Genetic predisposition to persistent fatigue after a diagnosis of colorectal cancer.

Introduction
Cancer-related fatigue (fatigue) is characterized by a prolonged and distressing subjective state marked by a heightened sense of physical, emotional, or cognitive tiredness or exhaustion associated with cancer or its treatment that is not proportional to recent activity and interferes with usual functioning.1 Fatigue may start before treatment and typically intensifies during treatment such as radiation2 and chemotherapy.3 The prevalence of treatment-related fatigue varies from 25% to 99%, and is influenced by patient characteristics, treatments, and assessment methods,4,5 with 30% to 60% of patients reporting moderate to severe fatigue.4,5 Although fatigue generally improves within a year after treatment, one-quarter to one-third of survivors experience persistent fatigue for months or even years after successful treatment completion.6
The pathophysiology of fatigue is multifactorial and complex,7,8 likely involving various biological mechanisms such as anemia, cytokine dysregulation, hypothalamic-pituitary-adrenal axis dysfunction, serotonin neurotransmitter imbalance, metabolic disturbances, and muscle deconditioning.9,10 Despite this complexity, most studies on the biological mechanisms of fatigue have focused narrowly on immune and inflammatory variables, underscoring the need for a deeper understanding of fatigue’s pathophysiology to identify new therapeutic targets.
The role of inherited genetic variation in fatigue among patients with cancer remains unclear. Twin studies suggested that fatigue has a heritability ranging from 6% to 50%.11,12 A few studies have explored the genetics of proinflammatory cytokine activity as a contributor to fatigue, often focusing on single-nucleotide variants (SNVs, formerly single-nucleotide polymorphisms) in candidate genes such as IL-1B, IL-6, and TNF.13-19 Only a limited number of genome-wide association studies (GWASs) conducted in noncancer fatigue have been reported.20-24
Our study extends previous research by focusing on fatigue in patients with colorectal cancer (CRC), one of the most common cancers. Advances in early detection and treatment have transformed CRC from a fatal disease into one that is increasingly curable, resulting in a large and growing population of long-term survivors (ie, 1.8 million by 2026).25 Fatigue is the most frequently reported adverse symptom among patients with CRC,26 which can persist for many years after diagnosis.27,28 To conduct a comprehensive study of fatigue in patients with CRC, we employed multiple approaches to gain insights into the underlying genetic predisposition, including GWASs, fine mapping of known loci, and integrative analyses.

Methods

Study sample
Participants were recruited from the ColoCare Study (ClinicalTrials.gov identifier NCT02328677), an international cohort study of patients with CRC. Eligible individuals were at least 18 years old, newly diagnosed with primary colon (International Statistical Classification of Diseases, Tenth Revision codes C18.0-C18.9) or rectal cancer (codes C19.9, C20.9, and C21.8) at any stage (I-IV), and receiving treatment at one of 6 US sites (Fred Hutchinson Cancer Center, H. Lee Moffitt Cancer Center, University of Tennessee Health Science Center, Washington University School of Medicine, Huntsman Cancer Institute, or Cedars-Sinai Medical Center) or at Heidelberg University Hospital in Germany. Patients were enrolled shortly after diagnosis, typically before surgery or treatment initiation, with baseline data collected at the start of CRC treatment. Detailed participant demographics have been published elsewhere.29-31 In this analysis, we present findings based on patients (n = 1219) who completed questionnaires on health-related quality of life, including fatigue, at one or more time points. Participants from the Fred Hutchinson Cancer Center did not complete relevant questionnaires and were excluded. Details on the collection of demographic and clinical characteristics are outlined in the Supplementary Methods. Approval for the study was obtained from the institutional review boards at each recruitment site, and all participants provided written informed consent.

Fatigue
Fatigue was measured at 5 time points: baseline (around treatment initiation) and 3, 6, 12, and 24 months after diagnosis, using the 3-item fatigue subscale from the European Organisation for Research and Treatment of Cancer Core Quality of Life Questionnaire.32-34 At each time point, the raw scores were converted to a total score ranging from 0 to 100,35 with higher scores indicating greater fatigue severity.
Because fatigue is a subjective measure, we considered 3 approaches to define it as an endpoint for our analysis. The primary approach was to use a summary measure of the trajectory of all reported values from diagnosis to survivorship (defined as mild, moderate, and severe), as reported by Li et al.32 To focus on the clinically meaningful outcome of severe fatigue (yes/no), we combined the moderate and mild fatigue groups as the referent category for the GWAS. In secondary analyses, we analyzed fatigue as a continuous variable, considering (1) the mean fatigue score (ie, average) across all time points and (2) the highest reported value (ie, worst) across any of the time points.

Genotyping
A total of 1220 samples were genotyped using the Illumina GDA_PGx_Psych-8_v1_20046747X383184_A1 arrays in the Molecular Genomics Core of the H. Lee Moffitt Cancer Center and Research Institute (Moffitt, Tampa, FL). Genotype calling was performed using GenomeStudio, version 2.0, software (Illumina) with the default parameters and clustering algorithm. Genotype data were processed and cleaned according to stringent quality control criteria at both the individual and SNV levels and imputed using the TOPMed Imputation Server. Details of the genotype data cleaning and preparation, population structure, and imputation are provided in the Supplementary Methods.

Statistical methods
Descriptive statistics for age and body mass index were computed, and t tests were used to compare these statistics by severe fatigue status (yes/no). Frequencies and percentages for demographic and clinical characteristics were calculated, and their associations with severe fatigue status were assessed using χ2 or Fisher exact tests, as appropriate. The GWAS was conducted to assess the risk of severe fatigue, using 8 577 063 directly genotyped, high-quality imputed SNVs and indels with a maximum allele frequency of at least 1% within our entire study population (n = 1066). The allelic dosage of each variant, under a log-additive genetic model, was evaluated in relation to severe fatigue risk (yes/no) and continuous measures of fatigue (average and worst levels) using the PLINK, version 2.0, toolset.36 For continuous scores, we used deciles to estimate the risk associated with each 10% increase in fatigue to improve interpretability. Per-allele effect size and 95% CIs were estimated using unconditional logistic regression, with adjustments made for age, sex, smoking status, tumor stage, and the first 2 principal components (PCs) to account for global ancestry. Age, sex, and global ancestry are commonly included as covariates in GWASs to account for their widespread influence on genetic associations. In addition, smoking status and tumor stage were selected as covariates because both factors are strongly associated with severe fatigue (P < .001). To assess potential treatment confounding, we conducted sensitivity analyses by adding adjuvant and neoadjuvant treatment status as covariates. As an additional sensitivity analysis, we restricted the sample to White participants to reduce potential bias from population stratification, repeating the association analyses with the same model specifications. The threshold for statistical significance in the discovery GWAS was set at P = 5 × 10−8. We accessed the Ensembl genome browser using the biomaRt package in R (R Foundation for Statistical Computing) to annotate SNVs from our GWAS and identified the nearest genes for each SNV. Replication of genetic associations was assessed using logistic regression for 43 previously known SNVs,13-18,  37-51 and fine mapping was conducted using the National Institutes of Health Sum of Single Effects algorithm on important regions identified in our study. To integrate our GWAS results with gene expression data, we performed a transcriptome-wide association study using the S-PrediXcan and MultiXcan models. S-PrediXcan estimates genetically regulated gene expression in a single tissue using prediction models from reference transcriptome datasets and evaluates its association with fatigue.52 MultiXcan extends this approach by using the substantial sharing of expression quantitative trait loci across tissues and contexts, improving gene discovery through multivariate regression, which accounts for tissue correlation structures.53 We applied Genotype-Tissue Expression, version 8, gene expression data for liver and whole blood, followed by colocalization analysis to identify shared genetic signals, as detailed in the Supplementary Methods. Figure S1 illustrates the workflow.

Results

Characteristics of the study population
Among 1219 patients, 378 experienced severe fatigue over the course of their disease (31.0%) (Table 1). There was a statistically significant difference in the distribution of severe fatigue by race (P = .01), with a higher percentage of White patients (92.6% vs 89.6%) and a lower percentage of Black or African American patients (3.4% vs 7.7%) reporting severe fatigue compared to patients not experiencing severe fatigue. Current smoking status was also significantly associated with severe fatigue (18.1% vs 9.6%, P < .001). Patients with advanced cancer stages (III and IV) were also significantly more likely to experience severe fatigue (64.3% vs 50.3%, P < .001).

Discovery
Our GWAS analysis included 8 577 063 genetic variants with a maximum allele frequency of at least 1%, all of which passed stringent quality control measures. The genomic control inflation factor (λ) remained below 1.007 for all models, indicating sufficient adjustment for population stratification (Figure S2). Manhattan plots are shown in Figure 1.
We identified a statistically significant locus on chromosome 4 near the LINC02505 gene for severe fatigue (rs6531463: odds ratio [OR] = 3.25, P = 3.88 × 10−8; rs4263411: OR = 3.31, P = 5.51 × 10−8; rs6811957: OR = 3.27, P = 7.00 × 10−8). On chromosome 3, variants near the NEK10 (rs73055756: OR = 0.59, P = 2.75 × 10−6) and SLC4A7 (rs11129284: OR = 0.59, P = 3.18 × 10−6) genes were also associated with severe fatigue. The SNVs near NWD2 on chromosome 4, including rs4832900 (OR = 0.50, P = 1.40 × 10−6) and rs6531534 (OR = 0.50, P = 1.40 × 10−6), showed suggestive associations (Table S1 and Figure 1).
When modeling mean fatigue, significant loci were identified on chromosome 3p24.1, close to the NEK10 and SLC4A7 genes. Notably, rs7617094 (OR = 0.57, P = 2.21 × 10−8), rs11129284 (OR = 0.56, P = 2.42 × 10−8), and rs73055754 (OR = 0.56, P = 2.42 × 10−8) within NEK10 were the most significant associations. Overall, 10 significant SNVs were identified at this locus, all associated with lower mean fatigue levels, suggesting a protective effect (OR range = 0.56-0.57, P ≤ 5 × 10−8) (Table S2 and Figure 1). In addition, several variants showed suggestive associations (P < 1 × 10−6), including 3p24.1 and 5q13.1 (Table S3 and Figure 1). We identified no significant SNVs that reached genome-wide significance (P < 5 × 108) when examining the worst fatigue level. Further adjustment for treatment exposure revealed that the overall pattern of associations remained consistent, suggesting that treatment did not significantly affect the top genetic signals (Figure S3).
In the ancestry-restricted analysis of self-reported White participants, the overall pattern of associations remained consistent with the primary results, suggesting that our findings are not driven by population stratification (Figure S4).

Replication of candidate fatigue-related variants
A total of 43 fatigue-associated SNVs had previously been identified from the published literature. Four SNVs exhibited nominal significance (P < .05) (Table S4).

Independent susceptibility alleles
Fine mapping with Sum of Single Effects identified likely causal variants across 2 loci—4p14 and 4p15.2—for severe fatigue. The 4p15.2 region contained the smallest credible set of 2 SNVs (rs76754501 posterior inclusion probability of 82.96%; rs114674171 posterior inclusion probability of 14.51%). In the 4p14 region, 25 SNVs were identified within the credible set, primarily annotated as long, intergenic, nonprotein-coding RNAs and positioned near the gene LINC02505. These SNVs had varying posterior inclusion probabilities (range, 4%-6.78%) (Figure 2 and Table S5). When we examined average fatigue, we also identified credible sets of SNVs in 4p14 as well as in 5q13.1 and 3p24.1 (Figure 3 and Table S6). The regions 5q13.1 and 3p24.1 were also identified when examining the worst level of fatigue (Figure 4 and Table S7).

Integrative analyses
Using S-PrediXcan, we identified statistically significant associations between the predicted expression of NEK10 in the liver and risk of fatigue (P < 4.31 × 10−6) (Table 2). Furthermore, S-MultiXcan highlighted several putative genes associated with fatigue: SDCCAG8, PRKCI, P2RY12, PPP1R16A, LACTB2, SLFN5, C17orf80, and NEK10 (P < 2.42 × 10−6). Using the COLOC package, NEK10 showed strong evidence of colocalization across all definitions of fatigue. For severe fatigue, the posterior probability for H4 (PP.H4) was 92.1%, with a high shared variant probability confirmed by sensitivity analysis (H4 > 0.9). Similarly, for average fatigue, NEK10 exhibited a PP.H4 of 95.2%, and for worst fatigue, a PP.H4 of 95.3%, both passing sensitivity criteria. For SLC4A, moderate evidence for colocalization was observed (PP.H4 values ranged from 53.8% to 61.7%) (Table 3 and Figure S5).

Discussion
Our study represents a large-scale GWAS of fatigue in patients with CRC, an understudied population with respect to fatigue.54 We identified significant associations with variants near LINC02505 on chromosome 4 and NEK10 and SLC4A7 on chromosome 3. Multiple lines of evidence, including integrative analysis and colocalization analysis, further implicated NEK10 in the risk of persistent fatigue.
The genetic and biological mechanisms underlying fatigue, particularly chronic fatigue syndrome, have recently garnered significant interest, with several studies highlighting key biological pathways, such as mitochondrial dysfunction,55,56 immune system dysregulation,55-57 metabolic disturbances,44 and the role of noncoding RNAs.57,58 Our findings provide modest evidence for several of these pathways in persistent fatigue among patients with CRC. Other studies, including a study by Hoogland and colleagues44 on patients with hematologic cancer undergoing allogeneic hematopoietic cell transplantation, identified variants in genes regulating lipid metabolism and circadian rhythms. Cameron et al.15 found that genetic variants, particularly those in immune-related genes, were strongly associated with heightened fatigue in patients with breast cancer undergoing chemotherapy.
Our GWAS identified the strongest association with severe fatigue near LINC02505 on chromosome 4. Although the specific functional role of LINC02505 in fatigue remains unclear, long, intergenic, noncoding RNAs have been implicated in regulatory processes, including inflammation, immune response, and cellular stress pathways.59-62 Long, intergenic, noncoding RNAs can modulate inflammatory pathways and immune responses, such as intermittent hypoxia and sleep apnea, helping explain their relevance in conditions such as chronic fatigue.63-65 Studies have shown that long, intergenic, nonprotein-coding RNAs such as NEAT1 and linc00899 may also regulate these immune pathways, influencing inflammation and immune dysregulation, which are commonly implicated in fatigue.63,66,67 In addition, previous findings from the ColoCare Study by Himbert et al.62 demonstrated that inflammation and angiogenesis-related biomarkers are associated with fatigue in patients with CRC, suggesting a biological link between systemic inflammation and fatigue severity. Our findings build on this result by identifying genetic variants associated with severe fatigue, suggesting a potential interplay between genetic susceptibility and inflammation pathways.
We provided evidence supporting a link between NEK10 and fatigue. NEK10, a member of the NIMA-related kinase family, is a serine/threonine kinase involved in critical cellular functions such as cell cycle regulation, DNA damage repair, apoptosis, and microtubule organization.68 Although the exact mechanisms underlying its association with fatigue remain unclear, NEK10 plays a key role in the G2/M cell cycle transition.69 Dysregulation of NEK10 can cause cell cycle arrest or premature mitosis, potentially leading to cellular stress and energy depletion.69  NEK10 also participates in mitochondrial metabolism and the DNA damage response, and its dysfunction has been linked to ciliary abnormalities and breast cancer.68 Primary cilia are important for sensing extracellular signals and regulating pathways related to proliferation, metabolism, and inflammation.68,70,71 Disruption of ciliary signaling may result in cellular dysfunction, chronic inflammation, and systemic fatigue.68
The SLC4A7 gene, located on chromosome 3, emerged as a potential locus associated with fatigue in our study. It encodes the sodium bicarbonate cotransporter 3, a member of the SLC4 family, which is involved in regulating cellular pH and maintaining ionic homeostasis.72 This transporter plays a key role in acid-base balance and intracellular pH regulation, processes that are essential for normal physiological function.72 Disruptions in cellular ion homeostasis and pH regulation have been implicated in fatigue-related disorders because these processes are critical for cellular function and energy metabolism.73 This finding is consistent with research by Pheasant and colleagues,74 which emphasized the role of cellular energy regulation and ion transport in the pathogenesis of fatigue. The association of SLC4A7 with fatigue may therefore reflect disturbances in cellular metabolic pathways, potentially leading to impaired muscle function and persistent fatigue in patients with cancer.
Among the 43 SNVs analyzed in the replication analysis, 4 variants, including rs2243248 (IL4), rs2134794 (ERCC3), rs11630479 (IGF1R), and rs670757 (GRB14), reached nominal significance. These SNVs have been previously linked to immune response, DNA repair, and growth factor signaling, processes that are relevant to cancer and fatigue.75-78 None of these SNVs overlap with the genes identified in our study, however, such as LINC02505, NEK10, and SLC4A7, possibly because of factors such as differences in study populations, phenotype definitions, statistical power, genetic heterogeneity, or environmental and lifestyle factors. In addition, fatigue likely has a polygenic basis, with multiple small-effect loci contributing to its manifestation.22
The transcriptome-wide association study results enhance our study by providing functional insights into the genetic basis of severe fatigue in patients with CRC. Although our GWAS identified novel SNVs associated with fatigue, transcriptome-wide association studies bridge the gap between genetic variation and gene expression, highlighting potential regulatory mechanisms.79 Notably, transcriptome-wide association studies support the involvement of genes such as LINC02505, NEK10, and SLC4A7, which were identified in our GWAS but had not been previously linked to fatigue in patients with cancer. By demonstrating that genetically predicted expression levels of these genes are associated with fatigue, transcriptome-wide association studies strengthen their biological relevance and suggest that altered gene regulation may contribute to symptom severity.

Strengths and limitations
We did not include functional validation of the variants identified to confirm the biological mechanisms underlying these associations. Future studies integrating multiomics approaches, such as transcriptomics and metabolomics, may be helpful in providing an even more comprehensive understanding of the biological pathways related to cancer-associated fatigue. A limitation of this study is the lack of harmonized treatment data across cohorts, which prevented adjustment for potential confounding effects of cancer therapies on fatigue. In addition, our data included patients with all subtypes of CRC receiving various treatments. Although some may view this as a limitation, we believe that the heterogeneity of the samples enhances the generalizability of our findings. Furthermore, independent replication of our study results is necessary, particularly across different cancer types and treatments. Nevertheless, we employed a robust analysis approach to examine candidate genes associated with fatigue in patients with CRC. Fatigue is commonly reported in this patient population, and current treatment options are limited. Behavioral and psychosocial interventions demonstrate benefits,80-83 but they can be time-intensive, which limits uptake, compliance, and maintenance. Therefore, identifying high-risk patients based on genetic susceptibility would help target preventive treatment strategies more effectively.

Supplementary Material
djaf140_Supplementary_Data