Abstract

Background

Pancreatic ductal adenocarcinoma (PDAC) is projected to be the second leading cause of cancer-related deaths within the next decade and continues to have a dismal 5-year survival rate of 9%.¹ Almost 80% of patients are diagnosed at locally advanced or metastatic stages and current treatment options are mostly limited to cytotoxic agents. Even with seemingly curative resection for patients presenting with localised disease many patients suffer early recurrence, suggesting an underlying propensity for early dissemination and resistance to available treatments.²

Elucidation of tumour cell phenotypes as they evolve during therapy can provide predictive insights into emergence of resistance to standard of care therapies in PDAC. A recently applied strategy to profile such dynamic changes in real time has involved the study of disseminated cancer cells, better known as circulating tumour cells (CTCs).³ As obtaining serial biopsies from primary or metastatic lesions in PDAC is clinically challenging, liquid biopsies have gained traction in recent years for understanding tumour kinetics under therapy. As opposed to circulating tumour DNA (ctDNA) and exosomes, CTCs provide the added benefit of representing the phenotypic characteristics of their tumour of origin.⁴

Classically, CTC isolation has involved the use of epithelial biomarkers such as cytokeratin and epithelial cell adhesion molecule (EpCAM) through immunogenic capture techniques such as CellSearch®, and their enumeration has shown to be a useful clinical prognosticator in breast, prostate, lung and colorectal cancers.^5–8 However, a predictive model utilising CTCs as a biomarker has yet to demonstrate strong evidence for clinical validity in PDAC.⁹ Specifically, the role of CTC sub-populations, including those with non-epithelial characteristics, in clinical stratification remains to be elucidated through a comprehensive unbiased approach. Additionally, other CTC phenotypes such as circulating tumour microemboli (CTMs) may represent unique tumour capabilities such as increased metastatic propensity.^10,11 As such, uncovering the dynamic nature of evolving CTC phenotypes correlating to tumour progression and therapeutic response is of clinical importance but poorly characterised, mostly due to a limitation of current enumeration strategies that only utilise epithelial and hematopoietic markers, restricting the definition of a CTC and potentially biasing its quantification. Ultimately, whether certain sub-populations of CTCs may be more clinically relevant than others remains undetermined.^12,13

In order to perform a comprehensive phenotypic characterisation of PDAC CTCs, we used an epitope-independent isolation approach utilising a microfluidics-based dielectrophoresis enrichment strategy, followed by a multiplexed immunostaining assay.^14–16 This approach provides the opportunity to profile the morphologic and phenotypic characteristics of these cells while bypassing the enrichment of epithelial cells. We evaluated its capability for the isolation of CTCs with different phenotypes and their clinical significance in a longitudinal cohort of PDAC patients.

Methods

Study population

Between April 2015 and January 2018, 74 patients were consented at MD Anderson Cancer Center (MDACC) under protocol number PA14-0867 and PA11-0670. Cohort details are shown in Table 1. The study was performed in accordance with standard ethical guidelines approved by the institutional review board (IRB) and in accordance with the Declaration of Helsinki. All patients had clinically and histologically confirmed localised or metastatic PDAC.

Table 1

Patient demographics and clinicopathological characteristics.

Cohorts characteristics	Patients
Total patient number	74
Peripheral blood samples	272
Mean age	62.1 years (range 31–82 years)
Gender
Men	42 (56.8%)
Women	32 (43.2%)
Median follow-up time per patient	15.4 months
Localised disease	41 (66.4%)
Metastatic	33 (45.6%)
Localised group: neoadjuvant treatment	36/41 (87.8%)
AJCC staging
I	9 (12.2%)
II	22 (29.7%)
III	10 (13.5%)
IV	33 (44.6%)
Metastasis side
Liver	23 (69.7%)
Lung	7 (21.2%)
Others (peritoneum, distant lymph nodes)	3 (9.1)
Tumour location
Head	56 (75.7%)
Tail/Body	18 (24.3%)
Treatment status
Baseline treatment-naïve	65
Baseline treated (palliative and neoadj.)	9

Open in a separate window

Results

Dielectrophoretic based isolation of heterogeneous CTC sub-populations

DEP-FFF enables epitope-independent isolation of viable CTCs based on physical cell properties including the diameter and surface rigidity.^14–16 This allows for the capture of distinct sub-populations of CTCs within an hour of blood processing time. In order to demonstrate our capture efficiency, we performed spike-in experiments utilising a PDAC cell line, PANC-1, into healthy human peripheral blood mononuclear cells (PBMCs). Spike-ins of 10, 100, 200, 500, 1000 and 1,200 (n=3 replicates for each) cells revealed recovery rates of up to 91% (average recovery rate 89.1%) with high linearity (R²=0.9873, Fig. 1a). Our ability to capture and characterise multiple unbiased CTC sub-populations was also tested by adding different PDAC cell lines with known phenotypes of interest, including epithelial (HPAFII), mesenchymal (MIAPaCa-2) and stem cell-like (PANC-1 spheroids) subtypes. These cell lines were successfully “re-identified” following DEP isolation, where cells were stained with anti-CD45, EpCAM, pan-cytokeratin (PanCK), vimentin and CD133 antibodies, as well as DAPI (Fig. 1b). Based on characterisation of isolated PDAC cells using DEP and imaging flow cytometry with distinct epithelial, mesenchymal and stem-cell markers, we defined four distinct sub-populations in the peripheral blood of patients with PDAC using expression of markers listed above—epithelial (E-CTC), mesenchymal (M-CTC), bi-phenotypic or partial epithelial-mesenchymal transition (pEMT-CTC) and stem cell-like (SC-CTC)—based on markers of interest (Table 2). CTCs were further curated based on characteristic size (>14μm) and distinct nuclear structure covering >80% of the cell surface (Fig. 1c). In addition to single CTCs, this technique was able to isolate CTCs associated with lymphocytes as well as clusters of CTCs (Fig. 1d, e).

Open in a separate window

Fig. 1

Capture efficiency of epitope-independent microfluidics device.

a Scatter plot showing known cell numbers added to PBMC and number of cells recovered from DEP-FFF, b Brightfield and immunofluorescence images of marker proteins of different cell lines recovered using DEP-FFF and analysed by imagestream flow cytometry. BF Brightfield, Pan-CK pan-cytokeratin, Vim vimentin, EpCAM epithelial cell adhesion molecule, DAPI 4′,6-Diamidin-2-phenylindol and CD45 Protein tyrosine phosphatase, receptor type C. Brightfield and immunofluorescence images of CTC populations (c) CTCs lymphocytes (d) and CTC–CTC clusters (e) identified in PDAC patients by ImageStream.

Table 2

Defined subtypes to classify patient-derived CTCs.

Phenotypes identified	Markers
E-CTC	CD45−, EpCAM+ and/or Pan-CK+, DAPI+
pEMT-CTC	CD45−, EpCAM+ and/or Pan-CK+, Vim+, DAPI+
M-CTC	CD45−, EpCAM−, Vim+, DAPI+
SC-CTC	CD45−, CD133+, DAPI+

Open in a separate window

CTC counts in overall PDAC cohort

Overall CTC detection rate across all samples, and including single CTCs and clusters, was 76.8% (209/272 samples, range 0–692 cells/7.5ml of blood). Mean number of CTCs per 7.5ml of blood was 29±70.6 (range 0–692). Across all samples, mean numbers of CTCs were as follows: E-CTC: 3.5±1.1 cells/7.5ml (12.2%), M-CTC: 9.3±1.5 cells/7.5ml (31.9%), pEMT-CTC: 13.8±2.1 cells/7.5ml (47.7%) and SC-CTC: 2.4±0.5 cells/7.5ml (8.2%) (Fig. 2a).

Open in a separate window

Fig. 2

Number of CTCs identified in PDAC patients.

a Scatter plot showing total counts in five phenotypically distinct populations across all collected samples (n=272) or at baseline treatment naïve status (n=65). b Last column shows all counts. Prognostic value of CTCs at baseline treatment-naïve status (n=65). c Progression-free survival comparing patients with more vs. less than 50% Prop-pEMT (p=0.018). d Overall survival comparing patients with more vs. less than 50% Prop-pEMT (p=0.63).

In treatment naïve patients at baseline (n=65), mean total CTC number was 41.2±62.7 and per sub-type: E-CTC-1.5±0.6 cells/7.5ml (3.7%), M-CTC 13.8±2.9 cells/7.5ml (33.5%), pEMT-CTC 21.0±4.05 cells/7.5ml (51.1%) and SC-CTC 4.8±2.1 cells/7.5ml (11.7%) (Fig. 2b). CTC clusters were identified in 32/65 (49.2%, range 0–18 clusters) treatment-naive patients with a mean 2.6±12.6 cells/7.5ml in all draws and 2.8±4.6 cells/7.5ml in BTN patients only (Fig. 2a,​2a,bb).

Notably, there was no difference in total CTC counts at baseline (n=65) based on tumour location, stage and size (all p>0.05, Supplementary Table 1, Supplementary Fig. 1A-D). However, the proportion of pEMT-CTC relative to total CTC count (“Prop-pEMT”) was significantly higher in M1 vs. M0 patients (mean 52.4% vs 28.4%, p=0.004, Supplementary Fig. 1E, Supplementary Table 1). Additionally, “Prop-pEMT” increased with stage (stage I: 24.4±26.9%, stage II: 28.5±23.5%, stage III: 32.4±35.1%, stage IV: 52.0±33.5%, p=0.0024, Supplementary Fig. 1F, Supplementary Table 1) and with distal tumour location (head: 33.5±29.1%, tail/body: 54.2±35.2%, p=0.04) Supplementary Fig. 1G Supplementary Table 1).

Liquid biopsies of all patients (n=139) while on therapy demonstrated that those with progressive disease based on RECIST criteria, compared to those with blood draws on therapy but without progression had significantly more total CTCs (p=0.0162), E-CTCs (p=0.0489), M-CTCs (p=0.0087) and higher Prop-pEMT (p=0.0410) (Supplementary Fig. 2 and Supplementary Table 2).

Baseline treatment-naïve (BTN, n=65) patients with a higher proportion of pEMT-CTC50% showed a significantly worse progression-free survival (PFS) compared to patients with a lower pEMT-CTC proportion on univariate analysis (p=0.020, Fig. 2c, Supplementary Table 3B, 4). On multivariate analysis, proportion of pEMT-CTC remains significantly associated to poorer PFS (HR 1.98, 95% CI 1.05–3.73, p=0.036) after adjusting for other factors (Supplementary Table 3A-C). Proportion of pEMT-CTC showed no impact on overall survival in BTN patients (Fig. 2d). In addition, disease stage had a significant impact on PFS (p-value=0.009), but was not a significant factor (p-value=0.190) in the multivariable model after adjusting for pEMT-CTC, log transformed SC, and age (Supplementary Table 3C). No other subgroup showed a significant prognostic value at BTN status (all, p>0.05; Supplementary Table 3A-B and Supplementary Table 5) including total pEMT counts at different cut-offs (Supplementary Table 6).

CTC counts in localised disease

Among patients with localised disease, mean CTC counts (including single CTCs and clusters) dropped significantly following resection across total, pEMT, EMT subgroups and clusters (Fig. 3a–f), but did not drop to zero in any patient. Among resected patients, seven (7/22, 31.8%) had an early recurrence within 12 months after surgical resection. Patients with “Prop-pEMT”>50% in preoperative blood draws (n=22) were at higher risk for an early recurrence (4/7 (57.1%) versus 1/15 (6.7%), p=0.02, Fig. 3g). Similarly, patients with preoperative “Prop-pEMT” >50% showed a worse overall survival (OS) after resection (677 days versus not reached p=0.01, Fig. 3h). None of the other CTC sub-populations showed a significant impact on early recurrence rate or OS (all p>0.05, Supplementary Table 7).

Open in a separate window

Fig. 3

Comparison of total counts in sub-populations.

a total CTCs, b pEMT-CTC, c M-CTC, d E-CTC, e SC-CTC, f CTC clusters before and after surgery (n=22). Prognostic value of CTCs at draw closest to surgery. g Bar graph plotting Prop-pEMT50% vs. recurrence status (Fisher exact test, p=0.021). h Kapla–Meier plot (overall survival from surgery) comparing patients with vs. without 50% Prop-pEMT (p=0.01). i Overall survival comparing patients with at least one pEMT-CTC vs none (p<0.001).

Molecular characterisation of different CTC sub-populations

To further characterise isolated CTCs, we performed droplet digital PCR (ddPCR) for mutant KRAS as previously described,¹⁸ in captured cells from a patient with metastatic PDAC following progressive disease in the liver. Cells were isolated by DEP-FFF from a single draw at the time of progression. They were then stained as described and FACS sorted in either pools of 10 cells prior to amplification (CTC-2, Vim⁺ and CTC-3, Vim⁺), or amplified as single cells (CTC-1, E⁺/CK⁺/Vim^+/−) depending on the number of cells captured. Whole-genome amplification of sorted cells was performed with the REPLI-g Single Cell Kit (Qiagen) per the manufacturer’s protocol. In addition, three pools of 10 CD45+ cells were also isolated and amplified and used as wild-type control and paired tumour tissue from the metastatic side was used as positive control (Table 3). Mutant KRAS was detected in all three CTCs (mutant allele fraction (MAF) 1.27, 84.3 and 78.4, respectively) and in the tumour sample (MAF 50.22) while all three CD45+ cells were KRAS negative. Additionally, to better characterise the M phenotype and exclude other cell populations that also expressed Vimentin, CD31⁺/Vim⁺ and FAP⁺/Vim⁺ cells were analysed in three clinical samples (Supplementary Fig. 3). In all three cases, we did not detect a single FAP⁺/Vim⁺ or CD31⁺Vim⁺ cells instead, we detected ~1% of CD45− cells expressing CD31 or FAP in combination with Vim and CK.

Table 3

Molecular characterisation of CTC by ddPCR.

Sample	Number of cells	Phenotype	MAF mutant KRAS by ddPCR
CTC-1	10	CD45−/Vim+	1.27
CTC-2	10	CD45−/Vim+	84.3
CTC-3	1	CD45−/E+/CK+/Vim+/−	78.4
Positive control	Pellet	Tissue	50.22
Negative control 1	10	CD45+	WT
Negative control 2	10	CD45+	WT
Negative control 3	10	CD45+	WT

Open in a separate window

Discussion

In visceral tumours, such as PDAC, repeated tumour sampling for therapy monitoring and molecular profiling is rarely performed in the clinical setting due to their invasive nature and cost. Thus, strategies involving liquid biopsies have gained traction for their minimally invasive approach in cancer diagnosis, prognostication, therapy stratification and measuring response to therapy. In this study, we interrogated if longitudinal characterisation of CTC sub-populations may better resolve the prognostic and predictive utility of this liquid biopsy compartment. We also describe CTC phenotypes that may be most relevant to metastasis and chemoresistance, which have been previously correlated with the presence of cancer stem cells or cells with epithelial-mesenchymal transition (EMT) profiles.^19,20

Biased immunogenic capture of CTCs based on epithelial markers potentially excludes cells that have biological implications in the process of metastasis and chemoresistance.^21,22 Even among epithelial markers typically used for CTC capture, such as EpCAM or cytokeratins, there is no consensus regarding specific marker(s) that may more effectively identify clinically relevant CTCs. Utilising DEP-FFF, which allows for isolation of particles based on their physical properties, we isolated CTCs through an antigen independent approach. In doing so, we effectively isolated phenotypically distinct CTC populations that may be grouped into four major sub-populations based on our multiplexed immunostaining assay: E-CTC, M-CTC, pEMT-CTC and SC-CTC

Similar to previous work,^23–26 the most common sub-population of PDAC CTCs expressed a combination of epithelial and mesenchymal markers (pEMT-CTC), representing neoplastic cells arrested in a “bi-phenotypic” state as they are shed from their tumour of origin.^21,22,27 Increasing evidence suggests that EMT is a highly dynamic process, inclusive of intermediate cell states (“pEMT”), such that only a minority of cancer cells transition to a pure mesenchymal state.^22,27,28 Notably, neoplastic cells with a partial-EMT phenotype show greater invasiveness and therapeutic resistance,²² which coincides with our findings of this CTC sub-population correlating with both stage at presentation, as well as with progression on therapy. Specifically, we demonstrate that “prop-pEMT” (the proportion of CTCs with pEMT phenotype overall CTCs) are higher in patients with metastases at diagnosis. Additionally, whereas there are no effective methods for stratification of early recurrence risk in resectable PDAC, we demonstrate that a high “prop-pEMT” at the time of surgery is correlated with early recurrence. Importantly, CTC numbers rarely dropped to zero even after complete surgical resection, which is in-line with recent studies.^5,25 This raises the concern that micro-metastatic lesions may have already been established even in patients with apparently curative resection and begun releasing CTCs themselves. Additionally, once overt metastasis is detectable, we show that a single EpCAM/vimentin co-expressing cell, representing a specific pEMT status, in treatment-naive patients is associated with a significantly worse overall survival. As previously described, we also did not detect differences in CTC total counts across PDAC stages, which may depict the aggressive nature of PDAC and its ability to metastasise from its onset.^2,26

While most clinical correlates of adverse outcome were associated with pEMT-CTCs, we identified at least one parameter associated with the population of Vim⁺/EpCAM⁻ CTCs (M-CTC). This M-CTC sub-population was significantly higher at radiographic progression compared to non-progression in metastatic patients under treatment. CTCs predominantly expressing vimentin have been associated with a worse prognosis or aggressive disease in other cancer types^6,7 while vimentin only expressing CTCs are seen in rarer numbers in the context of PDAC.^5,25,26 Taken together with pEMT-CTCs as described above, the presence of these two populations may portend a dynamic picture of CTCs that exist in the spectrum of EMT and the ability for tumours to adapt to chemotherapeutic pressures.

Among weaknesses of our study, we understand the limits of our characterisation approach in that only a certain number of markers can be included within the Imagestream analysis. We attempted to maximise our approach by including as many conjugated markers as we could reliably and consistently use. As such, we are insightful of the fact that the markers selected may not necessarily be sensitive or specific enough to characterise the entire population of these CTC subtypes. For example, we could only include CD133 and Vimentin as markers to characterise ‘stem cell-like” and mesenchymal populations, respectively. We believe that further studies should be able to take a more comprehensive approach to populations, like those of stem cells by including markers such as CD44, CD24, ALDH and RORγ.²⁹ Additionally, we have not fully characterised the size limit of our device in terms of its ability to isolate CTC clusters and thus feel it is important to interpret these results with caution.

In conclusion, our study identifies at least four distinct sub-populations of CTCs within a single blood draw of PDAC patients. Although such CTC subtypes have been previously described individually,^13,19 our study provides a comprehensive characterisation of such populations at baseline treatment naïve status and during longitudinal follow-up. Of all CTC sub-populations, we have isolated in PDAC through an antigen independent retrieval method, the pEMT phenotype showed the greatest predictive and prognostic value in both resectable and metastatic disease. The biological underpinnings of this pEMT sub-population warrants further study to understand the cellular mechanisms driving their aggressive potential and therapeutic resistance, particularly in contrast to M-CTCs. Our detailed phenotypic characterisation of CTCs is important as specific sub-populations, rather than total CTC number, in circulation may have greater clinical relevance and provide insights into treatment response.²⁰

Supplementary information

Acknowledgements

Author contributions

A.M., P.A.G., H.A. Study Design: A.M., P.A.G., H.A., A.S., V.B., D.U.K. Provision of clinical samples and information: B.S., G.R.V., M.H.K. Samples processing: A.S., V.B., P.G., D.U.K. Microfluidics device design: P.G., F.A.S.L. Sample isolation: A.S., V.B., P.A.G., D.U.K., N.K. Samples staining and cytometry: A.S., V.B., P.A.G., D.U.K. Data analysis: J.H., F.A.S.L., J.J.L., W.Q., Y.S. P.A.G. Manuscript writing: A.S., V.B. All authors read and approved the final manuscript.

Ethics approval and consent to participate

The study was performed in accordance with standard ethical guidelines approved by the institutional review board (IRB), protocol numbers PA11-0670 and PA15-0014 at MD Anderson Cancer Center, and in accordance with the Declaration of Helsinki. Patients provided their informed consent to participate in this study.

Data availability

All data generated and analysed during the current study are available from the corresponding author on reasonable request.

Competing interests

A.M. receives royalties for a pancreatic cancer biomarker test from Cosmos Wisdom Biotechnology, and this financial relationship is managed and monitored by the UTMDACC Conflict of Interest Committee. A.M. is also listed as an inventor on a patent that has been licensed by Johns Hopkins University to Thrive Earlier Detection. The remaining authors declare no competing interests.

Funding information

This research was supported in part by the Cancer Prevention and Research Institute of Texas (CPRIT) (No. RP160517), NCI P50 CA221707, U01 CA196403 and U01 CA200468 to A.M. N.K. was supported by the CPRIT Research Training Program (No. RP170067). V.B. was supported by the CPRIT Research Training Program (Nos. RP140106 and RP170067) and NCI (Nos. T32CA217789-03 and U54CA096297). J.J.L. was supported by the National Institutes of Health (NIH) (No. T32CA009599). A.S. was supported by the German Research Foundation (SE-2616/2-1). D.U.K. was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. 2018R1C1B5086234).

Footnotes

References