KLF6 across the Common Fund Data Ecosystem: an automated implementation of CFDE Use Case 6 linking MoTrPAC, HuBMAP/HRApop, LINCS and GTEx

Abstract

We implemented CFDE Use Case 6 ("knowledge about a gene") as an automated, reproducible pipeline and ran it for the transcription factor KLF6 (KLF transcription factor 6), starting from the intronic variant rs3750861. The pipeline maps a variant to its closest protein-coding gene (Ensembl), annotates the gene (MyGene.info), and queries four Common Fund data resources: MoTrPAC (effect of exercise training), HuBMAP via the Human Reference Atlas population atlas HRApop (cell types with high expression), LINCS L1000 (small molecules that induce expression) and GTEx (tissue expression). KLF6 is broadly and highly expressed (GTEx v8 median 5.4–264.7 TPM across 54 tissues, highest in Nerve Tibial). It is up-regulated in 391 LINCS chemical consensus signatures, and 7 of the 10 best-supported inducers inhibit PI3K, AKT or mTOR. In 6-month endurance-trained rats, KLF6 was not training-regulated in any of 22 tissue × assay × feature combinations at MoTrPAC's 5% FDR (lowest: subcutaneous white adipose RNA, FDR 0.21). In HRApop, KLF6 ranks among the top expressed genes of 14 cell-type summaries, led by LYVE1+ macrophages. All values are computed from public APIs; we state which steps needed substitute sources.

Introduction

The NIH Common Fund Data Ecosystem (CFDE) defines use cases that combine several Common Fund programs to answer one question. Use Case 6 asks what CFDE resources know about a single gene, with KLF6 as the worked example: given a gene or a variant, report the effect of exercise on its expression (MoTrPAC), the single cells that highly express it (HuBMAP), small molecules that induce it (LINCS) and its tissue expression (GTEx), and combine them in one interactive report.

KLF6 is a Krüppel-like zinc-finger transcription factor and tumor suppressor (UniProt Q99612; aliases BCD1, CBA1, COPEB, CPBP, GBF, PAC1, ST12, ZF9). Here we implement the workflow end to end and report what each resource returns for KLF6.

Methods

The pipeline is a Python 3.12 script with one R 4 helper. Each step calls a public API; no step needs credentials. Results were generated on 2026-10-14.

Input and variant mapping

An rsID is resolved with the Ensembl REST variation endpoint (GRCh38). Protein-coding genes within ±500 kb are fetched with the overlap/region endpoint, and the closest one is chosen (distance 0 if the variant lies inside the gene). A gene symbol input skips this step.

Gene annotation (MyGene.info)

MyGene.info v3¹ supplies identifiers, the RefSeq summary, GO biological process terms and HomoloGene orthologs. The rat ortholog (Entrez 58954) links the human gene to MoTrPAC.

Exercise (MoTrPAC)

We used the differential-analysis tables of the MoTrPAC PASS1B 6-month endurance training study in F344 rats² from the MotrpacRatTraining6moData repository: transcriptomics (19 tissues) and global proteomics (7 tissues). We selected the rat ortholog's features via FEATURE_TO_GENE_FILT and took, for each tissue × assay × feature, log2 fold changes for 1, 2, 4 and 8 weeks of training versus sedentary controls in each sex, plus MoTrPAC's selection_fdr (the per-feature training F-test, IHW-adjusted). A feature counts as training-regulated if selection_fdr < 0.05.

Single cells (HuBMAP / HRApop)

HuBMAP's gene-to-cell-type service (scFind, scfind.hubmapconsortium.org) did not accept connections during this study, and the HuBMAP data products are 11–13 GB .h5mu files per organ. We therefore queried HRApop, the Human Reference Atlas population atlas built from HuBMAP and CELLxGENE single-cell data³, through its SPARQL endpoint (lod.humanatlas.io/sparql, graph https://purl.humanatlas.io/graph/hra-pop). HRApop cell summaries keep the top expressed genes per cell type, so we listed every cell-summary row in which KLF6 (matched by Ensembl ID) appears, with its dataset, organ, annotation method (Azimuth, pan-human Azimuth, CellTypist), cell count and mean expression.

Small molecules (LINCS L1000)

We downloaded two Enrichr⁴ gene-set libraries derived from LINCS L1000⁵: LINCS_L1000_Chem_Pert_Consensus_Sigs (consensus up and down gene sets per compound) and LINCS_L1000_Chem_Pert_up (individual cell line × dose × time signatures). Inducers are compounds whose consensus up-set contains KLF6 and whose down-set does not. We ranked them by the number of individual signatures that also list KLF6 as up-regulated, matching compound names in signature labels. Compounds labeled only by BRD identifiers in that library get a count of 0, so the ranking undercounts them. Target classes in Table 2 are manual literature annotations, not computed.

Tissue expression (GTEx)

Median gene TPM per tissue came from the GTEx⁶ Portal API v2 (medianGeneExpression, dataset gtex_v8, GENCODE ENSG00000067082.14).

Report

All results go into one JSON file, which a renderer embeds into a self-contained interactive HTML report. The report shows the workflow graph as navigation, a MoTrPAC heatmap, the HRApop and LINCS tables (filterable) and sortable GTEx bars.

Results

Variant to gene

rs3750861 (chr10:3,782,241, alleles C/A/T, most severe consequence: intron variant) lies inside KLF6 (ENSG00000067082). KLF6 is the only protein-coding gene within ±500 kb, so the mapping is unambiguous.

Tissue expression (GTEx)

KLF6 is expressed in all 54 GTEx tissues. Table 1 lists the 10 highest and the 3 lowest.

Table 1. GTEx v8 median expression of KLF6.

Small molecules that induce KLF6 (LINCS)

391 compound consensus signatures list KLF6 as up-regulated and 624 as down-regulated; 0 list it in both. 524 individual signatures list it as up-regulated. Of the 10 best-supported inducers, 7 target PI3K, AKT or mTOR (Table 2), and 11 PI3K/AKT/mTOR inhibitors appear among all inducers.

Table 2. Top 15 LINCS L1000 compounds that induce KLF6.

Effect of endurance training (MoTrPAC)

No tissue × assay pair reached selection_fdr < 0.05 for rat Klf6 (Table 3). The lowest FDRs were in subcutaneous white adipose (RNA, FDR 0.21) and adrenal (RNA, FDR 0.52). No single sex × week contrast exceeded |log2FC| = 0.7, and none of the nominal p-values below 0.05 survives the training test (Table 3).

Table 3. MoTrPAC training response of rat Klf6, 8 lowest selection FDRs.

Cell types with high KLF6 expression (HRApop)

KLF6 appears among the top expressed genes in 14 HRApop cell-type summaries from heart, ovary, skin, small intestine (Table 4). All of them come from CELLxGENE datasets (Heart Cell Atlas, Tabula Sapiens); KLF6 is not among the top genes of any HuBMAP-sourced summary. Mean expression values use annotation-method-specific scales and should not be compared across methods.

Table 4. HRApop cell-type summaries in which KLF6 is a top expressed gene.

Discussion

Computed. Everything in the Results and Tables 1–4 except the target-class column of Table 2 comes directly from the public APIs listed in Methods.

Interpretation (hypotheses, not tested here). The enrichment of PI3K/AKT/mTOR inhibitors among KLF6 inducers suggests that KLF6 transcription is held down by PI3K–AKT–mTOR signaling, or rises as part of a stress or growth-arrest response to its inhibition. This fits KLF6's reported role as a growth-suppressive transcription factor, but L1000 signatures come mostly from cancer cell lines at 6–24 h and do not show a direct mechanism.

The MoTrPAC null result does not rule out an exercise effect. KLF6 is an immediate-early gene, and PASS1B tissues were collected 48 h after the last bout, which measures adaptation to training rather than the acute response. MoTrPAC's acute-exercise studies would be the right resource to test an acute effect.

Limitations. (1) HuBMAP was reached only indirectly. HRApop stores top genes per cell type, so absence from Table 4 does not mean absence of expression, and the HuBMAP-derived summaries never list KLF6. When scFind is back online, this step should query it directly. (2) LINCS ranking uses name matching and undercounts compounds labeled by BRD ID. (3) MoTrPAC data are from rats; we assume ortholog equivalence. (4) The use case title mentions metabolomics; MoTrPAC metabolomics does not map to genes, so we used transcriptomics and proteomics.

Data and code availability

All inputs are public: Ensembl REST (https://rest.ensembl.org), MyGene.info (https://mygene.info), GTEx Portal API v2 (https://gtexportal.org/api/v2), Enrichr libraries (https://maayanlab.cloud/Enrichr), HRApop (https://lod.humanatlas.io/sparql) and MotrpacRatTraining6moData (https://github.com/MoTrPAC/MotrpacRatTraining6moData). The pipeline (pipeline.py, motrpac_extract.R, render_report.py) is available from the operator on request.