Introduction
Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest tumor diseases worldwide [
1], with poor prognosis (5 years survival for all stages is only 10%) [
2] and low probability for curative surgery (< 20%) [
3]. The two major reasons for treatment failure of PDAC are late diagnosis and a highly complex tumor microenvironment that reduces therapeutic effects. The PDAC microenvironment consists of diverse populations of embedded cancer and other associated cells (i.e., fibroblastic, stellate, endothelial, neuronal, immune), remaining ducts and substantial extracellular matrix.
PDACs are sparsely vascularized and, due to abundant matrix formation, often exhibit elevated intra-tumoral pressure and thus vascular collapse [
4,
5]. Poor perfusion of PDAC has a significant clinical impact, as it contributes to reduced drug delivery and the development of drug resistance. Therefore, several agents targeting the stromal compartment have been developed and applied in clinical trials [
6,
7], mostly in combination with gemcitabine (GEM), which has remained the standard of care treatment for decades [
8]. Unfortunately, none of the tested combinations improved clinical outcome, including targeting of hyaluronic acid (HA) to reduce intra-tumoral pressure and inhibition of the Hedgehog pathway to block interactions between tumor and stromal cells [
9‐
11]. Hence, despite intensive research and a great number of clinical trials, management of patients with PDAC still lacks personalized protocols and instead relies on chemotherapy [
7]. The two first-line chemotherapeutic options for patients with advanced PDAC are a combination of four cytotoxic agents (i.e. fluorouracil, leucovorin, irinotecan and oxaliplatin; mFOLFIRINOX) [
12] or a combination of nab-paclitaxel (Abraxane, A) and GEM [
13].
Recent studies have shown that GEM therapy of PDAC affects both cancer cells and the tumor microenvironment, including vascularization. In particular, GEM has been reported to accumulate in stroma rich tumors [
14]. Preclinical investigation in clinically relevant endogenous mouse models of PDAC revealed that active metabolites of GEM are captured by cancer-associated fibroblasts (CAFs), pancreatic stellate cells (PSC), and M2-polarized macrophages, which possibly reduces its cytotoxic effect on the cancer cells [
15,
16]. Inconclusive results have been reported for GEM regarding endothelial cells and tumor vascularization. For example, human endothelial cells were highly sensitive to low concentrations of GEM in vitro [
17]. Reduced perfusion was noted after GEM treatment in a human PDAC xenografts as well as in endogenous murine tumors harboring
KRAS and
TP53 mutations [
18]. However, other reports have shown increased perfusion in response to GEM treatment in murine and human tumors in vivo [
19‐
21].
Increasing appreciation of the molecular and histopathological heterogeneity in PDAC has led to careful re-evaluation of imaging-derived biomarkers for non-invasive differentiation of PDAC subgroups and detection of individual therapy response. Imaging is routinely performed after systemic injection of contrast agents (CA) to visualize its relative regional accumulation as a surrogate of local blood supply and tissue composition. This clinical approach, to some extent, disregards the complexity of the underlying systemic (i.e. CA injection rate, heart rate, blood pressure, kidney function, etc.) and regional (i.e. vascularity, perfusion, permeability, tissue composition, etc.) determinants of CA biodistribution, which are greatly simplified therein. Computed tomography (CT) is the standard method for diagnosis and response monitoring of PDAC patients in clinical routine. CT is performed statically, after intravenous injection of iodine-based CA, which accumulates locally and causes differences in X-ray attenuation, quantified in Hounsfield Units (HU). Magnetic resonance imaging (MRI) is typically performed pre-clinically. In dynamic contrast enhanced (DCE)-MRI, a series of T
1-weighted images, or T
1 maps, is acquired before, during, and after intravenous injection of Gd
3+-based CAs with time resolution of 3–5 s. Although CT and DCE-MRI are different imaging techniques, both have been used to describe regional tissue morphology and physiology, including vascularity and perfusion and changes thereof [
22‐
24]. In PDAC, HU ratios (HUr) of tumor tissue, normalized to the aorta or tumor adjacent pancreas, have shown a positive correlation with desmoplastic stroma and a negative correlation with tumor cellularity and patient survival [
14,
25‐
27]. For DCE-MRI, commonly used parameters are the volume transfer constant (K
trans) and the extravascular extracellular volume fraction (v
e) determined using Tofts pharmacokinetic modeling [
28]. Differences in tumor perfusion and permeability, measured by K
trans, were also correlated with biological variations in tissue composition, response to therapy and clinical outcome in PDAC [
29‐
32]. However, such modeling introduces a high level of complexity and uncertainty. In contrast, area under the time–to–signal curve, e.g., 60 or 90 s after the arrival of the contrast bolus (iAUC60, iAUC90), has been used as a less accurate albeit feasible alternative. AUCs derived from signal intensities in T
1-weighted images or CA concentrations calculated from T
1 (=1/R
1) maps are parameters that, similarly to CT HU, reflect a combination of blood flow, vessel size and density, permeability, and functionality of the micro-vascular network [
22,
33]. For example, AUC60 has been correlated directly to perfusion measurements and K
trans estimations in murine and human studies [
23,
34] and suggested as a surrogate perfusion biomarker. In a study on differently vascularized tumors, iAUC90 was reported to be even more sensitive to vascular changes under anti-angiogenic treatment than K
trans [
22]. In addition, AUC is widely established for perfusion measurement in preclinical trials of small animals, where measuring the arterial input function (AIF) is particularly difficult and error prone [
35]. To account for inter-subject variability, both imaging techniques, with their respective parameters, rely on normalization (i.e., the aorta in patient CT and spinal muscle in mouse MRI) [
27,
36].
In this study, our aim was to establish imaging biomarkers based on CA accumulation and investigate their potential to detect differences in tissue composition and changes caused by chemotherapy in PDAC. For patient studies, we analyzed CA accumulation in routine pre-operative CT (HUr) and DCE-MRI (signal intensity of tumor to aorta ratios (SIt/a)). To investigate the effects of chemotherapy on CA accumulation, we used clinically relevant murine models and the standard preclinical imaging technique DCE-MRI (AUC60r). Using murine histopathological samples, we investigated GEM treatment effects on the cellular level. Finally, we analyzed GEM treatment effects on CA accumulation in human PDAC.
Discussion
Tumor morphology, physiology, and clinical outcome reveal substantial heterogeneity in PDAC with high tumor cellularity as an indicator of poor prognosis [
14,
25,
37]. Consequently, to facilitate personalized treatment, identification of non-invasive biomarkers for differentiation of tumor cellularity in PDAC would be clinically highly valuable. Here, we show that CA accumulation varies significantly in PDAC depending on tumor morphology, enabling non-invasive binary classification of tumor cellularity in vivo. In addition, our study reveals unappreciated effects of standard of care GEM treatment on CA accumulation that correlate with tumor vascularity and small molecule delivery. This finding has important implications for GEM-based combination treatment.
High tumor cellularity, non-invasively detected by diffusion weighted (DW)-MRI and CA accumulation in CT, has previously been identified as a predictor of poor outcome in PDAC [
25,
26,
37,
41]. We further expand on this classification and suggest the biomarkers HUr, SI
t/a, SI
t/m, and AUC60r for a non-invasive stratification of tumor cellularity subgroups. Here, we show that both CT- and MRI-derived perfusion-related values can be used for patient stratification. In line with our observations, the particularly aggressive transcriptionally defined squamous PDAC subtype has been shown to have the highest difference between tumor and normal pancreas in CA accumulation and was associated with a higher number of secondary poor-prognosis mutations such as
TP53 or
SMAD4 and with consequently worse survival [
26]. Moreover, the same study reported that stage IV patients revealed lower base line tumor CA accumulation compared to early stage operable patients [
26]. In our study, the chemotherapy treated PDAC patient cohort also showed a lower base-line HUr value in comparison to the resected cohort, and murine tumors revealed decrease in AUC60r over time. These observations may be explained by the fact that as tumors progress, cellularity increases, which in turns leads to poorer vascularization and collapsed vessels due to high intra-tumoral pressure. Cumulative, these processes lead to reduced perfusion and CA accumulation in later stage PDAC, which is then reflected in lower HUr and AUC60r values. Therefore, routine CT- or MR-based non-invasive perioperative prediction of tumor cellularity by means of CA accumulation presents a widely applicable, biologically meaningful, and clinically relevant stratification strategy with great potential in routine patient care.
A direct correlation of in vivo imaging and ex vivo histopathology data is highly desirable yet challenging [
40]. Especially in hPDAC, where more than 80% of patients are initially not eligible to surgery, imaging-correlated histopathology analyses are rare, and results are often obscured by the effects of neo-adjuvant chemotherapy. Here, we use GEMM of PDAC, which exhibit complex extracellular matrix on the background of an intact immune system, and recapitulate well the human condition. GEMM of PDAC were used in many meaningful preclinical trials [
4,
5,
15,
16,
37,
38] with in-/ex-vivo sample co-registration [
5,
37,
38,
40] and subsequent molecular tumor characterization [
4,
5,
15,
16,
38,
42]. For example, manual co-registration of diverse imaging methods such as ultrasound, transmission electron microscopy, or DCE-MRI with corresponding histopathology allowed profound investigation of micro- and macro-vascular tumor architecture, treatment-induced vascular and stromal changes, and drug delivery in
KPC tumors [
5,
18,
42]. Moreover, endogenous mPDAC is, similar to hPDAC, highly resistant to GEM—only 5–10% of KPC tumors show response—allowing preclinical drug trials of high predictive value [
18,
38,
42]. To fully leverage this potential, we used GEMM of different mutational backgrounds to optimally mirror the inter-tumoral heterogeneity that is noted in hPDAC [
37] and found striking similarities in the behavior of the tumor perfusion-related biomarker AUC60r. Analogous to hPDAC, we observed substantial heterogeneity in CA enhancement in the analyzed mPDAC population: low cellularity tumors revealed high CA accumulation, whereas high cellularity tumors showed low CA accumulation. In addition, we provide a direct correlation of imaging-derived biomarker AUC60r with functionality of vascular compartment measured by the amount of open vessels calculated in the imaging-corresponding axial slice of the tumor tissue.
Reduced CA accumulation in PDAC, compared to the normal surrounding pancreas, has long been noted and presents one of the diagnostic clues in clinical routine patient management. Consequently, it has led to the introduction of therapies intended to normalize intra-tumoral pressure, vascularization, and drug delivery [
43]. Indeed, therapy trials with stroma targeting agents have shown improved perfusion, monitored non-invasively by contrast enhanced ultrasound [
4,
42] and DCE-MRI [
18] in mPDAC as well as in hPDAC [
30]. Subsequent reduced vascular collapse and increased drug delivery lead to prolonged survival of the treated animals [
4,
5,
18,
42]. These studies confirm the high relevance of the vascular compartment and the potential role of perfusion-sensitive imaging derived biomarkers in treatment monitoring. Unfortunately, despite positive pre-clinical and early clinical trials, none of the stroma modifying agents has reached routine clinical application thus far [
9,
11,
44].
Despite intermediate efficacy and a poor long-term outcome, GEM remains one of the standard of care treatment choices in hPDAC. Therefore, understanding its impact on tumor vasculature is an important interrogation. We observed a strong decrease in the DCE-MRI-derived perfusion-sensitive marker AUC60r following GEM-based treatment in mPDAC. This effect was most prominent in the initially well perfused mPDAC
low tumors. Human endothelial cells have been reported to be considerably more sensitive to GEM than are pancreatic tumor cells, both in vitro and in orthotopic xenotransplants [
17]. In line with this observation, we noted a reduction in vascular proliferation in GEM-treated murine tumors. Other phenomena, such as apoptosis, were not analyzed, since GEM has been reported to specifically reduce proliferation in human and murine pancreatic tumor cells in vivo [
42] and in human endothelial cells in vitro [
17]. We further provide evidence that the reduction in vascular proliferation is of functional relevance, showing a reduced delivery of the systemically administered DNA-damaging substance cisplatin.
Our study also provides indirect evidence of the deteriorating effects of GEM on tumor vasculature in hPDAC by showing a significant decrease in CA accumulation in response to GEM-based, but not FOLFIRINOX-based, treatment. Similar to experiments in mPDAC, no significant differences between the treatments were observed in tumors with an initially poor CA accumulation. Nevertheless, in a subset of subjects, there was an increase in CA accumulation during GEM treatment, possibly related to low dosage (i.e., metronomic) delivery, which has previously been reported to normalize tumor vasculature by increased levels of the angiogenesis inhibitor thrombospondin-1 [
45]. In addition, a direct effect of GEM on tumor cells reduces intra-tumoral pressure [
19,
20] and likely increases tumor perfusion, which has previously been noted in GEM/Abraxane responders [
29]. However, the latter study was also conducted in patients with advanced PDAC, and no histological correlation nor sub-grouping was performed.
There are several limitations to the interpretation of results of this work. The analyzed hPDAC cohort with available imaging-correlated tissue samples was small in size and our observations from it require further prospective testing. The differences in the baseline CA accumulation between human cohorts may reflect more advanced disease within the cohort of study h2, resulting in lower initial perfusion. Further biopsy- and imaging-guided neoadjuvant trials may provide the necessary information on that matter and help to adjust the proposed cutoff for patient stratification. CA accumulation is dependent on but not equivalent to perfusion, and therefore prospective trials including imaging-tissue correlation experiments are necessary to confirm or discard our hypothesis of GEM-induced perfusion effects in hPDAC. In addition, due to poor soft tissue contrast and limited time resolution of standard of care contrast CT in our mouse models, this work used different CA and acquisition methods in the human and murine imaging studies, which limits absolute comparability and direct translational impact. Nevertheless, the similar trends in patient-measured CT-derived HUr and MRI-derived SI
t/a values and murine perfusion related biomarkers support our hypothesis. Furthermore, a clinical study in patients with liver metastases of colorectal cancer confirmed that MRI biomarkers of vascular function, including iAUC60, correlate with structural features of tumor vessels detected by CT [
22]. In addition, an improved technical setup for consecutive DCE-MRI and CT imaging of subcutaneous murine tumors was able to directly link functional MRI parameters to a structural CT-derived vascular volume parameter in vivo, which was confirmed by intra-vital microscopy and 3D ex vivo validation [
22,
46].
In conclusion, we tested and applied two clinical imaging-derived perfusion-related biomarkers (AUC60r and HUr) for the non-invasive differentiation of cellularity related subgroups in murine and human PDAC. Applied to therapy response monitoring of PDAC, we observed a decrease in CA accumulation in response to GEM treatment. In addition, we provide evidence that GEM interferes with endothelial cell proliferation, further aggravating pre-existent hypo-perfusion of PDAC in vivo. Our observations thus suggest tumor CA accumulation as a biomarker for tumor stratification based on cellularity and for the longitudinal monitoring. Blockage of tumor neo-vascularization may partially explain the failure of gemcitabine-based combinatorial regiments previously observed in human PDAC trials.
Declarations
Competing interests
All authors declare they have no financial interests, except:
1. J.T.S. reports research funding from BMS, Celgene and Roche/Genentech, consulting and personal fees from AstraZeneca, Bayer, BMS, Falk Foundation, Immunocore, MSD Sharp Dohme, Novartis, Roche, Servier and holds ownership in Pharma15 (all outside the submitted work).
2. Wilko Weichert has attended Advisory Boards and served as speaker for Roche, MSD, BMS, AstraZeneca, Pfizer, Merck, Lilly, Boehringer, Novartis, Takeda, Bayer, Amgen, Astellas, Eisai, Illumina, Siemens, Agilent, ADC, GSK, and Molecular Health. WW receives research funding from Roche, MSD, BMS, and AstraZeneca (all outside the submitted work).
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.