PCPro: a clinically accessible, circulating lipid biomarker signature for poor-prognosis metastatic prostate cancer

Background Using comprehensive plasma lipidomic profiling from men with metastatic castration-resistant prostate cancer (mCRPC), we have previously identified a poor-prognostic lipid profile associated with shorter overall survival (OS). In order to translate this biomarker into the clinic, these men must be identifiable via a clinically accessible, regulatory-compliant assay. Methods A single regulatory-compliant liquid chromatography-mass spectrometry assay of candidate lipids was developed and tested on a mCRPC Discovery cohort of 105 men. Various risk-score Cox regression prognostic models of OS were built using the Discovery cohort. The model with the highest concordance index (PCPro) was chosen for validation and tested on an independent Validation cohort of 183 men. Results PCPro, the lipid biomarker, contains Cer(d18:1/18:0), Cer(d18:1/24:0), Cer(d18:1/24:1), triglycerides and total cholesterol. Within the Discovery and Validation cohorts, men who were PCPro positive had significantly shorter OS compared to those who were PCPro negative (Discovery: median OS 12.0 months vs 24.2 months, hazard ratio (HR) 3.75 [95% confidence interval (CI) 2.29–6.15], p < 0.001, Validation: median OS 13.0 months vs 25.7 months, HR = 2.13 [95% CI 1.46–3.12], p < 0.001). Conclusions We have developed PCPro, a lipid biomarker assay capable of prospectively identifying men with mCRPC with a poor prognosis. Prospective clinical trials are required to determine if men who are PCPro positive will benefit from therapeutic agents targeting lipid metabolism.


INTRODUCTION
Prostate cancer is the second most common cancer and fifth highest cause of cancer death in men worldwide [1].Despite new treatments for metastatic castration-resistant prostate cancer (mCRPC), including taxane chemotherapy, androgen receptor signalling inhibitors (ARSI), poly-ADP ribose polymerase (PARP) inhibitors and lutetium-177-prostate-specific membrane antigen [2], men will either have intrinsic resistance or develop treatment resistance.New biomarkers and treatment strategies are needed.The long-term control of mCRPC requires strategies targeting multiple hallmarks of cancer including neoplastic cells, tumour microenvironment, inflammation, genetics and systemic metabolic factors (including lipid metabolism) [3].
Most circulating ceramides are derived from the liver [8] and are elevated in systemic inflammation [9].However, prostate cancer cells also express the appropriate biosynthetic enzymes and may produce ceramides that are transported into the circulation [5,10].Ceramides may contribute to therapeutic resistance and tumour growth through their conversion into pro-survival sphingosine-1phosphate (S1P) [8].Signal transduction pathways mediated by S1P promote cancer cell proliferation, migration, invasion and regulate lymphocyte trafficking by acting on S1P-specific receptors present on immune cells and cancer cells [8].Therefore, we hypothesise that a poor prognostic plasma lipid profile consisting of elevated sphingolipids is indicative of abnormalities in sphingolipid metabolism which contribute to treatment resistance and prostate cancer progression [7].Inhibition of the ceramide-S1P signalling axis was able to suppress cancer growth [7,8].Thus, the clinical outcomes of men with a poor prognostic lipid profile composed of sphingolipids may be improved by sphingolipidtargeting therapies.
However, in order to integrate selection of patients with a poor prognostic lipid profile into clinical trials, these men must be identifiable via a clinically accessible, regulatory-compliant assay.The 3LS is measured using a high-throughput LC-MS method, which lacks standardisation and validation via regulatory or industry standards [11,12].Reproducibility of LC-MS methods between laboratories remains challenging [13] and development of an assay according to National Pathology Accreditation Advisory Council (NPAAC) guidelines is required for clinical implementation [12].
This study aims to develop an accurate prognostic plasma lipid biomarker assay in accordance with the NPAAC guideline [12], that can identify men with mCRPC who have poor OS.Our study consisted of two parts-firstly, optimising a single LC-MS assay capable of quantifying a panel of candidate lipids and ensuring that the assay is accurate, precise, robust and regulatory-compliant [11].Secondly, running the assay on plasma samples from two cohorts of patients with mCRPC, to develop and validate a risk-score model from the optimum combination of lipids capable of identifying men with poor prognosis.

Quantitation of plasma lipids
Plasma from the Discovery and Validation Cohorts were analysed using three methods.Method 1: Candidate lipids were measured in plasma using a targeted LC-MS assay, developed with quantitation based on calibration standards of reference plasma and adjustment with stable isotope internal standards (S2).Method 2: Total cholesterol, high-density lipoprotein (HDL) and triglycerides were measured by enzymatic colorimetric assays using the COBAS 8000 analyser (module C702) (Roche).Method 3: Highthroughput lipidomic analysis of >300 lipids with relative quantitation was performed using LC-MS as described previously (S3) [4,6].Results from the high-throughput lipidomic analysis were used to identify samples with the 3LS.

Statistical analysis
Statistical analysis was performed with the software R v4.1.1 (referenced in S4) and IBM SPSS v27.OS was calculated from the date of treatment commencement to death and censored at date of last follow-up if death had not occurred.
The sum of the variables of the Cox regression formula (i.e., the lipid concentrations [X] multiplied by the natural logarithm of the hazard ratio (HR) (i.e. the coefficient, [β]) of each variable -β 1 X 1 + β 2 X 2 + … + β n X n ) was used as the risk score.A high score indicates poor prognosis.To determine the optimal cut-point of the score that designates if a person has a good or poor prognosis, the risk score was calculated for each person in the Discovery cohort.The range of scores from the median to the 70 th percentile were selected as candidate cut-points for evaluation.The 70 th percentile was chosen as the maximum score for evaluation because the proportion of men in the Discovery cohort who were designated as poor prognosis by the 3LS was 30%.The clinical outcomes of the risk groups produced by these candidate cut-points were evaluated using the C-statistic (Cox regression) and the log-likelihood (Weibull regression) (R package 'survival' v3.2-13) (i.e., men in the total cohort were split into those with scores above and below each cut-point, and survival outcomes were compared between the two groups).The cut-point that gave the highest C-statistic and log-likelihood was chosen (optimal points coincided).Model performance was assessed within the Validation cohort.
Receiver Operating Characteristic (ROC) area under the curve (AUC) was used to assess the model's ability to predict the 3LS (SPSS).

Study cohorts
The characteristics of the Discovery (105 men) and Validation (183 men) cohorts are summarised in Fig. 1, Table S2.2.1.Plasma from the Discovery cohort were obtained at baseline of taxane treatment, which was first line treatment for almost all participants.In contrast, plasma from the Validation cohort were obtained at baseline of ARSI for 70% of the men and 71% of them were first line.

Comparison between targeted and high-throughput assays
None of the lipids display collinearity with each other, except for Cer(d18:1/24:0) and Cer(d18:1/22:0) with Pearson's R = 0.76 (S5), thus both lipids were never together in the same model during model development.

Model development (Discovery cohort)
Twelve models consisting of different combinations of candidate lipids were derived (Table 2, S6).Ceramide ratios were included to account for noise reduction and LC-MS variance.Furthermore, ceramide ratios were reported as robust indicators of cardiovascular risk and are less dependent on clinical characteristics [17].Models 5 and 6 had the highest C-statistic (0.660) and were selected for further evaluation.The optimal cut-points of the risk scores that defines if a person has good or poor prognosis for models 5 and 6 were −1.1903 and −0.817 respectively (S7).
Analyses of progression-free survival (PFS) by cohort, therapy or treatment line showed that men who were PCPro-positive had shorter PSA-PFS or radiographic-PFS compared to men who were PCPro-negative (p < 0.05, S9).

Comparison of model to 3LS and clinicopathological factors
A high percentage of PCPro-positive patients have the 3LS (Discovery cohort: 68%; Validation cohort: 70%).The ROC AUC of the ability of PCPro to predict 3LS was 0.869 for the Discovery cohort and 0.751 for the Validation cohort (S10).
When PCPro was modelled with clinicopathological factors in multivariable Cox regression, PCPro was an independent predictor of OS (p < 0.05) in both the Discovery and Validation cohorts.Alkaline phosphatase was an additional independent predictor in the Discovery cohort and haemoglobin and albumin were additional independent predictors in the Validation cohort (S11).In bivariable Cox regression analysis, PCPro was an independent predictor of OS whereas diabetes status was not.PCPro was not associated with the presence of diabetes (S11.3).Other cardiovascular risk factors including age, weight, BMI and statin-use were not associated with OS (S11).

DISCUSSION
In summary, we have developed and validated PCPro, a novel lipid-based risk-score associated with poor prognostic mCRPC.PCPro effectively identified men with shorter OS in two independent cohorts.The prognostic ability of PCPro remained evident when cohorts were stratified by therapy or treatment line.PCPro can be performed in hospital laboratories and may be used to identify men with mCRPC for prospective studies of metabolic therapy.The National Cancer Institute's strategy for biomarker discovery demands that following assay development and analytical validation, a biomarker must progress through a clinical validation pathway through to commercialisation and regulatory approvala process that only a fraction achieve [18].Although biomarker research is academically interesting [19], for biomarkers to be clinically useful, they must enact clinical decisions that improve patient outcomes [20].Development of a biomarker using an NPAAC-concordant assay provides a pipeline towards clinically meaningful implementation.
An advantage of PCPro over high-throughput LC-MS is the use of unique stable isotope-labelled internal standards for each lipid.Internal standards account for recovery variance and matrix effects [13].Ideally, an internal standard should be used for each species analysed by LC-MS, however this is often prohibited by cost and availability when research assays include high lipid numbers [21,22].Internal standards were used for each analyte measured in PCPro, whereas in the high-throughput assay only a single internal standard was used for the entire ceramide class [23].A further advantage is that PCPro includes variables that are inversely correlated with survival (i.e. higher Cer(d18:1/24:0), total cholesterol and triglycerides are associated with longer OS, whereas higher Cer(d18:1/18:0) and Cer(d18:1/24:1) are associated with shorter OS).The inclusion of these variables may be related to their biology and adjusts for confounding effects, as ceramide, cholesterol and triglyceride synthesis and metabolism are coregulated [24][25][26].
Our previous work showed that men with mCRPC with alterations in their lipidomic profile have poor OS [4][5][6][7].The alteration in the lipidome extends well beyond the lipids included in PCPro.It would be impractical to include all these lipids, and rather, PCPro acts as an indicator for underlying metabolic changes.
The precision-oncology era is characterised by personalised therapeutics, and the development of specific biomarkers are crucial to delivering treatment to those who benefit most, sparing non-responders the cost and side-effects of treatment [27].Our next step is to integrate PCPro into clinical trials, to select patients to receive ceramide-targeting therapies in addition to standard care.Potential ceramide-targeting therapies are sphingosine kinase (SPHK) inhibitors, which inhibit the conversion of ceramide into S1P.SPHK inhibitors display anti-cancer effects in vitro and in mouse models, including prostate cancer models [8,28,29].Our previous work showed that SPHK inhibitors overcome enzalutamide resistance in prostate cancer cell lines and explants [7].Statins are able to decrease circulating levels of ceramides; however, retrospective studies of statins in addition to standard care for prostate cancer showed mixed results in improvement of clinical outcomes [30][31][32].Perhaps treating patients who are PCPro-positive with metabolic therapies will enrich for response.We hypothesise that PCPro has potential as a predictive biomarker, not just a prognostic tool, and this will be evaluated in prospective clinical trials.

CONCLUSIONS
We have developed PCPro, a novel lipid biomarker capable of prospectively identifying men with mCRPC with shorter OS.We hypothesise that this poor prognostic circulating lipid profile is metabolically actionable though drug and lifestyle interventions.
PCPro will allow us to identify men for prospective clinical trials of agents targeting lipid metabolism.

Fig. 1
Fig. 1 Patient cohorts and analysis strategy.mCRPC metastatic castration-resistant prostate cancer, LC-MS Liquid Chromatography-Mass Spectrometry, LASSO least absolute shrinkage and selection operator.

Fig. 2
Fig. 2 Kaplan-Meier survival analysis of overall survival by PCPro in the Discovery and Validation cohorts.A Survival in the Discovery cohort, B Survival in the Validation cohort, C Survival in those treated with ARSI in the Validation cohort, D Survival in those treated with taxane chemotherapy in the Validation cohort, E Survival in those treated with first line treatment in the Validation cohort, and F Survival in those treated with second line treatment in the Validation cohort.ARSI androgen receptor signalling inhibitor, HR hazard ratio, mo months, OS overall survival.

Table 1 .
Univariable Cox regression of log2 transformed lipid species concentrations as measured on the targeted and high-throughput assays, in the Discovery and Validation cohorts.
Blank cells indicate lipids which were not measured on the high-throughput lipidomic assay.Cer ceramide, PC phosphatidylcholine, HD high-density lipoprotein, CI confidence interval.

Table 2 .
Details of each of the prognostic models investigated.