Comprehensive Identification of the Bovine KLF Gene Family and Its Functional Regulation in Muscle Development: Insights from Single-Nuclei Transcriptomics

Simple Summary

Beef production is important for global food security. Understanding the genetic factors that control muscle growth and meat quality in cattle is key to improving breeding strategies. This study focused on a family of genes called KLF transcription factors, which are known to be master regulators of growth and metabolism. We systematically identified 14 KLF genes in the cattle genome. Using advanced single-nuclei RNA sequencing technology, we mapped the activity of these genes across different cell types (such as muscle fibers, fat cells, and immune cells) in the muscle tissue of two cattle breeds: Angus (beef breed) and Holstein (dairy breed). We discovered that the activity of these KLF genes varies greatly between cell types and between breeds. Using a machine learning approach, we identified KLF6, KLF9, KLF10, and KLF12 as key genes that may drive the differences in muscle development between the two breeds. Our findings provide valuable genetic targets for future breeding programs aimed at enhancing beef quality and yield.

Abstract

The Krüppel-like factor (KLF) family of transcription regulators plays pivotal roles in adipogenesis, myogenesis, and metabolism. While comprehensively studied in humans and mice, its characterization in cattle remains limited, especially within the skeletal muscle niche. This study aimed to systematically characterize the KLF family in Bos taurus and elucidate its role in breed-specific muscular development. We employed an integrated approach of comparative genomics and single-nucleus RNA sequencing (snRNA-seq) on longissimus dorsi muscle from Angus (ANG, beef breed) and Holstein (HST, dairy breed) cattle. Phylogenomic analysis identified 14 KLF genes, revealing evolutionary conservation and potential functional divergence. snRNA-seq delineated 11 distinct cell populations and uncovered cell-type-specific expression patterns of KLFs. Further machine learning based analysis pinpointed KLF6, KLF9, KLF10, and KLF12 as key global drivers of transcriptional differences between breeds, while KLF6 was identified as a major cell-type-specific contributor in lymphatic endothelial cells. Our work provides a foundational resource for understanding the KLF family in cattle and identifies promising candidate genes for improving meat production traits through molecular breeding.

1. Introduction

Beef is a key source of high-quality protein and essential nutrients (such as iron, zinc, and vitamin B12) and occupies a central position in the global dietary structure [1]. With population growth and rising living standards, global demand for beef continues to climb. Therefore, ensuring the stability of beef production and improving its yield and quality are crucial to meeting market demand and promoting social and economic development.

In cattle production, biological and pathological factors (such as diseases affecting health and growth) and genetic traits (determining growth potential, muscle fiber type, and fat distribution) jointly influence skeletal muscle growth and development and fat deposition efficiency [2]. Molecular breeding technology, with its advantages of precisely improving genetic traits and simultaneously enhancing meat yield and quality, has become a hot topic in cattle genetic improvement research.

Skeletal muscle, as the main component of beef, not only plays a key role in the movement and metabolism of cattle, but also serves as the foundation of meat production [3]. However, muscle growth and development are precisely controlled by numerous genes and their regulatory networks, making it a highly complex biological process. The Krüppel-like factor (KLF) family comprises a group of fundamental transcription regulators whose members mediate various important biological functions through a DNA-binding domain containing three zinc finger structures [4]. A total of 18 KLF genes, designated KLF1 through KLF18, have been identified in the human genome [5]. Research has shown that KLF genes have diverse functions in different species: for example, KLF4, KLF2, KLF3, and KLF6 are highly expressed in monocytes, macrophages, and dendritic cells in mice, while KLF1 is essential for red blood cell development [6]. In pig and human studies, KLF members are also involved in regulating muscle growth, metabolic processes, and cell proliferation and migration [7,8,9]. The role of the KLF gene in adipogenesis has been extensively studied. KLF2 has been shown to be a key negative regulator of adipogenesis. Recently, a study on single cells showed that overexpression of KLF2 in 3T3-L1 cells can inhibit PPARγ, thereby significantly reducing intracellular lipid accumulation [10]. KLF3 is a member of the KLF family, primarily responsible for adipocyte differentiation and fat deposition. Studies have shown that miR-32-5p can affect fat production by downregulating KLF3 in adipose tissue and intramuscular fat [11]. Due to the knockout of KLF3 by siRNAs, bovine fat production can be inhibited [12]. There are also studies indicating that KLF3 can inhibit the differentiation of porcine adipocytes in vitro by downregulating adipogenesis markers (including PPARG, C/EBPA, and FABP4) [11]. KLF6 is an important transcription factor in skeletal muscle development [13]. Research indicates that the KLF6 gene may serve as a potential candidate marker gene for beef breed improvement in QinChuan cattle through marker-assisted selection [14]. Research has shown that KLF9 is a novel regulatory factor induced by glucocorticoids (GC) in skeletal muscle. Its function is directly related to muscle atrophy: overexpression of KLF9 induces muscle atrophy, while knockout of its expression promotes muscle hypertrophy. These findings establish the position of KLF9 as a key regulator of skeletal muscle mass and reveal its important role in the GC-induced muscle atrophy pathway [15]. KLF4, KLF5, KLF6, KLF7, KLF10, and KLF15 are implicated in regulating skeletal muscle growth and development in mice [5,8,16,17,18], strongly suggesting that this family also plays a key role in muscle development in cattle. Although the KLF family has been systematically identified and functionally studied in other species (such as pigs and mice), there is still a significant lack of research on the bovine KLF gene family.

Given the established roles of KLF6, KLF9, KLF10, and KLF12 in muscle and fat production in other mammals, we hypothesize that these key KLF members are the core regulatory factors of transcriptional differences in the skeletal muscle microenvironment between beef and dairy breeds, thereby affecting muscle phenotype and meat production traits.

Due to the consistency of skeletal muscle tissue, which is a complex microenvironment composed of multiple cell types, most existing studies have been conducted using Bulk RNA-seq, which cannot reveal the precise expression profile and functional differentiation of KLF genes in specific cell types. The emergence of single-cell transcriptome sequencing (scRNA-seq or snRNA-seq) technology has provided an unprecedented powerful tool for deconstructing tissue heterogeneity at the single-cell level, mapping cell-type-specific expression profiles, and analyzing gene regulatory networks [19,20]. There are significant differences in muscle growth rate and fat deposition among different cattle breeds, such as Angus (ANG) and Holstein (HST), which are, respectively, beef and dairy breeds. However, the cellular-level molecular mechanisms driving these differences, particularly the contribution of the KLF family, have not yet been revealed at single-cell resolution.

This study aims to systematically elucidate the regulatory mechanisms of the bovine KLF transcription factor family in the development of the longissimus dorsi (LD) and the formation of breed differences through the integration of multi-omics strategies. The study first conducted a comprehensive identification and evolutionary analysis of the bovine KLF family, elucidating its phylogenetic relationships, structural characteristics, and genomic distribution patterns; subsequently, using snRNA-seq data, it mapped the expression profiles of KLF genes in different cell types of the LD in ANG and HST cattle, revealing their expression heterogeneity and breed-specific differences. Finally, combining machine learning and SHAP analysis, the study identified key KLF genes influencing muscle phenotype and inferred their regulatory networks. This study is the first to elucidate the functional mechanisms of the bovine KLF family at the single-nuclei level, providing important theoretical foundations for understanding the molecular basis of muscle development and molecular breeding in beef cattle.

2. Materials and Methods

2.1. Identification of KLF Family

To systematically identify and characterize the KLF gene family in bovine (Bos taurus) and enable comparative evolutionary analysis, we retrieved protein sequences from four representative mammalian species: Homo sapiens, Mus musculus, Sus scrofa, and Bos taurus. All sequences were downloaded from the UniProtKB database (Release 2023_08) using taxon-specific filters and the query term “Krüppel-like factor”. Only manually reviewed entries were retained. For each gene, the longest isoform encoding a complete DNA-binding domain was selected to ensure consistency and representativeness across species.

This process identified 18 KLF members in Homo sapiens, 17 in Mus musculus, 13 in Sus scrofa, and 14 in Bos taurus. The complete list of UniProt accession IDs for all KLF orthologs analyzed in this study is provided in Table S2. To further ensure the accuracy and reliability of the bovine KLF family set, all candidate sequences were validated by confirming the presence of the characteristic triple C₂H₂ zinc finger domain using Pfam domain analysis. Chromosomal locations and gene annotations for bovine KLFs were based on the ARS-UCD1.2 genome assembly and orthology information from human KLF nomenclature. The final high-confidence, multi-species KLF protein sequence dataset was used for subsequent phylogenetic reconstruction.

2.2. Structural and Evolutionary Analysis of KLF Genes: Protein Alignment, Phylogeny, and Chromosomal Location

To reconstruct the phylogeny of KLF genes, a multiple sequence alignment of all identified orthologs from the four species was generated using the MUSCLE algorithm [21].

To conduct this study, longissimus dorsi (LD) muscle samples were harvested from 6 ANG and 6 HST cattle provided by Xuyi Animal Husbandry Co., Ltd., Bayannur City, Inner Mongolia Autonomous Region, China. All animals were healthy adult males with similar age (36 months) and body weight (550–600 kg), raised under standardized feeding conditions. Tissue samples were harvested immediately post-mortem by trained personnel using aseptic techniques. For RNA integrity preservation, samples were flash-frozen in liquid nitrogen within 5 min of collection and stored at −80 °C until further processing. All animal procedures were approved by the Animal Ethics Committee of Inner Mongolia Agricultural University (Approval No. DK20231115; Date: 15 November 2023).

2.3. Isolation of Nuclei from Bovine LD Tissue

snRNA-seq was performed by personnel from MGI Tech Co., Ltd. (Shenzhen, China) Laboratory using the DNBelab C Series Single-Cell Library Prep Kit (TaiM 4, No. [940-001818-00,16RXN]) following the manufacturer’s protocol. The procedure involved: (1) droplet-based encapsulation of single nuclei with barcoded gel beads [22], (2) reverse transcription within droplets to generate barcoded cDNA, (3) cDNA amplification and purification, and (4) library construction for sequencing. LD samples were surgically extracted and rapidly frozen in liquid nitrogen. The intact nuclei were then isolated and purified with modifications to standard protocols. Briefly, the frozen tissue was minced into approximately 5 mm × 5 mm pieces and homogenized on ice in Nuclei EZ Lysis Buffer (Sigma-Aldrich, St. Louis, MO, USA; Catalog No. NUC101) for 10 min. The homogenate was filtered through a 40-μm cell strainer, and the nuclear suspension was centrifuged at 500× g for 5 min at 4 °C [23]. The pellet was resuspended in PBS to concentrate the nuclei. Nuclear integrity and concentration were assessed under a fluorescent microscope. Finally, the nuclear concentration was adjusted to approximately 1000 nuclei per microliter for library preparation [24].

2.4. Preparation of Single-Nuclei Libraries and Sequencing

The snRNA-seq libraries were prepared with the DNBelab C4 Single-nucleus RNA Library Prep Kit (MGI Tech Co., Ltd.) following the vendor’s protocol (DNBelab C User Manual V3.0) [25]. Briefly, isolated nuclei were first encapsulated in barcoded gel beads using the DNBelab C4 instrument to form Gel Bead in-emulsion (GEM) reactions. After emulsion breaking (performed with Recovery Agent, MGI), the released beads carrying 10 bp cell-barcodes (CB1 and CB2) and a 10 bp UMI were collected and subjected to reverse transcription (RT) at 42 °C for 90 min using RT Enzyme Mix (MGI). Second-strand synthesis was followed by 14 cycles of cDNA amplification with Amplification Enzyme Mix (MGI). Amplified cDNA was purified twice with 0.6× DNBelab Clean Beads, and 200–400 bp fragments were size-selected for library construction. Indexed sequencing libraries were generated with the DNBelab C4 Library Prep Module (MGI) using 10–12 cycles of PCR and purified again with 0.8× Clean Beads. Final libraries were quantified on a Qubit 4 fluorometer with the Qubit ssDNA Assay Kit (Invitrogen, Waltham, MA, USA; Cat. No. Q10212) and verified on a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) [26]. Paired-end sequencing (read-1 = 30 bp, read-2 = 100 bp) was performed on a DNBSEQ-T7 platform at China National GeneBank (Shenzhen, China). All experimental steps were carried out according to the manufacturer’s instructions (DNBelab C User Manual V3.0; MGI, 2025).

2.5. Single-Nucleus Data Analysis

Raw sequencing data were processed using the DNBC4tools (v2.1.0) specific to the MGI DNBelab C platform. First, a custom reference genome index was built using the “DNBC4tools mkref” function with the ARS-UCD1.2 bovine genome assembly. The pipeline was then applied to demultiplex raw data, extract cellular barcodes and UMIs, and generate the single-nuclei gene expression matrix. Subsequent analysis was performed in Seurat (v4.4.0) [27] for quality control, normalization, and cell clustering. Cells were filtered based on the following thresholds: nFeature_RNA > 200, nCount_RNA < 3000, and percent.mt < 10%. Following filtering, a total of 59,563 nuclei from ANG cattle and 51,835 nuclei from HST cattle were retained for downstream analysis. Table S1 provides a summary of sequencing and quality metrics for each sample, including total reads, mean reads per nucleus, mapping rate, and median genes per nucleus. The data were normalized using the NormalizeData function in Seurat with the default normalization method (“LogNormalize”) and a scale factor of 10,000. To mitigate batch effects, we applied Harmony [28] integration using the RunHarmony function with default parameters and the batch covariate set to ‘orig.ident’. Scaling and PCA were performed using the ScaleData function (regressing out the nCount_RNA and percent.mt variables) and RunPCA function, (using the top 2000 variable features and the first 50 principal components for downstream analysis), regressing out effects of UMI count per cell and mitochondrial content. For visualization and clustering, we conducted uniform manifold approximation and projection (UMAP) based on the first 50 principal components using the RunUMAP function. Cell clusters were identified using the FindClusters function with a resolution parameter of 0.2 and the Leiden algorithm. To identify marker genes across clusters or between conditions, differential expression analysis was performed using the FindAllMarkers function in Seurat, which implements the Wilcoxon rank sum test. This function provides two key statistical outputs for each gene: a probability value (p-value) adjusted for multiple testing using the Bonferroni correction (default in Seurat) to control the family-wise error rate (FWER), and an effect size measure, reported as the area under the curve (AUC). The AUC quantifies the discriminative power of the gene, representing the probability that a randomly selected cell in the target cluster has a higher expression level than a cell in other clusters (where 0.5 indicates no discriminative ability, and 1.0 or 0.0 indicates perfect separation). Differentially expressed genes (DEGs) were defined as those meeting a Bonferroni-adjusted p-value < 0.05, an absolute log₂ fold change |log₂FC| > 0.25, a minimum percentage of expressing cells (min.pct) > 0.1, and an effect size threshold of |AUC − 0.5| > 0.1 to ensure biological relevance beyond statistical significance.

2.6. Cell Type Annotation

We first used scMayoMap [29] software to automatically annotate single-nuclei data to preliminarily identify possible cell types for each cluster. Subsequently, to further improve the accuracy of annotations, we adopted manual annotations as a supplementary strategy. Specifically, we systematically summarized marker genes for muscle cell types from published literature [30]. Next, we developed a specialized R script to compare and integrate the automatically annotated results with manually collected muscle cell marker genes. The specific operation is to obtain the intersection of the top 25 marker genes in each cluster and the muscle cell type marker genes in the literature, in order to determine more accurate candidate cell types. We normalized the average log₂FC values of each marker gene for each candidate cell type in the intersection set to individual scores, and added up the scores of all marker genes as a comprehensive score to evaluate the likelihood of each cell type. A high score means that the cluster is more likely to belong to the corresponding cell type. Finally, by integrating previous research reports, we ultimately determined the cell type.

2.7. Explainable Machine Learning

To quantify the regulatory significance of the 14 KLF genes in distinguishing cellular identities and breed-specific expression patterns, we trained an XGBoost classifier on snRNA-seq data from bovine LD. To quantify the regulatory significance of the 14 KLF genes in distinguishing cellular identities and breed-specific expression patterns, we trained an XGBoost classifier on snRNA-seq data from bovine LD. SHAP (SHapley Additive exPlanations) values were computed to quantify the effect size and direction of each gene’s contribution to the model’s prediction. The mean absolute SHAP value for each gene was used to rank its global feature importance, while the sign and magnitude of individual SHAP values were used to interpret the direction and strength of its effect in specific cellular contexts. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC = 0.705) with 95% confidence intervals derived from bootstrapping (n = 1000). The AUC metric was prioritized due to its robustness to class imbalance and focus on overall discriminatory power, aligning with the study’s goal of identifying genes with cross-condition predictive utility. Gene rankings were validated through repeated training with different random seeds, and only genes with consistent SHAP-based contributions across iterations were considered robust and reported in the final interpretation of results.

3. Results

3.1. Characterization and Evolutionary Analysis of the KLF Gene Family

A total of 62 KLF genes were identified across four species based on the presence of the conserved zf-H2C2 domain: 18 in Homo sapiens (Hs), 17 in Mus musculus (Mm), 13 in Sus scrofa (Ss), and 14 in Bos taurus (Bt); the protein sequence is shown in Table S3. To investigate the evolutionary relationships among KLF family members, an NJ phylogenetic tree was constructed using protein sequences of these 62 genes (Figure 1). Phylogenetic analysis classified the genes into three distinct groups (I–III) based on sequence homology and branching patterns. Group I (red, 12 genes) contained orthologs such as Mm-KLF3/12/8, Hs-KLF3/12/8, Ss-KLF3/12/8, and Bt-KLF3/12/8. Group II (blue, 4 genes) included Mm-KLF5, Hs-KLF5, Ss-KLF5, and Bt-KLF5. The remaining 46 genes formed Group III (purple), representing a larger divergent cluster.

3.2. Genomic Distribution of Bovine KLF Genes

The 14 identified bovine KLF genes were unevenly distributed on 11 chromosomes (Figure 2). Chromosomes 7, 8, and 12 contain two KLF genes each, while the remaining 8 chromosomes contain one gene each. KLF1 and KLF2 are located on chromosome 7, KLF4 and KLF9 are located on chromosome 8, and KLF5 and KLF12 are located on chromosome 12. A single gene is located on chromosome 2 (KLF7), chromosome 6 (KLF3), chromosome 11 (KLF11), chromosome 13 (KLF6), chromosome 14 (KLF10), chromosome 21 (KLF13), chromosome 22 (KLF15), and chromosome X (KLF8).

Table 1. KLF gene family information in Bovine (B. taurus).

Gene Name	Gene ID	Protein ID	CDS ID	Chr	Genomic Location	Exon	AA	Mass (Da)
KLF1	ENSBTAG00000006083	ENSBTAP00000043340	ENSBTAT00000046003	7	12,671,273–12,675,059	3	372	39,444
KLF2	ENSBTAG00000052504	ENSBTAP00000074931	ENSBTAT00000108559	7	6,685,278–6,687,052	2	78	8035
KLF3	ENSBTAG00000017488	ENSBTAP00000057281	ENSBTAT00000085769	6	57,957,697–57,974,562	5	346	38,943
KLF4	ENSBTAG00000020355	ENSBTAP00000072107	ENSBTAT00000077353	8	97,175,407–97,178,550	3	145	16,518
KLF5	ENSBTAG00000002129	ENSBTAP00000002751	ENSBTAT00000002751	12	47,799,775–47,817,130	4	544	50,477
KLF6	ENSBTAG00000015188	ENSBTAP00000020207	ENSBTAT00000020207	13	44,597,409–44,601,156	2	309	33,456
KLF7	ENSBTAG00000044097	ENSBTAP00000066321	ENSBTAT00000078679	2	95,426,838–95,439,630	2	300	32,220
KLF8	ENSBTAG00000005852	ENSBTAP00000067755	ENSBTAT00000071858	X	93,269,040–93,444,986	5	248	26,365
KLF9	ENSBTAG00000016229	ENSBTAP00000021593	ENSBTAT00000021593	8	46,587,347–46,611,150	2	244	27,219
KLF10	ENSBTAG00000014396	ENSBTAP00000067318	ENSBTAT00000076027	14	61,770,518–61,775,159	4	483	53,028
KLF11	ENSBTAG00000046218	ENSBTAP00000063288	ENSBTAT00000080078	11	87,549,525–87,558,666	4	523	55,735
KLF12	ENSBTAG00000044007	ENSBTAP00000053615	ENSBTAT00000061307	12	48,526,715–48,823,511	7	408	44,613
KLF13	ENSBTAG00000051090	ENSBTAP00000057849	ENSBTAT00000079732	21	27,672,129–27,709,397	2	296	31,917
KLF15	ENSBTAG00000008313	ENSBTAP00000063598	ENSBTAT00000125115	22	60,616,031–60,648,782	2	394	41,542

provides detailed information on the genome coordinates, protein germplasm (UniProt), and coding sequences (CDS) of all 14 genes.

3.3. Analysis of Conserved Motifs and Protein Domains

An integrative approach combining phylogenetic reconstruction (Figure 3A) with analyses of gene architecture and conserved motifs was employed to elucidate the evolutionary relationships among KLF genes from B. taurus, H. sapiens, M. musculus, and S. scrofa. Through hidden Markov model (HMM)-based profiling, we identified three characteristic sequence motifs within the KLF protein family. Among these, Motif 1 and Motif 2 exhibited high conservation across most members. A notable exception was observed in bovine KLF2 and KLF8 (Bt-KLF2/8), which lacked all three motifs, implying a distinct evolutionary pathway for these genes (Figure 3B). Furthermore, closely related genes within the same phylogenetic subgroups displayed consistent motif patterns, suggesting potential functional similarities. Conserved domain analysis revealed 27 distinct protein domains across the 56 KLF genes examined (Figure 3C). Genes clustering within the same phylogenetic group demonstrated shared domain architectures, highlighting the structural conservation within subfamilies.

3.4. Analysis of Evolutionary Conservation and Functional Interaction Maps

Genomic distribution analysis (Figure 4A) revealed that all 14 bovine KLF genes exhibited low gene density in both middle and outer chromosomal regions. Notably, Bt-KLF13 and Bt-KLF9 displayed a syntenic relationship (red linkage), while no other KLF pairs showed collinearity. To explore functional associations, a protein–protein interaction (PPI) network was constructed (Figure 4B), identifying 11 KLF proteins with putative interactions. These interactions are connected to 10 functional genes (e.g., CEBPB, UTF1, ESRRB), suggesting potential regulatory crosstalk.

3.5. Prediction of the Tertiary Structure and Transmembrane Region of KLF Family Proteins

Homology modeling using the SWISS-MODEL server generated tertiary structure predictions for 14 bovine KLF proteins (Figure 5). The three-dimensional structure of proteins consists of motifs and domains. The activity and function of proteins are not only determined by the primary structure of protein molecules, but also closely related to their unique spatial structure. Incorrect spatial structure in proteins can lead to decreased protein function or even inactivation, which may also result in a series of mutations.

Transmembrane helix prediction was performed for all bovine KLF proteins using the TMHMM software (https://services.healthtech.dtu.dk/service.php? TMHMM2.0, accessed on 14 August 2025). Figure 6 shows the use of TMHMM to predict the transmembrane regions of 14 KLF protein sequences in bovine. The results show that none of the bovine KLF proteins contain transmembrane helical domains, as predicted by TMHMM. This finding is consistent with the well-established role of KLF transcription factors as intracellular proteins that function within the nucleus. The absence of transmembrane domains confirms that KLF proteins are not integral membrane proteins; however, subcellular localization (e.g., nuclear versus cytoplasmic) and potential secretion pathways would require further experimental validation using tools designed to predict signal peptides (e.g., SignalP) or nuclear localization signals (NLS).

3.6. Overview of KLF in the snRNA Seq Dataset of Bovine LD

snRNA-seq was performed on LD samples from 3-year-old ANG and HST cattle to investigate the KLF gene family distribution. Following quality control, 111,398 cells were retained for downstream analysis (59,563 ANG; 51,835 HST). Unsupervised clustering identified 16 cell populations (Figure 7A), which were initially annotated into 9 cell types using the scMayoMap automated tool (Figure 7B). Manual refinement using established skeletal muscle markers (Pax7 for satellite cells, MYH1 and MYH2 for myofibers) further resolved 11 distinct cell types (Figure 7C), with corresponding marker genes validated in Figure 7D [30]. Cell type composition analysis, based on six biological replicates per breed, revealed notable but statistically underpowered trends in breed-specific differences, likely due to high inter-individual variability (Figure 7E; Table S4). The most substantial trend was a higher proportion of myofiber in ANG cattle (mean ± SD: 32.9 ± 36.1%) compared to HST cattle (17.9 ± 21.2%), representing a large absolute difference (mean diff. = 15.0%, 95% CI: −24.3 to 54.4) with a medium effect size (Cohen’s d = 0.51), although this did not reach conventional statistical significance (p = 0.40). Conversely, multiple cell types—including FAP-1 (mean diff. = −5.0%, d = −0.84), Pericyte cells (mean diff. = −0.8%, d = −0.76), and Adipocytes (mean diff. = −0.5%, d = −1.00)—exhibited trends toward greater abundance in HST cattle, as indicated by negative mean differences and medium-to-large effect sizes. The complete statistical results, including confidence intervals and effect sizes for all cell types, are provided in Table S4. CellChat analysis demonstrated differential intercellular communication patterns between breeds, with ANG cattle exhibiting enhanced FAP-1-FAP-1 interactions and HST cattle showing increased LEndoC-Adipocyte crosstalk (Figure 7F,G).

In order to analyze the cell-type-specific regulatory mechanism of the KLF transcription factor family in the bovine LD microenvironment and reveal the molecular regulatory differences between meat breed (ANG) and dairy breed (HST), we conducted an analysis of the expression of the KLF gene family in 11 cell types of ANG and HST cattle LD (Figure 8A). Overall, there was no heterogeneity in the expression of the KLF gene family in different cell types of breeds, but there were differences in expression patterns. We found that KLF6 had the highest expression level in LEndoC of two breeds, with higher expression in VEndoC of HST cattle than in ANG cattle. KLF6 was also expressed higher in FAP, Satellite cell, Macrophage, Pericyte cell, T cell, and Adipocyte of ANG cattle than in HST cattle. KLF1 and KLF2 are expressed at low levels in all cell types of both varieties. The KLF12 gene is also expressed in all cell types, and its expression is higher in Myofiber, FAP-1, VEndoC, Macrophage, Pericyte cell, LEndoC, and Adipocyte of ANG cattle than in HST cattle, while the opposite is true in other cell types. Please refer to Table S5 for detailed expression levels. Using the AddModule Score algorithm, we calculated the KLF scores for each cell type and merged 14 KLF genes into one gene set. Surprisingly, VEndoC had the highest KLF score, followed by macrophages and LEndoC, followed by T cells, while other cell types had lower scores (Figure 8B,C, Table S6).

3.7. SHAP Explains the Contribution of the KLF Gene

To elucidate the regulatory roles of KLF genes in bovine muscle development, we performed SHAP analysis on snRNA-seq data from ANG and HST cattle (Figure 9). An XGBoost classifier achieved moderate predictive performance (AUC = 0.705, 95% CI: 0.68–0.73) for breed classification (Figure 9A). Global feature importance analysis revealed KLF9, KLF10, and KLF12 as the top three contributors (mean|SHAP| = 0.16, 0.14, and 0.12, respectively), suggesting these genes drive transcriptional differences between breeds (Figure 9B, Table S7).

Heatmap visualization demonstrated breed-specific contribution patterns: KLF genes showed positive SHAP values (HST-enriched) versus negative values (ANG-enriched) across cell types (Figure 9C). Notably, KLF6 exhibited the highest cell-type-specific contribution in LEndoC (mean SHAP = 0.231), exceeding other cell types by 2.3-fold (Figure 9D). Elevated KLF6 contributions were also observed in macrophage (0.144), adipocyte (0.132), and VEndoC (0.118), correlating with its expression patterns identified in prior analyses.

4. Discussion

Over the past few decades, extensive research has been conducted on the biological functions of the KLF gene family in mammals, including roles in adipogenesis, myogenesis, tumorigenesis, and the regulation of cellular and tissue metabolism [31,32,33,34,35]. However, systematic research on the expression patterns and regulatory functions of the Bos taurus family in the complex cellular ecosystem of the muscle tissue microenvironment, an important agricultural species, is still relatively scarce. This study comprehensively identified the bovine KLF gene family for the first time by integrating bioinformatics, comparative genomics, and single-nuclei transcriptomics. The expression profiles of the KLF gene family in the longest dorsal muscle of ANG and HST cattle were plotted at single-nuclei resolution, revealing its expression heterogeneity, breed-specific regulatory patterns, and potential functions. This provides new insights into the role of the KLF family in cattle muscle development and meat quality trait formation.

The KLF family is a group of highly conserved zinc finger transcription factors that bind GC-rich DNA sequences and play pivotal roles in proliferation, differentiation, and metabolic regulation [36]. It is important to clarify that the conserved motifs identified de novo by MEME analysis in this study are distinct from the canonical C₂H₂ zinc finger DNA binding domain that defines the KLF family. The triple C₂H₂ domain is a well-characterized, conserved structural unit responsible for GC-rich DNA binding, and its presence was a prerequisite for classifying a protein as a KLF member (validated by our Pfam analysis). In contrast, the MEME-predicted motifs represent additional conserved sequence patterns outside the DNA-binding domain. These motifs may underlie novel functional divergences, such as differential protein–protein interaction or subcellular localization signals that contribute to the functional specialization of KLF members across evolutionary lineages and cell types. Our integrated single-nuclei transcriptomic and machine learning analysis revealed the cell-type-specific regulatory landscape of KLF genes in bovine muscle, with KLF6 emerging as a key player in endothelial and adipocyte populations. The high abundance of KLF6 in adipocytes, particularly in the meat breed ANG, aligns with its well-established role as a positive regulator of adipocyte differentiation [37,38]. This suggests that its elevated expression may contribute to the enhanced intramuscular fat deposition (marbling) characteristic of Angus cattle. Furthermore, its significant expression in VEndoC and LEndoC resonates with its documented function in vascular remodeling [38], implying a potential role in modulating blood vessel and lymphatic network formation within the muscle microenvironment. The uniquely high SHAP value of KLF6 specifically in LEndoC cells, far exceeding that in other types, strongly suggests a previously underappreciated and critical function in bovine lymphatic biology, potentially mediating immune cell trafficking or metabolic waste clearance in a breed-specific manner. On the contrary, KLF12 is upregulated in multiple cell types such as myofiber and FAP-1 in ANG cattle. This cell-type-specific expression pattern strongly suggests that KLF members synergistically regulate processes such as muscle tissue development, immune microenvironment, angiogenesis, and fat deposition by exercising their functions in different cell populations.

An important aspect of this study is the integration of differential expression analysis with machine learning (XGBoost) to quantify the relative contributions of KLF genes in driving transcriptomic differences between breeds. SHAP analysis provides objective and quantitative evidence for identifying key regulatory factors. The global feature importance analysis consistently identified KLF9, KLF10, and KLF12 as the main contributors to distinguishing ANG and HST cattle varieties, indicating that they play a central role in shaping variety-specific transcription programs. More importantly, the decomposition of SHAP values at the cell type level reveals an unprecedented fine regulatory landscape: KLF6 exhibits highly specific contributions in lymphatic endothelial cells, with SHAP values far exceeding those of other cell types. This finding is consistent with the high expression level of KLF6 in this cell type, strongly suggesting that KLF6 may have a unique and critical regulatory function in bovine lymphatic endothelial cells, and may be a key molecular switch mediating immune microenvironment or lymphatic vessel function differences between breeds. Meanwhile, the significant contribution of KLF6 in macrophages and adipocytes suggests that it may jointly affect the final meat quality traits by regulating the interactions of multiple cell types.

One of the key findings of this study is the revelation of the variety specificity of KLF gene expression and regulation. Cell population composition analysis showed that the proportion of myofiber in the meat breed ANG was significantly higher than that in the dairy breed HST, while HST had a richer matrix–cell population (such as FAPs). This difference in cell composition is likely the cellular basis for the differences in meat production performance and meat quality between the two varieties. SHAP analysis further supports this conclusion from the perspective of functional contribution, revealing the differential driving effects of different KLF genes on variety characteristics under different cellular backgrounds. These results indicate that the KLF family is not equally involved in regulation but rather dominated by a few key members in specific cell types, leading to genetic differences between varieties. This provides a potential molecular mechanism for explaining the phenotypic differences between ANG and HST in muscle growth and fat deposition.

Based on protein interaction network prediction, bovine KLF protein interacts with key transcription factors known to regulate muscle and fat development, such as CEBPB and ESRRB. Recently, in-depth research on skeletal muscle RNA-seq data has shown that CEBPB can serve as a biomarker related to fat generation in muscle aging [39]. ESRRB is also known to be an important TF for myogenic development and stem cell function, and studies have shown that ESRRB promotes the transformation of fibroblasts into induced myogenic stem cells [40]. This suggests that bovine KLF may form complexes with such core regulatory factors and integrate into the complex transcriptional network that regulates muscle development and metabolism, thereby affecting the final economic traits.

The conclusion of this study is mainly based on bioinformatics prediction and correlation analysis, and its specific molecular mechanism still needs to be verified through experiments. For example, the potential functional changes caused by motif deletion in KLF2/8, as well as the key regulatory functions suggested by the high SHAP value of KLF6 in lymphatic endothelial cells, need to be further explored through functional experiments such as ChIP seq and CRISPR-Cas9 gene editing in vitro and in vivo. Furthermore, we note that our computational analysis did not include dedicated doublet detection or ambient RNA mitigation. While standard QC metrics (e.g., UMI/gene counts, mitochondrial percentage) were applied, future studies employing specialized tools (e.g., DoubletFinder, Scrublet, or DecontX) would help further refine the dataset and confirm the robustness of the findings, particularly in rare cell populations. In addition, this study focuses on a specific time point in adult cattle. If future research can cover multiple key developmental stages, it will be able to more dynamically reveal the regulatory role of the KLF family in the entire process of muscle development.

5. Conclusions

In summary, this study presents the first systematic single-nuclei atlas of the bovine KLF gene family within the skeletal muscle microenvironment. Our multi-omics approach, integrating comparative genomics and snRNA-seq data from over 110,000 high-quality nuclei, revealed cell-type-specific expression patterns and significant differential expression of KLF genes between meat (ANG) and dairy (HST) cattle breeds. Through explainable machine learning (XGBoost, AUC = 0.705), we quantitatively identified KLF6, KLF9, KLF10, and KLF12 as top contributors to transcriptional differences between breeds, with KLF6 demonstrating particular cell-type-specific regulatory potential in lymphatic endothelial cells. These statistically robust findings provide unprecedented insights into KLF functional diversity in bovine muscle biology and deliver high-confidence candidate targets for precision breeding. While this study establishes strong transcriptional associations, future research employing functional validation via CRISPR-Cas9 screening in bovine cellular models will be essential to confirm causal relationships and advance toward practical breeding applications.