The Psychoneurologic Symptom Cluster and Its Association With Breast Cancer Genomic Instability

Severe symptoms are associated with breast cancer and its treatment and can lead to decreased quality of life (Hamer et al., 2017). Between 28% and 52% of individuals with metastatic breast cancer experience the psychoneurologic (PN) symptom cluster, which includes pain, fatigue, sleep disturbance, anxiety, depression symptoms, and changes in cognitive function (Albusoul et al., 2017; Niklasson et al., 2017; Starkweather et al., 2013; Sullivan et al., 2018). A systematic review of multiple studies indicates that the PN symptom cluster occurs across the treatment trajectory (So et al., 2021). These symptoms may be present at the time of diagnosis or before the initiation of therapy, and pretreatment symptom severity may be predictive of the symptom experience throughout treatment (Fox et al., 2020; Grayson, Sereika, et al., 2023; Li et al., 2020; Tometich et al., 2019). Precision health research will enable the prediction of patients who are at highest risk for experiencing the PN symptom cluster and explain underlying biologic mechanisms. Accurate symptom prognostication may allow for more proactive symptom interventions, as opposed to reactive treatments once individuals report symptoms.

The National Cancer Institute (n.d.) defines genomic instability as the tendency of cancer cells to accumulate mutations and gross genomic anomalies such as chromosome number aberrations. Significant interactions exist between cancer genomic instability and systemic inflammation (Hanahan & Weinberg, 2011; Kawanishi et al., 2017). Inflammation, in turn, has been associated with the severity and occurrence of PN symptoms (Jehn et al., 2012; Ji et al., 2017; Khosravi et al., 2019; Mark et al., 2017; Schmidt et al., 2016). This leads to the hypothesis that cancer genomic instability may be predictive of the PN symptom cluster (Grayson, Cummings, et al., 2023). Associations have been observed between tumor features and cognitive performance in individuals with breast cancer (Koleck et al., 2017). In addition, KRAS variants in colorectal cancer have been associated with depression symptoms in older adults with colorectal cancer (Zhou et al., 2016), supporting that cancer genomics may be associated with symptoms. However, the relationship between cancer genomic variation, including genomic instability, and a broad range of PN symptoms remains largely unexamined. One aim of this study was to investigate the relationship between breast cancer genomic instability and PN symptoms.

Small fragments of cell-free DNA from cancer cells circulating in the bloodstream, known as circulating tumor DNA (ctDNA), show promise as a marker of cancer progression with the potential ability to enable selection of targeted therapies, monitor cancer therapeutic efficacy, and evaluate prognoses without invasive serial biopsies (Bettegowda et al., 2014; Crowley et al., 2013; Dawson et al., 2013; Diehl et al., 2008; Forshew et al., 2012; Murtaza et al., 2013; O’Leary et al., 2018; Phallen et al., 2017; Shen et al., 2018; Wang et al., 2016). Genomic alterations in tumor cells, such as DNA copy number alterations (CNAs), can be detected through ultra-low-pass whole-genome sequencing (ULP-WGS) of ctDNA (Chan et al., 2013; Dong et al., 2016; Heitzer et al., 2013; Homburger et al., 2019). In this study, data on CNAs from ctDNA as markers of cancer genomic instability were used to enable prediction and prognostication of PN symptoms based on tumor genomic factors that were previously available only via an invasive biopsy.

The purpose of this study was to phenotype the PN symptom cluster in individuals with metastatic breast cancer and associate PN symptom cluster phenotypes with cancer genomic instability as measured by ULP-WGS of ctDNA. In addition, this study served to investigate the relationship between markers of breast cancer genomic instability in ctDNA using ULP-WGS and PN symptoms. Blood was drawn at the time of recruitment for the parent study for ULP-WGS of ctDNA. Symptom data were collected from the entire cohort (N = 206), and genomic instability data were available from the parent study on a randomly selected subset of participants (N = 20), with subset sample size determined based on feasibility.

Methods and Variables

Sample and Setting

An observational, cross-sectional design was used to investigate PN symptom phenotypes and their relationship with cancer genomic instability in individuals with metastatic breast cancer. This study uses data from the ongoing parent study, Breast Disease Research Repository: Tissue and Bodily Fluid and Medical Information Acquisition Protocol (04-162), with the purpose to investigate the utility of ctDNA (via analysis with ULP-WGS and other methods) in monitoring progression of metastatic breast cancer and predicting survival. Participants were recruited to the parent study according to approved protocols at the time of their advanced breast cancer diagnosis and/or progression from nonmetastatic disease at the University of Pittsburgh Medical Center Magee-Womens Hospital in Pennsylvania. Participants were excluded from the parent study if the initiation of a new therapy or treatment occurred before consent and blood sample could be obtained. All participants recruited to the parent study at outset of this analysis were included.

PN Symptoms

Patient-reported measures of PN symptoms were collected using the MD Anderson Symptom Inventory (MDASI) via tablet at every clinic visit as part of an initiative by the advanced practice nurses to consistently measure symptoms. These symptom data were abstracted from the electronic health record by trained study staff for the clinic visit corresponding with the date of the first blood drawn after recruitment.

Symptom occurrence and severity were measured via the MDASI, an established patient-reported outcome instrument used to assess the severity and interference of cancer-related symptoms. The MDASI has been shown to be psychometrically valid in patients with breast cancer and in populations with other cancer types (Cleeland et al., 2000; Mendoza et al., 2013). For each symptom, participants were asked to rate symptoms at their worst in the past 24 hours from 0 (not present) to 10 (as bad as you can imagine). Subscale scores can be generated using the MDASI severity mean ratings. Although the MDASI assesses the severity of 13 core cancer-related symptoms, as well as selected symptoms from the MDASI Symptom Library, individual symptom ratings and subsets of symptoms may be used if specified a priori (Cleeland, 2009). As such, MDASI scores were collected for the six severity scales capturing the following six symptoms included in the PN symptom cluster: pain, fatigue, sleep disturbance, depression symptoms, anxiety, and decreased cognitive function.

Genomic Instability

Blood was collected at the time of study recruitment and immediately processed to plasma by the Pitt Biospecimen Core. ULP-WGS was performed at a core facility at the Broad Institute of the Massachusetts Institute of Technology and Harvard University. ctDNA was isolated from plasma. DNA input for library preparation was normalized to 25–50 ng, and library preparation was performed using KAPA HyperPrep Kit and Integrated DNA Technologies^® xGen Duplex Seq Adapters. Unique eight-base dual index sequences embedded within the p5 and p7 primers (purchased from Integrated DNA Technologies) were added during polymerase chain reaction testing. Library quantification was performed using the Invitrogen Quant-It™ broad range double-stranded DNA quantification assay kit. Each library was normalized to a concentration of 35 ng/mcl, and as many as 95 ULP-WGS samples were pooled together using equivolume pooling. The pool was quantified via quantitative polymerase chain reaction testing and normalized to the appropriate concentration to proceed to sequencing on a NovaSeq 6000. Data were analyzed using BamQC and Qualimap for quality control of the sequencing data. Raw data were aligned to reference genome (hg19) using BWA. CNVkit was used for gene annotation during CNA calling. CNAs were categorized as no change, amplification, or deletion.

In this analysis, CNA of the following six relevant oncogenes and tumor suppressor genes were included: TP53, MYC, erythroblastic leukemia viral oncogene homologue 2, ESR1, epidermal growth factor receptor, and KRAS. These genes were chosen because of their potential to be drivers of genomic instability in cancer. Gene functions and roles can be found in Table 1. Loss of tumor suppressor genes (deletion) can lead to decreased ability to respond to DNA damage. Amplification of oncogenes allows for clonal expansion and additional genomic changes through continued replication and clonal evolution (Kontomanolis et al., 2020). Because breast cancer genomic instability is the concept of interest, CNA of these oncogenes and tumor suppressor genes is an appropriate variable to represent the potential for the ongoing genomic changes in breast cancer cells that characterize genomic instability.

Individual, Disease, and Treatment Characteristics

A large range of potential confounding variables were abstracted from the electronic health record as part of the parent study to describe the sample characteristics. Variables included body mass index (BMI), age, menopausal status, smoking history, alcohol history, treatment history, and disease characteristics. Treatment history included dichotomous variables to indicate whether the participant received chemotherapy, radiation therapy, hormone therapy, and/or immunotherapy during the course of their treatment.

Disease characteristics included estrogen receptor, progesterone receptor, and HER2 status, as well as tumor type and location. Sites of metastases were recorded as binary variables marked as positive or negative for metastases to bone, central nervous system, lung, liver, and lymph node, because these are the most commonly seen metastatic sites in this cohort. A single category for other metastases sites was also indicated as positive or negative.

Statistical Analysis

Data screening: All data analyses were performed using IBM SPSS Statistics, version 27.0. Statistical tests with a p value less than 0.05 were considered statistically significant. Missingness was screened for using frequency tables. Because of low missingness occurring at random, cases with missing data were excluded listwise (n = 4). Means and SDs were calculated for the entire sample for continuous variables, and frequencies and percentages were calculated for categorical variables to describe the sample. An overall PN symptom severity score was calculated as a mean of all six symptom subscales (pain, fatigue, sleep disturbance, anxiety, depression symptoms, and decreases in cognitive function) to aid in sample description because the PN symptom cluster was hypothesized to be unidimensional.

Histograms of symptom severity data were generated and visually inspected, along with normality statistics, to assess data distribution. Because of a significant floor effect in symptom data, statistical methods for symptom phenotyping that were robust to non-normality were selected. One multivariate outlier was identified based on Mahalanobis distance on the six symptom severity variables and excluded based on a cutoff from a chi-square distribution with six degrees of freedom. This omission resulted in a final sample size of 201 participants included in the analysis, with 19 participants in the subsample with ULP-WGS data.

Factor analysis: Exploratory factor analysis was used to characterize the clustering of the severity of the six included self-reported symptoms: pain, fatigue, sleep disturbance, anxiety, depression symptoms, and decreases in cognitive function. For exploratory factor analysis, a factor loading of less than 0.4 was considered strong loading, and eigenvalues were used to determine the number of factors to be extracted. Principal axis factoring was used as the extraction method because of its robustness to violations of normality. When more than one factor was extracted, varimax rotation with Kaiser normalization was applied.

Hierarchical clustering: In addition, hierarchical cluster analysis by participant using the six symptom severity scores was used to generate clusters of participants representing different symptom phenotypes. This was completed using Ward’s method and squared Euclidean distance because of its robustness to noise and outliers. Visual analysis of the dendrogram was used to determine the optimal number of clusters. Factor analysis and hierarchical clustering were performed using the data from the larger sample, excluding patients with missing data and multivariate outliers (N = 201).

Disease, treatment, and individual characteristics: Disease, treatment, and individual characteristics were described for identified symptom phenotype groups. Multiple logistic regression, with symptom phenotype based on hierarchical cluster analysis as the dependent variable, was then performed for all collected disease, treatment, and individual characteristic variables to identify potential covariates and confounders individually. Odds ratios (ORs) with 95% confidence intervals (CIs) were generated for all potential covariates’ and confounders’ prediction of symptom phenotype group membership. Variables found to be significant predictors of the symptom phenotype group were then evaluated jointly.

To investigate potential differences in the subsample of randomly selected participants who had ULP-WGS data (n = 19) and those who were not selected for ULP-WGS analysis (n = 182), independent sample t tests were performed on all continuous type variables, including symptom severity data. In addition, chi-square independence of variance tests were performed to investigate potential between-group differences on all categorical variables, including symptom phenotype groupings resulting from application of hierarchical cluster analysis.

CNA of clinically relevant genes: In the subsample with ULP-WGS data (n = 19), binary logistic regression was performed with symptom phenotype as the dependent variable, with CNA of each of the six selected genes as predictors. Confounding variables and covariates identified in prior regression with the whole sample were controlled for when investigating CNA of selected genes. The assumption of linearity of the logit for continuous type predictor variables (e.g., age, BMI) was assessed using a Box–Tidwell procedure, and any potential multicollinearity was screened for using variance inflation factors. Because of the low number of participants with ULP-WGS data and the novelty of these data, CNA of selected genes by PN symptom phenotype were also described.

Results

Sample Characteristics

The final sample consisted of 201 individuals with metastatic breast cancer aged an average of 50.17 years (SD = 11.15), with an average BMI of 28.7 kg/m² (SD = 7.28). The majority of the sample was White (n = 172, 86%), non-Hispanic (n = 200, 99%), and diagnosed with ductal carcinoma (n = 159, 79%). The most common sites of metastasis in the sample were liver (n = 72, 36%) and bone (n = 68, 34%). All individual symptoms had severity scores ranging from 0 to 10. The symptom with the highest mean severity score was fatigue (mean = 4, SD = 2.93), and the symptom with the lowest mean severity score was depression symptoms (mean = 2, SD = 2.68).

Factor Analysis

Exploratory factor analysis of the six symptom severity scores extracted a single underlying factor when an eigenvalue greater than 1 was used as the criterion for the number of factors extracted. Factor loadings from this exploratory factor analysis can be found in Table 2. The single extracted factor explained 70% of variance with an eigenvalue of 3.7.

However, when the eigenvalue criteria were relaxed to allow values less than 1, two factors were extracted. Factor loadings after varimax rotation with Kaiser normalization can be found in Table 3. These factor loadings describe a first factor underlying pain, fatigue, and sleep disturbance, as well as second underlying depression symptoms, anxiety, and decreases in cognitive function. In addition, although decreases in cognitive function had a higher loading value for the second factor (which also included anxiety and depression symptoms), the loading value of decreases in cognitive function onto the pain–fatigue–sleep disturbance factor was also greater than the cutoff of 0.4. The two extracted factors explained 77% of variance with an eigenvalue of 0.9.

Hierarchical Clustering

Visual inspection of the dendrogram produced by hierarchical clustering using Ward’s method suggested a two- or three-group solution. The two-group solution identified a symptom phenotype with mild severity across all symptoms (n = 124) and a symptom phenotype with moderate to high severity across all symptoms (n = 77). The three-group solution described a generally low symptom severity phenotype (n = 124), a generally moderate symptom severity phenotype (n = 53), and a symptom phenotype with severe mood-related symptoms (n = 24). The three-group solution was described based on the potential of the description of the phenotype with severe mood-related symptoms to be clinically informative. Symptom severity for each phenotype identified by the three-group solution is described online in Supplemental Figure 1, and characteristics of each of the three phenotypes are detailed in Table 4.

Because of a low sample size (n = 19) in the subsample, with ULP-WGS outcomes resulting in singularities in the Hessian matrix in regression analysis when using the three-group solution, the two-group solution was used in the subsequent regression analysis. The authors acknowledge that three-symptom phenotypes likely exist in this population based on the results, and the use of the two-group solution in subsequent analysis is only because of constraints of data and sample size. The two-group solution combines the moderate symptom and severe mood-related symptom phenotypes from the three-group solution to combat the smaller number of participants in these groups. The resulting groups from the two-group solution used as dependent variables are the mild PN symptom phenotype group and the moderate to severe PN symptom phenotype group.

Disease, Treatment, and Individual Characteristics

In binomial logistic regression including the whole sample (N = 201), the only variables found to be significant predictors of symptom phenotype when included univariately were BMI (OR = 1.041, 95% CI [1, 1.083], p = 0.048) and smoking status (OR = 2.133, 95% CI [1.161, 3.922], p = 0.014). When both were included in the regression model, smoking status remained a significant predictor of symptom phenotype (OR = 2.031, 95% CI [1.097, 3.758], p = 0.024), whereas BMI was not a significant predictor of symptom phenotype (OR = 1.037, 95% CI [0.996, 1.08], p = 0.08). Therefore, smoking status was the only confounding variable controlled for when investigating CNA of selected genes as predictors of PN symptom phenotype.

CNA of Selected Genes

When comparing the subgroup with ULP-WGS results (n = 19) against the rest of the sample (n = 182), the groups differed by chance alone on symptom phenotype membership (χ² = 8.051, p = 0.005), with those who had ULP-WGS results more likely to fall into the moderate to severe symptom phenotype than those who did not.

When included individually in logistic regression, of the six selected genes, only the CNA status of TP53 was a significant predictor of symptom phenotype membership (OR = 27.5, 95% CI [1.996, 378.837], p = 0.013), with deletion of TP53 predicting membership in the moderate to severe PN symptom phenotype. When controlling for smoking status, CNA of TP53 was still a significant predictor of symptom phenotype, with those in a deletion in TP53 about 35 times more likely to belong to the moderate to severe symptom phenotype than to the mild symptom phenotype (OR = 35.219, 95% CI [1.973, 628.543], p = 0.015). The Nagelkerke R² of the final model, including smoking status and TP53, was 0.56.

Given the small sample size and novel nature of this analysis, descriptives (including frequencies and percentages) of CNA of selected genes by symptom phenotype were calculated (see Table 5). In the moderate to severe symptom group, 85% (n = 11) of participants had a TP53 deletion, while only 17% (n = 1) of participants in the mild symptom phenotype group had a TP53 deletion.

Discussion

This study aimed to phenotype PN symptoms in patients with metastatic breast cancer, as well as to associate PN symptom phenotypes with breast cancer genomic instability using ULP-WGS of ctDNA. These results from exploratory factor analysis suggest that although the PN symptoms likely form a single cluster, it is possible that they are divided into two separate clusters: the pain–fatigue–sleep disturbance cluster and the mood-related cluster that includes depression symptoms, anxiety, and decreases in cognitive function. This result is consistent with the findings of a systematic review of symptom clusters in patients with breast cancer, which reported that some studies found a single PN symptom cluster, whereas others found that PN symptoms split between two similar clusters (So et al., 2021). Many of the studies included in this review were conducted in early-stage breast cancer, and this study in metastatic breast cancer suggests that PN symptoms cluster similarly in this population. Additional research on PN symptoms in the metastatic breast cancer population is needed to elucidate whether these PN symptoms are indeed a single cluster in this population, or if they are separated into the pain–fatigue–sleep disturbance cluster and the mood-related cluster.

Results of factor analysis are complemented by findings from hierarchical clustering, seeking to describe distinct PN symptom phenotypes. Hierarchical clustering revealed three distinct symptom phenotypes: one with generally mild symptoms, one with moderate symptom severity across all six PN symptoms, and one phenotype characterized by severe mood-related symptoms (e.g., depression symptoms, anxiety, decreased cognitive function). This finding mirrors the symptom clustering from factor analysis, where exploratory factor analysis suggested a single cluster of six symptoms but also produced a two-cluster solution when statistical criteria for factor extraction were relaxed. Although additional research is needed to accurately quantify the relationships between PN symptoms, these findings complement the current science suggesting that PN symptoms co-occur but may exist in two separate clusters (the pain–fatigue–sleep disturbance cluster and the mood-related cluster) (So et al., 2021).

Disease, treatment, and individual characteristics found to be related to PN symptom phenotype membership were smoking history and BMI. Having a history of smoking and higher BMI were associated with a higher belonging to a more severe PN symptom phenotype group when evaluated individually. This is consistent with prior research on PN symptoms (Mark et al., 2017; Ratcliff et al., 2021; Syrowatka et al., 2017; Yang et al., 2022). Interestingly, although prior research has found chemotherapy and radiation therapy to be associated with more severe PN symptoms in individuals with breast cancer (Grayson, Sereika, et al., 2023; Hsu et al., 2017; Voute et al., 2020; Whisenant et al., 2020), the current study did not find any relationship between prior treatment modalities and PN symptom phenotype. This is potentially because of the research population being diagnosed with metastatic breast cancer as opposed to early-stage breast cancer.

A novel finding of this study was the relationship between breast cancer TP53 deletions and membership in a moderate to severe symptom phenotype group. TP53 has several important roles related to genomic instability, including regulation of DNA repair, cell cycle arrest, and apoptosis (Kastenhuber & Lowe, 2017). As such, a deletion of TP53 is indicative of ongoing accumulation of DNA variants in breast cancer. This deletion’s association with a more severe PN symptom phenotype indicates that cancer genomic instability may be implicated in PN symptom development. This relationship suggests a potential mechanism by which PN symptoms may develop before cancer diagnosis, which has been observed in pancreatic cancer (Kenner, 2018), providing a link between cancer progression and PN symptom development. Additional research is needed to further characterize more specific cancer genomic changes associated with PN symptom development. Replicating this research in larger sample sizes and including more cancer genetic variables, such as non-CNA variants or overall variation load, may suggest specific biologic pathways to target for symptom relief interventions. This direction of inquiry has the potential to increase quality of life for patients with cancer through proactive treatment and development of precision health interventions based on novel therapeutic targets.

Limitations

This study was not without limitations. Most notably, the sample size with ULP-WGS data was very small (n = 19). Because of the fruitful findings regarding TP53 deletions in this sample size, it is warranted to pursue this type of research in a larger cohort. In addition, it is worth noting that the subsample with ULP-WGS data had a significantly larger membership in the moderate to severe symptom phenotype than the group without ULP-WGS data. Biases may have existed in the random selection of the subsample included in ULP-WGS analysis, and may have affected the findings. Future research should seek to address these opportunities for improvement by continuing to study cancer genomics in relation to PN symptoms in larger sample sizes. These investigations should also include more information on genes related to DNA damage and repair to illustrate the relationship between genomic instability and symptom development.

Implications for Nursing

PN symptoms co-occur in individuals with metastatic breast cancer. Nurses should carefully assess patients who report pain, fatigue, sleep disturbance, anxiety, depression, or decreased cognitive function for concurrent symptoms. When one symptom is assessed, the nurse should recognize that the others may also be present, and interventions to address the symptom cluster as a whole should be considered.

The development of these symptoms may be affected by underlying biologic processes related to cancer genomics, indicating that nurses should assess patients for PN symptoms before the initiation of the cancer treatments often implicated in symptom development. With additional evidence, nurses should even consider whether PN symptoms may be an indication for cancer screening.

Conclusion

Cancer genomic markers may be useful in predicting symptoms and selecting effective symptom management interventions. To date, cancer genomic testing may be used in clinical settings for disease prognosis and selection of targeted treatments, and is increasingly becoming the standard of care (Litton et al., 2019). With more evidence, nurses may also be able to use the results of these genomic tests to identify individuals at risk for developing severe PN symptoms and advocate for proactive interventions for these individuals.

Analysis of ctDNA from blood or liquid biopsies is also increasingly used in clinical cancer care. The results of this research have the potential to broaden the clinical utility of these tests to encompass symptom prediction, enabling nurses to intervene proactively for patients at risk for severe symptoms and promote quality of life.

About the Authors

Susan C. Grayson, PhD, RN, was, at the time of this writing, a PhD candidate, Susan M. Sereika, PhD, is a professor and associate dean for research and education support services, and Yvette P. Conley, FAAN, PhD, is a professor and associate dean for research and scholarship, all in the School of Nursing at the University of Pittsburgh in Pennsylvania; Adrian V. Lee, PhD, is a professor in the Department of Pharmacology and Chemical Biology in the School of Medicine at the University of Pittsburgh and the director of the Institute for Precision Medicine in Pittsburgh, PA; Steffi Oesterreich, PhD, is a professor and vice chair in the Department of Pharmacology and Chemical Biology in the School of Medicine at the University of Pittsburgh and co-director of the Women’s Cancer Research Center at Magee-Womens Research Institute in Pittsburgh, PA; Theresa A. Koleck, PhD, RN, is an assistant professor and Margaret Q. Rosenzweig, PhD, FNP-BC, AOCNP^®, FAAN, is a professor, both in the School of Nursing at the University of Pittsburgh; Tiantong Liu, PhD, is a physician at China–Japan Friendship Hospital in the School of Medicine at Tsinghua University in Beijing; and Susan W. Wesmiller, RN, PhD, FAAN, is an associate professor in the School of Nursing at the University of Pittsburgh. This research was funded, in part, by Susan G. Komen and the Breast Cancer Research Foundation and the University of Pittsburgh Center for Research Computing (RRID:SCR_022735) through the High Throughput Computing Cluster, which is supported by the National Institutes of Health (S10OD028483). Grayson was supported by the National Institute of Nursing Research grant for Targeted Research and Academic Training Program for Nurses in Genomics (T32NR009759). Grayson, Sereika, Conley, Lee, Oesterreich, Koleck, Rosenzweig, and Wesmiller contributed to the conceptualization and design. Grayson and Liu completed the data collection. Grayson, Sereika, and Koleck provided statistical support. Grayson, Sereika, Conley, Lee, Rosenzweig, Liu, and Wesmiller provided the analysis. All authors contributed to the manuscript preparation. Grayson can be reached at graysonsc@nih.gov, with copy to ONFEditor@ons.org. (Submitted January 2024. Accepted April 4, 2024.)