Abstract
Monogenetic disorders that cause cerebellar ataxia are characterized by defects in gait and atrophy of the cerebellum, however, patients often suffer from a spectrum of disease, complicating treatment options. Spinocerebellar autosomal recessive 16 (SCAR16) is caused by coding mutations in STUB1, a gene that encodes the multi-functional enyzme CHIP (C-terminus of HSC70-interacting protein). The spectrum of disease found in SCAR16 patients includes a wide range in the age of disease onset, cognitive dysfunction, increased tendon reflex, and hypogonadism. Although SCAR16 mutations span the multiple functional domains of CHIP, it is unclear if the location of the mutation contributes to the clinical spectrum of SCAR16 or with changes in the biochemical properties of CHIP. In this study, we examined the associations and relationships between the clinical phenotypes of SCAR16 patients and how they relate to changes in the biophysical, biochemical, and functional properties of the corresponding mutated protein. We found that the severity of ataxia did not correlate with age of onset; however, cognitive dysfunction, increased tendon reflex, and ancestry were able to predict 54% of the variation in ataxia severity. We further identified domain-specific relationships between biochemical changes in CHIP and clinical phenotypes, and specific biochemical activities that associate selectively to either increased tendon reflex or cognitive dysfunction, suggesting that specific changes to CHIP-HSC70 dynamics contributes to the clinical spectrum of SCAR16. Finally, using linear models and Monte Carlo simulations, our data support the hypothesis that further inhibiting mutant CHIP activity lessens disease severity and may be useful in the design of patient-specific targeted approaches to treat SCAR16.
Introduction
Ataxia is a general term used to describe a loss of coordination. Ataxia can be caused by a variety of diseases, including metabolic disorders, vitamin deficiencies, peripheral neuropathy, cancer, or brain injuries. In addition to deterioration in movement and balance, ataxia can be accompanied by a spectrum of secondary disorders, including impairments in speech, vision, and cognitive ability. Ataxia is most often caused by the progressive deterioration of the cerebellum, known as cerebellar ataxia (CA), of which there are several causes: hyperthyroidism, alcoholism, stroke, multiple sclerosis, and traumatic injury. Additionally, there are known genetic mutations that are thought to cause CA, and these forms of CA are classified by their inheritance patterns (1). CA mutations are inherited most commonly in an autosomal recessive manner (estimated prevalence is 7 per 100,000). CA can also manifest as an autosomal dominant disorder (estimated prevalence is 3 per 100,000), in addition to less prevalent X-linked or mitochondrial form of inheritance. Most forms of autosomal dominant CAs are caused by polyglutamine expansions within a protein coding region, in contrast to autosomal recessive CAs that are caused by conventional mutations within the coding region (1). The age of onset, prognosis, accompanying symptoms vary both among and within the genetic forms of CA, and importantly, there are currently no front-line medications for CA (2).
Spinocerebellar ataxia autosomal recessive 16 (SCAR16, MIM 615768) is a recessive form of cerebellar ataxia with a wide ranging disease spectrum, that can also include hypogonadism, cognitive dysfunction, dysarthria, and increased tendon reflex (2). Using whole exome sequencing, we identified a mutation in STUB1 in two patients initially diagnosed with ataxia and hypogonadism (3). Subsequently, numerous clinical reports identified STUB1 mutations in patients with ataxia, confirming our initial identification of a new disease (3–10). Remarkably, STUB1 mutations were found in nearly 2% of Caucasian patients with degenerative ataxia, and these mutations appeared to be specific to the ataxia phenotype and not rare ubiquitous polymorphisms (8). STUB1 encodes the multi-functional enzyme CHIP (C-terminus of HSC70-interacting protein), recognized as an active member of the cellular protein quality control machinery and has multiple functions as both a chaperone (11, 12), co-chaperone (13, 14), and ubiquitin ligase enzyme (15, 16). As a chaperone, CHIP can cause structural changes to proteins to either maintain solubility or increase specific activity. As a cochaperone, CHIP directly interacts with heat shock proteins (HSP) and can aid in the stabilization and refolding of HSP-bound substrates. Conversely, as a ubiquitin ligase, CHIP ubiquitinates terminally-defective proteins and targets them for degradation by the ubiquitin proteasome system.
SCAR16 mutations span the three functional domains of CHIP (Figure 1A): the N-terminal TPR domain that binds HSPs, the coiled-coiled domain that is important for dimerization, and the C-terminal Ubox domain that is responsible for the ubiquitin ligase function (2). Currently, it is not known if the location of these mutations mediate specific aspects of the SCAR16 spectrum. Equally so, it is not known how changes in CHIP properties caused by substitution mutations relate to clinical phenotypes. In this report, we combined clinical data provided by numerous reports and a recent report that characterized the biochemical repercussions of several of these SCAR16 disease mutations (Figure 1A) (17). Our approach allowed us to identify the specific biochemical changes to CHIP that are coupled to SCAR16 clinical characteristics. We developed linear models and used simulations to identify which properties of mutant CHIP proteins may impact disease severity. Defining the relationship between changes in specific features of CHIP and disease phenotypes may reveal new clues to both the spectrum of this disease, and ultimately, guide precision medicine-based strategies to treat SCAR16.
Materials and Methods
SCAR16 patient data
Clinical data were obtained from published reports (Table S1) (3–10). One measure of disease severity is the score from the Scale for the Assessment and Rating of Ataxia (SARA). When SARA scores were not implicitly stated, SARA was imputed based on the clinical report as indicated in Table S1 (18).
CHIP mutation data
All biophysical and biochemical properties of CHIP proteins with disease-associated substitution mutations were obtained from published data (Table S2) (17). HSP70 ubiquitination at 37 °C or 25 °C was measured by densitometry analysis and represented by the total amount of HSP70 that was modified by ubiquitination; wild-type (WT) CHIP ubiquitinated 73% or 81% of the HSP70 total in the reaction, respectively.
Statistical Analysis
All analyses were performed using JMP Pro (v14.2.0). Continuous and categorical clinical variable distributions were analyzed using the Shapiro-Wilk W test or chi-squared test, respectively. Bivariate analysis was performed using either t test, one-way ANOVA, linear regression, or contingency analysis using Fisher’s exact test, depending on the two variables being compared. A false discovery rate of <10% (Benjamini-Hochberg) was used to control for multiple test correction for each dependent variable (Table S3), with raw P values reported. Post hoc tests, when applicable, are described in the figure legends. Partial least squares regression was initially performed using all variables, using leave-one-out cross validation and then refined by retaining all variables with VIP values > 0.8 (19, 20). The model coefficients for the original data are provided in the corresponding data tables.
Monte Carlo simulations were performed by modeling SARA and age of onset using a multiple Y partial least squares regression equation using the indicated variables (19–22). Targets for SARA and age of onset improvement were set at one standard deviation from the mean. Simulations were run using all X variables as random with a normal distribution. 5000 baseline simulations were used to fit the original data. The simulation was adjusted to maximize desirability via an additional 5000 simulations to identify parameters that maximize improvement of both Y variables (23).
Data Availability
Table S1 contains clinical variables. Table S2 contains the biochemical data. Table S3 contains the adjusted P values. All supplementary files are available at the UNC digital repository, DOI: 10.17615/8dqf-e678.
Results
Clinical variable analyses from SCAR16 patients
Phenotypic data were collated from multiple clinical reports, encompassing 24 patients (Table S1) (3–10). Several variables had skewed distributions (Table 2), most notably, hypogonadism was found in only four patients, whereas over 70% of the patients suffered from increased tendon reflex and/or cognitive dysfunction. There was nearly an equal number of males and females, as well as homozygous and hemizygous patients. The median clinical score for ataxia (SARA) was 18.5, a value associated with moderate dependence for daily activities (18), and the median age of onset was 17 years of age.
Bivariate analysis
We explored the co-penetrance of three reported SCAR16 clinical phenotypes, cognitive dysfunction, increased tendon reflex, and hypogonadism. There was no association between cognitive dysfunction and increased tendon reflex (Fisher’s exact test, P = 1.000, suggesting that these two phenotypes are distinct from each other. Hypogonadism was only reported in four patients, limiting the power of the contingency analysis, however, all four patients with hypogonadism also had cognitive dysfunction (Fisher’s exact test, P < 0.0001).
We next examined the relationship between two measures of disease severity, the age of onset and SARA, a reliable and valid clinical measure of ataxia (18). We hypothesized that more severe ataxia would correlate with an earlier age of onset. Counter to our hypothesis, there was no correlation between age of onset and ataxia (Figure 1B), indicating that these two variables were independent of each other. We further explored the relationships of age of onset and SARA with the other clinical variables reported in SCAR16 patients (Table 2). Patients with homozygous mutations had a reported disease onset 12 years earlier than patients with compound heterozygous mutations (Figure 1C). However, unlike other genetic forms of ataxia caused by trinucleotide repeats (24, 25), there were no associations between either homozygozity or age of onset with either SARA or any of the remaining clinical variables (Figure 1D, Table S2, S3). In contrast, when the associations between SARA and the remaining clinical variables were measured (Table 3), we found that SCAR16 patients with cognitive dysfunction scored 10 points higher on the SARA assessment (Figure 1E). Additionally, patients with European ancestry had the highest average SARA (Figure 1F, median = 32) whereas there was no difference between those with Han Chinese or Middle Eastern/North African ancestry (median = 15 and 14.5, respectively). These data demonstrate that SCAR16 patients with cognitive dysfunction had more severe deficits in motor function and that other genetic factors may potentially influence the effect of CHIP mutations on the severity of ataxia.
Multivariate analysis
Given the covariance structure of the patient variables and the mixture of continuous and nominal variables, we employed partial least squares regression to model ataxia severity (using SARA) as a function of the clinical characteristics. This modeling technique permits the usage of correlated explanatory phenotype variables in describing the distribution of SARA. Leave-one-out cross-validation identified three factors: ancestry, cognitive dysfunction, and tendon reflex (Figure 2A) that explained 54% of the variation in SARA (Figure 2B). Cognitive dysfunction had the largest estimate in the regression analysis (Equation 1, CD = cognitive dysfunction, TR = tendon reflex, Y = yes, N = no, EAS = East Asian ancestry, EUR = European ancestry). Also, European decent and increased tendon reflex contributed to higher SARA. Our model suggests that genetic factors associated with ancestry may influence the severity of SCAR16, even to a greater extent than homozygosity, age of onset, and sex, as these later variables did not contribute to the predictive power of the modeling (Table 3).
Changes in CHIP biochemistry caused by SCAR16 mutations
A recent study analyzed the effect of SCAR16 mutations on the properties of CHIP at the protein level (17). We used these data to look for relationships between various biophysical and biochemical properties to better understand the effect of these mutations on CHIP function. All continuous data were normally distributed except for KD, which did not meet normalcy testing even after various transformations, therefore, all analyses were conducted using non-transformed data. As expected, there was a strong positive correlation between ubiquitin chain formation (a readout of the alignment between CHIP and the E2 enzyme) (17) and the extent of HSP70 ubiquitination (ρ = 0.64). Also, HSP70 ubiquitination was inversely correlated with the KD between CHIP and a peptide containing the HSP70 binding motif (ρ = −0.76). We observed a positive correlation between Bmax and KD regarding CHIP interactions with the HSP70 peptide (ρ = 0.66). This unexpected positive correlation suggests that some mutant CHIP proteins, including CHIP-T246M, may have an increased binding capacity towards chaperones, consistent with our previous studies (3, 26). Throughout the initial correlation analyses (Figure 3A) the mutant CHIP proteins appeared to cluster based on the domain that harbors the mutation. This observation was confirmed via hierarchical clustering of mutant CHIP proteins and their corresponding biochemical parameters. Mutant proteins clustered primarily by the domain harboring the mutation (Figure 3B), consistent with the premise that the domain affected by the mutation may have differential actions on CHIP activities.
The impact of the location of SCAR16 mutations on CHIP biochemistry
We hypothesized that the location of the mutation would differentially impact the multiple functions of CHIP to either bind to chaperones and/or mediate polyubiquitin chain formation. To test this hypothesis, the variability in biochemical properties of CHIP mutations were tested for associations with the domain harboring the mutation. This analysis revealed three associations at a FDR < 10% (Table 4). First, binding studies between CHIP and the HSP70 tail peptide revealed that Ubox mutations had increased binding capacity towards HSP70 (Figure 3C); however, there was no difference in binding affinity across domains (Figure 3D). Further, as shown by (17), all mutant CHIP proteins maintained some capacity to polyubiquitinate HSP70, except for two Ubox mutants, M240T and T246M (Figure 3E). Moreover, every Ubox mutant had a reduced capacity to form polyubiquitin chains, perhaps due to altered interactions with E2 enzymes, whereas TPR and CC mutations were not defective in these same conditions (Figure 3F). These data suggest that Ubox mutations, in particular, affect functions that span both the co-chaperone and ubiquitin ligase functions of CHIP.
Determining the relationship between patient phenotypes and the altered biochemical properties of mutant CHIP proteins
Finally, we analyzed the clinical phenotypic data from SCAR16 patients and the corresponding biochemical characteristics of mutant CHIP proteins to determine if clinical variables were related to biochemical changes caused by the coding mutations. This approach has several limitations. First, only substitution mutations were characterized biochemically; therefore, phenotype-functional associations from indel mutations are not included in this analysis. Second, given the recessive and sometimes hemizygous nature of SCAR16, these analyses were performed on a per allele basis. Thus, in the case of hemizygous patients, this analysis cannot distinguish if one mutation has a more functional role than the other regarding the clinical manifestations of SCAR16. However, the majority of hemizygous genotypes include one allele that results in a pre-terminal stop codon, predicted to be degraded by non-sense mediated RNA decay (Table S1).
Connecting the location of CHIP mutations with SCAR16 clinical variables
Oneway ANOVA was used to measure the relationship between the domain location harboring the mutation and the continuous clinical variables. There was no association of either age of onset, SARA, or SARAadj with the domain location of the mutation (P = 0.51, 0.86, or 0.14, respectively). However, there was a highly skewed distribution between the mutation location and cognitive dysfunction (Figure 4). Most notably, 94% of alleles with Ubox mutations associated with cognitive dysfunction, with no difference in the frequency of cognitive dysfunction between TRP or CC allele mutations (59% of TPR or CC alleles associated with cognitive dysfunction). In constrast, hypogonadism and increased tendon reflex were equally distributed in regards to the location of CHIP mutations (Fisher’s exact test = 0.15 and 1.00, respectively). These data suggest that domain-specific changes in CHIP function may contribute to the clinical spectrum of SCAR16, particularly Ubox mutations that are linked with cognitive dysfunction.
Bivariate analysis of the biochemical changes caused by CHIP mutations with cognitive dysfunction and tendon reflex
Changes in the biochemical properties of mutant CHIP may influence the clinical spectrum of SCAR16. As discussed above, there was no link between cognitive dysfunction and tendon reflex phenotypes in SCAR16 patients. As such, the relationships between the corresponding biochemical properties of mutant CHIP proteins encoded by disease alleles and these two independent phenotypes were analyzed. Tm and steady-state expression levels of mutant CHIP proteins did not associate with either cognitive dysfunction (P = 0.57 and 0.58) or increased tendon reflex (P = 0.28 or 0.62). However, cognitive dysfunction associated with 35% lower activity in CHIP/E2-mediated ubiquitin chain formation (Figure 5A), suggesting this activity is essential to maintain cerebellar cognition. CHIP functions primarily as a dimer, and seven of the 13 characterized mutations maintain oligomeric distributions similar to CHIP-WT (17). Interestingly, dimeric forms of mutant CHIP associated with increased tendon reflex (Figure 5B). Therefore, loss of ubiquitin ligase function may be a prominent contributor to cognitive dysfunction, whereas CHIP that still forms dimers, and perhaps some degree of altered CHIP function, may contribute to increased tendon reflex. Accordingly, aspects of the CHIP-HSC70 interaction changed reciprocally when compared to either cognitive dysfunction or tendon reflex. Alleles from SCAR16 patients with cognitive dysfunction had increased HSP70 binding capacity (delta = 14.3 μmol per min, (Figure 5C) with no change in KD (Figure 5D). In contrast, alleles from SCAR16 patients with increased tendon reflex corresponded to mutant CHIP proteins with similar binding capacity (Figure 5E) and affinity to HSP70 (Figure 5F) as wild-type CHIP. Although only reaching marginal significance, we again observed opposite effects with cognitive dysfunction and tendon reflex regarding the ability of CHIP to ubiquitinate HSP70 (Figure 5G, 5H), effects that are consistent with the patterns in CHIP/E2-mediated ubiquitin chain formation (Figure 5A) and HSP70 binding (Figure 5C – 5F). Differential biochemical activities of CHIP linked with these clinical phenotypes support the concept that altered CHIP-HSC70 dynamics, caused by disease mutations, contribute to the clinical spectrum of SCAR16.
Multivariate modeling of disease onset and SARA as a function of the biochemical properties of mutant CHIP
We predicted that the biochemical changes to CHIP via disease mutations would affect both the age of onset and disease severity. First, bivariate analysis of age of onset and SARA with each of the biochemical properties of the mutant alleles (Table 5) identified that alleles encoding mutant CHIP proteins with lower activity and higher Bmax associated with an earlier age of onset. Additionally, alleles encoding mutant CHIP proteins forming higher-order oligomers and reduced binding affinities associated with lower SARA scores. These data suggest that non-functional forms of CHIP (higher-order oligomers and decreased HSC70 binding affinity) results in less severe disease as opposed to mutant CHIP proteins that still maintain normal tertiary structure and binding activities towards chaperones. Multivariate linear models using partial least squares were developed for both age of onset (Figure 6A, Equation 2) and SARA (Figure 6B, Equation 3) as functions of the biochemical properties of mutant CHIP proteins. Tm and %expression were the only variables that did not contribute to the modeling of either onset or SARA (VIP > 0.8). The remaining five variables accounted for 70% of the variance in the age of onset; however, SARA modeling was less powerful, accounting for only 20% of the variance (Table 5).
Effects of CHIP stability on disease onset and severity
One approach in designing therapies for SCAR16 would be to restore CHIP activity. Kanack et al. hypothesized that mutant CHIP proteins could be stabilized by decreasing the reaction temperature below the Tm of the mutant proteins (17). It was subsequently demonstrated that decreasing the reaction temperature to 25 °C, partially recovered CHIP activities (17). Statistical analysis of these data demonstrated that decreasing the reaction temperature increased HSP70 ubiquitination (Figure 7A) and partially recovered some of the in vitro binding alterations seen in mutant CHIP proteins (Figure 7B, 7C). We used the data from experiments performed at 25 °C, including HSP70 ubiquitination, KD Bmax, in the biochemical models of onset and SARA to examine the impact on these two phenotypes of SCAR16. Increasing mutant CHIP activity led to an earlier predicted age of onset (Figure 8A) with the mean of the model predictions equaling a six year lower age of onset (Figure 8B). Mutant CHIP parameters at 25 °C had a less dramatic impact on SARA (Figure 8C) and predicted a small increase in SARA (Figure 8D). Considering the effects on both SARA and age of onset, the effect of increasing mutant CHIP activities, by lowering the reaction temperature, predicted a more severe disease.
Ultimately, we were interested in identifying changes in CHIP properties and activities that may result in delaying onset and decreasing severity (SARA). Monte Carlo simulations were performed using a combined partial least squares regression model of both age of onset and SARA, incorporating the four continuous biochemical variables identified in Table 4. The simulation that fit the original SCAR16 patient data indicated decreases in both ubiquitination-related activities and a modest reduction in binding affinity compared to wild-type CHIP (Table 6). Next, we set the improvement targets at one σ from the mean, equating to a 10.7 year increase in age of onset and a 10 point decrease in SARA. The results of the simulation were consistent towards both Y variables: further inhibiting mutant CHIP activity towards HSP70, either by decreasing HSP70 ubiquitination and/or reducing the binding affinity to HSP70, would both delay onset and severity of SCAR16 (Figure 9). The simulations predicted a delay in the age of onset by 7.2 years and an 8.2 point decrease in SARA by decreasing the affinity of mutant CHIP proteins to HSP70 to 17 μM and decreasing the HSP70 ubiquitination to 7.2%, relative to wild-type CHIP (Table 6). Not surprisingly, there was a strong negative correlation between the amount of HSP70 ubiquitination and KD of the CHIP-HSP70 interaction (Figure 3A, ρ = −0.76), suggesting that blocking the interaction mutant CHIP with HSP70 represents a therapeutic opportunity that can be explored in future studies.
Discussion
Mutations that cause SCAR16 span the three functional domains of the multi-functional enzyme CHIP (Figure 1A). Prior to this study, it was not known if the location of these mutations associate with specific aspects of the clinical spectrum exhibited by SCAR16 patients. Equally so, it was not known how changes in CHIP properties caused by substitution mutations related to clinical phenotypes.
First, we found that in addition to cognitive dysfunction and increased tendon reflex, genetic background may influence the severity of SCAR16 (Figure 1E & F, Figure 2A, and Table 3), whereas homozygozity was unrelated to ataxia severity (Figure 1D). It is possible that quality and access to health care and environmental factors could confound the observation of ancestry on SCAR16 severity, however, there could be other genetic factors that could lessen or exacerbate the loss or change in CHIP function. Identifying modifiers of CHIP function may provide additional therapeutic opportunities.
Ultimately, we were interested in identifying mutation-specific effects on CHIP function and the spectrum of SCAR16, and we found two primary classes of mutations. Overall, Ubox mutations had a more dramatic effect on the overall loss of CHIP function (Figure 3) and strongly associated with cognitive dysfunction in SCAR16 patients (Figure 4 and 5). In contrast, mutations with more modest effects on CHIP function, primarily the mutations located in the TPR and CC domains, were linked to the increased tendon reflex seen in SCAR16 patients (Figure 5). Therefore, mutations that retain this intact, but slightly diminished activity towards HSP70, appear to drive the increased tendon reflex pathology.
These data allowed us to generate linear models (Figure 6 and Table 5) to both identify properties of CHIP that may be useful to target in SCAR16 patients and to test experimental data. The activity of several mutant CHIP proteins could be recovered by lowering the reaction temperature (Figure 7), however, our models predicted that mutants with higher activities resulted in more severe disease phenotypes (Table 3). As such, the experimental data of increased mutant CHIP activity also predicted an earlier age of onset and more severe ataxia (Figure 8). Using a combined model of both age of onset and ataxia severity, our simulation results were consistent with the idea that inhibiting mutant CHIP interactions with HSP70 predict a later age of onset and less severe ataxia (Figure 9, and Table 6).
It remains unclear is why Ubox mutations, with disrupted ubiquitin-related activities, strongly associate with cognitive dysfunction. Ubox mutations also have higher Bmax, and we previously observed that the Ubox mutant CHIP-T246M pulled down more HSC70 and HSP70 compared to CHIP-WT in multiple cell models, and in our engineered mouse line that expresses CHIP-T246M from the endogenous locus (3, 26). One possibility is that Ubox mutants still bind E2 enzymes, but the inability to transfer ubiquitin disrupts E2 function, and perhaps activity of E2 enzymes towards other E3 ligases. Alternatively, the propensity to form higher-order oligomers (17, 26) and changes in solubility (26) could also affect the function of proteins that still interact with Ubox mutants.
Overall, our data suggest that inhibiting the interaction between mutant CHIP and HSP70 chaperones could be used as a targeted approach in cognitively normal patients with TPR and CC domain mutations. In contrast, SCAR16 patients with cognitive dysfunction and Ubox mutations may benefit from the use of molecular chaperones to prevent the oligmerization of these mutant CHIP proteins. CHIP impacts several cellular pathways, and identifying the CHIP-dependent pathways that may contribute to the specific pathologies in SCAR16, such as necroptosis (27), IGF1 (28), mitophagy (29), autophagy (30, 31), or water balance (32), may also uncover therapeutically relevant targets. Additionally, gene therapy approaches are also be applicable to SCAR16. The obvious solution is to use gene editing approaches to correct these mutations, however, given the recessive nature of the disease, delivery of a functional copy of CHIP may also be beneficial (33, 34); alternatively, antisense oligonucleotide therapy that can downregulate mutant CHIP protein levels may also prove to be an an effective approach (35).
Additional SCAR16 patients with new, uncharacterized mutations continue to be reported (9, 36–39), in addition to a possible variant that functions in a dominant manner (40). With additional clinical data and with the advent of new molecular, cellular, and preclincal models to study CHIP function, it is likely that precision-based approaches could be developed based on the specific mutation or the specific loss of function. Finally, by looking more broadly at the various autosomal recessive ataxias, additional themes and targets that could be effective across multiple ataxia diseases may also come to light in the years to come (41).
ACKNOWLEDGEMENTS
We thank members of the Schisler Laboratory, including Rebekah Sanchez-Hodge for a critical review of the manuscript, Kalleen Kelley, and the McAllister Heart Institute administration team. All authors approved the final version of the manuscript and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.
Footnotes
Format changes to entire document, including a separate discussion section instead of a combined results/discussion section. Title was updated.