Immunofluorescence Profiling of Collagen Subtypes is a Predictor of Treatment Outcomes in Pancreatic Cancer

Abstract

Desmoplasia in pancreatic ductal adenocarcinoma (PDAC) is characterized by elevated levels of tumor collagen. Desmoplasia restricts drug delivery in PDAC, contributes to treatment resistance, and is associated with poor survival outcomes. We have previously shown that photodynamic therapy (PDT)-based treatment remediates desmoplasia in orthotopic PDAC tumors by reducing second harmonic generation signals from collagen by >90% and by reducing collagen alignment by >10³-fold. Remediating desmoplasia correlated with improved survival outcomes in mice. To understand the phenomenon at a fundamental level, it is important to dissect the impact of therapy on collagen subtypes. In this study, we demonstrate that immunofluorescence profiling of collagen subtypes I, II, III and IV in PDAC tumors 72 h following multiple treatment regimens is predictive of long-term outcomes. Treatment regimens include nanoliposomal irinotecan chemotherapy (nal-IRI; akin to ONIVYDE^™), a combination of nal-IRI chemotherapy with PDT encapsulated in a single photoactivable multi-inhibitor liposome (PMIL) and an EGFR-targeted PMIL construct (TPMIL). Results show that the relative tumor content of Collagen I, II and III was inversely correlated with overall survival (P≤0.0013, P≤0.0001, P≤0.0011, respectively), while, surprisingly, the relative tumor content of Collagen IV was directly correlated with overall survival (P ≤ 0.0001). Similar relationships were observed between the relative tumor content of collagen subtypes and the residual tumor volume at day 88 following treatment. Considering that the relationship between collagen subtypes and treatment outcomes is observed across multiple treatment regimens, immunofluorescence profiling at 72 h following treatment appears to be predictive of tumor growth inhibition and survival in PDAC. Early immunofluorescence collagen subtype profiling may therefore aid in treatment personalization and may inform the dosimetry and scheduling of combination regimens for PDAC, such as chemotherapy and emerging PDT-based combinations to maximize patient survival benefit.

Introduction

Collagen is the predominant extracellular matrix (ECM) protein in pancreatic ductal adenocarcinoma (PDAC) and is synthesized by both stromal cells and, to a lesser extent, PDAC cells.^1,2 Desmoplasia is a fibrotic reaction that results in elevated collagen deposition in the tumor. Desmoplasia restricts drug delivery, mechanistically contributes to treatment resistance, and contributes to an immunosuppressive microenvironment^3–6 Collagen itself contributes to >90% of all signaling between the ECM and tumor cells in PDAC throughout all stages of the disease.² Tumor collagen itself plays a paradoxical role in tumor development, progression, invasion, and metastasis. While reducing the ECM content in PDAC has proven to improve responses to gemcitabine, nanomedicines, and other therapies in pre-clinical models, clinical results have shown very limited improvements.^4,7–11 Recently, introducing the stromal targeting drug losartan to FOLFIRINOX chemotherapy and nivolumab immunotherapy in locally advanced PDAC provided no benefit for patient resectability, progression-free survival, overall survival, or pathological complete response rates.¹² These findings further point to the complex role of the stroma in PDAC.

Photodynamic therapy (PDT), along with the sub-therapeutic photodynamic priming (PDP) phenomenon, is an important tool in the management of pre-cancers, cancers, and non-cancer indications.¹³ PDT is a light-activated modality, whereby the photoactivation of a light-sensitive compound (a photosensitizer, PS), results in the production of biomodulatory and therapeutic reactive molecular species, including singlet oxygen. PDT is currently in phase 2 trials for unresectable, locally advanced PDAC (NCT03033225), and has been shown to be safe and effective at inducing tumor tissue necrosis in a light-dependent manner.^14,15 Importantly, a finite zone of tumor tissue extends beyond the PDT-induced necrosis, which comprises “primed” tumor tissue that is in response to the PDP effect in areas where the light fluence is not high enough to destroy the tumor tissue.^16,17 This PDP zone can be imaged using CT texture analysis in patients.¹⁷ We have previously shown that a therapeutic dose of PDT (0.5 mg/kg PS, 150 J/cm² of 690 nm light) using cetuximab-targeted liposomes containing a lipid conjugate of the PS benzoporphyrin derivative (BPD) induced significant PDAC tumor necrosis.¹⁸ PDT also significantly reduced tumor collagen, as determined by the Masson’s trichrome stain. We have also shown that a lower PDP dose (0.25 mg/kg PS, 50 J/cm² of 690 nm light) using EGFR-targeted photoactivable liposomes (TPMILs) containing the chemotherapeutic irinotecan controlled orthotopic PDAC tumor growth by >90% and doubled overall survival.¹⁹ Importantly, PDP using TPMILs remediated desmoplasia in vivo by reducing collagen density by >90% and disrupted collagen alignment by >10³-fold, as determined by second harmonic generation imaging (SHG). Importantly, PDP-induced disruption of collagen alignment was directly correlated with survival. It is well established that the photochemical species generated during PDT and PDP can chemically react with collagen,^20,21 causing a stiffening or softening of collagen matrices in vitro, depending on the dose applied.^22,23 This in addition to biomodulatory effects on stromal cells provides a mechanistic basis for the pronounced effects PDT-based therapy has on tumor collagen.²⁴

It is known that collagen subtypes play varying, yet oftentimes paradoxical, roles in PDAC. Historically, Collagen I has been understood to promote the malignant phenotype in PDAC.²⁵ However, recent studies have shown that depletion of collagen I may promote PDAC tumor progression and immunosuppression.²⁶ Not much is known about the role of collagen II in PDAC. It is also known that collagens I and III down-regulate E-cadherin in PDAC, thereby promoting metastatic behavior.²⁷ Collagen IV is a non-fibrillar collagen and is a major component of the basement membrane. Collagen IV expression in PDAC has been found to support cancer cell proliferation and apoptosis resistance in vitro, and promotes migration and metastasis in vivo.^28,29 SHG imaging, which we conducted previously,¹⁹ is limited to structurally intact fibrillar collagens I and II, it is therefore unable to resolve Collagens III and IV.³⁰ Therefore, immunofluorescence provides an alternative route for analyzing PDAC tumor collagen with the advantage of being able to distinguish between various subtypes.

To date, not much is known about the impact of treatment-induced changes in the tumor composition of collagen subtypes in PDAC. Here, we leverage previously published data on orthotopic PDAC tumor responses to PDP using TPMILs and their untargeted PMIL counterpart, as well as nanoliposomal irinotecan, which resembles the approved formulation Onivyde.¹⁹ Using immunofluorescence, we profile the treatment-induced changes to the tumor collagen subtypes and correlate them with various treatment response metrics, such as residual tumor volume and survival. We present a method for acute immunofluorescence analyses of PDAC tumors only 72 h after initiation of therapy and evaluate the predictive nature of collagen subtype composition in PDAC.

Methods

PDAC Tumor Treatment

Orthotopic desmoplastic PDAC tumors comprising MIA Paca-2 pancreatic cancer cells and patient-derived pancreatic cancer-associated fibroblasts (PCAFs) were treated in our previously published study in athymic Swiss Nude mice.¹⁹ Mice were treated with photoactivated EGFR-targeted TPMILs, a photoactivated untargeted liposomal construct (PMILs), or nanoliposomal irinotecan resembling the approved formulation Onivyde. Details of the treatment groups are described in Table 1, which also specifies the data point designations that are presented in Figures 1–4. Treatment groups comprised 3-5 animals per arm.

Table 1.

Data point designations in Figures 1–4 of this study, their respective treatment group assignment, the respective PDT dose parameters, irinotecan dosages, and cetuximab dosages.

Data Point Designation	Treatment Arm	Photosensitizer Equivalent Dose (mg/kg, i.v.)	Irinotecan Equivalent Dose (mg/kg, i.v.)	Cetuximab Equivalent Dose (mg/kg, i.v.)	Light Fluence (J/cm2)	Light Irradiance (mW/cm2)	Drug-Light Interval (h)
a	No Treatment Control	n.a.	n.a.	n.a.	n.a.	n.a.	n.a.
b	Photactivable multi-inhibitor liposomes (PMILs)	0.25	9.6	0	50	100	12
c	Nanoliposomal irinotecan	n.a.	20	n.a.	n.a.	n.a.	n.a.
d	EGFR-Targeted photoactivable multi-inhibitor liposomes (TPMILs)	0.25	9.6	1.7	50	100	12

Figure 1.

Representative pseudo-colored immunofluorescence images of untreated control MIA PaCa-2 + PCAF pancreatic tumor tissue sections stained for collagens I, II, III and IV. (Scale bars are 500 μm).

Figure 4.

Scatter plots representing the correlations between median survival and the percentages of tumor collagen subtypes, with respect to total tumor collagen content. (Data is mean ± S.E.M. Data point designations of treatment groups a-d are detailed in Table 1. P values (two-tailed test) correspond to the statistical significance of the Pearson’s r correlation calculated using GraphPad Prism v10.01)

Second Harmonic Generation (SHG) Imaging

SHG imaging has been used extensively in the literature to quantify collagen density and collagen alignment in preclinical and clinical tissue.^30–33 We previously used SHG imaging to determine collagen density and collagen alignment from cryopreserved unfixed PDAC tumor tissue sections following the treatments described in Table 1.¹⁹ SHG imaging was performed in backward scattered mode using an Olympus FV1000 multiphoton microscope with a tunable Mai Tai laser operating at 880 nm. Tissue was imaged using a 40× water-immersion objective and signals were collected using a 400–460 nm SHG emission filter set. Tumor analysis was performed by random sampling of 10 regions of interest (n = 4–5 mice per group; 40–50 regions of interest per arm).

Immunofluorescence Staining and Image Analysis

Treated mice were allowed to recover for 72 h and were humanely euthanized. PDAC tumors were extracted, bisected and cryosectioned at a thickness of 5 μm per section. Tissue sections were then fixed with acetone, stained using primary and fluorescently labeled secondary antibodies, and counterstained with DAPI (Table 2).

Table 2.

A summary of the primary and fluorescently labeled secondary antibodies used in this study to image Collagens I-IV in tumor tissue sections.

Antibody	Supplier	Catalog Number	Species Reactivity	Source Species
α-Collagen I	Cell Signaling Technology	39952S	Human	Rabbit
α-Collagen II	Abcam	Ab34712	Human (and others)	Rabbit
α-Collagen III	Southern Biotech	1330-01	Human (and others)	Goat
α-Collagen IV	Abcam	Ab6586	Human (and others)	Rabbit
α-Rabbit-Alexa 647	Invitrogen	A32795TR	Rabbit	Donkey
α-Goat-Alexa 488	Invitrogen	A32814TR	Goat	Donkey

Immunofluorescence images of the mounted tumor tissue sections were acquired using the Hamamatsu NanoZoomer slide scanner. Image analysis was performed using ImageJ (https://imagej.net/) by randomly selecting 10 regions of interest (ROIs) within each tumor section and quantifying collagen signals by pre-processing threshold values with Huang thresholding. Image acquisition settings were consistent for each tumor collagen subtype across all treatment groups. Unprocessed fluorescence images of the collagen subtypes were analyzed and the mean signals for each ROI were obtained using ImageJ. The collagen subtype fraction (F) for each collagen subtype was first calculated by dividing the mean ROI intensities for each collagen subtype by the mean ROI intensities of the same subtype in untreated tumors (Equation 1). The purpose of deriving F is to normalize the differences in fluorescence signal intensities across the different collagen subtypes, which naturally arise from variations in secondary antibody fluorescence intensities. The percentage of each collagen subtype (% Collagen_ST) in every tumor was then obtained by diving F for each collagen subtype with the sum of F for all collagen subtypes in a given tumor, which was then multiplied by 100 (Equation 2),.

$$\mathit{collagen\; subtype\; fraction}\left(F\right)=\frac{{\mathit{mean\; collagen\; signal}}_{\mathit{TREATMENT\; GROUP\; x}}}{{\mathit{mean\; collagen\; signal}}_{\mathit{UNTREATED\; CONTROLS}}}$$ Equation 1:

$$\%{\mathit{Collagen}}_{\mathit{ST}}=\frac{F_{\mathit{Collagen\; x}}}{\sum \left(F_{\mathit{Collagen\; I}}+F_{\mathit{Collagen\; II}}+F_{\mathit{Collagen\; III}}+F_{\mathit{Collagen\; IV}}\right)}\times 100$$ Equation 2:

% Collagen_ST values were then plotted against treatment outcome metrics that were generated from our previously published study.¹⁹ Treatment outcome metrics included Residual Tumor Volume at d 88, Collagen Angular Anisotropy (as determined by Second Harmonic Generation Imaging), Collagen Density (as determined by Second Harmonic Generation (SHG) Imaging), Overall Survival, Median Survival, and Progression-Free-Survival. Correlation analyses were performed as described below.

Statistical Analysis

All statistical analyses of data in this study were performed using GraphPad Prism v10.01. The mean of individual data points corresponding to the treatment outcome metrics (above) from individual animals were plotted as scatter plots using GraphPad Prism v10.01 against individual % Collagen_ST values generated from Equations 1 and 2. Scatter plots were presented as mean ± S.E.M. values for clarity, as presented throughout the study. Simple linear regressions were used for all scatter plots. GraphPad Prism’s Correlation Analysis function was used to compute Pearson’s r correlation coefficient values. Pearson’s r was computed for X vs. every Y dataset. The statistical significance of the Pearson’s r correlations was automatically calculated using a two-tailed test during the correlation analysis.

Results and Discussion

Immunofluorescence staining of collagens I, II, III and IV (Figure 1) was used in this study to evaluate how acute treatment-induced changes in PDAC tumor collagen subtypes could be predictive of treatment outcomes. Immunofluorescence staining was performed on tumor cryosections from our previously published study¹⁹, which were harvested 72 h following treatment. These treatments conducted in our previous study included photoactivated EGFR-targeted TPMILs, a photoactivated untargeted liposomal construct (PMILs), and nanoliposomal irinotecan resembling the approved formulation Onivyde (Table 1).¹⁹ Relative collagen levels were analyzed from the immunofluorescence staining in this study and were correlated with previously published treatment outcome metrics.¹⁹

As shown in Figure 2, correlation analysis between residual the tumor volume at d 88 after initiation of treatment, and the percentages of tumor collagen subtypes at 72 h after initiation of treatment revealed that a statistically significant direct correlation was found for collagen I (r = 0.3127, P = 0.0012) and collagen III (r = 0.2849, P = 0.0016). Furthermore, a statistically significant inverse correlation was found for collagen IV (r = −0.4123, P < 0.0001). Collagen II content appears to have no predictive value for residual tumor volume. These findings suggest that acute treatment-induced reductions in collagens I and II, and acute treatment-induced elevations in collagen IV could be predictive of a lower tumor volume in PDAC as the disease progresses.

Figure 2.

Scatter plots representing the correlations between residual tumor volume at d 88 and the percentages of tumor collagen subtypes, with respect to total tumor collagen content. (Data is mean ± S.E.M. Data point designations of treatment groups a-d are detailed in Table 1. P values (two-tailed test) correspond to the statistical significance of the Pearson’s r correlation calculated using GraphPad Prism v10.01)

Correlation analysis in Figure 3 between progression-free survival (survival days at which tumors were ≤100 mm³) and the percentages of tumor collagen subtypes at 72 h after initiation of treatment revealed that a statistically significant direct correlation was found for all collagen subtypes. An inverse correlation was found for collagen I (r = −0.3024, P = 0.0038), collagen II (r = −0.5785, P < 0.0001) and collagen III (r = −0.2851, P = 0.004). Furthermore, a statistically significant direct correlation was found for collagen IV (r = −0.4967, P < 0.0001). These correlations suggest that acute treatment-induced reductions in collagens I, II, and III, and acute treatment-induced elevations in collagen IV could be predictive of prolonged progression-free survival in PDAC.

Figure 3.

Scatter plots representing the correlations between progression-free survival and the percentages of tumor collagen subtypes, with respect to total tumor collagen content. (Data is mean ± S.E.M. Data point designations of treatment groups a-d are detailed in Table 1. P values (two-tailed test) correspond to the statistical significance of the Pearson’s r correlation calculated using GraphPad Prism v10.01)

Figure 4 demonstrates that statistically significant correlations between median survival and the percentages of tumor collagen subtypes were only observed for collagen II (inverse correlation, r = −0.5771, P < 0.0001) and collagen IV (inverse correlation, r = 0.3403, P = 0.0007). Collagens I and III appear to have no predictive value for predicting median survival. These results suggest that acute treatment-induced reductions in collagen II and acute treatment-induced elevations in collagen IV could be predictive of extended median survival in PDAC.

As shown in Figure 5, correlation analysis between overall survival and the percentages of tumor collagen subtypes at 72 h after initiation of treatment revealed that statistically significant correlations were found for all collagen subtypes tested. Statistically significant inverse correlations were found for collagen I (r = −0.3350, P = 0.0013), collagen II (r = −0.5261, P < 0.0001), and collagen III (r = −0.3226, P = 0.0011). A statistically significant direct correlation was also observed for collagen IV (r = 0.5353, P < 0.0001). As such, the results suggest that acute treatment-induced reductions in collagens I, II and III, as well as acute treatment-induced elevations in collagen IV, could be predictive of prolonged overall survival in PDAC.

Figure 5.

Scatter plots representing the correlations between overall survival and the percentages of tumor collagen subtypes, with respect to total tumor collagen content. (Data is mean ± S.E.M. Data point designations of treatment groups a-d are detailed in Table 1. P values (two-tailed test) correspond to the statistical significance of the Pearson’s r correlation calculated using GraphPad Prism v10.01)

There were no significant correlations, direct or inverse, between the percentage of each collagen subtype determined by immunofluorescence in this study, and the SHG imaging-derived collagen metrics from our previously published study¹⁹ (Table 3, Table 4). The SHG imaging-derived collagen metrics from our previously published study are collagen density and collagen angular anisotropy. A summary of the Pearson’s r correlations in this study and the corresponding statistical significance P values are summarized in Table 3 and Table 4, respectively.

Table 3.

Strength of the direct and inverse correlations between tumor collagen subtype compositions at 72 h following therapy (determined in this study using immunofluorescence) and various treatment outcome metrics from our previously published study¹⁹.

	Pearson’s r
	Collagen I	Collagen II	Collagen III	Collagen IV
Residual Tumor Volume at d 88	0.3127	0.1075	0.2849	−0.4123
Progression-Free-Survival	−0.3024	−0.5785	−0.2851	0.4967
Median Survival	−0.1660	−0.5771	−0.1590	0.3403
Progression-Free-Survival	−0.3024	−0.5785	−0.2851	0.4967
Collagen Angular Anisotropy (SHG Imaging from previous study 19 )	−0.0100	−0.1785	0.1394	−0.1124
Collagen Density (SHG Imaging from previous study 19 )	−0.1142	0.1086	0.0899	0.0043

Table 4.

Statistical significance of the direct and inverse correlations between tumor collagen subtype compositions at 72 h following therapy (determined in this study using immunofluorescence) and various treatment outcome metrics from our previously published study¹⁹.

	P Value
	Collagen I	Collagen II	Collagen III	Collagen IV
Residual Tumor Volume at d 88	0.0012	0.2871	0.0016	<0.0001
Progression-Free-Survival	0.0038	<0.0001	0.0040	<0.0001
Median Survival	0.1178	<0.0001	0.1141	0.0007
Overall Survival	0.0013	<0.0001	0.0011	<0.0001
Collagen Angular Anisotropy (SHG Imaging from previous study 19 )	0.9197	0.0756	0.1289	0.2315
Collagen Density (SHG Imaging from previous study 19 )	0.2462	0.2819	0.3287	0.9638

P values in bold typeface correspond to P ≤0.05.

The absence of a relationship between immunofluorescence profiling of collagen subtypes in this study and the SHG imaging-derived collagen metrics is to be expected, as both techniques are inherently different in what they analyze. Immunofluorescence detects the presence of collagen subtype molecular epitopes, while SHG imaging is a second-order nonlinear optical process that detects ordered polycrystalline macromolecules that lack a center of symmetry, such as Collagens I/II.³⁴ Previous studies in the literature have also demonstrated the same discrepancy between tumor histological staining, namely using Masson’s Trichrome stain, and SHG imaging-derived collagen metrics on density and alignment.³¹

It is well established that Collagen IV, a non-fibrillar network-forming collagen, is the major component of the basement membrane.³⁵ The turnover and degradation of Collagen IV, amongst other basement membrane collagens, has been associated with the malignant phenotype.³⁶ Degradation of Collagen IV is directly associated with invasion of the basement membrane and cancer cell metastasis. Furthermore, Collagen IV has also been associated with dose dependent effects on angiogenesis; moderate concentrations (~30 μg/ml) of Collagen IV stimulated neovascularization, while higher concentrations >300 μg/ml stabilized neovessels in a 3D aorta model.³⁷ Our data suggests that treatment-induced elevations in collagen IV are the most predictive of favorable treatment outcomes, which could be attributed to a reduction in PDAC invasion and metastasis, as well as a potential inhibitor effect on angiogenesis. Interestingly enough, soluble collagen IV has been found to be directly correlated with poor prognosis in PDAC patients.³⁸ Our findings, therefore, could suggest that treatment-induced elevations in collagen IV may elicit the opposite effect on prognosis, and that high baseline levels of collagen IV may be mechanistically and pathologically distinct. Alternatively, our findings could suggest that soluble collagen IV and collagen IV in the extracellular matrix itself have a distinct, contradictory effect on prognosis. Our findings may also suggest that, along with treatment-induced elevations in collagen IV, concomitant treatment-induced reductions in collagens I-III are needed for the favorable treatment outcomes that we observe. This alternative hypothesis may also be supported by the fact that collagen IV is the only subtype that we investigate that is non-fibrillar, is abundant in the basement membrane, and plays an important role in containing invasive malignant tissue. Future studies will explore the relationship between treatment-induced changes in tumor collagen subtypes and angiogenesis, as well as the rates of metastases, which our previous data suggests impacts survival more than tumor volume alone.¹⁶

Conclusions

The abundance, density, organization, and subtype-composition of tumor collagen play important, complex roles in PDAC disease progression and survival. While long-term changes in the extracellular matrix, such as radiation- and chemotherapy-induced fibrosis in tumor tissue and healthy tissue, are well reported, acute changes in tumor collagen composition following treatment are yet to be fully explored. This study leverages the findings of our recent study,¹⁹ which shows that PDT-based treatment combinations in mice bearing orthotopic desmoplastic PDAC tumors significantly reduce second harmonic generation signals from tumor collagen and disrupt tumor alignment by >10³-fold. Importantly, treatment-induced disruptions in tumor collagen alignment were directly correlated with improved survival. Here, we explore the relationship between collagen subtype composition and several PDAC treatment outcome metrics. By using immunofluorescence staining of tumor tissues only 72 h after treatment, we demonstrate that treatment-induced elevations in collagen IV consistently predict lower residual tumor volume 88 d after treatment, and prolonged survival. To a lesser extent, we also show that treatment-induced reductions in collagens I, II, and III can be predictive of improved long-term treatment outcomes. The significance of the findings we present here is broad and can be potentially applied to various therapeutic strategies, including chemotherapy and PDT-based combinations. We propose future comprehensive studies exploring whether changes in collagen subtype composition can be predictive of outcomes for additional pancreatic cancer regimens, such as FOLFIRINOX, gemcitabine, stereotactic body radiotherapy (SBRT), and immunotherapy combinations. Furthermore, evaluating collagen subtype composition using immunohistochemistry and proteomic analyses could provide insights into the analytical techniques that provide the greatest predictive value.

The approach we present here, which could be applied to assess PDAC tumor tissue collagen subtypes in patient tumor biopsies, could prove to be invaluable in optimizing, personalizing, and iteratively adapting combination treatments to offer the greatest survival benefit to PDAC patients. These findings also warrant further investigation into the role of acute changes in collagen subtype composition following treatment in various other solid malignancies, which also exhibit extensive desmoplasia.

• Acute changes in tumor collagen is predictive of outcomes in pancreatic cancer

• Collagen subtypes exhibit differential changes following treatment

• Acute reductions in proportions of Collagens I-III predict favorable outcomes

• Acute elevations in the proportion of Collagen IV predicts favorable outcomes