High KYNU Expression Is Associated with Poor Prognosis, KEAP1/STK11 Mutations, and Immunosuppressive Metabolism in Patient-Derived but Not Murine Lung Adenocarcinomas

Simple Summary

Lung adenocarcinoma is a common and deadly form of lung cancer, and better tools are needed to predict how aggressive the disease will be and which treatments may work best. In this study, we focused on a metabolic enzyme called kynureninase (KYNU), which showed two distinct patterns of activity across lung tumors. We found that high KYNU activity often signals a poor prognosis and is linked to specific genetic changes in cancer cells. Interestingly, KYNU appears to affect how tumors interact with the immune system and how they process certain nutrients. Our findings also suggest that current mouse models may not accurately reflect how KYNU works in human lung tumors, which is important for designing better therapies. By understanding KYNU’s role, we hope to guide future research in cancer metabolism and improve treatment strategies.

Abstract

Background/Objectives: We aimed to discover genes with bimodal expression linked to patient outcomes, to reveal underlying oncogenotypes and identify new therapeutic insights in lung adenocarcinoma (LUAD). Methods: We performed meta-analysis to screen LUAD datasets for prognostic genes with bimodal expression patterns. Kynureninase (KYNU), a key enzyme in tryptophan catabolism, emerged as a top candidate. We then examined its relationship with LUAD mutations, metabolic alterations, immune microenvironment states, and expression patterns in human and mouse models using bulk and single-cell transcriptomics, metabolomics, and preclinical model datasets. Pan-cancer prognostic associations were also assessed. Results: Model-based clustering of KYNU expression outperformed median-based dichotomization in prognostic accuracy. KYNU was elevated in tumors with KEAP1 and STK11 co-mutations but remained a strong independent prognostic marker. Metabolomic analysis showed that KYNU-high tumors had increased anthranilic acid, a catalytic product, while maintaining stable kynurenine levels, suggesting a compensatory mechanism sustaining immunosuppressive signaling. Single-cell and bulk data showed KYNU expression was cancer cell-intrinsic in immune-cold tumors and myeloid-derived in immune-infiltrated tumors. In murine LUAD models, Kynu expression was predominantly immune-derived and uncoupled from Nrf2/Lkb1 signaling, indicating poor model fidelity. KYNU’s prognostic associations extended across cancer types, with poor outcomes in pancreatic and kidney cancers but favorable outcomes in melanoma, underscoring the need for lineage-specific considerations in therapy development. Conclusions: KYNU is a robust prognostic biomarker and potential immunometabolic target in LUAD, especially in STK11 and KEAP1 co-mutated tumors. Its cancer cell-intrinsic expression and immunosuppressive metabolic phenotype offer translational potential, though species-specific expression patterns pose challenges for preclinical modeling.

1. Introduction

Lung cancer remains the leading cause of cancer-related deaths worldwide, with lung adenocarcinoma (LUAD) being the most common subtype, accounting for 40% of cases [1]. Advances in early diagnosis, staging, and multimodal treatments—including surgery, radiotherapy, and chemotherapy—have expanded LUAD treatment options. The identification of oncogenic drivers, development of targeted therapies, and integration of immune checkpoint blockade (ICB) and antibody-based therapies have further improved survival in many LUAD molecular subsets. However, significant challenges remain, including the need for robust biomarkers to guide the selection of optimal systemic therapies for individual patients, improving the long-term durable responses to both ICB and targeted therapies, preventing metastases, and overcoming resistance, which frequently occurs in all classes of systemic therapy [1]. In lung cancer clinical translational research, many therapeutic advances have been driven by the identification of a potential biomarker (e.g., mutated EGFR and KRAS) and their validation in patient-derived preclinical models (e.g., cell lines, xenografts) and mouse models (e.g., genetically engineered mouse models, GEMMs). These models have been instrumental in demonstrating the functional significance of biomarkers by showing that targeting them produces an anti-tumor effect. Hence, assessing the functional significance of biomarkers and their compatibility with traditional syngeneic and patient-derived (xenograft) mouse models is essential for facilitating clinical translatability.

Over the past decade, large-scale gene expression studies have advanced biomarker discovery by linking oncogenotypes to gene expression programs and patient outcomes. Just as LUADs can have “bimodal” oncogenotypes (e.g., presence or absence of TP53, KRAS, or EGFR mutations) that influence their pathogenesis, pathophysiology, and response to treatment, genes exhibiting dramatic bimodal expression patterns (very high or very low) may also play a key role in LUAD biology and therapeutic response [2]. Leveraging our Lung Cancer Explorer (LCE, http://lce.biohpc.swmed.edu/) [3], an integrated resource of expression and clinical data from >6700 patients across 56 studies, we applied Gaussian mixture modeling to identify genes exhibiting bimodal expression patterns [4]. These distributions suggest potential oncogenotype-driven gene reprogramming, revealing genes that may serve as strong prognostic markers and have functional significance.

Here, we identify KYNU expression, which encodes a metabolic enzyme in the kynurenine pathway, as a bimodally distributed prognostic biomarker in LUAD. Through a meta-analysis of 23 LUAD datasets, we show that bimodal stratification of KYNU expression more accurately captures prognosis compared with median-based dichotomization. High KYNU expression correlates strongly with KEAP1/STK11 co-mutations in LUADs, enhancing the precision of prognostic models beyond mutation status alone. Our metabolomics analysis of LUAD cell lines revealed a distinct KYNU-associated metabolic profile, including elevated anthranilic acid, increased NADP, and depletion of niacinamide. Furthermore, we identify a translational gap in murine preclinical models: KYNU’s biology cannot be effectively modeled in traditional genetically engineered mouse systems, limiting its preclinical functional study. Finally, extending our computational analysis to other cancers, we find high KYNU expression is also associated with worse outcomes in many cancer types, including pancreatic cancer, but is associated with better prognosis in melanoma. These findings highlight the importance of cancer lineage-specific considerations when developing therapeutic strategies targeting the kynurenine pathway.

2. Materials and Methods

2.1. Determination of Bimodal Distribution

We implemented Gaussian mixture model clustering with two pre-defined clusters and equal variances using the “Mclust” function from the R package mclust [4]. The bimodal index was computed using the “bimodalIndex” function from the R package BimodalIndex, which fits data to a two-component mixture model with equal variance.

2.2. Meta-Analysis of Lung Adenocarcinoma Studies for Survival Association with Gene Expression

The collection and reprocessing of data from lung cancer studies with both gene expression and clinical data was described previously [3]. Among 56 available datasets, we selected 23 datasets that included both overall survival data and tumor gene expression data for LUAD patients. Each dataset contained at least 10 patients, totaling 3114 cases. For each gene, we performed survival association analysis using two approaches: (1) Bimodal clustering-based dichotomization, comparing survival between high and low expression groups, and (2) Continuous gene expression analysis, using sample-wise z-score standardized values. We used a univariate Cox proportional hazard regression model to assess the association between gene expression and overall survival, estimating hazard ratios and p-values for each dataset. A random effects meta-analysis was then performed to estimate summary hazard ratios and p-values across datasets.

2.3. Processing of Gene Expression Data

Transcript per million (TPM) RNA-seq data for CCLE [5] were downloaded from the DepMap portal under version 19Q1 (https://depmap.org/portal/download/). Only cell lines with “tcga_code” annotated as “LUAD” from the CCLE annotation table were included for analyses. Gene-normalized RNA-seq data from TCGA LUAD, processed by RSEM (RNA-Seq by Expectation-Maximization) algorithm, were downloaded from Firebrowse (http://firebrowse.org/) (doi:10.7908/C19P30S6). FPKM RNA-seq data for CPTAC were downloaded from dbGaP with accession number phs001287 and retrieval date 3 June 2020. Data preprocessing included quantile normalization and/or log transformation, applied as appropriate.

2.4. Processing of Mutation Data

Mutation data for CCLE were downloaded from the DepMap portal as “CCLE_DepMap_18q3_maf_20180718.txt”. Mutation data for TCGA were retrieved from Firebrowse on 14 September 2017 by running the “Analyses.Mutation.MAF” function from the R package “FirebrowseR” [6]. For both CCLE and TCGA, mutations with variant annotations such as “Silent”, “Intron”, or “IGR” were removed. Mutation data for CPTAC were downloaded from dbGaP with accession number phs001287 with a retrieval date of 3 June 2020. Mutations annotated as “LOW” in the “IMPACT” field or have a non-empty “GDC_FILTER” were removed. For downstream analyses, we used the union of mutations detected by at least one of the following variant callers: MuTect, MuSE, PinDel, Sniper, or VarScan.

2.5. Screen for Oncogenotype Associated with High KYNU Expression

The Cancer Gene Census (CGC) catalog of genes implicated in cancer was retrieved from the COSMIC website (https://cancer.sanger.ac.uk/census) on 2 September 2019. In each LUAD mutation dataset (CCLE, TCGA, and CPTAC), we analyzed only CGC-listed genes with mutations detected in at least 9 independent samples. A Mann–Whitney U test was used to compare KYNU expression between mutants and wild-type samples for each gene. We applied Benjamini–Hochberg procedures for multiple comparisons but focused on nominal p-values, given the relatively small sample size.

2.6. Determination of KEAP1/STK11 Status in Different Datasets

In addition to mutation data, we incorporated additional molecular datasets to assess the functional status of STK11 and KEAP1 across our study cohorts. For CCLE, gene fusion, translocation, and copy number data were downloaded as “CCLE_Fusions_20181130.txt”, “CCLE_translocations_SvABA_20181221.xlsx”, and “public_19Q1_gene_cn.csv” files, respectively, from the DepMap portal. Translocation or fusion events involving STK11 or KEAP1 at the breakpoint were classified as “loss” for the tumor suppressor gene status. Additionally, copy number values smaller than −1 were also indicative of loss. Since LKB1 protein levels showed a bimodal distribution in reverse phase protein array (RPPA) data, we performed model-based clustering to classify the cell lines. Those with LKB1 RPPA levels below the model-determined threshold were annotated as LKB1 loss. For TCGA, “TCGA_genomic_alterations.tsv” was downloaded from cBioPortal [7]. We classified mutations as loss events if they were not annotated as “AMP” (amplification), “not profiled”, or “no alteration”. For CPTAC, we relied solely on mutation data to define loss status without incorporating additional molecular data.

2.7. Pathway Analyses for KEAP1/STK11 Associated Gene Expression

Gene set enrichment analysis (GSEA) [8] was used for three sets of pathway analyses comparing genes differentially expressed in samples with or without mutations in STK11, KEAP1, or both. The GSEA R script was adapted from the original script downloaded from MSigDB [8] (https://www.gsea-msigdb.org/gsea/downloads.jsp). The signal-to-noise ratio was used as the gene ranking metric. Gene ontology terms of biological processes (also from MSigDB) were used as the input geneset library. Results from CCLE, TCGA, and CPTAC datasets were reviewed. To identify genes consistently upregulated in both KEAP1/STK11 co-mutants and samples with mutations in KEAP1-only or STK11-only, we first identified gene sets with nominal p-values below 0.05 and positive normalized enrichment scores (NES) across all three datasets. From each of the “GO_POLYKETIDE_METABOLIC_PROCESS” (shared between “KEAP1mutSTK11mut-KEAP1wtSTK11wt” and “KEAP1mutSTK11wt-KEAP1wtSTK11wt” results) and “GO_CAMP_CATABOLIC_PROCESS” (shared between “KEAP1mutSTK11mut-KEAP1wtSTK11wt” and “KEAP1wtSTK11mut-KEAP1wtSTK11wt” results) genesets, genes appear in multiple leading-edge lists were selected for heatmap visualization. Similarly, for genes downregulated in KEAP1/STK11 co-mutants in TCGA and CPTAC (but not CCLE), the most frequently appearing genes from the leading-edge list of geneset “GO_MHC_PROTEIN_COMPLEX_ASSEMBLY” were used in the heatmap. The most frequent macrophage-related genes appeared in the leading-edge lists from two gene sets: “GO_MACROPHAGE_COLONY_STIMULATING_FACTOR_SIGNALING_PATHWAY” and “GO_RESPONSE_TO_MACROPHAGE_COLONY_STIMULATING_FACTOR”. Enrichment plots were generated using modified functions from the R package “fgsea” [9] to overlay results across datasets. The p-values were empirically estimated using permutated sample labels [8].

2.8. Tumor Immune Infiltrate Association Analyses

Estimates of TCGA tumor immune infiltrates, generated using different computational algorithms by Li et al. [10], were downloaded from the supplementary table available on the publisher’s website. TCGA LUAD tumor samples were classified into KYNU-high and KYNU-low groups, and the association between KYNU expression and immune infiltrate estimates was assessed by Pearson correlation within each group. For each algorithm, we examined gene signatures from the original publication or software repository to exclude signatures containing KYNU. We identified that KYNU was part of the monocyte lineage signature in MCPcounter and, therefore, excluded MCPcounter scores from our analyses [11].

2.9. Pathway Analysis for KYNU-Associated Genes

In TCGA and CPTAC datasets, we assessed the correlation between KYNU expression and the expression levels of all other genes in the genome using Pearson correlation. This analysis was performed for all samples in each dataset, a subgroup of samples without KEAP or STK11 mutations and the remaining group of samples with KEAP1 and/or STK11 mutations. Multiple comparison adjusted p-values were calculated by Benjamini–Hochberg procedures. Genes with adjusted p-values less than 0.05 in both the TCGA and CPTAC datasets were selected for pathway enrichment analysis. A hypergeometric test was applied to identify enriched canonical pathways using gene sets from MSigDB [8]. Multiple comparison-adjusted p-values were further generated for the pathway p-values.

2.10. Analysis of scRNA-Seq Data from Healthy Human Lung

The Travaglini_2020 dataset consists of single-cell RNA sequencing (scRNA-seq) data from healthy human lung tissue [12]. Processed count data and cell type annotations provided by the authors were downloaded from Synapse (Accession ID: syn21041850). For this study, FACS-sorted SmartSeq2 data were used. Cell types with fewer than 10 cells were excluded from analyses. Library size normalization was performed using the “library.size.normalize” function from the R package “phateR” [13]. Log₂-transformed expression data were used for visualization. Cell types in the figure were ordered based on their average KYNU expression.

2.11. Molecular Features Associated with KEAP1/STK11 Oncogenotypes or KYNU Expression

Metabolomics data from CCLE [5] were downloaded from the DepMap portal as “CCLE_metabolomics_20190502.csv”. CPTAC proteomics data were extracted from a supplementary table of the original CPTAC LUAD paper [14]. To assess molecular associations, we applied Pearson correlation or one-way ANOVA, depending on the data type and distribution.

2.12. Pan-Cancer Survival Analyses

TCGA pan-cancer survival data were downloaded from the supplementary table “NIHMS978596-supplement-1.xlsx” in Liu et al. [15]. TCGA pan-cancer RNA-seq data, reprocessed by Toil pipeline [16], was downloaded as “tcga_RSEM_Hugo_norm_count” from Xena (https://tcga.xenahubs.net). In addition to cancer types, sample types (primary tumor, blood, metastasis) were taken into consideration for patient stratification.

2.13. Other R Packages Used for Analyses

All analyses were conducted in R (version 4.2.2) [17]. The following R packages were used: data wrangling: openxlsx [18], data.table [19], dplyr [20], plyr [21], tidyverse [22], reshape2 [23], statistical analyses: stats [17], survival [24], meta [25], survminer [26] graph visualization: ggplot2 [27], GGally [28], ggridges [29], ggrepel [30], grid [17], patchwork [31], cowplot [32], highcharter [33], ComplexHeatmap [34], viridis [35], DescTools [36], RColorBrewer [37], scales [38] table visualization: finalfit [39], kableExtra [40], webshot [41], and formattable [42].

3. Results

3.1. KYNU mRNA Expression Is Bimodally Distributed in Lung Adenocarcinoma, Associated with Protein Expression, and Its High Expression Is Associated with Poor Prognosis

We analyzed gene expression from 3114 patients across 23 lung adenocarcinoma (LUAD) datasets and identified KYNU as a top bimodally distributed gene with high expression associated with poor prognosis (Figure 1a). Other bimodally expressed genes identified in our analyses, such as FOXM1 [43,44], MELK [45], RGS20 [46,47,48], and OIP5 [49], have been previously reported to associate with clinical outcomes and play functionally significant roles in lung adenocarcinoma. Notably, while many top prognostic genes exhibit strong coexpression patterns, KYNU and RGS20 expression remain largely independent, showing only weak correlations with expression of the other bimodally expressed genes. This suggests that they may contribute to LUAD pathophysiology through distinct, non-overlapping mechanisms (Figure S1). KYNU encodes kynureninase, an enzyme that links the tryptophan catabolism pathway to de novo NAD synthesis. Its expression in LUAD exhibits a distinct bimodal distribution (Figure 1b), and survival analysis revealed that model-based clustering dichotomization by Gaussian mixture models outperforms median-based approaches in capturing KYNU’s prognostic significance (Figure 1c,d; Table S1). Importantly, using DNA molecular annotation data from 74 LUAD cell lines, we observed no correlation with KYNU copy number or mutations with KYNU mRNA expression levels (Figure 1e). Importantly, using proteomic data from CPTAC LUAD [14], we confirmed that KYNU’s bimodal expression at the mRNA level is highly concordant (r = 0.88) with its protein-level expression (Figure 1f).

Kynurenine, a metabolite in the tryptophan catabolism pathway, has been well-characterized as an immunosuppressive molecule [50,51], acting as an endogenous ligand for the aryl hydrocarbon receptor (AhR) [52]. AhR activation promotes a tolerogenic immune environment through multiple mechanisms [53], including driving T cell differentiation into regulatory T cells [54]. Given that KYNU enzymatically degrades kynurenine, its association with poor prognosis was unexpected, as it would seemingly counteract kynurenine’s immunosuppressive effects. However, the activation of the kynurenine pathway in cancer cells may lead to local tryptophan depletion, restricting its availability to immune cells and thereby suppressing anti-tumor immunity [55]. Additionally, increased kynureninase activity may support tumor growth by enhancing de novo synthesis of nicotinamide adenine dinucleotide (NAD), a critical cofactor in bioenergetics and redox homeostasis (Figure 1g). The complexities of the kynurenine pathway’s role in cancer pathogenesis and anti-tumor immunity are well recognized and have been extensively reviewed by Gouasmi et al. [56].

To contextualize KYNU within related pathways, we examined the following gene sets: “Kynurenine Metabolites Suppress T-Cells in Cancer Immune Escape” from Elsevier Pathway Collection [57], “Tryptophan Catabolism” from Reactome [58], and “Reactive Oxygen Species Pathway” from the MSigDB Hallmark database [8]. Among these pathways, KYNU expression stood out because of its consistent bimodal distribution across studies and its stronger association with poor prognosis compared with the expression of other pathway members (Figure S2). This suggests that KYNU expression is regulated in a distinct manner, potentially linking it to tumor aggressiveness.

3.2. KYNU Expression Is Upregulated in KEAP1/STK11 LUAD Co-Mutants

To identify genetic determinants of KYNU upregulation, we analyzed datasets that report both gene expression and mutation data, including CCLE [5], TCGA [59], and CPTAC [14]. Across all three datasets, KEAP1 and STK11 mutations were consistently associated with elevated KYNU expression (Table S2).

We further examined structural variations and assessed the impact of KRAS, KEAP1, and STK11 alterations on KYNU expression. KYNU expression was particularly elevated in cells harboring both KEAP1 and STK11 mutations, a combination that was well represented across the datasets (Figure 2a). Although KRAS frequently co-mutates with KEAP1 and STK11 [60], we did not observe significant KYNU upregulation by KRAS mutation status alone in two of the three datasets. Notably, KYNU expression was more pronounced in patient metastasis-derived cell lines with STK11 and KEAP1 mutations (Figure S3a). The higher frequency of KEAP1/STK11 co-mutations in patient-derived cell lines compared with tumor datasets (Figure 2b; Table S2b) suggests a potential selective advantage for in vitro establishment, possibly because of enhanced glutamine utilization and reactive oxidative stress resistance [61]. This may also be explained by kynurenine pathway activation, which has been implicated in promoting anchorage-independent survival, potentially supporting cell viability during in vitro replating [62].

This trend was particularly evident in a panel of isogenic HCC515-derived cell lines generated from a prior study [63]. Cells with single KEAP1 or STK11 mutations exhibited modest KYNU upregulation compared with wild-type cells, while KEAP1 and STK11 co-mutant cells displayed the highest KYNU expression. This pattern highlights a clear additive effect of the two mutations on KYNU regulation (Figure 2c, left). To assess whether this effect was unique to KYNU, we compared its expression with canonical NRF2 and LKB1 pathway targets: NQO1 (regulated by NRF2) and PDE4D (regulated by LKB1). Unlike KYNU, NQO1 and PDE4D showed unidirectional regulation: NQO1 was elevated only in KEAP1 mutants, while PDE4D was increased only in STK11 mutants (Figure 2c, middle and right). The absence of a stepwise pattern for these genes underscores the distinct regulatory mechanism of KYNU, which integrates inputs from both pathways.

Beyond comparing KYNU expression by driver mutation status, we further assessed its expression relative to other NRF2 targets (G6PD, NQO1) and LKB1 targets (CPS1, PDE4D) across datasets. While G6PD and NQO1 exhibited a strong correlation, along with CPS1 and PDE4D, KYNU displayed only a modest correlation with these pathway genes (Figure S4a). Although G6PD and CPS1 also demonstrated bimodal expression patterns, survival analysis showed that their prognostic significance, when assessed using model-based clustering, was less pronounced than that of KYNU (Figure S4b–e compared with Figure 1d). These findings collectively highlight KYNU as a shared downstream target of NRF2 and LKB1 signaling that stands out for its selective upregulation in KEAP1/STK11 co-mutants, which may contribute to its strong prognostic value.

3.3. KYNU Expression Provides Prognostic Value Independent of KEAP1/STK11 Co-Mutations in LUAD

Previous studies have shown that co-mutations of KRAS/STK11 or KRAS/KEAP1 are associated with significantly shorter survival in NSCLC patients [64], with KEAP1/STK11 co-mutations conferring an even worse outcome [65]. Using survival and mutation data from 1258 LUAD patients in the MSK-IMPACT dataset [66], we found that patients with co-mutations in KEAP1/STK11 had markedly shorter survival compared with those with mutations in only one of these genes, regardless of KRAS status (Figure 3a–c). Patients with single or co-occurring STK11 and KEAP1 mutations had similar distributions in sex and primary versus metastatic tumor status compared with the mutation-negative group. However, never-smokers were markedly underrepresented among STK11 and/or KEAP1-mutant patients (0.1–0.8%), whereas they constituted 23% of the mutation-negative cases (Figure 3d). To further investigate the frequency and relationship of KEAP1 and STK11 mutations in LUAD, we analyzed AACR GENIE data from 5709 LUAD patients. KEAP1/STK11 co-mutations were present in 11% of cases, with an approximately equal distribution between KRAS-mutant and KRAS-wild-type patients (Figure 3e). Interestingly, KEAP1/STK11 co-mutations occurred more frequently than KRAS/KEAP1 or KRAS/STK11 co-mutations, suggesting a functional synergy between NRF2- and LKB1-regulated pathways in tumor progression. To further explore the oncogenic landscape of tumors harboring KEAP1/STK11 co-mutations without concurrent KRAS mutations, we evaluated alterations in key growth signaling pathways driven by RTKs, RAS/MAPK, or PI3K/AKT/mTOR pathway genes (Figure 3f). Among the 483 tumors with KEAP1/STK11 co-mutations and wild-type KRAS, no known alterations in these pathways were detected in 250 tumors (>50%), indicating that KEAP1/STK11 co-mutations may be sufficient to drive oncogenesis in a large subset of patients. This supports a distinct oncogenic mechanism in this subgroup, potentially involving metabolic reprogramming and stress response deregulation, independent of canonical RTK pathway activation.

Interestingly, in the TCGA LUAD cohort, we found that high KYNU expression predicts poor survival, regardless of KEAP1/STK11 co-mutation status (Figure 4a,b). Specifically, among the 29 patients with KEAP1/STK11 co-mutations, those with low KYNU expression (n = 15) exhibited much longer median survival—comparable to patients without KEAP1/STK11 mutations. Conversely, 33 out of 354 patients without KEAP1/STK11 mutations displaying high KYNU expression experienced shorter median survival (Figure 4b). To further validate the prognostic significance of KYNU expression, we performed a multivariate analysis controlling for KEAP1/STK11 status, smoking status, age, gender, and TNM stage. KYNU expression remained significantly associated with worse overall survival in the TCGA LUAD cohort (Figure 4c). Validation with an independent dataset—the CPTAC LUAD cohort—using univariate analyses confirmed that high KYNU expression is consistently associated with poor survival outcomes, regardless of KEAP1/STK11 co-mutation status (Figure 4d,e). These findings suggest that while KYNU expression is co-regulated by LKB1 loss and KEAP1 mutation-mediated NRF2 activation, it provides additional prognostic value beyond the mutation status of STK11 and KEAP1.

3.4. Tumor-Intrinsic and Microenvironmental Sources of KYNU Expression Underlying Its Bimodal Distribution in LUAD

Given previous reports linking KEAP1/STK11 co-mutations to an immune cold phenotype in LUAD [67], we sought to explore the relationship between KYNU expression and tumor immune infiltration. As a first step, we examined the expression of PTPRC (encoding CD45), a pan-leukocyte marker, to evaluate whether total immune cell presence varied with KYNU expression. A biphasic relationship was observed between PTPRC and KYNU expression across multiple LUAD datasets. Specifically, a positive correlation was evident only in KYNU-low tumors, suggesting distinct immune microenvironments between KYNU-low versus KYNU-high tumors (Figure 5a). In TCGA LUAD KYNU-low tumors, immune infiltrate score estimates [10] revealed the strongest correlations between KYNU expression and macrophage scores (Figure 5b). Additional positive correlations were observed with neutrophils, myeloid dendritic cells, and plasmacytoid dendritic cells, suggesting that KYNU expression in KYNU-low tumors may originate from myeloid cells. This pattern was further corroborated by our re-analysis of scRNA-seq data from healthy human lung tissue [12], which confirmed high KYNU expression in multiple myeloid cell subsets, particularly macrophages and plasmacytoid dendritic cells (Figure 5c). In contrast, KYNU expression was minimal in epithelial cell types, except for goblet cells. Notably, goblet cells also exhibited high levels of KMO, IDO1, and IDO2 expression, suggesting higher kynurenine pathway activity. We extended this analysis to two independent scRNA-seq LUAD datasets [68,69], which showed that KYNU expression was primarily restricted to myeloid lineage cells (Figure 5d,e).

In contrast, KYNU-high tumors exhibited an inverse relationship between KYNU expression and immune infiltrate estimates in TCGA LUAD (Figure S5a), suggesting reduced immune infiltration. To further investigate the molecular landscape, we performed gene set enrichment analysis (GSEA) comparing KEAP1/STK11 co-mutants (double mutants), single mutant, and wild-type LUAD samples. The results revealed that rather than forming a completely distinct signature, double mutants exhibit a convergence of metabolic gene programs observed in single mutants, particularly those involved in tumor-intrinsic processes such as cAMP catabolism and polyketide metabolism (Figure S5b). Notably, MHC II-related antigen presentation pathways, which are associated with the tumor microenvironment (TME), were significantly downregulated in double mutants (Figure S5b,c). Furthermore, macrophage-related gene expression was uniformly low in both single and double mutants, supporting the conclusion that KYNU expression in KYNU-high tumors originates from cancer cells rather than infiltrating immune cells (Figure S5c, black box).

Together, these findings demonstrate that KYNU expression represents two distinct biological contexts in LUAD. In KYNU-low tumors, expression is primarily driven by myeloid cell infiltration, whereas in KYNU-high tumors, it predominantly reflects expression by cancer cells. This duality underscores the potential of KYNU as a biomarker for stratifying immune microenvironments in LUAD and guiding tailored therapeutic strategies. Given the observed dichotomy in KYNU expression contexts, we next explored whether these distinct transcriptional profiles are associated with underlying metabolic differences.

3.5. KYNU Metabolomics Association in Patient-Derived LUAD Lines Identifies a Compensatory Metabolic Mechanism That Provides a Basis for a LUAD Immune Suppressive Microenvironment

With the discovery that high KYNU expression correlates with KEAP1/STK11 mutations, we recognized a paradox. On the one hand, kynurenine is a well-known immunosuppressive metabolite that should be degraded by high KYNU expression [54]. On the other hand, STK11 and KEAP1 mutations are strongly associated with an immunosuppressive tumor microenvironment [60,67]. Consistent with this paradox, our analysis revealed an inverse correlation between KYNU expression and immune infiltrates in KYNU-high tumor samples (Figure 6a). To explore this further, we examined CCLE metabolomics data [70] from 72 LUAD cell lines to identify metabolic profiles associated with KYNU upregulation (Table S3). We identified several metabolites correlated with KYNU expression as a result of KEAP1 or STK11 mutations, including reduced glutathione and 1-methylnicotinamide (Figure S6). The increase in glutathione is likely driven by KEAP1 loss (Figure S6a,b), as NRF2 activation induces the expression of genes encoding the catalytic (GCLC) and modifier (GCLM) subunits of glutamate-cysteine ligase, the rate-limiting enzyme in GSH synthesis [71]. Similarly, the increase in 1-methylnicotinamide could be attributed to the STK11 mutations (Figure S6c,d), as LKB1 regulates the enzyme Nicotinamide N-methyltransferase (NNMT), which catalyzes this metabolic conversion [72]. We have validated these relationships in both CCLE RNA-seq data and in CPTAC LUAD tumor proteomics data (Figure S6e–h).

Notably, anthranilic acid, the product of kynureninase, exhibited the strongest positive correlation with KYNU RNA expression (Figure 6a). While its levels varied significantly across KEAP1/STK11 oncogenotypes (Figure 6b), a multivariate linear model controlling for KEAP1/STK11 mutations confirmed that KYNU expression independently drives anthranilic acid upregulation (p-value = 0.0029) (Figure 6c). In contrast, kynurenine, the substrate of kynureninase, showed no correlation with KYNU expression (Figure 6d–f). The absence of kynurenine depletion, despite increased kynureninase activity, suggests a compensatory mechanism that sustains kynurenine levels, possibly through enhanced kynurenine synthesis. This maintenance of kynurenine levels in the face of high KYNU expression, coupled with anthranilic acid upregulation, provides a basis for immune suppression in the LUAD tumor microenvironment.

Beyond anthranilic acid, we identified additional metabolites associated with KYNU expression. NADP levels exhibited a positive correlation (r = 0.41, p = 3 × 10⁻⁴), while niacinamide (nicotinamide) levels showed a negative correlation (r = −0.3, p = 1 × 10⁻² (Figure 6g–l). The upregulation of NADP may be influenced by the activation of de novo NAD synthesis via the kynurenine pathway. Increased NADP availability could enhance redox homeostasis and anabolic metabolism, supporting tumor survival. In contrast, niacinamide depletion with higher KYNU expression suggests a reduced reliance on NAD salvage pathways. Since niacinamide is a key precursor for NAD synthesis in the salvage pathway, its depletion may indicate a metabolic shift favoring de novo NAD biosynthesis from tryptophan-derived metabolites rather than the recycling of niacinamide. Together, these findings suggest that KYNU-driven metabolic reprogramming alters NAD metabolism, which also potentially favors pathways that support tumor adaptation to stress and immune evasion.

3.6. Potential Translational Challenges in Using Genetically Engineered Mouse Models of LUAD to Study the Role of KYNU in Lung Cancer Pathogenesis and Therapy Development

Selecting appropriate preclinical models is critical for uncovering the functional role of KYNU in LUAD. While human data demonstrate a gradual increase in KYNU expression in tumors with KEAP1 or STK11 mutations, with the highest levels observed in co-mutated tumors, translating these findings into genetically engineered murine models remains a challenge. The abundance of KYNU in myeloid cells suggests a potential role in immune regulation. An obvious question is the tumor cell-autonomous role of KYNU expression in LUAD growth and survival. Importantly, in patient-derived LUAD cell lines, KYNU expression showed no correlation with sensitivity to CRISPR- or RNAi-mediated depletion of KYNU in DepMap data [73]. In fact, in DepMap, only 3/1178 human cancer cell lines showed KYNU dropout in CRISPR screens, and a significant gene effect was only found in endometrioid ovarian cancer. These findings highlight the need for in vivo LUAD models to explore KYNU’s interactions with the TME and its broader immune-modulatory roles.

To assess the feasibility of syngeneic models, we analyzed scRNA-seq data from LKR13 syngeneic mouse models, which include Kras, Kras/Keap1, Kras/Lkb1, and Kras/Keap1/Lkb1 mutant tumors [67]. Across all genotypes, Kynu expression was absent (Figure 7a). Further examination of autochthonous tumors from genetically engineered mouse models (GEMMs) using previously published data [63] also revealed a discrepancy with patient tumors (Figure 7b). In these models, Kynu expression was highest in tumors wild-type for Keap1/Stk11 and decreased in Keap1 and/or Stk11 mutants, contradicting the patterns observed in human LUAD. Given the previously noted reduction in macrophage-related gene expression in KEAP1 and/or STK11 mutant patient tumors, these findings suggest that Kynu expression in GEMMs is primarily driven by macrophage infiltration rather than cancer cell-intrinsic expression.

We extended our analysis to pan-cancer syngeneic mouse model data [74] and consistently observed low Kynu expression in tumors across all tested models (Figure 7c). Conversely, Kynu was highly expressed in immune-rich tissues, such as the spleen and lymph nodes, reinforcing the notion that Kynu expression in mice is primarily immune cell-derived. However, in CCLE data of human patient-derived cancer cell lines, many non-LUAD cancer cell lines exhibited robust cell-intrinsic KYNU expression (Figure 7d). In these cell lines, KYNU was upregulated even in the absence of KEAP1/STK11 co-mutation, indicating that KYNU regulation varies across cancer lineages.

Collectively, our data indicate a lack of robust, cancer cell-intrinsic Kynu expression in mouse models, highlighting a species-specific difference that poses challenges for preclinical modeling of KYNU’s role in LUAD using syngeneic mouse models.

3.7. Cancer Lineage-Specific Prognostic Considerations for KYNU and Kynurenine Pathway

Expanding beyond LUAD, we investigated KYNU’s prognostic associations across additional cancer types (Figure 8a). Notably, high KYNU expression was linked to better outcomes in two melanoma subtypes: uveal melanoma (UVM) and metastatic skin cutaneous melanoma lesions (SKCM.met) (Figure 8b). This association parallels the failure of IDO inhibitor trials in melanoma, prompting us to investigate the prognostic impact of IDO1 expression across pan-cancer datasets. Strikingly, IDO1, along with other kynurenine pathway genes previously targeted in melanoma (IDO2 and TDO2), also correlated with better outcomes in SKCM.met (Figure 8c,d). These findings suggest that failed clinical therapeutic targeting of the kynurenine pathway in melanoma may have an explanation, as its components appear to be associated with favorable prognoses in this context.

Conversely, in many cancer types, similar to findings in LUAD, high KYNU expression was associated with worse outcomes. Notably, kidney chromophobe cancer (KICH), pancreatic adenocarcinoma (PAAD), and thymoma (THYM) exhibited distinct bimodal KYNU expression distributions (Figure 8e). The separation between high- and low-KYNU groups in these tumor lineages was particularly striking, with hazard ratios exceeding 10. This dramatic outcome divergence underscores KYNU’s prognostic significance in these cancers and suggests its potential as a biomarker for aggressive disease phenotypes, particularly in PAAD. Taken together, these findings highlight the complexity of KYNU’s prognostic role, which varies across different cancer types.

4. Discussion

Kynurenine, a key metabolite in the tryptophan catabolism pathway, is well known for its immunosuppressive effects, including the suppression of T-cell proliferation and effector functions [75]. This has driven therapeutic strategies aimed at inhibiting upstream enzymes or delivering exogenous kynureninase to deplete kynurenine levels [76]. However, our findings highlight an apparent paradox: high KYNU expression, which facilitates kynurenine breakdown, correlates with poor prognosis in LUAD. This contradiction suggests that kynureninase’s role extends beyond simply reducing kynurenine-driven immunosuppression. Similarly, Fahrmann et al. and León-Letelier et al. reported that elevated KYNU, regulated by NRF2 activation, is linked to an immunosuppressive tumor microenvironment and poor outcomes across multiple cancer types [77,78]. For a comprehensive overview of therapeutic strategies targeting individual enzymes in the kynurenine pathway, refer to the recent review by León-Letelier et al. [79]. These findings support the hypothesis that kynureninase’s biological impact on cancer is multifaceted, potentially involving metabolic reprogramming and immune evasion mechanisms.

In LUAD, STK11 and KEAP1 mutations have been implicated in promoting aggressive tumor behavior [63,67]. Our data reveal that KYNU expression is highest in LUAD tumors with KEAP1/STK11 co-mutations, reflecting an independent yet incremental regulation by the NRF2 and STK11/LKB1 pathways. This pattern is distinct from canonical NRF2 or LKB1 targets, which are regulated primarily by a single pathway. For example, NQO1 is specifically elevated in KEAP1 mutants, while PDE4D is upregulated in STK11 mutants. The unique regulation of KYNU suggests it serves as an integrative node in metabolic pathways governed by NRF2 and LKB1. While NRF2 activation promotes redox homeostasis by enhancing antioxidant responses and NAD biosynthesis, STK11/LKB1 loss disrupts mitochondrial function and elevates reactive oxygen species (ROS), creating an increased demand for NAD to counteract oxidative stress. By linking tryptophan catabolism to de novo NAD synthesis, kynureninase may help reconcile opposing metabolic pressures, ensuring cancer cell survival under conditions of redox and bioenergetic stress. This dual regulation underscores kynureninase’s pivotal role in shaping the metabolic landscape of aggressive LUAD. In addition to the association of high KYNU expression and KEAP1/STK11 mutations, the association between KYNU expression and poor prognosis remains significant even after controlling for KEAP1/STK11 mutation status and other clinical variables, reinforcing its independent prognostic value.

Our findings reveal that KYNU expression is upregulated in LUAD, yet kynurenine levels, which should be degraded by KYNU, in fact, remain stable, suggesting a compensatory mechanism that sustains immunosuppressive signaling despite increased KYNU expression. In contrast, anthranilic acid, the direct product of kynureninase, exhibited the strongest positive correlation with KYNU expression in LUADs, further supporting kynureninase enzymatic activity in these tumors. Additionally, we observed a shift in NAD metabolism, with higher NADP and lower niacinamide levels, suggesting a preference for de novo NAD biosynthesis over salvage pathways. This metabolic shift may enhance redox balance and metabolic flexibility in tumors.

A major limitation of our current study is that our findings are based on correlative analyses from LUAD cell line metabolomic and transcriptomic data rather than direct manipulation of STK11, KEAP1, or KYNU in controlled experimental settings. Future studies involving genetic or pharmacologic perturbation of these pathways in isogenic models will be essential to establish causal relationships and assess the therapeutic potential of targeting kynureninase-associated metabolic shifts. Additionally, while our analysis focused on anthranilic acid–the product of kynurenine cleavage by KYNU, the available CCLE metabolomics dataset did not measure 3-HK or 3-HAA. Thus, although our findings suggest active kynurenine-to-anthranilate conversion, we cannot exclude the possibility that KYNU also modulates alternative branches of the kynurenine pathway in lung cancer cells. Furthermore, while kynurenine is a known activator of AHR signaling, it is important to recognize that kynurenic acid–another potent AHR ligand, can be generated independently of kynurenine via IL4I1-mediated oxidation of tryptophan [80]. IL4I1 activity has been implicated in shaping immunosuppressive tumor microenvironments [81], suggesting that integrating IL41 expression and kynurenic acid measurements into future analyses may provide additional insights into the broader landscape of tryptophan metabolism and immune remodeling in LUAD. Of course, another major limitation and subject of future research is the direct demonstration that high KYNU expression is associated with similar metabolomic changes in LUADs in patients.

A major challenge in understanding kynureninase’s role in LUAD is the translational gap between human and murine models. Our analyses indicate that KYNU expression in murine tumors does not recapitulate the oncogenotype-driven, cancer cell-intrinsic expression observed in humans. Instead, in murine tumor models, Kynu is primarily expressed in myeloid cells, highlighting species-specific differences in its regulation. Our human scRNA-seq analysis revealed high KYNU expression in goblet cells (Figure 5c), which are specialized epithelial cells responsible for mucus secretion in the respiratory tract. This raises the question of whether LUAD that express high levels of KYNU have some connection to or derivation from goblet cells. While abundant in the human lung, goblet cells are notably rare in the mouse lung [82]. Furthermore, goblet cells increase in response to irritants such as cigarette smoke, leading to mucus hypersecretion and potential airway obstruction [83]. Given that KEAP1/STK11 co-mutated tumors are extremely rare in never-smokers (Figure 3d), it is possible that goblet cell hyperplasia induced by tobacco exposure plays an important role in the carcinogenesis of this subtype. Notably, the Laughney et al. dataset includes three tumors with KEAP1/STK11 co-mutations [69]. However, no significant increase in KYNU expression was observed in these co-mutant tumors, and the fraction of epithelial cells was relatively small (10%). This could indicate a technical limitation of the scRNA-seq workflow, where cancer cells with high mucin expression could be difficult to dissociate, potentially leading to an underrepresentation of KYNU-high epithelial cells. Supporting this possibility, a previous KYNU immunohistochemistry (IHC) study in LUAD tumors confirmed KYNU expression in cancer cells [78], suggesting that KYNU-high tumor cells may be selectively lost or underrepresented in single-cell dissociation protocols.

Our analysis revealed distinct biological contexts of KYNU expression in LUAD. In KYNU-low tumors, expression is predominantly myeloid-derived, correlating with higher immune infiltration, including macrophages and dendritic cells. Conversely, KYNU-high tumors exhibit cancer cell-intrinsic expression and a depleted immune microenvironment, highlighting kynureninase’s dual role as a marker for stratifying immune contexts in LUAD. Interestingly, 3-hydroxyanthranilic acid (3-HAA), a kynureninase product derived from both 3-hydroxy kynurenine (3-HK) and anthranilic acid (see Figure 1g for pathway), inhibits nitric oxide synthase (NOS) expression in macrophages [84], suggesting that under normal conditions, kynureninase may act as a feedback regulator, tempering macrophage activity to maintain immune balance. However, in cancer, this mechanism may be hijacked. Elevated KYNU expression in cancer cells could lead to excessive 3-HAA production, suppressing NOS expression and reactive nitrogen species (RNS) generation. This suppression could impair macrophage effector functions and contribute to an immunosuppressive tumor microenvironment. Additionally, de novo NAD synthesis from the kynurenine pathway plays a pivotal role in macrophage effector responses [85]. Kynureninase-driven depletion of NAD precursors may disrupt these responses, further compromising macrophage function. These dual impacts—on RNS generation and NAD metabolism—highlight kynureninase’s potential as a critical regulator of immune evasion in LUAD and underscore the therapeutic promise of targeting KYNU to restore immune function in the tumor microenvironment.

Beyond LUAD, KYNU’s prognostic significance varies strikingly across cancer types. For example, in melanoma, high KYNU expression is paradoxically associated with better outcomes, aligning with the failure of IDO inhibitor trials [86]. This suggests that prior clinical efforts targeting the kynurenine pathway may have focused on the wrong cancer type. In melanoma, where enzymes in the pathway appear protective, their role may not be representative of other cancers. In contrast, in cancers such as pancreatic adenocarcinoma (PAAD) and thymoma (THYM), KYNU-high tumors are associated with significantly worse survival, with some hazard ratios exceeding 10. The stark survival disparities between KYNU-high and KYNU-low groups in these cancers underscore the considerable untapped potential of targeting KYNU and the kynurenine pathway, particularly in contexts where they clearly drive tumor aggressiveness.

5. Conclusions

Our study reveals that KYNU is a robust, bimodally distributed prognostic biomarker in LUAD, with elevated expression strongly associated with KEAP1/STK11 co-mutations, reduced immune infiltration, and poor clinical outcomes. Metabolomic analyses suggest that high KYNU expression corresponds with active kynurenine-to-anthranilate conversion and altered NAD metabolism. Notably, we find that common murine LUAD models fail to recapitulate the cancer cell-intrinsic KYNU expression pattern observed in patients, underscoring a critical translational gap. These findings support the need for improved preclinical systems and highlight kynureninase and its downstream metabolites as potential therapeutic targets in LUAD and other kynureninase-driven cancers.