioritization. SPP1, TGFB1, OSM, and TNF were frozen for downstream assessment. Temporal evidence was evaluated in 26 independent GSE288444 bulk RNA-sequencing samples. Candidate-specific scTenifoldKnk analyses used five network seeds and empirical nulls of controls matched for expression, detectability, and network topology. Human evidence was assessed in 23 within-patient GSE26051 tendon pairs, including complete-pair exclusion sensitivity. Acute quadriceps and healthy Achilles datasets provided donor-aware contextual localization only. Final classes followed a non-weighted rule. Results Quality control retained 29,868 of 30,880 cells, resolving ten myeloid states and 14 mesenchymal states in seven receiver families. The myeloid system formed a resampling-stable but directionally unresolved chain; the full mesenchymal trajectory failed the prespecified concordance gate. Across 9,600 CellChat tasks, 390 canonical receiver-family ligand–receptor axes were assembled, and NicheNet retained 12 primary-core bundles in six reporting families. All four ligands showed replicated murine temporal effects, although receiver-program coherence varied. Perturbation profiles were stable across seeds, but none exceeded its candidate-specific matched null or achieved four-of-five-seed receiver-program enrichment. Within-patient results supported SPP1 most strongly and TGFB1 more moderately; OSM and TNF were opposite in direction and imprecise. External-human context did not change the classes: SPP1 Tier 1, TGFB1 Tier 2, OSM Tier 3, and TNF Tier 4. Conclusions SPP1 showed the strongest cross-layer directional coherence and is the leading candidate for experimental testing. TGFB1 remained context dependent, OSM exploratory, and TNF a falsification comparator. Negative matched-null results constrain claims of exceptional cell-intrinsic network centrality without excluding extracellular signaling hypotheses. Protein localization, receptor engagement, and replicated functional perturbation are required before causal or therapeutic conclusions.摘要 公共单细胞数据集可用于筛选肌腱修复过程中巨噬细胞-基质信号通路的候选靶点，但可能会夸大其时间重复性与因果关联。本研究旨在明确，经过重复小鼠时间序列验证、匹配对照的网络扰动分析、同一患者配对人类样本分析以及限定范围的外部人类样本评估后，哪些配体候选物仍能获得可靠的实验支持。 方法 本研究采用了在配体候选物受限的下游分析前已固化的标准化分析方案。我们对来自GSE288443数据集的4个合并小鼠肌腱损伤单细胞文库（时间点分别为损伤后0天、1天、7天及28天，每个时间点对应1个文库）分别开展细胞状态重构与文库内CellChat–NicheNet优先级排序分析。选定SPP1、TGFB1、OSM及TNF作为下游验证的候选配体。利用26个独立的GSE288444批量RNA测序（bulk RNA-sequencing）样本评估候选物的时间序列表达证据。针对每个候选配体的scTenifoldKnk分析采用5个网络种子，并使用与表达水平、检测灵敏度及网络拓扑结构相匹配的对照经验零假设。我们在23对同一患者配对的GSE26051肌腱样本中评估人类样本的验证证据，其中包含完整样本对的排除敏感性分析。急性股四头肌与健康跟腱数据集仅用于供体特异性上下文定位分析。最终的候选物分级采用非加权规则。 结果 质控步骤共保留30880个细胞中的29868个，在7个受体家族中解析出10个髓系细胞状态与14个间充质细胞状态。髓系细胞系统形成了重采样稳定但方向未明确的链式分化轨迹；完整的间充质细胞轨迹未通过预先设定的一致性阈值筛选。在共计9600项CellChat分析任务中，共组装得到390条经典受体家族配体-受体信号轴，NicheNet则在6个报告家族中保留了12个核心信号模块。尽管不同候选物的受体程序一致性存在差异，但4个候选配体均在小鼠实验中表现出可重复的时间序列表达效应。不同网络种子的扰动分析结果均保持稳定，但没有任何一个候选配体的分析结果超过其匹配的对照零假设，也未达到5个种子中4个种子的受体程序富集标准。同一患者配对样本的分析结果最强烈支持SPP1，对TGFB1的支持程度中等；OSM与TNF的分析结果方向相反且置信度不足。外部人类样本的验证结果未改变候选物的分级：SPP1为1级，TGFB1为2级，OSM为3级，TNF为4级。 结论 SPP1表现出最强的跨层方向一致性，是当前最具潜力的实验验证候选靶点。TGFB1仍表现出上下文依赖性，OSM尚处于探索阶段，而TNF则作为证伪性对照使用。匹配对照的零假设阴性结果限制了关于细胞内在网络核心性的过度主张，但并未排除细胞外信号转导假说的可能性。在得出因果关联或治疗应用相关结论之前，需验证蛋白质定位、受体结合以及可重复的功能扰动实验结果。