THBS2 as a prognostic biomarker for patients diagnosed with metastatic pancreatic ductal adenocarcinoma

Patients newly diagnosed with metastatic pancreatic ductal adenocarcinoma generally have poor survival, with heterogeneous rates of progression. Biomarkers that could predict progression and/or survival would help inform patients and providers as they make care decisions. In a previous retrospective study, we discovered that circulating thrombospondin-2 (THBS2) could, in combination with CA19-9, better distinguish patients with PDAC versus healthy controls. Here we evaluated whether THBS2 levels, previously not known to be prognostic, were associated with outcome in 68 patients at time of diagnosis of metastatic PDAC. Specifically, we interrogated the association of THBS2 level, alone or in combination with CA19-9, with progression by 90 days and/or survival to 180 days. The results indicate that elevated THBS2 levels alone, at the time of a metastatic PDAC diagnosis, can identify patients with a shorter time to death and thus help patients and providers when planning treatment.


INTRODUCTION
Pancreatic ductal adenocarcinoma (PDAC) is often diagnosed at advanced stages of disease, with unresectable tumors that respond poorly to systemic therapy [1,2].The low incidence of PDAC and the absence of a known genetic or familial predisposition in most cases make it challenging to develop biomarkers with necessary sensitivity and specificity for early detection [3][4][5].In addition, recent studies indicate that PDAC may progress more rapidly [6] than previously appreciated [7,8].
Indeed, the challenges of early detection have led to the suggestion that resources should be shifted towards better predictors of progression and survival for clinical management of individuals already diagnosed with PDAC [9].Thus, the present study is to assess blood biomarkers for discriminating between outcomes for patients with metastatic PDAC (mPDAC).
Carbohydrate antigen 19-9 (CA19-9) has been used as a diagnostic and prognostic marker for PDAC, but has insufficient power for broad utility [10].We previously discovered candidate PDAC biomarkers with an induced

Research Paper
This article has been corrected.Correction in: Oncotarget.2022; 13:1187-1187.www.oncotarget.compluripotent stem cell line, derived from a recurrent, late stage PDAC tumor [11].In a retrospective study of controls and PDAC patients with various stages of disease at the time of diagnosis, plasma thrombospondin-2 (THBS2) levels, combined with serum CA19-9 [12,13], could discriminate PDAC from controls with an overall specificity of 98% at a sensitivity of 87% [14].Yet, in a subsequent analysis of prospectively collected samples, neither THBS2, CA19-9, nor the combination were able to sensitively predict PDAC up to one year prior to a clinical diagnosis [15].
Definitive surgical therapy is not an accepted option for mPDAC, and therefore these individuals are often offered systemic chemotherapy with a variety of potential regimens [16].There is considerable heterogeneity in the time to disease progression and death that has been difficult to predict for each patient [17].A biomarker that prognosticates outcome and potentially predicts treatment response for mPDAC patients would help indicate the urgency for treatment, whether to treat at all, possible stratification onto a clinical trial, and patient's planning that inevitably accompanies a PDAC diagnosis.Given the demonstrated utility for detecting PDAC at the time of clinical presentation, here we tested THBS2 and CA19-9 at time of diagnosis as prognostic indicators for mPDAC.

Study cohort
Among the 68 study patients (Table 1), 54 (79.4%) patients had died (median = 224 days, range: 8 to 1085 days) and 14 were alive at last follow up (353 days; 206 to 1085).The 180-day mortality rate was 30.9% (21 of 68 patients).Of the 68 study patients, 62 (91%) had progression of PDAC with a median of 234 days (range: 8 to 1085); the remaining 6 patients without progression had a median follow-up time of 353 days (range: 213 to 598).The 90-day progression rate was 45.6% (31 of 68 patients).There were 18 patients who had both progression at 90 days and survived less than 180 days.One characteristic, first line therapy, differed significantly between patients alive or dead at 180 days (p = 0.041).There were no significant differences in age, sex, race, smoking, or diabetes status at blood draw primary location of the tumor, ECOG performance status, first line therapy or sites of metastasis between patients with and without progression at 90 days.

Combination of THBS2 and CA19-9
Multivariable logistic regression analyses were used to evaluate whether a combination of THBS2 and log2-CA19-9 values would be a useful biomarker of death by 180 days and/or progression by 90 days.Table 2 presents the univariable and multivariable regression models.In these analyses, THBS2 and log2-CA19-9 measurements were each normalized by subtracting their respective sample means and dividing by their sample standard deviations.In the univariate models the THBS2 coefficients for both death by 180 days and progression by 90 days were statistically significant (p < 0.001 and p = 0.009, respectively).However, the log2-CA19-9 coefficient was significant only for death by 180 days (p = 0.008).For death by 180 days, the AUC for THBS2 alone (0.803) and the AUC for the best linear predictor from the model with THBS2 and log2-CA199 was higher (0.843), but these AUCs were not significantly different (p = 0.224).

DISCUSSION
Although considerable effort has been focused on improving the detection of PDAC at an early, potentially curable stage, most patients have metastatic disease at diagnosis.As a result of low rates of response and poor outcomes for many of these patients, therapeutic decision making is difficult and could be improved by a prognostic biomarker.Circulating GPC1-expressing exosomes, several circulating RNAs, and circulating tumor DNA have all been proposed as potential prognostics; however, none have been clinically implemented [17].CA19-9 and CEA may be used as indicators of response [10], though neither is utilized as a baseline prognostic.
We focused on endpoints with high clinical relevance: progression by 90 days, the standard period for first radiographic monitoring of patients on therapy, and death by 180 days, the prognostic requirement for hospice care.Knowing a patient's likelihood of such poor prognoses may change the decision-making of patients and physicians, leading to differing levels of aggressiveness in treatment, opting for experimental therapies, or forgoing treatment.Similarly, the information may help physicians set expectations for clinical course and outcomes with patients and their families.Further studies on biomarker dynamics during treatment should also be informative.
Though this study is limited by small sample size and was performed at a single institution, the results warrant further study to confirm findings, validate cut points, and demonstrate replicability.Further, inclusion of other identified prognostic biomarkers and risk factors in a multi-analyte/variable clinical model may improve utility.Additionally, research into the biology underlying the THBS2 prognostic signal may reveal targetable or other measurable differences in good and poor prognostic groups.  1 The intercepts in the univariate regression models for 180-day death were −2.864 and −3.237 for THBS2 and Log2-CA19-9, respectively. 2The intercepts in the univariate regression models for 90-day progression were −1.386 and −0.029 for THBS2 and Log2-CA19-9, respectively. 3Interaction was not significant in the multivariate model for 180-day death (p = 0.420) or for 90-day progression (p = 0.106).

Blood collection and processing
Blood samples were collected from therapy-naive individuals with mPDAC at diagnosis at the University of Pennsylvania Abramson Cancer Center between 2015 and 2018.This prospective nonrandomized observational study (IRB #822028) was approved by the University of Pennsylvania Institutional Review Board and all individuals who had a blood sample collected signed informed consent.The ages, sex, and other subject demographics are in Table 1.Blood was collected in K2EDTA collection tubes and plasma supernatants were isolated after each of two centrifugation steps (1,600 g × 0 min and 3,000 g × 15 min), after which it was aliquoted and banked at −80°C.

Primary outcomes
The study focused on two clinically relevant outcomes.The first, death by 180 days, was reached if the patient died within 180-days of blood sampling.The second, progression by 90 days, was reached if the patient had clinician-defined progression (based on radiographic or clinical evidence) or died within 90-days of blood sampling.

Biomarkers
THBS2 replicate values were determined from banked plasma using a commercial ELISA kit (Bio-Techne) as described [15].CA19-9 was measured on banked plasma samples at the Hospital of the University of Pennsylvania Clinical Endocrinology Laboratory via the Elecsys CA19-9 Immunoassay run on a Cobas e601 platform (Roche) by electrochemiluminescence per the manufacturer's instructions.The operator was blinded to sample identity.For statistical analyses, CA19-9 values were transformed using the binary logarithm (base 2) due to the large range of the clinical values (4.5 to 793,700).

Statistical methods
The distributions of clinical factors at diagnosis, including age and sex, stratified by progression by 90 days and death by 180 days were compared using either Fisher's exact test (2 categories) or an exact chi-square test (>2 categories, Table 1).The area under the receiver operating curve (AUC) was computed with its 95% confidence interval (CI) for each biomarker and outcome.Potential binary cut points for THBS2 and CA19-9 were selected by maximizing Youden's index and were used to calculate the sensitivity/ specificity, and positive/negative predictive values of the two tests.Linear logistic regression prediction models were developed for each outcome to investigate whether THBS2 and log2 CA19-9, and potentially their interaction, would be more predictive than either of the two biomarkers alone.Hosmer-Lemeshow goodness-of-fit test was used to evaluate model lack of fit.For each outcome, THBS2 AUCs and 95% CIs were computed for patients based on their firstline therapy, either FOLFIRINOX (n = 25) or gemcitabine plus nab-paclitaxel with or without hydroxychloroquine (n = 36) received as part of a clinical trial [18]).P-values less than 0.05 were considered statistically significant.SAS Version 9.4 and NCSS 2021 were used for statistical analyses.

Table 1 : Description of the study cohort by 180-day death and 90-day progression All patients Death by 180 days p-value Progression by 90 days
* At time of blood draw.