Inflammatory Pathway Genes Associated with Inter-Individual Variability in the Trajectories of Morning and Evening Fatigue in Patients Receiving Chemotherapy

1. Introduction

During chemotherapy (CTX), over 45% of patients experience clinically meaningful levels of fatigue that decrease their ability to tolerate treatments, engage in social relationships, and maintain regular work activities [1]. However, a growing body of evidence demonstrates that inter-individual variability exists in fatigue severity across cancer diagnosis [2–4] and treatments [5, 6]. In addition, recent work from our group [7–11] and others [4, 12] demonstrates that morning and evening fatigue are distinct yet related symptoms. Some of this inter-individual variability is explained by different phenotypic characteristics that distinguish between higher levels of morning (e.g., higher body mass index (BMI), lack of regular exercise, higher state anxiety) and evening (e.g., being white, higher years of education, child care responsibilities) fatigue [9, 10]. In addition to phenotypic differences, preliminary evidence suggests that variations in cytokine genes are associated with inter-individual differences in morning (e.g., tumor necrosis factor alpha (TNFA) [13] and evening (e.g., interleukin (IL)4 [14] and IL6 [15]) fatigue severity.

While considered to be multi-factorial, a growing body of evidence suggests that cytokine dysregulation, as well as many other neuroinflammatory processes may modulate fatigue severity in a number of chronic conditions [16–19]. Increased knowledge of the mechanisms that underlie fatigue is essential for the development of effective treatments for this devastating symptom. However, no definitive conclusions can be drawn from studies that evaluated associations between fatigue severity and various biomarkers of cytokine dysregulation (for reviews see [18, 19]).

1.1 Associations between fatigue and serum markers of inflammation

To examine this cytokine dysregulation hypothesis, several studies evaluated the associations between fatigue severity and serum cytokine levels. To date, results are inconclusive, with some studies finding positive associations between fatigue severity and circulating levels of TNF-α [20, 21] and IL-6 [18, 20, 22–29] and others finding no associations with TNF-α[18, 22, 30–33], IL-6 [30, 34–38], and IL-4 [25, 36, 38]. These inconsistent results may be related to the challenges associated with the measurement of serum cytokines, as well as circadian variations in cytokine levels [39].

An alternative approach is to measure circulating levels of biomarkers of immune activation (e.g., cellular receptors) [40]. Again, these results are inconclusive. Some studies found positive associations between fatigue and changes in serum levels of IL-1 receptor antagonist (IL-1ra)[24, 35, 41, 42], soluble TNF receptor II (sTNF-RII) [41, 43, 44], sTNF-RI [45], and sIL-6R [37, 46, 47]. However, other studies found no associations between fatigue severity and changes in serum levels of IL-1ra [27, 34, 36], sTNF-RII [36, 42], and sIL-6R [46]. Of note, none of these studies evaluated for associations between diurnal variations in fatigue severity and changes in these serum markers.

1.2 Changes in gene expression and fatigue

Another approach to examine the role of inflammation in fatigue is to evaluate for changes in the expression of inflammatory genes. To date, seven studies have evaluated for changes in gene expression associated with fatigue severity in oncology patients [34, 48–53]. Five of these studies [34, 48, 50, 52, 53] examined changes in gene expression related to inflammation/immune function. In four of these studies higher levels of fatigue were associated with upregulation of genes that regulate cytokine production (i.e., interferon alpha-inducible protein 27 (IFI27) [48], α-synuclein (SNCA) [52], IL1 [34], IL6 [34], IL4 [50]). In another study [53], differentially perturbed cytokine pathways were associated with higher levels of evening fatigue. However, across these studies only mean or evening fatigue scores were evaluated. In addition, the sample sizes for these studies were relatively small (i.e., 15 [49] to 137 [50] patients).

1.3 Associations between fatigue and variations in cytokine genes

A third approach that can be used to examine the role of inflammation in fatigue is to evaluate for associations between fatigue severity and variations in cytokine genes. Single nucleotide polymorphisms (SNP) in TNFA [54, 55], IL6 [37, 56, 57], and IL1RA [58] were associated with increased levels of fatigue. To date, only three studies evaluated for associations between variations in cytokine genes and diurnal variations in fatigue severity [13–15]. In a study of oncology patients (n=185) and their family caregivers (n=103), SNPs in TNFA (i.e., rs1800629, rs3093662) and IL6 (i.e., rs4719714) were associated with higher levels of morning and evening fatigue [13–15]. Additionally, a polymorphism in IL4 rs2243248 was associated with lower levels of evening fatigue [14]. While the studies cited above provide preliminary evidence that variations in cytokine genes are associated with diurnal variations in fatigue severity, two of the studies evaluated only one polymorphism [13–15] and none of them evaluated oncology patients undergoing CTX.

While evidence exists for the role of cytokine dysregulation as a modulator of neuroinflammation, recent studies found other pathways and processes that contribute to the development of inflammation (e.g., the mitogen-activated protein kinase (MAPK) pathway [59], and inflammasomes [19, 60]). However, the contribution of these pathways to fatigue severity in oncology patients undergoing CTX has not been evaluated. Increased knowledge of whether additional inflammatory pathways are associated with diurnal variations in fatigue severity would enhance our understanding of the various mechanisms that contribute to this devastating symptom.

Recently, we identified common and distinct phenotypic characteristics for morning [9] and evening [10] fatigue severity in oncology patients undergoing CTX. This study extends these findings to identify associations between variations in genes associated with a variety of inflammatory processes and the severity of morning and evening fatigue. Since genes interact with one another [61], the polymorphisms that were evaluated were grouped into common inflammatory pathways to provide insights into the role of related genes and the severity of morning and evening fatigue. The purposes of this study, in a sample of oncology patients with breast, gastrointestinal (GI), gynecological (GYN), or lung cancer who received two cycles of CTX, were to determine whether variations in genes involved in inflammatory processes were associated with inter-individual variability in initial levels as well as in the trajectories of morning and evening fatigue.

2. Methods

2.1 Patients and settings

Some of the details of the phenotypic [9–11, 62] and genotypic [63, 64] methods used in this study are published elsewhere. In brief, patients were recruited from two comprehensive cancer centers, one Veteran’s Affairs hospital, and four community-based oncology programs. Patients with a diagnosis of breast, GI, GYN, or lung cancer were eligible to participate if they were ≥18 years of age; had received CTX within the previous four weeks; were scheduled to receive at least two additional cycles of CTX; were able to read, write, and understand English; and gave written informed consent.

2.2 Instruments

Patients completed a demographic questionnaire, the Karnofsky Performance Status (KPS) scale [65], and the Self-Administered Comorbidity Questionnaire (SCQ) [66]. In addition, patients completed a number of questionnaires to evaluate anxiety [67], depression [68], and sleep disturbance [69].

Fatigue was evaluated using the 18 item Lee Fatigue Scale (LFS) that assesses physical fatigue and energy [70]. Each item was rated on a 0 to 10 numeric rating scale (NRS). Total fatigue and energy scores were calculated as the mean of the 13 fatigue and the 5 energy items, with higher scores indicating greater fatigue severity and higher levels of energy, respectively. Using separate LFS questionnaires, patients were asked to rate each item based on how they felt within 30 minutes of awakening (i.e., morning fatigue and morning energy) and prior to going to bed (i.e., evening fatigue and evening energy). The LFS has established cutoff scores for clinically meaningful levels of fatigue (i.e., ≥3.2 for morning fatigue, ≥5.6 for evening fatigue) [70]. The LFS is easy to administer, relatively short, and has well established validity and reliability [70]. As noted in previous reports (9,10), the Cronbach’s alphas were .95 for evening fatigue, .95 for morning fatigue, .93 for evening energy, and .95 for morning energy.

2.3 Study Procedures

Each of the sites’ Institutional Review Board approved the study. All patients provided written informed consent. Patients completed study questionnaires in their homes a total of six times over two cycles of CTX (prior to CTX administration (i.e., recovery from previous CTX cycle, assessments 1 and 4), approximately 1 week after CTX administration (i.e., acute symptoms, assessments 2 and 5), approximately 2 weeks after CTX administration (i.e., potential nadir, assessments 3 and 6)).

3. Genomic Analyses

3.1 Blood collection and genotyping

Genomic deoxyribonucleic acid (DNA) was isolated from peripheral blood mononuclear cells (PBMCs), using the PUREGene DNA Isolation System (Invitrogen, Carlsbad, CA). DNA was quantitated with a Nanodrop Spectrophotometer (ND-1000) and normalized to a concentration of 50 nanograms/microliter (diluted in 10 mM Tris/1 mM EDTA). Genotyping was performed using the HumanExome Array-12 v1.1 on the Infinium Beadchip genotyping platform which provides focused coverage in the coding regions (i.e., exons) of genes (Illumina, San Diego, CA). Data were processed according to the standard protocol using GenomeStudio (Illumina, San Diego, CA).

3.2 Candidate gene and SNP selection

Candidate genes were selected based on evidence in the literature of an association between each gene and fatigue. SNPs representing these genes were selected from the genome-wide SNP array.

Quality control filtering excluded SNPs with call rates of <95%. SNPs with less than (i.e., monomorphic) or more than (i.e., tri- or tetra-allelic) two alleles were excluded. Allele counts at different loci were assumed to be independent. The 309 SNPs among the 64 candidate genes that passed all quality control filters and whose occurrence rates were evaluated in this sample are listed in Supplemental Table 1. The genes are grouped within their common inflammatory pathways based on a review of the literature as well as the description of each gene’s function and pathway found in the National Center for Biotechnology Information (NCBI) gene database and pathcards (http://pathcards.genecards.org). The following inflammatory pathways are represented: chemokines (3 genes); cytokines (12 genes); inflammasome (11 genes); Janus kinase/signal transducers and activators of transcription (JAK/STAT, 10 genes); mitogen-activated protein kinase/jun amino-terminal kinases (MAPK/JNK, 3 genes); nuclear factor-kappa beta (NFkB, 18 genes); and NFkB and MAP/JNK (7 genes).

4. Statistical analyses

4.1 Demographic and clinical data

The sample’s demographic and clinical characteristics and symptom severity scores at enrollment were determined with descriptive statistics and frequency distributions. These analyses were done using the Statistical Package for the Social Sciences (SPSS) version 22 [71].

4.2 Genetic data

Gene counting determined allele and genotype frequencies. To be included in subsequent evaluations, each SNP needed to have a total of six occurrences of the rare allele (i.e., heterozygous or homozygous) in order not to over- or under- estimate the effect of the rare allele. After applying this criterion, 93 SNPs among 49 genes were evaluated as potential predictors of inter-individual variability in morning and evening fatigue. Liability scores composed of the number of rare allele occurrences across all SNPs for each candidate gene were generated by summing the number of rare alleles carried by each patient.

To minimize confounding due to population stratification, ancestry informative markers (AIMs) identified with principal component (PC) analysis were used in subsequent analyses [72–75]. Approximately 3,468 AIMS were included in this analysis. To adjust for potential confounding due to population substructure (i.e., race/ethnicity) the first three PCs were included as covariates in the hierarchical linear modeling (HLM) analyses.

4.3 HLM Analysis

Details of the HLM analysis are published elsewhere [9, 10]. In brief, HLM based on full maximum likelihood estimation was performed in two stages to evaluate the effects of individual SNPs and liability scores on initial levels as well as on changes over time in the severity of morning and evening fatigue [76]. Morning and evening fatigue were evaluated in separate HLM analyses. Since the six assessments encompassed two cycles of CTX, a piecewise model strategy was employed to evaluate the pattern of change in morning and evening fatigue over time. The first piece (PW1) modeled change over time during the first CTX cycle (i.e., Assessments 1, 2, and 3). The second piece (PW2) modeled change during the second CTX cycle (i.e. Assessments 4, 5, and 6).

Then, inter-individual differences in the piecewise trajectories of morning and evening fatigue were examined by modeling the individual change parameters (i.e., intercept and slope parameters) as a function of proposed predictors at level 2. First, each of the SNPs and liability scores that passed the quality control filters was evaluated in an exploratory analysis to determine whether it would result in a better fitting model if it alone was added as a predictor. To improve estimation efficiency and construct a parsimonious model, SNP predictors with a t value of <2.0 were excluded from subsequent model testing.

Each of the SNPs and liability scores, identified in the exploratory analyses (Tables 1 and ​and2),2), were entered into the model that controlled for self-reported and genomic estimates of race and ethnicity to predict each individual change parameter in morning or evening fatigue severity. Only SNPs that maintained a statistically significant contribution were retained in the final models. A p-value of <.05 indicated statistical significance.

Consistent with our previous studies [14, 63, 77–80], recommendations from the literature [81, 82], rigorous quality controls for genomic data and the exploratory nature of our analyses, adjustments were not made for multiple testing. Since significant SNPs and liability scores identified in the exploratory analysis were evaluated further in the HLM analyses that controlled for population stratification (i.e., genomic and self-reported estimates of race and ethnicity), and other variations in the same gene, the significant independent genetic associations reported are unlikely to be due solely to chance.

4.4 Estimation of polymorphism function

Based on the results of the HLM analyses, the SNPs associated with inter-individual differences in the initial levels or trajectories of morning or evening fatigue were evaluated for their potential impact on protein function using two bioinformational tools (i.e., Sorting Intolerant From Tolerant (SIFT) algorithm [83] and Polymorphism Phenotyping v2 (PolyPhen-2) [84]). SIFT predicts whether a SNP in a coding region results in an amino acid substitution that may affect protein function. This prediction is based on an analysis of the conservation of amino acid residues in sequence alignments of closely related sequences [84]. PolyPhen-2 compares the SNP to sequence-based and structure-based predictive features to predict the functional significance of the SNP [84]. Results from the bioinformational tools are described in the discussion.

5. Results

5.2 Changes in morning fatigue severity

HLM was used to examine how morning fatigue scores changed within the two cycles of CTX, controlling for self-reported and genomic estimates of race and ethnicity. The estimates for the initial piecewise model are presented in Table 4. Since the model was unconditional (i.e., no covariates included in the model), the average morning fatigue severity score at enrollment (i.e., 3.011 on a 0 to 10 scale) represents the intercept. The estimated linear piecewise rates of change were 1.192 and 0.532 (both p<.0001) for piecewise linear 1 and piecewise linear 2, respectively. The estimated quadratic piecewise rates of change were −.599 and −.153 (both p<.0001) for piecewise quadratic 1 and piecewise quadratic 2, respectively. Figure 1A displays the unconditional model for mean morning fatigue scores over the two cycles of CTX.

Figures 1

A–F – Unconditional piecewise model of mean morning fatigue scores for six assessment points over two cycles of chemotherapy (A). Influence of the recessive model (TT+TC vs. CC) of the rare C allele in IL12B rs3213094 on the inter-individual differences in the intercept for morning fatigue (B). Influence of the dominant model (CC vs. CA+AA) of the rare A allele in TNFA rs1041981 on the slope parameters for morning fatigue (C). Influence of the recessive model (AA+AG vs. GG) of the rare G allele in NOD2 rs2076756 on the inter-individual differences in the intercept for morning fatigue (D). Influence of the recessive model (GG+GC vs. CC) of the rare C allele in NLRP5 rs471979 on the slope parameters for morning fatigue (E). Influence of the dominant model (TT vs. TC+CC) of the rare C allele in NLRP6 rs74044411 on the slope parameters for morning fatigue (F).

Table 4

Hierarchical Linear Models for Morning Fatigue

Cytokine Genes
Morning Fatigue and IL12B rs3213094	Coefficient (SE)
	Unconditional Model	Final Model
Fixed effects
Intercept	3.011 (.099)+	3.008 (.099)+
Ethnicitya
Black versus White	−.845 (.917)	−.735 (.913)
Asian versus White	.214 (.675)	.265 (.672)
Hispanic versus White	.328 (.331)	.348 (.329)
Ancestry informative markers principal componentsb
PC1	−.102 (.104)	−.105 (.103)
PC2	−.052 (.088)	−.031 (.088)
PC3	.008 (.039)	.008 (.039)
PW1 – linear	1.192 (.148)+	1.191 (.148)+
PW1 – quadratic	−.599 (.071)+	−.599 (.071)+
PW2 – linear	.532(.096)+	.531 (.096)+
PW2 – quadratic	−.153 (.031)+	−.153 (.031)+
Time invariant covariates
Intercept - IL12B rs3213094		−.760 (.319)*
Variance components
In intercept	3.552+	3.512+
Goodness-of-fit deviance (parameters estimated)	11026.531695 (13)+	11020.881398 (14)
Model comparison χ (df)		5.650 (1)*
Morning Fatigue and TNFA rs1041981		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.011 (.099)+
Ethnicitya
Black versus White		−.807 (.917)
Asian versus White		.187 (.675)
Hispanic versus White		.304 (.331)
Ancestry informative markers principal componentsb
PC1		−.106 (.103)
PC2		−.045 (.088)
PC3		.015 (.039)
PW1 – linear		1.186 (.148)+
PW1 – quadratic		−.597 (.071)+
PW2 – linear		.533 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
PW1 – linear - TNFA rs1041981		−.641 (.278)*
PW1 – quadratic - TNFA rs1041981		.264 (.131)*
Variance components
In intercept		3.549+
Goodness-of-fit deviance (parameters estimated)		11019.804052 (15)
Model comparison χ (df)		6.728 (2)*
Inflammasome Pathway Genes
Morning Fatigue and NLRP5 rs471979		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.011 (.099)+
Ethnicitya
Black versus White		−.852 (.918)
Asian versus White		.212 (.676)
Hispanic versus White		.310 (.332)
Ancestry informative markers principal componentsb
PC1		−.101 (.104)
PC2		−.051 (.088)
PC3		.004 (.039)
PW1 – linear		1.190 (.148)+
PW1 – quadratic		−.598 (.071)+
PW2 – linear		.533 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
PW1 – linear - NLRP5 rs471979		−3.292 (1.258)*
PW1 – quadratic - NLRP5 rs471979		1.323 (.595)*
Variance components
In intercept		3.566+
Goodness-of-fit deviance (parameters estimated)		11016.309635 (15)
Model comparison χ (df)		10.222 (2)*
Morning Fatigue and NLRP6 rs74044411		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.010 (.099)+
Ethnicitya
Black versus White		−.727 (.914)
Asian versus White		.229 (.672)
Hispanic versus White		.339 (.330)
Ancestry informative markers principal componentsb
PC1		−.128 (.103)
PC2		−.068 (.087)
PC3		.011 (.039)
PW1 – linear		1.186 (.148)+
PW1 – quadratic		−.597 (.071)+
PW2 – linear		.534 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
Intercept
PW1 – linear - NLRP6 rs74044411		−1.675 (.785)*
PW1 – quadratic - NLRP6 rs74044411		.539 (.373)
Variance components
In intercept		3.522+
Goodness-of-fit deviance (parameters estimated)		11010.712889 (15)
Model comparison χ (df)		15.819 (2)**
Morning Fatigue and NOD2 rs2076756		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.010 (.099)+
Ethnicitya
Black versus White		−.916 (.912)
Asian versus White		.280 (.671)
Hispanic versus White		.359 (.329)
Ancestry informative markers principal componentsb
PC1		−.096 (.103)
PC2		−.068 (.087)
PC3		.009 (.039)
PW1 – linear		1.192 (.148)+
PW1 – quadratic		−.599 (.071)+
PW2 – linear		.532 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
Intercept - NOD2 rs2076756		−.960 (.383)*
Variance components
In intercept		3.507+
Goodness-of-fit deviance (parameters estimated)		11020.293155 (14)
Model comparison χ (df)		6.239 (1)*
JAK/STAT Pathway Genes
Morning Fatigue and IL4R Liability Score		Coefficient (SE)
		Final Model
Fixed Effects
Intercept		3.010 (.099)+
Ethnicitya
Black versus White		−.772 (.917)
Asian versus White		.233 (.674)
Hispanic versus White		.336 (.331)
Ancestry informative markers principal componentsb
PC1		−.103 (.103)
PC2		−.061 (.088)
PC3		.012 (.039)
PW1 – linear		1.191 (.148)+
PW1 – quadratic		−.599 (.071)+
PW2 – linear		.532 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
Intercept
PW1 – linear - IL4R Liability Score		−.401(.168)*
PW1 – quadratic - IL4R Liability Score		.167 (.079)*
Variance components
In intercept		3.548+
Goodness-of-fit deviance (parameters estimated)		11019.542001 (15)
Model comparison χ (df)		6.990 (2)*
NFkB pathway Genes
Morning Fatigue and IL17RB Liability Score		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.011 (.099)+
Ethnicitya
Black versus White		−.843 (.913)
Asian versus White		.256 (.672)
Hispanic versus White		.324 (.330)
Ancestry informative markers principal componentsb
PC1		−.111 (.103)
PC2		−.037 (.088)
PC3		.013 (.039)
PW1 – linear		1.191 (.148)+
PW1 – quadratic		−.599 (.071)+
PW2 – linear		.531 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
Intercept - IL17RB Liability Score		−.352 (.162)*
Variance components
In intercept		3.519+
Goodness-of-fit deviance (parameters estimated)		11021.825332 (14)
Model comparison χ (df)		4.706 (1)*
Morning Fatigue and TNFRSF10A rs17620		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.010 (.099)+
Ethnicitya
Black versus White		−.873 (.916)
Asian versus White		.235 (.674)
Hispanic versus White		.330 (.331)
Ancestry informative markers principal componentsb
PC1		−.099 (.103)
PC2		−.073 (.088)
PC3		.012 (.039)
PW1 – linear		1.194 (.148)+
PW1 – quadratic		−.601 (.071)+
PW2 – linear		.535 (.096)+
PW2 – quadratic		−.154 (.031)+
Time invariant covariates
PW1 – linear - TNFRSF10A rs17620		.402 (.189)*
PW1 – quadratic - TNFRSF10A rs17620		−.193 (.093)*
PW2 – linear - TNFRSF10A rs17620		.197 (.127)
PW2 – quadratic - TNFRSF10A rs17620		−.051 (.041)
Variance components
In intercept		3.544+
Goodness-of-fit deviance (parameters estimated)		11018.710558 (17)
Model comparison χ (df)		7.821 (4)
Morning Fatigue and TNFRSF10D Liability Score		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.011 (.099)+
Ethnicitya
Black versus White		−.810 (.917)
Asian versus White		.203 (.675)
Hispanic versus White		.323 (.331)
Ancestry informative markers principal componentsb
PC1		−.106 (.104)
PC2		−.048 (.088)
PC3		.008 (.039)
PW1 – linear		1.195 (.148)+
PW1 – quadratic		−.601 (.071)+
PW2 – linear		.529 (.096)+
PW2 – quadratic		−.152 (.031)+
Time invariant covariates
PW2 – linear - TNFRSF10D Liability Score		−.211 (.094)*
PW2 – quadratic - TNFRSF10D Liability Score		.057 (.033)
Variance components
In intercept		3.552+
Goodness-of-fit deviance (parameters estimated)		11018.923385 (15)
Model comparison χ (df)		7.608 (2)*
Morning Fatigue and TNFRSF11A Liability Score		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.011 (.099)+
Ethnicitya
Black versus White		−.866 (.914)
Asian versus White		.172 (.673)
Hispanic versus White		.330 (.330)
Ancestry informative markers principal componentsb
PC1		−.107 (.103)
PC2		−.046 (.087)
PC3		.005 (.039)
PW1 – linear		1.193 (.148)+
PW1 – quadratic		−.600 (.071)+
PW2 – linear		.535 (.096)+
PW2 – quadratic		−.154 (.031)+
Time invariant covariates
PW1 – linear - TNFRSF11A Liability Score		−.381 (.141)*
PW1 – quadratic - TNFRSF11A Liability Score		.174 (.070)*
PW2 – linear - TNFRSF11A Liability Score		−.130 (.097)
PW2 – quadratic - TNFRSF11A Liability Score		.022 (.031)
Variance components
In intercept		3.532+
Goodness-of-fit deviance (parameters estimated)		11009.606255 (17)
Model comparison χ (df)		16.925 (4)*
Morning Fatigue and TNFRSF14 rs2234163		Coefficient (SE)
		Final Model
Fixed effects
Intercept		3.011 (.099)+
Ethnicitya
Black versus White		−.903 (.913)
Asian versus White		.112 (.673)
Hispanic versus White		.311 (.330)
Ancestry informative markers principal componentsb
PC1		−.104 (.103)
PC2		−.055 (.087)
PC3		.010 (.039)
PW1 – linear		1.192 (.148)+
PW1 – quadratic		−.599 (.071)+
PW2 – linear		.532 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
Intercept - TNFRSF14 rs2234163		1.426 (.638)*
Variance components
In intercept		3.514+
Goodness-of-fit deviance (parameters estimated)		11021.551571 (14)
Model comparison χ (df)		4.980 (1)*
Morning Fatigue and TNFRSF21 Liability Score		Coefficient (SE)
		Final Model
Intercept		3.012 (.099)+
Ethnicitya
Black versus White		−.815 (.913)
Asian versus White		.281 (.672)
Hispanic versus White		.310 (.330)
Ancestry informative markers principal componentsb
PC1		−.104 (.103)
PC2		−.069 (.088)
PC3		.001 (.039)
PW1 – linear		1.194 (.148)+
PW1 – quadratic		−.601 (.071)+
PW2 – linear		.531 (.096)+
PW2 – quadratic		−.153 (.031)+
Time invariant covariates
Intercept TNFRSF21 Liability Score		.416 (.183)*
Variance components
In intercept		3.517+
Goodness-of-fit deviance (parameters estimated)		11021.369941 (14)
Model comparison χ (df)		5.162 (1)*

^{Self-reported ethnicity – represented by three “dummy” coded variables}

^{Ancestry informative markers - represented by the first three PCs (i.e., PC1, PC2, PC3)}

^p<.05,

^p<.001,

^p<.0001

Abbreviations: df = degrees of freedom; IL4R = Interleukin 4 receptor; IL12B= Interleukin 12B; IL17RB = Interleukin 17 receptor B; JAK/STAT = Janus kinase/signal transducers and activators of transcription; NFkB = nuclear factor-kappa beta; NLRC5 = Nod-Like receptor family, caspase recruitment domain containing 5; NLRP5 = Nod-Like receptor family, pyrin domain containing 5; NLRP6 = Nod-Like receptor family, pyrin domain containing 6; NOD2 = Nucleotide-Binding oligomerization domain containing 2; PC=principal component; PW1 = piecewise 1 PW2 = piecewise 2; rs = reference SNP cluster identification number; SE = standard error; TNFA = Tumor necrosis factor alpha; TNFRSF10A = Tumor necrosis factor receptor superfamily member 10A; TNFRSF10D = Tumor necrosis factor receptor superfamily member 10D; TNFRSF11A = Tumor necrosis factor receptor superfamily member 11A; TNFRSF14 = Tumor necrosis factor receptor superfamily member 14; TNFRSF21 = Tumor necrosis factor receptor superfamily member 21

5.3 Genomic predictors of inter-individual differences in morning fatigue

Table 4 shows the final HLM models for morning fatigue. For the cytokine genes, two SNPs were associated with inter-individual differences in morning fatigue. Figure 1B illustrates the adjusted initial level of morning fatigue (i.e., intercept) based on a recessive model for IL12B rs3213094 (i.e., TT+TC vs. CC). Figure 1C illustrates the predicted changes in the trajectory of morning fatigue (i.e., slope) based on a dominant model for TNFA rs1041981 (i.e., CC vs. CA+AA).

For the inflammasome pathway genes, three SNPs were associated with changes in morning fatigue. Figure 1D illustrates the adjusted initial levels of morning fatigue based on the recessive model for nucleotide-binding oligomerization domain containing 2 (NOD2) rs2076756 (i.e., AA+AG vs. GG). Figures 1E and 1F respectively, illustrate the predicted changes in the trajectories of morning fatigue based on the recessive model for non-like receptor family, pyrin containing domain 5 (NLRP5) rs471979 (i.e., GG+GC vs. CC) and a dominant model for NLRP6 rs74044411 (i.e., TT vs. TC+CC).

For the JAK/STAT pathway, the liability score for all IL4R, was associated with inter-individual differences in the slope of morning fatigue (Figure 2A).

Figures 2

A–D – Influence of the liability score for the sum of the occurrence of the rare alleles in IL4R on the slope parameters for morning fatigue (A). Influence of the dominant model (AA vs. AG+GG) of the rare G allele for TNFRSF14 rs2234163 on the inter-individual differences in the intercept for morning fatigue (B). Influence of the liability score for the sum of the occurrence of the rare alleles in IL17RB, on the inter-individual differences in the intercept for morning fatigue (C). Influence of the liability score for the sum of the occurrence of the rare alleles in TNFRSF21 on the inter-individual differences in the intercept for morning fatigue (D).

For the NFkB pathway genes, two SNPs and four liability scores were associated with inter-individual differences in morning fatigue. Figure 2B illustrates the adjusted initial level of morning fatigue based on the dominant model for tumor necrosis factor receptor super family, member 14 (TNFRSF14) rs2234163 (i.e., AA vs. AG+GG). Figures 2C and 2D respectively, illustrate the adjusted initial levels of morning fatigue based on the liability scores for IL17RB and TNFRSF21. Figures 3A, 3B, and 3C, respectively, illustrate the predicted changes in the trajectories of morning fatigue for the additive model for TNFRSF10A rs17620 (i.e., TT vs. TC vs. CC) and the liability scores for TNFRSF10D and TNFRSF11A.

Figures 3

A–C – Influence of the additive model (TT vs. TC vs. CC) of the rare C allele for TNFRSF10A rs17620 on the slope parameters for morning fatigue (A). Influence of the liability score for the sum of the occurrence of the rare alleles in TNFRSF10D on the slope parameters for morning fatigue (B). Influence of the liability score for the sum of the occurrence of the rare alleles in TNFRSF11A on the slope parameters for morning fatigue (C).

5.4 Changes in evening fatigue severity

HLM was used to examine how evening fatigue scores changed within the two cycles of CTX, controlling for self-reported and genomic estimates of race and ethnicity. The estimates for the initial piecewise model are presented in Table 5. Since the model was unconditional (i.e., no covariates), the intercept represents the average evening fatigue severity score at enrollment (i.e., 5.310 on a 0 to 10 scale). The estimated linear piecewise rates of change were 0.601 and 0.394 (both, p<.0001) for piecewise linear 1 and piecewise linear 2, respectively. The estimated quadratic piecewise rates of change were −.306 and −.113 (both, p<.0001) for piecewise quadratic 1 and piecewise quadratic 2, respectively. Figure 4A displays the unconditional model for mean evening fatigue scores over two cycles of CTX.

Figures 4

A–D – Unconditional piecewise model of mean evening fatigue scores for six assessment points over two cycles of chemotherapy (A). Influence of the recessive model (TT+TC vs. CC) of the rare C allele in IL12B rs3213094 on the inter-individual differences in the intercept for evening fatigue (B). Influence of the liability score for the sum of the occurrence of the rare alleles in IL12B on the inter-individual differences in the intercept for evening fatigue (C). Influence of the dominant model (CC vs. CA+AA) of the rare A allele in TNFA rs1041981 on the slope parameters for evening fatigue (D).

Table 5

Hierarchical Linear Models for Evening Fatigue

Cytokine Genes
Evening Fatigue and IL12B rs3213094	Coefficient (SE)
	Unconditional Model	Final Model
Fixed effects
Intercept	5.310 (.090)+	5.307 (.090)+
Ethnicitya
Black versus White	−1.584 (.824)	−1.501 (.822)
Asian versus White	−.269 (.608)	−.231 (.606)
Hispanic versus White	−.348 (.298)	−.332 (.297)
Ancestry informative markers principal components
PC1	−.070 (.093)	−.072 (.093)
PC2	−.044 (.079)	−.029 (.079)
PC3	−.009 (.035)	−.008 (.035)
PW 1 – linear	.601 (.137)+	.601 (.137)+
PW 1 – quadratic	−.306 (.066)+	−.306 (.066)+
PW 2 – linear	.394 (.088)+	.394 (.088)+
PW 2 – quadratic	−.113 (.028)+	−.113 (.028)+
Time invariant covariates
Intercept IL12B rs3213094		−.573 (.288)*
Variance components
In intercept	2.867+	2.843+
Goodness-of-fit deviance (parameters estimated)	10399.632349 (13)+	10395.680232 (14)
Model comparison χ (df)		3.952 (1)*
Evening Fatigue and IL12B Liability Score		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.308 (.090)+
Ethnicitya
Black versus White		−1.579 (.820)
Asian versus White		−.322 (.605)
Hispanic versus White		−.347 (.297)
Ancestry informative markers principal componentsb
PC1		−.086 (.093)
PC2		−.021 (.079)
PC3		−.003 (.035)
PW 1 – linear		.600 (.137) +
PW 1 – quadratic		−.306 (.066) +
PW 2 – linear		.393 (.088) +
PW 2 – quadratic		−.112 (.028) +
Time invariant covariates
Intercept - IL12B Liability Score		−.258 (.115)*
Variance components
In intercept		2.838+
Goodness-of-fit deviance (parameters estimated)		10394.643652 (14)
Model comparison χ (df)		4.989 (1)*
Evening Fatigue and TNFA rs1041981		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.309 (.090)+
Ethnicitya
Black versus White		−1.548 (.822)
Asian versus White		−.293 (.606)
Hispanic versus White		−.369 (.298)
Ancestry informative markers principal componentsb
PC1		−.073 (.093)
PC2		−.038 (.079)
PC3		−.002 (.035)
PW 1 – linear		.596 (.137)+
PW 1 – quadratic		−.304 (.066)+
PW 2 – linear		.395 (.088)+
PW 2 – quadratic		−.113 (.028)+
Time invariant covariates
Intercept
PW 1 – linear - TNFA rs1041981		−.602 (.256)*
PW 1 – quadratic - TNFA rs1041981		.252 (.121)*
Variance components
In intercept		2.852+
Goodness-of-fit deviance (parameters estimated)		10392.936781 (15)
Model comparison χ (df)		6.696 (2)*
Inflammasome Pathway Genes
Evening Fatigue and CARD6 rs10512747		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.309 (.089)+
Ethnicitya
Black versus White		−1.591 (.819)
Asian versus White		−.350 (.605)
Hispanic versus White		−.397 (.297)
Ancestry informative markers principal componentsb
PC1		−.057 (.093)
PC2		−.043 (.078)
PC3		−.008 (.035)
PW 1 – linear		.599 (.137)+
PW 1 – quadratic		−.306 (.066)+
PW 2 – linear		.394 (.088)+
PW 2 – quadratic		−.113 (.028)+
Time invariant covariates
Intercept - CARD6 rs10512747		−.441 (.176)*
Variance components
In intercept		2.829+
Goodness-of-fit deviance (parameters estimated)		10393.393966 (14)
Model comparison χ (df)		6.238 (1)*
Evening Fatigue and NLRP4 rs17857373		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.308 (.090)+
Ethnicitya
Black versus White		−1.600 (.821)
Asian versus White		−.298 (.605)
Hispanic versus White		−.365 (.297)
Ancestry informative markers principal componentsb
PC1		−.061 (.093)
PC2		−.046 (.079)
PC3		−.003 (.035)
PW 1 – linear		.601 (.137)+
PW 1 – quadratic		−.306 (.066)+
PW 2 – linear		.394 (.088)+
PW 2 – quadratic		−.113 (.028)+
Time invariant covariates
Intercept - NLRP4 rs17857373		−.603 (.291)*
Variance components
In intercept		2.841+
Goodness-of-fit deviance (parameters estimated)		10395.340962 (14)
Model comparison χ (df)		4.291 (1)*
Evening Fatigue and NLRP6 rs74044411		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.309 (.090)+
Ethnicitya
Black versus White		−1.491 (.822)
Asian versus White		−.257 (.606)
Hispanic versus White		−.339 (.297)
Ancestry informative markers principal componentsb
PC1		−.093 (.093)
PC2		−.060 (.079)
PC3		−.007 (.035)
PW 1 – linear		.598 (.136)+
PW 1 – quadratic		−.306 (.066)+
PW 2 – linear		.395 (.088)+
PW 2 – quadratic		−.113 (.028)+
Time invariant covariates
Intercept
PW 1 – linear - NLRP6 rs74044411		−1.417 (.682)*
PW 1 – quadratic - NLRP6 rs74044411		.449 (.322)
Variance components
In intercept		2.847+
Goodness-of-fit deviance (parameters estimated)		10385.256709 (15)
Model comparison χ (df)		14.376 (2)**
Evening Fatigue and NOD2 rs2076756		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.308 (.089)+
Ethnicitya
Black versus White		−1.684 (.813)*
Asian versus White		−.177 (.599)
Hispanic versus White		−.304 (.294)
Ancestry informative markers principal componentsb
PC1		−.061 (.092)
PC2		−.067 (.078)
PC3		−.008 (.035)
PW 1 – linear		.601 (.137)+
PW 1 – quadratic		−.306 (.066)+
PW 2 – linear		.394 (.088)+
PW 2 – quadratic		−.113 (.028)+
Time invariant covariates
Intercept - NOD2 rs2076756		−1.350 (.342)+
Variance components
In intercept		2.777+
Goodness-of-fit deviance (parameters estimated)		10384.268252 (14)
Model comparison χ (df)		15.364 (1)+
MAP/JNK Pathway Genes
Evening Fatigue and IL17RD rs61742267		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.310 (.090)+
Ethnicitya
Black versus White		−1.539 (.821)
Asian versus White		−.265 (.605)
Hispanic versus White		−.319 (.297)
Ancestry informative markers principal componentsb
PC1		−.059 (.093)
PC2		−.044 (.079)
PC3		−.013 (.035)
PW 1 – linear		.601 (.137)+
PW 1 – quadratic		−.306 (.066)+
PW 2 – linear		.393 (.088)+
PW 2 – quadratic		−.112 (.028)+
Time invariant covariates
Intercept - IL17RD rs61742267		−.933 (.429)*
Variance components
In intercept		2.840+
Goodness-of-fit deviance (parameters estimated)		10394.915944 (14)
Model comparison χ (df)		4.716 (1)*
NfKB Pathway Genes
Evening Fatigue and IL17RB rs2232346 and IL17RB rs1043261		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.308 (.089)+
Ethnicitya
Black versus White		−1.573 (.817)
Asian versus White		−.179 (.603)
Hispanic versus White		−.344 (.296)
Ancestry informative markers principal componentsb
PC1		−.065 (.092)
PC2		−.037 (.078)
PC3		−.009 (.035)
PW 1 – linear		.600 (.137) +
PW 1 – quadratic		−.306 (.066) +
PW 2 – linear		.394 (.088) +
PW 2 – quadratic		−.113 (.028) +
Time invariant covariates
Intercept - IL17RB rs2232346		−.791 (.388)*
Intercept - IL17RB rs1043261		−1.456 (.584)*
Variance components
In intercept		2.809+
Goodness-of-fit deviance (parameters estimated)		10389.389046 (15)
Model comparison χ (df)		10.243 (2)*
Evening Fatigue and LTBR Liability Score		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.309 (.090)+
Ethnicitya
Black versus White		−1.562 (.826)
Asian versus White		−.277 (.609)
Hispanic versus White		−.346 (.299)
Ancestry informative markers principal componentsb
PC1		−.072 (.093)
PC2		−.044 (.079)
PC3		−.010 (.035)
PW 1 – linear		.604 (.136)+
PW 1 – quadratic		−.308 (.066)+
PW 2 – linear		.396 (.088)+
PW 2 – quadratic		−.113 (.028)+
Time invariant covariates
Intercept
PW 1 – linear - LTBR Liability Score		−3.188 (1.080)*
PW 1 – quadratic - LTBR Liability Score		1.540 (.513)*
Variance components
In intercept		2.885+
Goodness-of-fit deviance (parameters estimated)		10390.660476 (15)
Model comparison χ (df)		8.972 (2)*
Evening Fatigue and TNFRSF14 rs2234163		Coefficient (SE)
		Final Model
Fixed effects
Intercept		5.309 (.090)+
Ethnicitya
Black versus White		−1.648 (.818)
Asian versus White		−.383 (.605)
Hispanic versus White		−.366 (.296)
Ancestry informative markers principal componentsb
PC1		−.072 (.092)
PC2		−.048 (.078)
PC3		−.007 (.035)
PW 1 – linear		.600 (.137)+
PW 1 – quadratic		−.306 (.066)+
PW 2 – linear		.394 (.088)+
PW 2 – quadratic		−.113 (.028)+
Time invariant covariates
Intercept - TNFRSF14 rs2234163		1.590 (.572)*
Variance components
In intercept		2.821+
Goodness-of-fit deviance (parameters estimated)		10391.966279 (14)
Model comparison χ (df)		7.666 (1)*

^{Self-reported ethnicity – represented by three “dummy” coded variables}

^{Ancestry informative markers - represented by the first three PCs (i.e., PC1, PC2, PC3)}

^p<.05,

^p<.001,

^p<.0001

Abbreviations: CARD6 = Caspase recruitment domain family member 6; df = degrees of freedom; IL12B= Interleukin 12B; IL17RB = Interleukin 17 receptor B; IL17RD = Interleukin 17 receptor D; JAK = jaunts kinase transducers and activators of transcription; LTBR = Lymphotoxin beta receptor; MAP = mitogen-activated protein kinase; NFkB = nuclear factor-kappa beta; NLRP4 = Nod-Like receptor family, pyrin domain containing 4; NLRP6 = Nod-Like receptor family, pyrin domain containing 6; NOD2 = Nucleotide-Binding oligomerization domain containing 2; PC=principal component; PW1 = piecewise 1; PW2 = piecewise 2; rs = reference SNP cluster identification number; SE = standard error; TNFA = Tumor necrosis factor alpha; TNFRSF14 = Tumor necrosis factor receptor superfamily member 14.

5.5 Genomic predictors of inter-individual differences in evening fatigue

Table 5 shows the final HLM models for evening fatigue. For the cytokine genes, two SNPs and one liability score were associated with inter-individual differences in evening fatigue. Figures 4B and 4C respectively, illustrate the adjusted initial level of evening fatigue based on the recessive model for IL12B rs3213094 (i.e., TT+TC vs. CC) and the liability score for IL12B. Figure 4D illustrates the predicted changes in the trajectories of evening fatigue for a dominant model for TNFA rs1041981 (i.e., CC vs. CA+AA).

For the inflammasome pathway genes, four SNPs were associated with evening fatigue. Figures 5A, 5B, and 5C respectively, illustrate the adjusted initial levels of evening fatigue based on the additive model for capsase recruitment domain family member 6 (CARD6) rs10512747 (i.e., TT vs. TC vs. CC); the dominant model for NLRP4 rs17857373 (i.e., CC vs. CG+GG) and the recessive model for NOD2 rs2076756 (i.e., AA+AG vs. GG). Figure 5D illustrates the predicted changes in the trajectory of evening fatigue based on the dominant model for NLRP6 rs74044411 (i.e., TT vs. TC+CC).

Figures 5

A–E – Influence of the additive model (TT vs. TC vs. CC) of the rare C allele in CARD6 rs10512747 on the inter-individual differences in the intercept for evening fatigue (A). Influence of the dominant model (CC vs. CG+GG) of the rare G allele for NLRP4 rs17857373 on the inter-individual differences in the intercept for evening fatigue (B). Influence of the recessive model (AA+AG vs. GG) of the rare G allele for NOD2 rs2076756 on the inter-individual differences in the intercept for evening fatigue (C). Influence of the dominant model (TT vs. TC+CC) of the rare C allele for NLRP6 rs74044411 on the slope parameters for evening fatigue (D). Influence of the dominant model (AA vs. AG+GG) for the rare G allele for IL17RD rs61742267 on the inter-individual differences in the intercept for evening fatigue (E).

For the MAP/JNK pathway genes, Figure 5E illustrates the adjusted initial level of evening fatigue based on the dominant model for IL17RD rs61742267 (i.e., AA vs. AG+GG).

For the NFkB pathway genes, three SNPs and one liability score were associated with inter-individual differences in evening fatigue. Figure 6A illustrates the adjusted initial level of evening fatigue based on the dominant model for IL17RB rs2232346 (i.e., TT vs. TC+CC) and the recessive model for IL17RB rs1043261 (i.e., TT+TC vs. CC). Figures 6B and 6C respectively, illustrate the adjusted initial level of evening fatigue based on the dominate model for TNFRSF14 rs2234163 (i.e., AA vs. AG+GG) and the predicted change in the trajectory of evening fatigue based on the liability score for lymphotoxin beta receptor (LTBR).

Figures 6

A–C – Influence of the dominant model (TT vs. TC+CC) of the rare C allele for IL17RB rs2232346 and the recessive model (TT+TC vs. CC) of the rare C allele for IL17RB rs1043261 on the inter-individual differences in the intercept for evening fatigue (A). Influence of the dominant model (AA vs. AG+GG) of the rare G allele for TNFRSF14 rs2234163 on the inter-individual differences in the intercept for evening fatigue (B). Influence of the liability score for the sum of the occurrence of the rare alleles in LTBR on the slope parameters for evening fatigue (C).

5.1 Sample characteristics

As summarized in Table 3, of the 543 patients in the study, the majority of the patients were female, white, diagnosed with breast cancer, and were treated with CTX on a 21-day cycle. Most patients were well educated, married or partnered, and currently not employed. At enrollment, patients reported clinically meaningful sleep disturbance and anxiety levels. Morning and evening fatigue scores at enrollment were just below the cutoff for clinically meaningful levels (i.e., ≥3.2 for morning fatigue, ≥5.6 for evening fatigue) [70].

5.2 Changes in morning fatigue severity

HLM was used to examine how morning fatigue scores changed within the two cycles of CTX, controlling for self-reported and genomic estimates of race and ethnicity. The estimates for the initial piecewise model are presented in Table 4. Since the model was unconditional (i.e., no covariates included in the model), the average morning fatigue severity score at enrollment (i.e., 3.011 on a 0 to 10 scale) represents the intercept. The estimated linear piecewise rates of change were 1.192 and 0.532 (both p<.0001) for piecewise linear 1 and piecewise linear 2, respectively. The estimated quadratic piecewise rates of change were −.599 and −.153 (both p<.0001) for piecewise quadratic 1 and piecewise quadratic 2, respectively. Figure 1A displays the unconditional model for mean morning fatigue scores over the two cycles of CTX.

Figures 1

A–F – Unconditional piecewise model of mean morning fatigue scores for six assessment points over two cycles of chemotherapy (A). Influence of the recessive model (TT+TC vs. CC) of the rare C allele in IL12B rs3213094 on the inter-individual differences in the intercept for morning fatigue (B). Influence of the dominant model (CC vs. CA+AA) of the rare A allele in TNFA rs1041981 on the slope parameters for morning fatigue (C). Influence of the recessive model (AA+AG vs. GG) of the rare G allele in NOD2 rs2076756 on the inter-individual differences in the intercept for morning fatigue (D). Influence of the recessive model (GG+GC vs. CC) of the rare C allele in NLRP5 rs471979 on the slope parameters for morning fatigue (E). Influence of the dominant model (TT vs. TC+CC) of the rare C allele in NLRP6 rs74044411 on the slope parameters for morning fatigue (F).

5.3 Genomic predictors of inter-individual differences in morning fatigue

Table 4 shows the final HLM models for morning fatigue. For the cytokine genes, two SNPs were associated with inter-individual differences in morning fatigue. Figure 1B illustrates the adjusted initial level of morning fatigue (i.e., intercept) based on a recessive model for IL12B rs3213094 (i.e., TT+TC vs. CC). Figure 1C illustrates the predicted changes in the trajectory of morning fatigue (i.e., slope) based on a dominant model for TNFA rs1041981 (i.e., CC vs. CA+AA).

For the inflammasome pathway genes, three SNPs were associated with changes in morning fatigue. Figure 1D illustrates the adjusted initial levels of morning fatigue based on the recessive model for nucleotide-binding oligomerization domain containing 2 (NOD2) rs2076756 (i.e., AA+AG vs. GG). Figures 1E and 1F respectively, illustrate the predicted changes in the trajectories of morning fatigue based on the recessive model for non-like receptor family, pyrin containing domain 5 (NLRP5) rs471979 (i.e., GG+GC vs. CC) and a dominant model for NLRP6 rs74044411 (i.e., TT vs. TC+CC).

For the JAK/STAT pathway, the liability score for all IL4R, was associated with inter-individual differences in the slope of morning fatigue (Figure 2A).

Figures 2

A–D – Influence of the liability score for the sum of the occurrence of the rare alleles in IL4R on the slope parameters for morning fatigue (A). Influence of the dominant model (AA vs. AG+GG) of the rare G allele for TNFRSF14 rs2234163 on the inter-individual differences in the intercept for morning fatigue (B). Influence of the liability score for the sum of the occurrence of the rare alleles in IL17RB, on the inter-individual differences in the intercept for morning fatigue (C). Influence of the liability score for the sum of the occurrence of the rare alleles in TNFRSF21 on the inter-individual differences in the intercept for morning fatigue (D).

For the NFkB pathway genes, two SNPs and four liability scores were associated with inter-individual differences in morning fatigue. Figure 2B illustrates the adjusted initial level of morning fatigue based on the dominant model for tumor necrosis factor receptor super family, member 14 (TNFRSF14) rs2234163 (i.e., AA vs. AG+GG). Figures 2C and 2D respectively, illustrate the adjusted initial levels of morning fatigue based on the liability scores for IL17RB and TNFRSF21. Figures 3A, 3B, and 3C, respectively, illustrate the predicted changes in the trajectories of morning fatigue for the additive model for TNFRSF10A rs17620 (i.e., TT vs. TC vs. CC) and the liability scores for TNFRSF10D and TNFRSF11A.

Figures 3

A–C – Influence of the additive model (TT vs. TC vs. CC) of the rare C allele for TNFRSF10A rs17620 on the slope parameters for morning fatigue (A). Influence of the liability score for the sum of the occurrence of the rare alleles in TNFRSF10D on the slope parameters for morning fatigue (B). Influence of the liability score for the sum of the occurrence of the rare alleles in TNFRSF11A on the slope parameters for morning fatigue (C).

5.4 Changes in evening fatigue severity

HLM was used to examine how evening fatigue scores changed within the two cycles of CTX, controlling for self-reported and genomic estimates of race and ethnicity. The estimates for the initial piecewise model are presented in Table 5. Since the model was unconditional (i.e., no covariates), the intercept represents the average evening fatigue severity score at enrollment (i.e., 5.310 on a 0 to 10 scale). The estimated linear piecewise rates of change were 0.601 and 0.394 (both, p<.0001) for piecewise linear 1 and piecewise linear 2, respectively. The estimated quadratic piecewise rates of change were −.306 and −.113 (both, p<.0001) for piecewise quadratic 1 and piecewise quadratic 2, respectively. Figure 4A displays the unconditional model for mean evening fatigue scores over two cycles of CTX.

Figures 4

A–D – Unconditional piecewise model of mean evening fatigue scores for six assessment points over two cycles of chemotherapy (A). Influence of the recessive model (TT+TC vs. CC) of the rare C allele in IL12B rs3213094 on the inter-individual differences in the intercept for evening fatigue (B). Influence of the liability score for the sum of the occurrence of the rare alleles in IL12B on the inter-individual differences in the intercept for evening fatigue (C). Influence of the dominant model (CC vs. CA+AA) of the rare A allele in TNFA rs1041981 on the slope parameters for evening fatigue (D).