Advertisement

Analysis of a precision medicine approach to treating Parkinson's disease: Analysis of the DATATOP study

  • Sid E. O'Bryant
    Correspondence
    Corresponding author.University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA.
    Affiliations
    Institute for Translational Research, Department of Pharmacology & Neuroscience, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA
    Search for articles by this author
  • Melissa Petersen
    Affiliations
    Institute for Translational Research, Department of Pharmacology & Neuroscience, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA

    Department of Family Medicine, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA
    Search for articles by this author
  • Fan Zhang
    Affiliations
    Institute for Translational Research, Department of Pharmacology & Neuroscience, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA

    Department of Family Medicine, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA
    Search for articles by this author
  • Leigh Johnson
    Affiliations
    Institute for Translational Research, Department of Pharmacology & Neuroscience, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA
    Search for articles by this author
  • David Mason
    Affiliations
    Institute for Translational Research, Department of Pharmacology & Neuroscience, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA

    Department of Family Medicine, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA
    Search for articles by this author
  • James Hall
    Affiliations
    Institute for Translational Research, Department of Pharmacology & Neuroscience, University of North Texas Health Science Center, 3500 Camp Bowie Blvd, Fort Worth, Texas, 76107, USA
    Search for articles by this author
Open AccessPublished:November 25, 2021DOI:https://doi.org/10.1016/j.parkreldis.2021.11.027

      Highlights

      • Blood samples were analyses from the DATATOP trial.
      • Our precision medicine approach was highly accurate in predicting treatment outcomes (primary and secondary).
      • A precision medicine approach has tremendous potential to significantly improve patient outcomes.

      Abstract

      Introduction

      The aim of this study was to examine the potential application of a targeted proteomic predictive biomarker comprised predominantly of inflammatory proteins in distinguishing those who responded to a previously conducted clinical trial for Parkinson's disease (PD).

      Methods

      Plasma samples obtained from a biorepository were assayed from a total of n = 520 DATATOP (Deprenyl And Tocopherol Antioxidative Therapy Of Parkinsonism) clinical trial participants across treatment arms. Support vector machine analyses were conducted to distinguish responder status on primary (need for Levodopa) and secondary trial endpoints (UPDRS Motor and Total Scores).

      Results

      For the α-tocopherol and deprenyl placebo treatment arm (TOC), the targeted proteomic biomarker was able to distinguish responder status with an accuracy (area under the curve [AUC]) of 91% for the primary endpoint while it was 100% across secondary endpoints. For the deprenyl and α-tocopherol placebo treatment arm (DEP), the AUC was 93% for the primary endpoint and 99–100% for the secondary endpoints. For the combined treatment arm, AUC was 87% for the primary and 94–96% for the secondary endpoints.

      Discussion

      The targeted proteomic predictive biomarker was highly accurate in distinguishing responder status across treatment arms thereby supporting the application of a precision medicine approach to treating PD.

      Keywords

      1. Introduction

      Parkinson's disease (PD) is the second most common neurodegenerative disease affecting over 1% of people age 65 and over in the United States (U.S.) [
      • Tysnes O.B.
      • Storstein A.
      Epidemiology of Parkinson's disease.
      ]. The cost of PD to our society was reported to be $23 billion annually in the U.S. in 2005 [
      • Huse D.M.
      • Schulman K.
      • Orsini L.
      • Castelli-Haley J.
      • Kennedy S.
      • Lenhart G.
      Burden of illness in Parkinson's disease.
      ]. Considering the estimated 15% growth in the elderly U.S. population during the last decade, these costs can be expected to increase dramatically as the population ages. Despite advancements in our understanding of PD, current therapeutics remain limited with no disease modifying therapies currently available. In order to advance novel therapeutics in PD, the identification of biomarkers for PD is of the utmost importance [
      • Gerlach M.
      • Maetzler W.
      • Broich K.
      • et al.
      Biomarker candidates of neurodegeneration in Parkinson's disease for the evaluation of disease-modifying therapeutics.
      ]. It is our hypothesis that PD is a heterogeneous condition and therefore, a paradigm shift is required to make substantial advancements in therapeutic outcomes. Specifically, we hypothesize that specific subsets of PD patients can be treated with targeted therapeutics.
      While much of today's pharmacotherapy is “trial and error” [
      • Jorgensen J.T.
      Companion diagnostics: the key to personalized medicine. Foreword.
      ], precision medicine is a biomarker-guided medicine [
      • Hampel H.
      • O'Bryant S.E.
      • Castrillo J.I.
      • et al.
      Precision medicine - the golden gate for detection, treatment and prevention of Alzheimer's disease.
      ] that is designed to improve early and accurate diagnostics and therapeutics. The FDA defines precision medicine (also known as “personalized medicine”) as “an innovative approach to tailoring disease prevention and treatment that considers differences in people's genes, environments, and lifestyles. The goal of precision medicine is to target the right treatment to the right patients at the right time” [

      Precision Medicine | FDA.

      ]. In fact, biomarker guided therapies in oncology have resulted in drastically improved patient outcomes [
      • Jorgensen J.T.
      Clinical application of companion diagnostics.
      ]. A precision medicine approach has been proposed for numerous diseases [
      • Cresci S.
      • Depta J.P.
      • Lenzini P.A.
      • et al.
      Cytochrome p450 gene variants, race, and mortality among clopidogrel-treated patients after acute myocardial infarction.
      ]; however, few studies have provided direct empirical support. By leveraging previously conducted clinical trial biorepositories, it is possible to provide proof-of-concept data for the precision medicine approach [
      • O'Bryant S.E.
      • Zhang F.
      • Johnson L.A.
      • et al.
      A precision medicine model for targeted NSAID therapy in Alzheimer's disease.
      ]. In fact, we have previously utilized stored samples to demonstrate the utility of a precision medicine approach to treating Alzheimer's disease (AD) using NSAID therapy [
      • O'Bryant S.E.
      • Zhang F.
      • Johnson L.A.
      • et al.
      A precision medicine model for targeted NSAID therapy in Alzheimer's disease.
      ]. Here we sought to determine if a predictive biomarker could be generated to identify the specific subset of PD patients most likely to respond to a given therapy.
      The FDA defines a “Predictive Biomarker” as “a biomarker used to identify individuals who are more likely than similar patients without the biomarker to experience a favorable or unfavorable effect from a specific intervention or exposure [
      FDA-NIH Biomarker Working Group
      BEST (Biomarkers, EndpointS, and Other Tools) Resource Glossary.
      ].” It is our view that the lack of significant advancement in disease modifying clinical trials targeting PD is due to the “one-size-fits-all” approach. Based on our prior work [
      • O'Bryant S.E.
      • Edwards M.
      • Zhang F.
      • et al.
      Potential two-step proteomic signature for Parkinson's disease: pilot analysis in the Harvard Biomarkers Study.
      ], and the extant literature linking inflammation to PD, we hypothesize that inflammation could be a potential biological pathway that could be applied towards the generation of a predictive biomarker, which can facilitate targeted, precision medicine, interventions in PD.
      Numerous studies have implicated inflammation in PD [
      • Hakimi M.
      • Selvanantham T.
      • Swinton E.
      • et al.
      Parkinson's disease-linked LRRK2 is expressed in circulating and tissue immune cells and upregulated following recognition of microbial structures.
      ,
      • Santiago J.A.
      • Scherzer C.R.
      • Potashkin J.A.
      Network analysis identifies SOD2 mRNA as a potential biomarker for Parkinson's disease.
      ,
      • Li Y.J.
      • Oliveira S.A.
      • Xu P.
      • et al.
      Glutathione S-transferase omega-1 modifies age-at-onset of Alzheimer disease and Parkinson disease.
      ]. Neuroinflammation has been found to be significantly upregulated in PD in animal models as well as human [
      • Manocha G.D.
      • Floden A.M.
      • Puig K.L.
      • Nagamoto-Combs K.
      • Scherzer C.R.
      • Combs C.K.
      Defining the contribution of neuroinflammation to Parkinson's disease in humanized immune system mice.
      ]. Additionally, it has been demonstrated that the induction of PD-like symptoms in humanized CD34+ mice using MPTP can cause an associated neuroinflammatory response that was attenuated using FK506, an immunosuppressant drug. Based on these and other studies, it has been suggested that anti-inflammatory medications may have utility in PD [
      • Olsen A.L.
      • Riise T.
      • Scherzer C.R.
      Discovering new benefits fromold drugs with big data-promise for Parkinson disease.
      ,
      • O′neill E.
      • Harkin A.
      Targeting the noradrenergic system for anti-inflammatory and neuroprotective effects: implications for Parkinson's disease.
      ,
      • Wakade C.
      • Giri B.
      • Malik A.
      • et al.
      Niacin modulates macrophage polarization in Parkinson's disease.
      ,
      • Mandal S.
      • Das Mandal S.
      • Chuttani K.
      • Sawant K.K.
      • Subudhi B.B.
      Preclinical study of ibuprofen loaded transnasal mucoadhesive microemulsion for neuroprotective effect in MPTP mice model.
      ,
      • Joshi N.
      • Singh S.
      Updates on immunity and inflammation in Parkinson disease pathology.
      ]. In fact, one recent study demonstrated that probiotic intake impacted inflammatory gene expression in PD patients after 12-weeks [
      • Borzabadi S.
      • Oryan S.
      • Eidi A.
      • et al.
      The Effects of Probiotic Supplementation on Gene Expression Related to Inflammation, Insulin and Lipid in Patients with Parkinson's Disease: A Randomized, Double-Blind, Placebo-Controlled Trial.
      ] with another showing that probiotic supplementation reduced inflammation and decreased MDS-UPDRS scores [
      • Borzabadi S.
      • Oryan S.
      • Eidi A.
      • et al.
      The Effects of Probiotic Supplementation on Gene Expression Related to Inflammation, Insulin and Lipid in Patients with Parkinson's Disease: A Randomized, Double-Blind, Placebo-Controlled Trial.
      ]. Additionally, administration of omega-3 fatty acids with vitamin E has been shown to reduce inflammation and improve UPDRS scores in PD patients [
      • Taghizadeh M.
      • Jamilian M.
      • Mazloomi M.
      • Sanami M.
      • Asemi Z.
      A randomized-controlled clinical trial investigating the effect of omega-3 fatty acids and Vitamin E co-supplementation on markers of insulin metabolism and lipid profiles in gestational diabetes.
      ]. The DATATOP (Deprenyl And Tocopherol Antioxidative Therapy Of Parkinsonism) clinical trial found that deprenyl, an MAO-B inhibitor, 10 mg/day was able to delay time until disability, which warranted initiation of levodopa therapy; however, there was no benefit of deprenyl in postponing levodopa-related adverse events or extending life [
      • Shoulson I.
      • Datatop
      A decade of neuroprotective inquiry. Parkinson study group. Deprenyl and tocopherol antioxidative therapy of parkinsonism.
      ,
      • Datatop
      A multicenter controlled clinical trial in early Parkinson's disease: Parkinson study group.
      ,
      • Penney J.B.
      • Oakes D.
      • Shoulson I.
      • et al.
      Impact of deprenyl and tocopherol treatment on Parkinson's disease in DATATOP patients requiring levodopa.
      ] though problems have been noted with study design and trial endpoints.
      The goal of this study was to generate a proteomic-based predictive biomarker that could distinguish treatment response as defined by reaching primary and secondary endpoints in the previously conducted DATATOP trial. We hypothesize that our previously generated methods for identifying specific subgroups of clinical trial participants that responded to a particular therapy [
      • O'Bryant S.E.
      • Zhang F.
      • Johnson L.A.
      • et al.
      A precision medicine model for targeted NSAID therapy in Alzheimer's disease.
      ] can be further applied to help identify the subset of PD participants that were treatment responders in the DATATOP trial.

      2. Methods

      2.1 Participants and methods of DATATOP

      The sample included in this study were derived from available biorepository samples from participants enrolled in a previously conducted clinical trial for PD (DATATOP). The DATATOP [
      • Datatop
      A multicenter controlled clinical trial in early Parkinson's disease: Parkinson study group.
      ] trial enrolled a total of 800 participants over a 2-year time frame between September 1987 and November 1988 across multiple sites. Of those participants enrolled in the total DATATOP clinical trial, for the current study, the following trial participants were included for further analysis after removing some individuals with missing values: n = 134 from the α-tocopherol and deprenyl placebo treatment arm, n = 135 from the deprenyl and α-tocopherol placebo treatment arm, and n = 131 from the combined active deprenyl and α-tocopherol treatment arm. Of note, trial participants enrolled in the double placebo arm (n = 132) were only reported on in the supplemental analysis of the Results section. No significant group differences were observed in demographic characteristics (age, gender) between DATATOP trial participants across trial arms with biorepository serum samples versus those without (see Supplemental Table 1).
      Inclusion criteria for the trial included a diagnosis of PD (Hoehn and Yahr stage as I or II) within the past 5 years [
      • Datatop
      A multicenter controlled clinical trial in early Parkinson's disease: Parkinson study group.
      ]. Exclusion criteria included the presence of a severe tremor, dementia diagnosis, depression, or use of a symptomatic PD medication. Participants who met inclusion criteria were randomized into one of the following treatment arms: α-tocopherol (2000 IU/d) and deprenyl placebo (TOC), deprenyl (10 mg/d) and α-tocopherol placebo (DEP), active deprenyl and α-tocopherol (TOC and DEP), or double placebo [
      • Datatop
      A multicenter controlled clinical trial in early Parkinson's disease: Parkinson study group.
      ]. After 14 months, the trial was discontinued due to the positive effects shown among those receiving deprenyl on reducing PD symptoms and the study became an open-label administration for the remainder of the two-year trial [
      • Liu C.
      • Cholerton B.
      • Shi M.
      • et al.
      CSF tau and tau/Aβ42 predict cognitive decline in Parkinson's disease.
      ].
      The DATATOP trial primary endpoint was PD symptom progression suggestive of disability requiring dopaminergic therapy. Secondary endpoints included change in PD symptoms as measured by the Unified Parkinson's Disease Rating Scale (UPDRS) motor and total scores [
      • Datatop
      A multicenter controlled clinical trial in early Parkinson's disease: Parkinson study group.
      ]. The UPDRS total score is comprised of the sum of the motor, cognitive, and ADL subscales. Change in UPDRS score was calculated based on prior work [
      • Huang X.
      • Auinger P.
      • Eberly S.
      • et al.
      Serum cholesterol and the progression of Parkinson's disease: results from DATATOP.
      ]. For purposes of this study, the determination of responder status for secondary trial endpoints were made based on empirically defined cut-offs for clinically meaningful change scores in UPDRS motor and UPDRS total scores [
      • Shulman L.M.
      • Gruber-Baldini A.L.
      • Anderson K.E.
      • Fishman P.S.
      • Reich S.G.
      • Weiner W.J.
      The clinically important difference on the unified Parkinson's disease rating scale.
      ]. For UPDRS motor score, the clinically meaningful cut-off score was less than or equal to −2.5 for responders, while others (non-responders and adverse responders) were defined as those greater than or equal to −2.5 [
      • Shulman L.M.
      • Gruber-Baldini A.L.
      • Anderson K.E.
      • Fishman P.S.
      • Reich S.G.
      • Weiner W.J.
      The clinically important difference on the unified Parkinson's disease rating scale.
      ]. For UPDRS total score, the clinically meaningful cut-off score was less than or equal to −4.3 for responders, while others (non-responders and adverse responders) were defined as those greater than or equal to −4.3 [
      • Shulman L.M.
      • Gruber-Baldini A.L.
      • Anderson K.E.
      • Fishman P.S.
      • Reich S.G.
      • Weiner W.J.
      The clinically important difference on the unified Parkinson's disease rating scale.
      ]. For the primary trial endpoint, those participants who did not require Levodopa at 15 months were classified as responders while those who did require Levodopa at the end of the trail were classified as others (adverse responder).

      2.2 Proteomic assays

      All blood biomarker assays were conducted at the University of North Texas Health Science Center in the Institute for Translational Research (ITR) Biomarker Core. Preparation of samples for proteomic assay was conducted using the Hamilton Robotics StarPlus system while any re-aliquoting was conducted via the Hamilton easyBlood robotic system. Plasma samples were assayed via multi-plex biomarker assay platform using electrochemiluminescence (ECL). Plasma samples were assayed with commercially available assays for the following: C-reactive protein (CRP), soluble intercellular adhesion molecule 1 (sICAM1), vascular cell adhesion molecule 1 (sVCAM1), serum amyloid A (SAA), interleukin (IL)-6, IL-10, tumor necrosis factor alpha (TNF-α), IL-5, IL-7, Eotaxin-3, thymus and activation regulated chemokine (TARC), alpha 2 macroglobulin (A2M), beta 2 microglobulin (B2M), Factor 7 (FVII), tenascin C (TNC), fatty acid binding protein 3 (FABP-3), IL-18, thrombopoietin (TPO) and I-309 and were included based on prior work that validated their use in detecting neurogenerative diseases across assay platform, species and tissues [
      • O'Bryant S.E.
      • Xiao G.
      • Zhang F.
      • et al.
      Validation of a serum screen for Alzheimer's disease across assay platforms, species, and tissues.
      ,
      • O'Bryant S.E.
      • Edwards M.
      • Johnson L.
      • et al.
      A blood screening test for Alzheimer's disease.
      ]. The ITR lab has assayed over >20,000 samples on these markers using this system. Inter- and intra-assay variability has been excellent. Average CVs (>3000 samples) for these assays are all <10% with the majority being ≤5%. The ultrasensitive Simoa platform was also used to assay plasma Amyloid beta (Aβ) 40 and 42, total-tau, neurofilament light chain (Nf-L) and alpha-synuclein. The combined MSD and Simoa platform assays were reported on in the supplemental proteomic assay section of the Results, and reflect a broader proteomic biomarker profile. Our team has assayed >5000 samples using this platform and CVs remain ≤5%.

      2.3 Statistical analyses

      The predictive biomarker profile was generated using support vector machine (SVM) analyses. SVM is based on the concept of decision planes that define decision boundaries and serves as a classifier method by performing classification tasks through constructing hyperplanes in a multidimensional space that can separate cases of different class labels. SVM predictive biomarker profiles were based on responders versus others (non-responders, adverse responders) (i.e., only 2 groups). Treatment responder was defined as UPDRS (motor and total) score that was stable or improved at 15 months from baseline whereas non-responder was defined as any decline in UPDRS scores during this same time point. Treatment responder was also defined as not requiring Levodopa at 15 months while others (adverse responders) were those who required Levodopa. The purpose of this approach was to have a predictive biomarker that could selectively identify only those most likely to respond while all others would be ruled out. We conducted internal five-fold cross-validation within the sample with the SVM analyses. The SVM analyses were conducted with the e1071 package (v1.6-8) in R (v3.4.2). In order to build an SVM model to predict treatment response, the radial basis function kernel was used together with five-fold cross-validation, cost = 100 and gamma = 0.001. The original data was randomly partitioned into 5 equal sized subsamples. A single subsample was retained as a testing set and the remaining 4 subsamples were used as training sets. For each model, we ran the internal cross-validation randomly five times. This approach of cross-validation overcame overfitting of the prediction models. SVM does not assume normality and, therefore, raw baseline proteomic data were utilized. The analyses were restricted to treatment arms TOC, DEP and combined (TOC and DEP) as the goal was specifically to identify predictive biomarker of treatment response. The resulting biomarker classifier differs between the three study arms examined in three respects: 1) support vectors, 2) protein weight, and 3) other model parameters such as gamma, cost, and scale. These differences provide each classifier with prediction power for patients under each treatment arm. The prediction performance of using one arm to predict the other 2 arms is not as good as when we build and apply prediction models inside each arm. This could mean that a treatment may capture only a fraction of patients and are therefore, not representative of the whole patients under all three treatment arms. Due to the heterogeneity of PD, we stratified our patients into three groups: TOC, DEP, and DEP + TOC, and modeled each group separately first and then compared their performance. Additionally, to ensure that treatment response was what was being predicted, we combined multiple performance measures as the selection and weighting of measures in performance evaluation is critical. For example, using area under the curve (AUC) alone may not correctly evaluate if the biomarkers predict treatment response because AUC only measures the ranking of the probabilities, and therefore, it does not tell us if the probabilities are well calibrated. We therefore incorporated six measurement in order to better evaluate the performance: 1) precision/positive predictive value (PPV), 2) Accuracy, 3) Sensitivity, 4) Specificity, 5) negatively predictive value (NPV), and 6) AUC.

      3. Results

      3.1 TOC treatment arm

      The DATATOP study outlined several endpoints with the primary being the need for Levodopa and the secondary being change in UPDRS motor and total scores. When the full proteomic panel was applied to the primary endpoint for those randomized into the TOC treatment arm, it produced an AUC of 0.91 for distinguishing responder (did not require Levodopa at 15 months) versus others (required Levodopa) with a sensitivity of 0.86 and specificity of 0.88 with an optimized cut-off of 0.557 (Fig. 1). The addition of age and sex increased the detection accuracy only slightly as AUC reached 0.92 (sensitivity of 0.72, specificity of 0.96) with an optimized cut-off of 0.148. In the same treatment arm (TOC), when the full proteomic panel was applied to detect the secondary endpoint for change in UPDRS motor scores at 15 months from baseline, AUC, sensitivity, and specificity reached 1.00 for distinguishing those who responded and those who experienced no-response or an adverse response with an optimized cut-off of 0.852 (Fig. 1). The addition of age and sex did not improve the detection accuracy, which remained at the highest prediction performance. Again, in the same treatment arm (TOC), when applied to change in UPDRS total score from baseline to 15 months, the full proteomic panel reached an AUC, sensitivity, and specificity of 1.00 for distinguishing those who responded from those who experienced no-response or an adverse response with an optimized cut-off of 0.798 (Fig. 1). The addition of age and sex again did not improve detection accuracy, which remained at the highest prediction performance.
      Fig. 1
      Fig. 1Confusion matrix and variable importance plot for treatment arm receiving TOC for the full panel for distinguishing primary and secondary endpoints. NPV = negative predictive value; PPV = positive predictive value, AUC = Area under the receiver operating characteristic curve; TOC arm = α-tocopherol (2000 IU/d) and deprenyl placebo.

      3.2 DEP treatment arm

      When the full proteomic panel was applied to the primary endpoint for those randomized into the DEP treatment arm, the AUC reached 0.93 while the sensitivity reached 0.82 and specificity reached 0.94 with an optimized cut-off of −0.914 for distinguishing responders versus others (adverse responders) (Fig. 2). The addition of age and sex did not improve the detection accuracy as AUC reached 0.92 with an optimized cut-off of −0.694 (sensitivity of 0.89, specificity of 0.90). When the same proteomic panel was applied to the secondary endpoint of change in UPDRS motor scores at 15 months from baseline in the same treatment arm (DEP), the AUC reached 0.99, while sensitivity and specificity both reached 1.00 in distinguishing responders versus non-responders and adverse responders with an optimized cut-off score of 0.891 (Fig. 2). Addition of age and sex again did not improve detection accuracy, which remained at the highest prediction performance (AUC of 0.99, sensitivity of 1.00, specificity of 0.98). When applied to the other secondary endpoint of change in UPDRS total scores at 15 months from baseline again in the same treatment arm (DEP), AUC, sensitivity, and specificity all reached 1.00 with an optimized cut-off of 0.961 (Fig. 2). The addition of age and sex again did not improve detection accuracy, which remained at the highest prediction performance.
      Fig. 2
      Fig. 2Confusion matrix and variable importance plot for treatment arm receiving DEP for the full MSD panel for distinguishing primary and secondary endpoints. NPV = negative predictive value; PPV = positive predictive value, AUC = Area under the receiver operating characteristic curve; DEP arm = deprenyl (10 mg/d) and α-tocopherol placebo.

      3.3 Combined TOC and DEP treatment arm

      When the full proteomic panel was additionally applied to the primary endpoint for those randomized into the treatment arm of combined TOC and DEP, it produced an AUC of 0.87 with a sensitivity of 0.83 and specificity of 0.83 with an optimized cut-off of −0.881 for distinguishing responders from others (adverse responders) (see Fig. 3). The inclusion of age and sex improved the detection accuracy of the model as AUC reached 0.92 with an optimized cut-off score of −0.764 (sensitivity of 0.92, specificity of 0.79). Again, for the combined treatment arm (TOC and DEP), the same proteomic panel remained elevated when applied to the secondary endpoint of change in UPDRS motor scores from baseline to 15 months for distinguishing responders from non-responders and adverse responders with an AUC of 0.96, sensitivity of 0.93 and specificity of 0.99 with an optimized cut-off of 0.944 (Fig. 3). Inclusion of age and sex improved detection accuracy as AUC increase to 0.99 with an optimized cut-off of 0.768 (sensitivity of 0.85, specificity of 1.00). When applied to detect another secondary endpoint of change in UPDRS total score from baseline to 15 months for the same combined treatment arm (TOC and DEP), the same proteomic panel reached an AUC of 0.94, sensitivity of 0.64, and specificity of 1.00 with an optimized cut-off of 0.731 for distinguishing responders from non-responders and adverse responders (Fig. 3). The addition of age and sex improved detection accuracy some as AUC increased to 0.95 with an optimized cut-off score of 0.841 (sensitivity of 0.82, specificity of 1.00).
      Fig. 3
      Fig. 3Confusion matrix and variable importance plot for the combined treatment arm receiving both TOC and DEP for the full MSD panel for distinguishing primary and secondary endpoints. NPV = negative predictive value; PPV = positive predictive value, AUC = Area under the receiver operating characteristic curve; combined TOC + DEP arm.

      3.4 Supplemental analyses

      Additional supplemental analyses were also conducted to examine the utility of the proteomic predictive biomarker in distinguishing the need for Levodopa if the treatment arms were combined (DEP, TOC, and DEP + TOC) as well as for the placebo arm. Findings revealed that combination of groups (DEP, TOC, and DEP + TOC) yielded a lower performance than when each group was modeled separately (accuracy of 0.71, sensitivity of 0.67, specificity of 0.74, NPV of 70%, Precision/PPV of 72%, AUC of 0.77). For the placebo arm, similar results were found as for the DEP + TOC arm (accuracy of 0.85, sensitivity of 0.80, specificity of 0.88, NPV of 89%, Precision/PPV of 79%, AUC of 0.87).
      Supplemental analyses were also conducted to expand the MSD proteomic panel to include additional biomarkers specifically related to AD (Aβ40 and 42, total-tau, Nf-L) and PD (α-synuclein) pathology. The purpose of including AD/PD specific proteins to the proteomic panel (previously reported) was to see if this increased the detection accuracy of the biomarkers in distinguishing responder status. When this supplemental panel was applied to detect those requiring Levodopa at 15 months (adverse responders) from those who did not (responders) in the TOC treatment arm, the AUC increased to 0.93 with an optimized cut-off of 0.554 while sensitivity remained unchanged and specificity increased to 0.90. When applied to detect change in UPDRS motor and total scores in this same treatment arm (TOC) at 15 months from baseline reflecting responder versus others, AUC, sensitivity, and specificity all remained unchanged with the additional markers. For the DEP treatment arm, the supplemental proteomic panel increased AUC just slightly to 0.95 when applied to distinguish those who required Levodopa at 15 months (adverse responders) versus those who did not (responders) while sensitivity increased to 0.89 and specificity decreased to 0.92 with an optimized cut-off of −0.863. In this same treatment arm (DEP), the added biomarkers increased AUC and specificity to 1.00 for detecting change in UPDRS motor scores at 15 months from baseline (optimized cut-off for UPDRS motor score of 0.864). AUC, sensitivity, and specificity remained unchanged in this treatment arm for distinguishing responder status for UPDRS total scores. When applied to the combined treatment arm (TOC and DEP), the added biomarkers increased AUC to 0.91, sensitivity to 0.87, and specificity to 0.89 with an optimized cut-off of −0.916 when applied to distinguishing the primary endpoint for need of Levodopa at 15 months. For distinguishing responder status in this same combined treatment arm (TOC and DEP) based on change in UPDRS motor score from baseline to 15 months, the added biomarkers increased AUC, along with sensitivity and specificity to 1.00 with an optimized cut-off of 0.917. Similarly, for this same treatment arm (TOC and DEP), the supplemental proteomic panel increased AUC and sensitivity for distinguishing change in UPDRS total score from baseline to 15 months while specificity remained unchanged at 1.00 with an optimized cut-off of 0.9.

      4. Discussion

      The current results provide strong empirical support for the investigation into a precision medicine approach to treating patients suffering from PD. Our results revealed that the biomarkers were predictive of patients in the DATATOP study reaching the endpoint representing disease progression (need of Levodopa at 15 months) versus those who did not experience disease progression with AUCs ranging from 87 to 93%. Higher prediction accuracy was found when applying the same predictive biomarker for distinguishing responder status on secondary trial endpoints (UPDRS motor and total scores) with AUCs ranging from 94 to 100% with the applied optimized cut-off score. The addition of AD and PD specific biomarkers to the supplemental predictive biomarker increased detection accuracy only slightly (AUC increase of 2%) across treatment arms for the primary endpoint as well as for the secondary endpoint for those in the DEP (UPDRS motor score AUC increase of 1%) and combined (UPDRS motor score AUC increase of 5%, UPDRS total score AUC increase of 6%) treatment arms. Inclusion of demographic factors (age, sex) improved detection accuracy of the predictive biomarkers by 1–5% for the primary endpoint in the TOC alone or combined treatment arm as well as 1–3% for the secondary endpoints for the combined treatment arm only. Across treatment arms, the variable importance plots revealed a number of inflammatory proteins spanning both pro-inflammatory and general inflammation among the top markers when applied to trial endpoints.
      Our prior work demonstrates that inflammatory markers are highly relevant in detecting and discriminating PD from other neurodegenerative diseases. We conducted a study in the Harvard Biomarker Study Biorepository to assess the ability to discriminate PD from other neurodegenerative diseases by assaying n = 150 plasma samples (PD n = 50, “other neurodegenerative disease” n = 50 [AD n = 12, FTD n = 25, other n = 13], controls n = 50) [
      • O'Bryant S.E.
      • Edwards M.
      • Zhang F.
      • et al.
      Potential two-step proteomic signature for Parkinson's disease: pilot analysis in the Harvard Biomarkers Study.
      ]. The proteomic profile approach was highly accurate in discriminating PD from other neurodegenerative diseases with an AUC = 0.98. Additionally, multiple inflammatory markers were among the top 10 in the profile for discriminating PD from other neurodegenerative disease [
      • O'Bryant S.E.
      • Edwards M.
      • Zhang F.
      • et al.
      Potential two-step proteomic signature for Parkinson's disease: pilot analysis in the Harvard Biomarkers Study.
      ]. Therefore, our work, as well as that of others, support the notion of the importance of inflammation in PD. Combined, this work also provides empirical support for the use of biomarkers for the creation of novel diagnostic and therapeutic approaches for PD.
      In our prior work, we have also demonstrated the potential utility of a precision medicine approach to treating AD. We assayed baseline plasma samples (pre-randomization) from the ADCS AD anti-inflammatory trial [
      • Aisen P.S.
      • Schafer K.A.
      • Grundman M.
      • et al.
      Effects of rofecoxib or naproxen vs placebo on alzheimer disease progression: a randomized controlled trial.
      ] to test our hypothesis that the pro-inflammatory endophenotype would predict treatment response among AD patients. As was shown in our prior work [
      • O'Bryant S.E.
      • Zhang F.
      • Johnson L.A.
      • et al.
      A precision medicine model for targeted NSAID therapy in Alzheimer's disease.
      ], a targeted panel of markers could be utilized to generate a predictive biomarker for the identification of the specific subsets of patients who would most likely benefit from specified therapies. In that study, we analyzed data from the ADCS NSAID trial [
      • Aisen P.S.
      • Schafer K.A.
      • Grundman M.
      • et al.
      Effects of rofecoxib or naproxen vs placebo on alzheimer disease progression: a randomized controlled trial.
      ], a multicenter, randomized, double-blind, placebo-controlled parallel group trial with 1-year exposure to study medications. Individuals who met enrollment criteria with a diagnosis of probable AD in this trial were randomized to rofecoxib (25 mg once daily), naproxen (220 mg twice-daily) or placebo. In that study, our inflammatory-specific predictive biomarker was 97% accurate in identifying treatment response to naproxen and 98% accurate in identifying treatment response to rofecoxib [
      • O'Bryant S.E.
      • Zhang F.
      • Johnson L.A.
      • et al.
      A precision medicine model for targeted NSAID therapy in Alzheimer's disease.
      ].
      There are limitations to the current study. First, this is an initial proof-of-concept study and additional prospective studies are needed to create predictive biomarkers for any therapeutics. Secondly, it is possible that additional markers will improve the overall accuracy of the approach. While we had targeted proteomic markers a priori selected, we also assayed a broader panel for additional refinements of the model as needed. Another limitation to the study includes the DATATOP study drug Deprenyl, an MAO-B inhibitor, which has been shown to have symptomatic effects on PD [
      • Teo K.C.
      • Ho S.-L.
      Monoamine oxidase-B (MAO-B) inhibitors: implications for disease-modification in Parkinson's disease.
      ,
      • Miklya I.
      • Knoll B.
      • Knoll J.
      A pharmacological analysis elucidating why, in contrast to (-)-deprenyl (selegiline), alpha-tocopherol was ineffective in the DATATOP study.
      ], which may impact disease progression thereby potentially limiting the findings from the treatment group that included this study drug. Additional limitations exist with the PD diagnosis of those in DATATOP as some individuals may have been misdiagnosed, thereby impacting the findings for those who were classified as responders. This study also only sought to provide a comparison of overall performance for each treatment arm to response and did not seek to predict what the best PD therapy could be for a specific patient. Upcoming analyses are planned to add the clinical application analysis from this study in the future.
      Despite some of the limitations for the study, this work highlights the need to continue studying other previously conducted trials and biorepositories. With sufficient data, such work can easily set the stage for novel trials that selects patients using predictive biomarkers. This approach has substantial advantages. First, by enrolling only the subset of patients most likely to respond, the effect size increases and thereby the sample size reduces substantially. Second, if multiple subgroups are identified within the PD general population, a patient screening system can examine for multiple subgroups within any given patient to facilitate multiple trial enrollment. Third, this approach reduces patient burden by excluding patients who should not undergo investigational drugs. Fourth, this approach can expedite novel therapies to patients thereby providing patients with novel therapies and extending patent life for companies. Finally, an application of these methods to existing drugs (with known qualities to avoid preclinical and SAD/MAD studies), can be expedited to patients by leveraging the 505(b) (2) mechanism at the FDA [

      Applications Covered by Section 505(b)(2) | FDA.

      ]. Overall, the current results strongly support the possibility of a precision medicine approach to treating PD.
      The raw data utilized in this study has been made available to the MJFF with the goal of widespread use by the scientific community. Therefore, anyone interested can receive the data for follow-up analyses. Additionally, the interested investigator can reach out to the authorship team and will be provided with the proteomic data. Our goal with this study is to establish proof-of-concept of a precision medicine approach to PD therapeutics. We acknowledge that this is a first-step and other investigators may have better approaches to the data analytics and, therefore, we encourage use of the data.

      Acknowledgements

      Funding for this work was provided by the Michael J. Fox Foundation for Parkinson's Disease Research under award number 14448 . Data from this study can be obtained from the MJFF.

      Appendix A. Supplementary data

      The following is the Supplementary data to this article:

      References

        • Tysnes O.B.
        • Storstein A.
        Epidemiology of Parkinson's disease.
        J. Neural. Transm. 2017; 124: 901-905https://doi.org/10.1007/s00702-017-1686-y
        • Huse D.M.
        • Schulman K.
        • Orsini L.
        • Castelli-Haley J.
        • Kennedy S.
        • Lenhart G.
        Burden of illness in Parkinson's disease.
        Mov. Disord. 2005; (Published online)https://doi.org/10.1002/mds.20609
        • Gerlach M.
        • Maetzler W.
        • Broich K.
        • et al.
        Biomarker candidates of neurodegeneration in Parkinson's disease for the evaluation of disease-modifying therapeutics.
        J. Neural. Transm. 2012; (Published online)https://doi.org/10.1007/s00702-011-0682-x
        • Jorgensen J.T.
        Companion diagnostics: the key to personalized medicine. Foreword.
        Expert Rev. Mol. Diagn. 2015; 15: 153-156https://doi.org/10.1586/14737159.2015.1002470
        • Hampel H.
        • O'Bryant S.E.
        • Castrillo J.I.
        • et al.
        Precision medicine - the golden gate for detection, treatment and prevention of Alzheimer's disease.
        J. Prevent. Alzheimer's Dis. 2016; (Published online)https://doi.org/10.14283/jpad.2016.112
      1. Precision Medicine | FDA.

        • Jorgensen J.T.
        Clinical application of companion diagnostics.
        Trends Mol. Med. 2015; 21: 405-407https://doi.org/10.1016/j.molmed.2015.05.003
        • Cresci S.
        • Depta J.P.
        • Lenzini P.A.
        • et al.
        Cytochrome p450 gene variants, race, and mortality among clopidogrel-treated patients after acute myocardial infarction.
        Circulation: Cardiovascular Genetics. 2014; (Published online)https://doi.org/10.1161/CIRCGENETICS.113.000303
        • O'Bryant S.E.
        • Zhang F.
        • Johnson L.A.
        • et al.
        A precision medicine model for targeted NSAID therapy in Alzheimer's disease.
        J. Alzheim. Dis. 2018; 66: 97-104https://doi.org/10.3233/JAD-180619
        • FDA-NIH Biomarker Working Group
        BEST (Biomarkers, EndpointS, and Other Tools) Resource Glossary.
        2016https://doi.org/10.1093/neuonc/nox242
        • O'Bryant S.E.
        • Edwards M.
        • Zhang F.
        • et al.
        Potential two-step proteomic signature for Parkinson's disease: pilot analysis in the Harvard Biomarkers Study.
        Alzheimer's Dementia: Diagn. Assess. Dis. Monit. 2019; 11: 374-382https://doi.org/10.1016/j.dadm.2019.03.001
        • Hakimi M.
        • Selvanantham T.
        • Swinton E.
        • et al.
        Parkinson's disease-linked LRRK2 is expressed in circulating and tissue immune cells and upregulated following recognition of microbial structures.
        J. Neural. Transm. 2011; (Published online)https://doi.org/10.1007/s00702-011-0653-2
        • Santiago J.A.
        • Scherzer C.R.
        • Potashkin J.A.
        Network analysis identifies SOD2 mRNA as a potential biomarker for Parkinson's disease.
        PLoS One. 2014; (Published online)https://doi.org/10.1371/journal.pone.0109042
        • Li Y.J.
        • Oliveira S.A.
        • Xu P.
        • et al.
        Glutathione S-transferase omega-1 modifies age-at-onset of Alzheimer disease and Parkinson disease.
        Hum. Mol. Genet. 2003; (Published online)https://doi.org/10.1093/hmg/ddg357
        • Manocha G.D.
        • Floden A.M.
        • Puig K.L.
        • Nagamoto-Combs K.
        • Scherzer C.R.
        • Combs C.K.
        Defining the contribution of neuroinflammation to Parkinson's disease in humanized immune system mice.
        Mol. Neurodegener. 2017; (Published online)https://doi.org/10.1186/s13024-017-0158-z
        • Olsen A.L.
        • Riise T.
        • Scherzer C.R.
        Discovering new benefits fromold drugs with big data-promise for Parkinson disease.
        JAMA Neurology. 2018; (Published online)https://doi.org/10.1001/jamaneurol.2018.0345
        • O′neill E.
        • Harkin A.
        Targeting the noradrenergic system for anti-inflammatory and neuroprotective effects: implications for Parkinson's disease.
        Neural Regen. Res. 2018; (Published online)https://doi.org/10.4103/1673-5374.235219
        • Wakade C.
        • Giri B.
        • Malik A.
        • et al.
        Niacin modulates macrophage polarization in Parkinson's disease.
        J. Neuroimmunol. 2018; (Published online)https://doi.org/10.1016/j.jneuroim.2018.05.002
        • Mandal S.
        • Das Mandal S.
        • Chuttani K.
        • Sawant K.K.
        • Subudhi B.B.
        Preclinical study of ibuprofen loaded transnasal mucoadhesive microemulsion for neuroprotective effect in MPTP mice model.
        Iran. J. Pharm. Res. (IJPR). 2018; (Published online)https://doi.org/10.22037/ijpr.2018.2169
        • Joshi N.
        • Singh S.
        Updates on immunity and inflammation in Parkinson disease pathology.
        J. Neurosci. Res. 2018; (Published online)https://doi.org/10.1002/jnr.24185
        • Borzabadi S.
        • Oryan S.
        • Eidi A.
        • et al.
        The Effects of Probiotic Supplementation on Gene Expression Related to Inflammation, Insulin and Lipid in Patients with Parkinson's Disease: A Randomized, Double-Blind, Placebo-Controlled Trial.
        Archives of Iranian Medicine, 2018 (Published online)
        • Borzabadi S.
        • Oryan S.
        • Eidi A.
        • et al.
        The Effects of Probiotic Supplementation on Gene Expression Related to Inflammation, Insulin and Lipid in Patients with Parkinson's Disease: A Randomized, Double-Blind, Placebo-Controlled Trial.
        Archives of Iranian Medicine, 2018 (Published online)
        • Taghizadeh M.
        • Jamilian M.
        • Mazloomi M.
        • Sanami M.
        • Asemi Z.
        A randomized-controlled clinical trial investigating the effect of omega-3 fatty acids and Vitamin E co-supplementation on markers of insulin metabolism and lipid profiles in gestational diabetes.
        J. Clin. Lipidol. 2016; (Published online)https://doi.org/10.1016/j.jacl.2015.12.017
        • Shoulson I.
        • Datatop
        A decade of neuroprotective inquiry. Parkinson study group. Deprenyl and tocopherol antioxidative therapy of parkinsonism.
        Ann. Neurol. 1998; 44 (Published online): S160-S166
        • Datatop
        A multicenter controlled clinical trial in early Parkinson's disease: Parkinson study group.
        Arch. Neurol. 1989; (Published online)https://doi.org/10.1001/archneur.1989.00520460028009
        • Penney J.B.
        • Oakes D.
        • Shoulson I.
        • et al.
        Impact of deprenyl and tocopherol treatment on Parkinson's disease in DATATOP patients requiring levodopa.
        Ann. Neurol. 1996; (Published online)https://doi.org/10.1002/ana.410390107
        • Liu C.
        • Cholerton B.
        • Shi M.
        • et al.
        CSF tau and tau/Aβ42 predict cognitive decline in Parkinson's disease.
        Park. Relat. Disord. 2015; (Published online)https://doi.org/10.1016/j.parkreldis.2014.12.027
        • Huang X.
        • Auinger P.
        • Eberly S.
        • et al.
        Serum cholesterol and the progression of Parkinson's disease: results from DATATOP.
        PLoS One. 2011; (Published online)https://doi.org/10.1371/journal.pone.0022854
        • Shulman L.M.
        • Gruber-Baldini A.L.
        • Anderson K.E.
        • Fishman P.S.
        • Reich S.G.
        • Weiner W.J.
        The clinically important difference on the unified Parkinson's disease rating scale.
        Arch. Neurol. 2010; (Published online)https://doi.org/10.1001/archneurol.2009.295
        • Shulman L.M.
        • Gruber-Baldini A.L.
        • Anderson K.E.
        • Fishman P.S.
        • Reich S.G.
        • Weiner W.J.
        The clinically important difference on the unified Parkinson's disease rating scale.
        Arch. Neurol. 2010; (Published online)https://doi.org/10.1001/archneurol.2009.295
        • O'Bryant S.E.
        • Xiao G.
        • Zhang F.
        • et al.
        Validation of a serum screen for Alzheimer's disease across assay platforms, species, and tissues.
        J. Alzheim. Dis. 2014; 42: 1325-1335https://doi.org/10.3233/JAD-141041
        • O'Bryant S.E.
        • Edwards M.
        • Johnson L.
        • et al.
        A blood screening test for Alzheimer's disease.
        Alzheimer's Dementia: Diagn. Assess. Dis. Monit. 2016; 3: 83-90https://doi.org/10.1016/j.dadm.2016.06.004
        • Aisen P.S.
        • Schafer K.A.
        • Grundman M.
        • et al.
        Effects of rofecoxib or naproxen vs placebo on alzheimer disease progression: a randomized controlled trial.
        J. Am. Med. Assoc. 2003; (Published online)https://doi.org/10.1001/jama.289.21.2819
        • Teo K.C.
        • Ho S.-L.
        Monoamine oxidase-B (MAO-B) inhibitors: implications for disease-modification in Parkinson's disease.
        Transl. Neurodegener. 2013; 2: 19https://doi.org/10.1186/2047-9158-2-19
        • Miklya I.
        • Knoll B.
        • Knoll J.
        A pharmacological analysis elucidating why, in contrast to (-)-deprenyl (selegiline), alpha-tocopherol was ineffective in the DATATOP study.
        Life Sci. 2003; 72: 2641-2648https://doi.org/10.1016/s0024-3205(03)00174-7
      2. Applications Covered by Section 505(b)(2) | FDA.