PD-1 and PD-L2 expression predict relapse risk and poor survival in patients with stage III colorectal cancer

Abstract

Background

Immune responses have long been an area of interest in cancer research. In this study, the effects of programmed cell death-1 (PD-1) and its ligand (PD-L2) on the prognosis of colorectal cancer (CRC) were investigated.

Methods

Primary tumour specimens of stage III CRC patients operated between 2002 and 2013 were assessed for PD-1 and PD-L2 expression and various clinicopathological and prognostic factors.

Results

We observed a significant relationship between poor prognostic factors and PD-1/PD-L2 expression. These biomarkers were also found to serve as independent risk factors for LIR and MSI. In univariate analysis, relapse-free survival (RFS) and overall survival (OS) rates were found to be poor in PD-1 and PD-L2 positive patients. In multivariate analysis, these biomarkers were found to serve as independent poor prognostic factors for RFS and OS.

Conclusions

Our data indicate that PD-1 and PD-L2 may serve as independent prognostic survival parameters for CRC patients and may be employed for the design of targeted therapies.

Introduction

Colorectal cancers (CRCs) are among the deadliest cancers worldwide. Many risk factors have been associated with CRC development, including alcohol consumption, ulcerative colitis, tobacco use, ageing, sedentary lifestyle, chromosomal imbalances, microsatellite instability (MSI) and genetic mutations [1, 2]. Recent studies have shown that immunosuppression plays an important role in the development and progression of CRC [3]. More than half of all CRC cases are diagnosed at advanced stages, such as stage III. Although overall the clinical results have become better with advances in diagnostic methods and treatment approaches, relapses and death are still common in the advanced stage patient population [1–3]. Also, although current treatment regimens include adjuvant chemotherapy as a standard for advanced cases, the absolute improvement in terms of survival is low. On the other hand, the TNM system does not take into account any indicators other than traditional pathological staging parameters when deciding on treatment [3, 4]. However, some CRC subpopulations may benefit from targeted therapies and, as such, there is a pressing need for new biomarkers allowing better clinical management. Currently, programmed cell death-1 (PD-1) and its ligands (PD-L1/PD-L2) are among the most promising parameters.

The tumour microenvironment is a complex structure comprising different immune cells, including lymphocytes, cancer-associated fibroblasts, tumour-associated macrophages and dendritic cells. Accumulation of T lymphocytes is a common feature of many malignancies [5, 6]. PD-1 and its ligands are among the most important inhibitory molecules in T cells. The PD-1 pathway has amply been studied, and the effects of the immune checkpoint proteins PD-1, PD-L1 and PD-L2 on survival have been illustrated in many tumours [7–13]. Also, anti-PD-1 agents have successfully been applied to various cancers such as melanoma, CRC and several advanced-stage cancers [14–18]. Most of these studies have, however, focussed on the effects of PD-L1, whereas studies on whether PD-L2 is expressed in epithelial cancer types, as well as its functional, prognostic and therapeutic implications, are rather limited. Here, we aimed to assess PD-1 and PD-L2 expression in stage III CRCs and to investigate the role of these parameters on survival.

Materials and methods

Patient selection

Our study was designed in accordance with REMARK [19] recommendations and was approved by the ethics committee of Kırıkkale University Health Research Ethics Committee (2020.11.04). All procedures were performed in accordance with the 1964 Helsinki declaration and national/institutional ethics committee standards. Our study was carried out at the Kırıkkale University Faculty of Medicine, Department of Pathology. All patients who were operated for CRC between 2002 and 2013 were retrieved from the electronic database of Kırıkkale University Hospital. Clinical information was available from in total 330 cases. Twenty-three patients who died within 1 month postoperatively, or who had more than one tumour and/or who received neoadjuvant therapy were excluded from the study. Also, tumour tissue was scarce in paraffin blocks from 11 cases, whereas paraffin blocks were not available for 14 cases and 78 cases had tumour stages other than stage III. These were also excluded. As a result, we ended up with in total 212 patients. The clinical, pathological and survival data of these patients were obtained from the hospital database. The prognostic parameters used in this study were age, gender, location, size, pT-stage, grade, angiolymphatic invasion, perineural invasion, surgical margin, local inflammatory response (LIR), microsatellite instability (MSI) status, and tumour budding.

Tissue sample evaluation

Paraffin blocks and hematoxylin-eosin (H&E) stained slides of cases stored at room temperature were collected from the archives of the Department of Pathology. The numbers of tumour blocks from each patient differed and ranged from 1 to 19. H&E stained slides were re-examined and a block that best represented the tumour was subsequently selected. Three 4 μm thick sections were taken from this block and stained with anti-PD-1 and anti-PD-L2 antibodies and H&E. All sections were evaluated by two expert pathologists blinded from clinical and pathological information. The guidelines of the American Joint Cancer Classification Committee [20] were used for the evaluation.

Microscopic scoring

Evaluation of PD-1 and PD-L2 expression was performed semi-quantitatively using conventional microscopy (Nikon Eclipse E600, Switzerland) with a 20x objective. First, all sections were examined using an 4x objective to get an idea of the staining distribution. Areas of the tumour showing membranous or cytoplasmic staining (even focal or weak) were considered positive. If there were no clearly stained blue nuclei present, the immunohistochemical (IHC) staining was considered false and was not included in the scoring. Finally, all patients were divided into two groups (positive and negative).

Follow-up analysis

The day of surgery was considered as a basis for calculating survival and recurrence times. The total follow-up period of the cases was 11 years, but all events after 60 months were accepted as 60 months. The time between the day of surgery and the day of death was defined as overall survival (OS). The time between the day of surgery and the day of local and/or regional recurrence was defined as relapse-free survival (RFS).

Reproducibility assays

Interobserver agreement and tumour heterogeneity were evaluated for reproducibility. A kappa (к) test was used to evaluate the interobserver agreement and intra-class correlation (ICC) was used to evaluate tumour heterogeneity. The к value represents the ratio of variation between observers and was classified according to Landis and Koch [21]. The ICC value represents the ratio of variation between tumours [22]. The ICC value is expected to be high if heterogeneity is due to variability among different tumours (i.e., biological variation) and low if due to intra-tumour variability (i.e., heterogeneity).

Immunohistochemistry

Two sections of 4 μm thick were taken from the selected paraffin blocks. Tonsillar tissue was used as a positive control and normal colon tissue was used as a negative control. First, the sections were subjected to deparaffinization and rehydration, after which they were boiled in citrate buffer solution (pH = 8) for 10 min in a microwave (antigen recovery process) and cooled down to room temperature. Next, the sections were immersed in a 0.3% hydrogen peroxide-methanol solution for 10 min at room temperature (blocking endogenous peroxidase activity). Mouse monoclonal anti-PD-1 (Ventana, 1:100, clone: NAT105) and mouse monoclonal anti-PD-L2 (Dako, 1:50, clone: 22C3) antibodies were used as primary antibodies and incubated overnight at room temperature. On the next day, the sections were incubated with a secondary antibody (Dako) for 1 h at room temperature, after which 3,3′-diaminobenzidine (Dako) was applied for 5 min. Finally, hematoxylin (Merck, Darmstadt, Germany) was applied to the sections after which they were mounted with Pertex (Histolab, Göteborg Sweden).

Statistical analysis

Percentage, frequency, range, median and standard deviation were used while recording statistical values. Chi-square and logistic regression [95% CI and 1.0 odds ratio (OR)] tests were used for univariate and multivariate analyses of prognostic parameters, respectively. Spearman correlation test was used for correlation between estimates and Wilcoxon Signed-Rank test was used for differences. As mentioned above, the ICC test was used for tumour heterogeneity and the к test for interobserver agreement. Log-Rank and Cox regression [95% CI and 1.0 hazard ratio (HR)] tests were used for univariate and multivariate survival analyses, respectively. Survival curves were plotted using Kaplan-Meier analysis. SPSS 21.0 (IBM Institute, North Castle, NY, USA) was used for statistical analysis. P values < 0.05 were considered statistically significant.

Results and discussion

Of the 212 patient included, 131 (61.8%) were male and 81 (38.2%) were female. Median age and tumour size were 71 (range: 35–98 years) and 6.0 (range: 3–11 cm), respectively. Eighty (37.7%) of the cases were pT I-II, 92 (43.3%) were pT-III and 40 (18.8%) were pT IV. The tumours were located in the right colon in 75 (35.3%) of the cases and in the left colon in 137 (60.0%) of the cases.

All tissue sections were scanned at low magnification. Overall, the staining distribution was heterogeneous, being more prominent in invasive areas (Fig. 1). A significant difference was found between PD-1and PD-L2 expression and poor prognostic parameters (for PD-1: grade [p = 0.016], angiolymphatic invasion [p = 0.009], positive surgical margin [p = 0.004], LIR [p < 0.001] and MSI [p = 0.001]; for PD-L2: grade [p = 0.022], angiolymphatic invasion [p = 0.003], positive surgical margin [p = 0.001], LIR [p < 0.001] and MSI [p < 0.001]). Logistic regression analysis was performed for PD-1 and PD-L2 and, by doing so, they were found to serve as independent risk factors for LIR (for PD-1: OR = 2.28 [1.23–2.74], p = 0.003; for PD-L2: OR = 2.34 [1.28–2.79], p = 0.005) and for MSI (for PD-1: OR = 2.36 [1.17–2.58], p = 0.028; for PD-L2: OR = 2.47 [1.33–3.14], p = 0.035) (Tables 1 and 2).

Fig. 1.

Representative microscopic images of PD-1 and PD-L2 expression. All sections were first scanned at low magnification (4x) after H&E staining (a-b). Evaluation of PD-1 (c-d) and PD-L2 (e-f) expression was performed semi-quantitatively using IHC stained slides and a 20x objective. Abbreviations: PD-1: Programmed cell death-1, PD-L2: Programmed death-ligand 2, H&E: Hematoxylin and eosin, IHC: Immunohistochemistry

Table 1.

Relationship between clinicopathological features, PD-1 and PD-L2 expression

		PD-1			PD-L2
		Positive	Negative	P value	Positive	Negative	P value
Age				0.772			0.766
	< 71	38 (38%)	42 (36%)		35 (38%)	45 (36%)
	≥ 71	60 (62%)	72 (64%)		55 (62%)	77 (64%)
Gender				0.321			0.319
	Female	40 (40%)	39 (34%)		37 (41%)	42 (34%)
	Male	58 (60%)	75 (66%)		53 (59%)	80 (66%)
Localization				0.441			0.264
	Right	32 (32%)	43 (37%)		28 (31%)	47 (38%)
	Left	66 (68%)	71 (63%)		62 (69%)	75 (62%)
Size				0.253			0.559
	< 6 cm	41 (41%)	39 (34%)		36 (40%)	44 (36%)
	≥ 6 cm	57 (59%)	75 (66%)		54 (60%)	78 (64%)
pT-stage				0.780			0.766
	pT I-II	36 (36%)	44 (38%)		35 (38%)	45 (36%)
	pT III-IV	62 (64%)	70 (62%)		55 (62%)	77 (64%)
Grade				0.016*			0.022*
	Low/Moderate	43 (43%)	32 (31%)		40 (44%)	35 (28%)
	High	55 (57%)	82 (69%)		50 (56%)	85 (72%)
Angiolymphatic invasion				0.009*			0.003*
	No	48 (48%)	34 (28%)		45 (50%)	37 (30%)
	Yes	50 (52%)	80 (72%)		45 (50%)	85 (70%)
Perineural invasion				0.874			0.860
	No	38 (38%)	43 (37%)		35 (38%)	46 (37%)
	Yes	60 (62%)	71 (63%)		55 (62%)	76 (63%)
Surgical margin				0.004*			0.001*
	Negative	47 (47%)	33 (28%)		45 (50%)	35 (28%)
	Positive	51 (53%)	81 (72%)		45 (50%)	87 (72%)
LIR				< 0.001*			< 0.001*
	Low	46 (46%)	26 (22%)		43 (47%)	28 (22%)
	High	52 (54%)	88 (78%)		47 (53%)	94 (78%)
MSI status				0.001*			< 0.001*
	MMR-P	50 (51%)	33 (28%)		47 (53%)	36 (29%)
	MMR-D	48 (49%)	81 (72%)		43 (47%)	86 (71%)
Tumour budding				0.061			0.077
	Negative	45 (45%)	38 (33%)		41 (45%)	41 (33%)
	Positive	53 (55%)	76 (67%)		49 (55%)	81 (67%)

*Significance limit for p results was set at 0.05. Significant results are depicted in italics

Abbreviations: PD-1: Programmed cell death-1, PD-L2: Programmed death-ligand 2, MSI: Microsatellite instability, MMR-P: Mismatch repair proficiency, MMR-D: Mismatch repair deficiency, LIR: Local inflammatory response

Table 2.

Regression analysis of five parameters

	PD-1		PD-L2
	OR (95% CI)	P value	OR (95% CI)	P value
Grade	2.48 (0.69–3.96)	0.251	3.48 (0.67–3.82)	0.337
Angiolymphatic invasion	2.44 (0.65–3.81)	0.329	2.56 (0.77–3.71)	0.564
Surgical margin	2.84 (0.88–3.62)	0.119	2.52 (0.84–3.57)	0.152
LIR	2.28 (1.23–2.74)	0.003*	2.34 (1.28–2.79)	0.005*
MSI status	2.36 (1.17–2.58)	0.028*	2.47 (1.33–3.14)	0.035*

*Significance limit for p results was set at 0.05. Significant results are depicted in italics

Abbreviations: PD-1: Programmed cell death-1, PD-L2: Programmed death-ligand 2, OR: Odds ratio CI: Confidence interval, LIR: Local inflammatory response

The correlation results (for PD-1: r = 0.712, p < 0.001; for PD-L2: r = 0.676, p < 0.001) and difference results (for PD-1: r = 0.318, p < 0.001; for PD-L2: r = 0.354, p < 0.001) were generally reproducible. When examining tumour heterogeneity, it was found that most of the diversity was due to biological variation between different tumours. For example, the ICC value of 0.703 listed in Table 3 means that 70.3% of the total heterogeneity was due to biological variation between different tumours. The inter-observer agreement results were within the clinically useful range (for PD-1: 0.73; for PD-L2: 0.71 (Table 3).

Table 3.

Reproducibility of the research

	Kappa	ICC(95% CI)	Correlation/Difference
PD-1	0.73	0.703 (0.496–0.779)	0.712 / 0.318 (p < 0.001)
PD-L2	0.71	0.686 (0.509–0.764)	0.676 / 0.354 (p < 0.001)

*Significance limit for p results was set at 0.05

Abbreviations: PD-1: Programmed cell death-1, PD-L2: Programmed death-ligand 2, CI: Confidence interval, ICC: Intra-class correlation coefficient

During follow-up,168 of the patients relapsed (PD-1 positive = 127; PD-L2 positive = 117) and 152 died (PD-1 positive = 114; PD-L2 positive = 107). The 5-year OS and RFS rates were 22% and 21% in the PD-1 positive patient group and 21% and 20% in the PD-L2 positive patient group, respectively (Table 4).

Table 4.

Survival analysis related to PD-1 and PD-L2 expression

		Univariable Analysis		Multivariable Analysis
		OS	RFS	OS	RFS
		P value (5-year survival)	P value (5-year survival)	P value (HR 95% CI)	P value (HR 95% CI)
Age		0.358	0.292	NC	NC
	< 71	47%	50%	-	-
	≥ 71	27%	27%	-	-
Gender		0.516	0.476	NC	NC
	Female	47%	49%	-	-
	Male	29%	27%	-	-
Size		0.559	0.602	NC	NC
	< 6 cm	46%	47%	-	-
	≥ 6 cm	29%	30%	-	-
Localization		0.567	0.596	NC	NC
	Right	47%	49%	-	-
	Left	28%	29%	-	-
pT-stage		0.496	0.453	NC	NC
	pT I-II	46%	48%	-	-
	pT III-IV	29%	27%	-	-
Grade		0.172	0.167	NC	NC
	pT I-II	47%	49%	-	-
	pT III-IV	28%	28%	-	-
Angiolymphatic invasion		0.128	0.142	NC	NC
	No	50%	50%	-	-
	Yes	25%	27%	-	-
Perineural invasion		0.645	0.574	NC	NC
	No	45%	47%	-	-
	Yes	28%	30%	-	-
Surgical margin		0.008*	0.003*	0.021*	0.012*
	Negative	54%	55%	2.42	2.38
	Positive	21%	20%	(1.18–3.56)	(1.22–3.77)
LIR	0.001*	<0.001*	0.008*	0.001*
	Negative	54%	54%	2.48	2.21
	Positive	20%	21%	(1.27–3.23)	(1.38–3.44)
MSI		0.003*	0.001*	0.013*	0.005*
	MMR-P	51%	52%	2.36	2.21
	MMR-D	23%	24%	(1.25–3.49)	(1.32–3.57)
Tumour budding		0.772	0.658	NC	NC
	Negative	57%	47%	-	-
	Positive	28%	29%	-	-
PD-1	0.001*	< 0.001*	0.004*	0.001*
	Negative	57%	55%	2.63	2.61
	Positive	22%	21%	(1.33–3.85)	(1.47–3.58)
PD-L2		0.001*	0.001*	0.005*	0.004*
	Negative	55%	57%	2.72	2.66
	Positive	21%	20%	(1.40–4.31)	(1.51–4.29)

In univariate survival analysis, a significant difference was seen between the survival groups for PD-1 (RFS, p < 0.001; OS, p = 0.001) and PD-L2 (RFS, p = 0.001; OS, p = 0.001) expression. Subsequent multivariate survival analysis revealed that these two parameters served as independent poor survival parameters for RFS (for PD-1: HR = 2.61 [CI: 1.47–3.58], p = 0.001; for PD-L2: HR = 2.66 [CI: 1.51 to 4.29], p = 0.004) and OS (for PD-1: HR = 2.63 [1.33 to 3.85], p = 0.004; for PD-L2: HR = 2.72 [1.40 to 4.31], p = 0.005). Other independent poor prognostic parameters were LIR, MSI and surgical margin (Table 4, Fig. 2).

Fig. 2.

Kaplan-Meier survival analysis. Survival curves of PD-1 (a-b) and PD-L2 (c-d). The significance limit for p was set at 0.05. Abbreviations: PD-1: Programmed cell death-1, PD-L2: Programmed death-ligand 2

Cancer cells use various mechanisms to escape from the immune system. These include mechanisms such as hiding from the immune system by downregulating foreign tumour antigens, creating an immunosuppressive microenvironment by secreting anti-inflammatory cytokines, and silencing cancer-targeting immune cells by expressing negative regulators of the immune system [6, 7]. Although immune responses against tumour cells have been observed in many cancer types, these responses may be balanced by immunosuppressive factors. One of the main interactions between tumour cells and the immune system is based on PD-1 and its ligands, PD-L1 (B7-H1/CD274) and PD-L2 (B7-DC) [8–10]. The effector functions of T cells expressing PD-1 are reduced when a complex is formed with either of these two ligands.

It is widely accepted now that PD-L1 expressed by epithelial cancer cells inhibits anti-tumor responses and, thus, facilitates the development of cancer [11, 12], as has for example been shown in melanoma and ovarian cancer [23]. In a study encompassing 454 CRC patients, however, PD-L1 expression was observed in only 12% of the cases [24]. In another study encompassing 181 CRC patients, PD-L1 expression was observed in 16 cases and was not found to be associated with survival [25]. In addition, Song et al. [26] evaluated the prognostic significance of PD-L1 expression in 404 CRC cases, and found, upon multivariate analysis, that it did not serve as an independent prognostic factor. These studies indicate that PD-L1 expression may be seen in only a limited subset of CRC cases and may not serve as an independent prognostic factor.

PD-L2 expression was initially thought to occur only in dendritic cells and macrophages and was first studied in oesophagal cancers [11]. More recently, it was also found to be expressed in other tumours such as primary mediastinal large B cell lymphoma and melanoma [12, 27, 28]. Here, we found that PD-L2 is expressed in advanced stage CRCs. Also, a significant association with poor prognostic parameters and a worse survival were observed in PD-L2 positive CRC cases. This association was also observed upon multivariate analysis. As noted above, several reports have indicated that PD-L1 bears no independent prognostic significance in CRC [24–26]. The different prognostic effects of PD-L2 and PD-L1 expression on CRC may be related to their different molecular functions. For example, alternative secondary receptors for PD-L1 (CD80) and PD-L2 (RGMb) have been reported in the literature [29]. Although several studies have shown that these two ligands may suppress T lymphocyte responses, their effects may differ according to tissue and cancer type due to variability in the (micro-environmental) immune background [9]. It is not yet fully known which inhibitory molecules are mediated and which transcription factors are involved in the interactions between PD-1 and its ligands. Also, it is not clear which mechanisms overlap in the intracellular pathways affected by these two ligands. Our results may bring a novel perspective to PD-L2, i.e., agents that inhibit immune checkpoint interactions have become promising in the treatment of cancer patients, and PD-L2 may be a useful molecular marker for patient selection (Fig. 3).

Fig. 3.

Flow chart of the research and results/conclusions. Abbreviations: PD-1: Programmed cell death-1, PD-L2: Programmed death-ligand 2, CRC: Colorectal cancer, LIR: Local inflammatory response, HR: Hazard ratio, CI: Confidence interval, OR: Odds ratio, RFS: Relapse-free survival, OS: Overall survival

It is known that lymphocytes facilitate the recruitment of different cell types in the tumour microenvironment through receptors and, thereby, generate an effective immune response [30]. Although it is not clear whether the presence and subtypes of lymphocytes reflect specific host and/or tumour characteristics, it is considered prognostically important for many tumour types, including CRC [31, 32]. Additional studies have suggested that a prominent T cell infiltration serves as one of the most important CRC outcome markers [33]. Here, we found that the lymphocyte counts were significantly lower in PD-L2 expressing tumours, indicating that the associated pathways may be used during invasive CRC tumour stages. It also indicates that inhibition of these pathways may be a treatment option.

Although chromosomal instability plays a role in the development of many CRCs, approximately 12–15% is related to a deficiency in mismatch repair proteins (dMMR) responsible for microsatellite instability (MSI) [34]. MSI-associated CRCs are usually right-sided, mucinous and lymphocyte-rich, and have a better prognosis than other types of CRC [35]. In these tumours, up-regulation of T cell checkpoints and increased T lymphocyte infiltration have been observed [25, 36]. Since immune checkpoints are among the important factors allowing tumours to escape from the immune system, inhibition of these checkpoints may increase the host immune response, slow down tumour progression, and even cause tumour regression [37, 38]. There is limited information available on the prognostic effect of PD-L2 expression in MSI-associated CRCs, and the results so far have been variable. There are some reports, however, of significant positive results. Le Dung et al. [38] reported, for example, that 32% of dMMR-CRC patients treated with an anti-PD-1 agent responded positively. They also found that PD-L2 expression was associated with survival in dMMR-CRC patients. Although these findings are difficult to interpret, they may be related to the different modes of action of different PD-L2 pathways within the heterogeneous CRC spectrum.

Here, we have examined the tumour microenvironment, which has received much attention recently, and have investigated two promising parameters therein. We have selected our population from the most common CRC patient group to be fairly homogeneous. In addition, we have acted in accordance with REMARK recommendations at every stage of our study. The limitations of our study are as follows. Since the study was retrospective, it was limited in terms of sample difference. The samples we examined were small and do, therefore, not represent the entire tumour. Also, since our patients were treated according to guidelines from before 2015, there may be differences compared to current approaches.

Conclusions

From our data, we conclude that PD-1 and PD-L2 may serve as independent poor prognostic factors for stage III CRC. The relationship between these two factors and LIR and MSI may shed light on the role of the tumour microenvironment in CRC development and guide the design of novel targeted therapies. PD-1 and PD-L2 may serve as biomarkers that can be easily applied in daily practice, with high prognostic reproducibility.

Abbreviations

CRC

Colorectal cancer

PD-1

Programmed cell death-1

PDL-2

Programmed death-ligand 2

OS

Overall survival

RFS

Relapse-free survival

IHC

Immunohistochemistry

H&E

Hematoxylin and eosin

HPF

High power field

ICC

Intra-class correlation coefficient

HR

Hazard ratio

K

Kappa

SD

Standard deviation

CI

Confidence interval

MSI

Microsatellite instability

MMR

Mismatch repair

Funding

This work was supported by the Scientific Research Projects Coordination Unit of Kırıkkale University.

Compliance with ethical standards

Conflict to interest

The authors do not report any conflict of interest.