Elevated circulating PlGF/sFlt1 ratio is associated with neural invasion and predicts disease prognosis in PDAC

Given its release into the circulation, we retrospectively assessed circulating PlGF in patients with PDAC (please refer to Suppl. Table 1 for cohort description). Since soluble receptor Flt1 (sFlt1; sVEGFR1) binds PlGF [25], we also determined sFlt1 and calculated the ratio PlGF/sFlt1 (from here on referred to as PlGF/sFlt1circ) to estimate the fraction of unbound bioactive PlGF (please refer to Suppl. Figure 1A-H for individual results for PlGF and sFlt1). Compared to healthy controls, PlGF/sFlt1circ was elevated in sera of PDAC patients (Fig. 1A), but did not differ between patients with advanced, non-resectable versus resectable tumors (Fig. 1B).

Fig. 1

High circulating PlGF/sFlt1 serum ratio is associated with neural invasion and shorter survival in patients with PDAC undergoing curative-intent surgery. A PlGF/sFlt1circ is elevated in PDAC patients (n = 73) compared to healthy controls (ctr; n = 79). B PlGF/sFlt1circ does not differ between patients with non-resectable (locally advanced or metastatic tumors, PT; n = 37) and patients receiving curative-intent surgery (CS; n = 36). C In patients receiving curative-intent surgery, those with evidence of NI (Pn1; n = 20) exhibited higher PlGF/sFlt1circ than patients without NI (Pn0; n = 16). D and E Comparable PlGF/sFlt1circ ratios in patients with (D N1-2, n = 27) or without (N0; n = 9) lymph node metastasis and in patients with R0 (E n = 24) versus R1 resection (n = 12). F and G In Kaplan–Meier estimates PlGF/sFlt1circ above cut-off correlates with shorter OS in patients with resectable PDAC (F HR: 4.05; 95% confidence interval: 1.22 to 13.48; Log-rank p = 0.007; n = 17 below and n = 19 above cut-off), but not with palliative disease (G HR: 0.845; 95% confidence interval: 0.33 to 2.19; p = 0.696; n = 20 below and n = 17 above cut-off). H Tumor-related pain was quantified using visual analogue scales (VAS 0–10) and grouped into no or low (0–3, n = 20), or moderate to strong pain (4–10, n = 16). Shown are PlGF/sFlt1circ for the curative surgery cohort of patients. A-E and H scatter dot plots with median and interquartile range. *, P < 0.05; ****, P < 0.0001; ns, not significant

Notably, in patients undergoing curative-intent surgery, elevated PlGF/sFlt1circ was associated with the occurrence of NI (Fig. 1C), but not with lymph node metastasis (Fig. 1D) nor histological tumor-free resection margins (Fig. 1E).

Based on ROC analyses, a cut-off at 0.251 PlGF/sFlt1circ achieved optimal stratification of patients who were eligible for curative-intent surgery into subgroups with or without NI (Suppl. Figure 1I). Presurgical PlGF/sFlt1circ above this cut-off was linked to shorter overall survival (OS; Fig. 1F) and stronger neuropathic pain (Fig. 1H). In the palliative situation PlGF/sFlt1circ did not separate prognostic subgroups (Fig. 1G).

In contrast to PlGF/sFlt1circ, circulating VEGF in patients with PDAC did not differ from healthy controls, nor were VEGF levels associated with lymph node metastasis, NI, OS, and pain (Suppl. Figure 1 J-M). Moreover, PlGF/sFlt1circ did not correlate with circulating VEGF levels in matched samples from the same patients (rS = 0.261; p = ns).

Circulating PlGF/sFlt1 ratio predicts neural invasion in a prospective cohort of PDAC

The positive association between PlGF/sFlt1circ and NI was corroborated in a prospective validation cohort of PDAC patients undergoing curative-intent surgery. As before, we found PlGF/sFlt1circ elevated in patients with NI (Fig. 2A and Suppl. Figure 2), and tumors from patients with PlGF/sFlt1circ above the cut-off at 0.251 all exhibited NI (p = 0.0195; Fisher’s exact test). Again, PlGF/sFlt1circ did not correlate with the incidence and extent of lymph node metastasis (Fig. 2B and C) or tumor-free resection margins (Fig. 2D). Fittingly, a binominal logistic regression model identified PlGF/sFlt1circ as predictive of NI in both, the retrospective and prospective cohort, whereas neither the status of lymphangioinvasion nor resection margins improved the model (Suppl. Table 4 and 5).

Fig. 2

Circulating PlGF/sFlt1 ratio predicts neural invasion in a prospective cohort of PDAC patients undergoing curative-intent surgery. A PlGF/sFlt1circ is elevated in patients with NI (n = 32) compared to patients without NI (n = 9). B-D PlGF/sFlt1circ does not correlate with either incidence (B) or extent (C) of lymph node metastasis (B n = 12 for N0 and n = 29 for N1-2; C n = 22 for below and n = 19 for above median); or with R0 versus R1 resection margins (D n = 28 for R0, n = 13 for R1). **, P < 0.01; ns, not significant

PlGF and its receptor Nrp1 are expressed at the tumor-nerve interface in PDAC

Consistent with elevated PlGF/sFlt1circ in PDAC patients, PlGF mRNA-transcript levels were increased in human PDAC tissues (Fig. 3A) and orthotopic xenograft tumors (Fig. 3B, and own previous data [25]) compared to the adjacent healthy pancreas. Both, tumor cells and tumor stroma constitute sources of PlGF, as determined by species-specific quantification of murine and human PlGF proteins in xenograft tumors (Fig. 3B) and human PDAC cell lines (Fig. 3C). High levels of PlGF mRNA transcripts correlated with a higher score of stromal inflammation and desmoplasia in human PDAC (Suppl. Figure 3A and B), consistent with expression of PlGF in tumor-associated macrophages and cancer-associated fibroblasts [23, 25, 28, 29]. Moreover, conditioned supernatants from PDAC cell cultures induced PlGF expression in primary Schwann cells, but not in primary neurons (Fig. 3D).

Fig. 3

PlGF and its receptors are expressed at the tumor-nerve interface. A PlGF mRNA-transcript expression in human PDAC (n = 13) and healthy pancreas (ctr, n = 9). B Human (representing tumor cell derived) and murine (representing host derived) PlGF proteins in DANG orthotopic xenograft tumors (PDAC) and paired pancreas (ctr) determined using species-specific ELISA. C ELISA-based quantification of PlGF in supernatants of human PDAC cell line cultures (n = 3). D Primary neurons (PN) and Schwann cells (SC) from newborn mice were cultivated with control media (ctr) or conditioned media (CM) from MiaPaCa tumor cell cultures and PlGF mRNA expression determined (n = 3). Tumor cell conditioned supernatants induce PlGF expression in Schwann cells. E and F Representative IHC for the PlGF receptor Nrp1 in tissues of PDAC (E) and healthy pancreas (F). Intratumoral (in E) and intrapancreatic (in F) nerves are indicated by asterisks. Nrp1 expression in ductal epithelial cancer cells (arrowheads) and nerves (arrows). G and H mRNA transcripts and protein expression of Nrp1 and VEGFR1, respectively, in (G) human PDAC cell lines, the immortalized human pancreatic ductal epithelial cell line HPDE, as well as in (H) dorsal root ganglia (DRG), primary neurons (PN) and Schwann cells (SC; n = 3). *, P < 0.05; **, P < 0.01

Consistent with previous reports on expression of Nrp1 in tumor and stromal cells in PDAC [29, 35, 36], Nrp1 immunoreactivity was found on tumor cells, parenchymal nerves and on vascular endothelial cells, but was only weakly present in epithelial cells of healthy pancreas (Fig. 3E and F). Correspondingly, human pancreatic ductal epithelial (HPDE) cells lack Nrp1 mRNA and protein expression, which contrasts with abundant Nrp1 mRNA transcripts and protein levels in most PDAC cell lines (Fig. 3G). Experimentally, we found interaction of PlGF with the receptor Nrp1 required and sufficient to stimulate clonal growth of PDAC cells (Suppl. Figure 3C and D).

Notably, Nrp1 mRNA transcripts were also present in murine primary neurons, Schwann cells, and dorsal root ganglia (DRG; Fig. 3H), as well as in F11 neuron cultures (not shown). In contrast, VEGFR1 mRNA transcripts were rarely detectable among human PDAC cell lines (Fig. 3G), but expressed in DRGs, primary neurons, F11 neurons and Schwann cells (Fig. 3H; and not shown), which is consistent with the reported stromal expression pattern in human PDAC tissues [37].

Neural invasion of extratumoral nerves predicts early disease recurrence in PDAC

In routine pathology, NI is reported as a qualitative feature present in up to 90% of PDAC samples [2, 5, 9, 11]. To obtain a quantitative representation of NI, we performed morphometric analyses evaluating the presence, quality and area of tumor cell invasion of nerves, and the number and diameters of nerves as a measure of nerve density and hypertrophy (Fig. 4A and B and Suppl. Figure 4). Since NI can spread beyond the tumor boundaries, these parameters were recorded in standardized areas within the tumor (intratumoral; IT) and within the non-transformed pancreas outside the tumor margin (extratumoral; ET).

Fig. 4

Morphologic parameters of neural invasion and plasticity: Neural invasion of extratumoral nerves predicts early disease recurrence and overall survival. A Representative IHC images and magnifications of intratumoral (A1 and A2) and extratumoral NI (A3-A5; scale bars, 200 μm). Shown are perineural invasion (PNI; open arrows) and intraneural invasion (INI; filled arrows) of intratumoral (in A1 and A2) and extratumoral (intrapancreatic in A3; extrapancreatic in A4 and A5) nerves (asterisks). B Illustration summarizing morphologic parameters of neural plasticity and invasion (created with BioRender.com). Neural plasticity is quantified by number, caliber and area of intra-/extratumoral nerves (middle panel). Neural invasion is assessed by presence, localization (perineural vs. intraneural) and extent (circumference; score 1–4) of tumor cells within the neural space (right panel). C Incidence of tumors with (NI +) or without (NI-) neural invasion (n = 20; p = 0.19, Fisher’s exact test) in PDAC specimens (intratumoral) and corresponding adjacent healthy pancreas (extratumoral). D and E Data on DSF and OS was available for n = 17 patients. A high neural dissemination score of extratumoral nerves is associated with shorter DFS (D) and OS (E) in patients receiving curative-intent surgery. Kaplan–Meier estimates depict DSF and OS of patients with extensive neural invasion of extratumoral nerves (score 2; n = 6) versus absent or focal neural invasion (score 0 and 1, respectively; n = 11; HR: 4.57; 95% confidence interval: 1.09 to 19.22; Log-rank p = 0.0038 in D HR: 4.63; 95% confidence interval: 1.00 to 21.52; Log-rank p = 0.0004 in E). F Percentage of tumor-invaded nerves per total nerves in PDAC specimens (intratumoral) and adjacent healthy pancreas (extratumoral). G and H High fraction of invaded extratumoral nerves is associated with shorter DFS (G) and OS (H) in patients receiving curative-intent surgery. Kaplan–Meier estimates depict DSF and OS of patients with extratumoral nerve fractions above (n = 8) or below (n = 9) median (HR: 3.29; 95% confidence interval: 0.95 to 11.39; Log-rank p = 0.0210 in G; HR: 3.93; 95% confidence interval: 1.06 to 14.58; Log-rank p = 0.0013 in H). **, P < 0.01

NI of intratumoral nerves (IT-NI) occurred in almost all patients, but involvement of extratumoral nerves affected only 2/3 of patients (ET-NI; Fig. 4C). The extent of NI widely varied between patients and between intra- versus extratumoral areas, as indicated by the fraction of tumor-invaded nerves per total nerves (Fig. 4F), which was lower in extratumoral when compared to intratumoral regions.

Clinically, the presence and extent of ET-NI translated into an unfavorable prognosis in patients undergoing curative-intent surgery. Indeed, strong involvement of extratumoral nerves, as reflected by an extensive ET-NI dissemination score (Fig. 4D and E, and Suppl. Figure 4) and high fractions of tumor-invaded extratumoral nerves (Fig. 4G and H) correlated with shorter disease-free survival (DFS) and OS.

Expression of PlGF transcripts correlates with neural invasion of extratumoral nerves

We then assessed whether PlGF mRNA transcript levels correlated with the incidence and extent of NI in human PDAC tissues. Given the obvious clinical impact (see Fig. 4), our analyses focused on PlGF in relation to NI of extratumoral nerves (Fig. 5A-F). Tumors with ET-NI exhibited higher PlGF transcript levels as compared to tumors in which ET-NI was absent (Fig. 5A). Strikingly, all tumors with PlGF transcript levels > median showed ET-NI and exhibited a higher NI dissemination score, whereas tumors with low PlGF transcript levels either completely lacked ET-NI (50% of cases; Fig. 5B) or exhibited only low NI dissemination scores (Fig. 5C). Moreover, correlation analysis demonstrated a positive correlation between PlGF transcript levels and the fraction of invaded nerves (rS = 0.4996; p = 0.0294; Suppl. Figure 5F). Next, we related PlGF transcript levels to morphologic features of more advanced ET-NI, such as a wide circumferential range of perineural tumor cell invasion along the sheath of nerves (referred to as perineural invasion; PNI), and tumor cell invasion within the intraneural space of nerves, referred to as intraneural invasion (INI; illustrated in Fig. 4A and B). Tumors with ET-PNI exhibited higher mean PlGF transcript levels as compared to tumors without ET-PNI (Fig. 5D). Conversely, tumors with PlGF transcript levels > median exhibited a high ET-PNI score (Fig. 5E) and a larger ET-PNI area (Fig. 5F), as determined from the range of circumferential growth of tumor cells and the area of tumor cell clusters within the perineural sheath (scheme illustrated in Fig. 4B). Clinically, occurrence and extent of ET-INI both predicted early disease recurrence and a shorter OS (Suppl. Figure 5A-D). In sharp contrast, PlGF transcript levels did not correlate with either lymphangioinvasion (Fig. 5G), or incidence (Fig. 5H) or the extent of lymphatic metastasis (Fig. 5I).

Fig. 5

PlGF mRNA transcript levels correlate with the extent of neural invasion of extratumoral nerves. A-I Analyses refer to n = 20 PDAC samples that allowed for examination of extratumoral nerves. A PlGF mRNA transcript levels dependent on presence (NI +) or absence of neural invasion (NI-). B Incidence of NI in tumors with PlGF mRNA transcripts < median and > median. C Semiquantitative assessment of the NI dissemination as extensive (score 2), focal (score 1) or absent (score 0) in tumors with PlGF mRNA transcripts < median and > median. D PlGF mRNA transcript levels in tumors without (PNI absent) or with perineural invasion (PNI present). E Circumferential range of PNI was morphometrically determined and scored as 0 (absent), 1 (1/4 circumference), 2 (1/2 circumference), 3 (3/4 circumference), and 4 (whole circumference). Shown are PNI scores of affected nerves in tumors with PlGF mRNA transcripts < median and > median. F The PNI area fraction was determined by calculating the ratio of the PNI tumor cell area and the area of the corresponding nerve. G PlGF mRNA transcripts in tumors without (L0) and with (L1) lymphangioinvasion. H and I, Comparable PlGF mRNA expression in tumors with (N1-2) and without (N0) lymphatic metastasis (H) and, conversely, similar fractions of tumor infiltrated lymph nodes per total lymph nodes in tumors with PlGF mRNA transcripts < median and > median (I). *, P < 0.05; **, P < 0.01; ns, not significant

Circulating PlGF/sFlt1 ratio correlates with a more extended neural invasion

Given the correlation of tissue PlGF mRNA to several parameters that describe the extent of NI, we next asked, whether PlGF/sFlt1circ also reflects the extent of NI in a quantitative way. Therefore, we quantified NI on the basis of the most pertinent morphometric parameters in serum-matched tissue samples in the prospective cohort. Subsequent analyses revealed a significant correlation between PlGF/sFlt1circ and the global fraction of invaded nerves, when both intratumoral and extratumoral nerves were evaluated (rS = 0.3244, p = 0.0411; Suppl. Figure 5E).

Taken together, our morphometric analyses link PlGF/sFlt1circ and tissue PlGF expression to the presence and extent of NI in the retrospective and prospective PDAC cohorts and support the notion that PlGF/sFlt1circ may serve as a quantitative serum biomarker of NI, prompting us to experimentally explore the function of PlGF at the tumor-nerve interface.

PlGF mediates mutual chemoattraction between tumor cells and Schwann cells

Schwann cells physiologically act as conduits for subsequent axonal outgrowth, but were shown to temporarily disengage from the perineural sheath and bridge the space towards tumor cell colonies in cancer, thus promoting NI [14, 38, 39]. We therefore determined the effects of PlGF on the mutual chemoattraction of Schwann cells and tumor cells. Conditioned media from DANG and Panc1 tumor cell lines, which endogenously secrete PlGF (Fig. 3C), enhanced the directed migration of Schwann cells, while inhibition of PlGF by using anti-PlGF, but not control IgG1 antibodies abrogated this effect (Fig. 6A). Vice versa, conditioned supernatant from Schwann cell cultures stimulated the directed migration of Panc1 and Capan-2 tumor cells (Fig. 6B), while blocking PlGF using anti-PlGF antibodies abolished the directed migration of PDAC cells towards chemoattractant released by Schwann cells (Fig. 6B). Thus, PlGF constitutes a bidirectional chemoattractant acting on tumor cells and Schwann cells.

Fig. 6

PlGF mediates mutual chemoattraction between tumor cells and Schwann cells and stimulates neurite outgrowth. A Neutralizing anti-PlGF antibodies inhibit directed migration of Schwann cells from the upper transwell chamber towards conditioned media from DANG or Panc1 monolayers (lower chamber) as compared to IgG1 control (n = 3). B Anti-PlGF inhibits directed migration of Panc1 and Capan-2 cells towards chemoattractant stimuli from conditioned Schwann cell supernatants placed in the lower chamber (n = 3). C and D Whole primary DRGs (containing neurons and Schwann cells) from newborn mice were incubated with supernatants from various PDAC cells (C) or medium containing recombinant nerve growth factor (NGF), glial-derived nerve growth factor (GDNF) and PlGF (D). Overall neurite length was determined using NeuroQuant® software based on selective staining of primary neurons for neuron-specific β3-tubulin (n = 3–5). PlGF stimulates neurite outgrowth, whereas neutralizing antibodies to PlGF secreted by DANG cells inhibit neurite outgrowth. E–G Representative images of β3-tubulin stained primary neurons cultured with control media (E), DANG supernatant (F) and DANG supernatant with anti-PlGF (G). *, P < 0.05

Tumor derived PlGF regulates neural plasticity

To assess neural plasticity, we performed co-culture assays combining PDAC cell lines with primary cell cultures from DRGs, freshly isolated primary neurons and Schwann cells, or with cultures of F11 hybridoma neurons.

First, primary neurons were co-cultured with either human pancreatic ductal epithelial (HPDE) or PDAC cells in separate patches using IBIDI® inserts (Suppl. Figure 6A-C), and nascent neurites were visualized at 48 h. Notably, neurite length increased upon co-culture with HupT3 PDAC cells as compared to HPDE cells. Moreover, stimulation of primary neuron cultures with conditioned media from HupT3 and DANG cell lines induced mRNA of the growth-associated-protein (GAP)-43, a marker of neural outgrowth and regeneration (Suppl. Figure 6D). Thus, ex vivo co-culture approaches capture aspects of neural plasticity in PDAC.

Second, whole DRG primary cell cultures instead of purified neurons were used, since the inclusion of Schwann cells more closely reflects the tumor-nerve interface in vivo. PDAC cell supernatants variably stimulated overall neurite length (Fig. 6C), as did recombinant PlGF and two established neurotrophic factors, nerve growth factor (NGF) or glial-derived neurotrophic factor (GDNF; Fig. 6D), used as positive controls.

Importantly, neutralizing antibodies to the endogenously produced PlGF in conditioned media from DANG cell cultures significantly reduced neurite length (Fig. 6D-G) and abrogated the directed migration of F11 neurons towards DANG cell supernatants (Suppl. Figure 6E). Thus, PlGF supported cancer-mediated neural plasticity.

PlGF is induced following chemotherapy in vitro and in vivo

Clinical approaches to reduce tumor recurrence following curative-intent surgery focus on (neo)adjuvant chemotherapy. Therefore, we determined PlGF production in DANG xenograft tumors treated with the chemotherapeutic agent gemcitabine or vehicle. Chemotherapy increased PlGF expression in tumor epithelial cells and the stroma of PDAC xenografts (Fig. 7A). We also directly exposed primary neurons and Schwann cells to conditioned PDAC cell media supplemented with chemotherapy or vehicle and determined effects on PlGF. Chemotherapy concentration-dependently increased PlGF expression in Schwann cells, but not in primary neurons (Fig. 7B), confirming that chemotherapy enhances the availability of PlGF in the stromal compartment. In line with this observation, own previous data showed, that a combination treatment with anti-PlGF and chemotherapy potentiated growth inhibition of mouse orthotopic PDAC compared to monotherapies [25].

Fig. 7

Chemotherapy induces PlGF expression within the neural compartment of PDAC. A Treatment of mice bearing orthotopic DANG tumors with the chemotherapeutic agent gemcitabine induces PlGF production by tumor epithelial cells (human) and stromal cells (mouse) as determined using species-specific ELISA. B Conditioned tumor cell supernatant (CM) and gemcitabine dose-dependently induce PlGF expression in Schwann cells (n = 3–5). *,P < 0.05