Introduction

Immune checkpoint inhibitors (ICI) have revolutionized cancer treatment. These antibodies target immune checkpoint receptors, including programmed cell death 1 (PD-1) and cytotoxic T-lymphocyte-associated protein 4. They may be used alone or in combination. ICI are approved for the treatment of metastatic disease in several different cancers, and studies investigating effects in early disease stages are ongoing.1 The difference in efficacy among individuals remains poorly understood and potential biomarkers to predict treatment outcomes, as well as a better understanding of the biological mechanisms, are of major interest.

Malignant melanoma (MM) is an aggressive skin cancer arising from genetic mutations in melanoma cells.2 The introduction of ICI has improved median progression-free survival rates from around 3 months up to 11 months in patients with metastatic disease, and some patients experience long-term survival.3 4 However, many patients fail to respond optimally to ICI therapy. The identification of biomarkers that could be used to select optimal treatments for patients is therefore of great interest. High expression of programmed death-ligand 1 (PD-L1) within the tumor is often related to a better prognosis for ICI treatment.5 However, these markers lack accuracy, and it is still difficult to determine the individual patient response to treatment.6–8 The soluble forms of checkpoint receptors have been suggested as potential markers of prognosis and treatment response but are not established for clinical use.9 10

PD-1 is a transmembrane receptor expressed by activated T- and B-cells. The indigenous function of PD-1 is to control immune activity and maintain self-tolerance. On PD-1 pathway activation, T cell activity is reduced, and cytokine production, T cell proliferation, and cytotoxicity are diminished. PD-1 has two known ligands, PD-L1 and PD-L2. PD-L1 is widely expressed, also on cancer cells, whereas expression of PD-L2 is more restricted.11 12 PD-1 and its ligands are present in soluble forms. These can be measured in plasma, and especially the presence of soluble PD-1 (sPD-1) has been examined in multiple inflammatory conditions and cancers.13–16 The functional role of sPD-1 remains debatable, but plasma levels increase with the degree of inflammation.17

Expression of PD-L1 is upregulated in response to interferon γ (IFN-γ), and thereby immune activation. When PD-L1 expression is increased in cancer cells, it enables signaling through the PD-1 pathway, resulting in reduced T cell activity and potentially tumor cells escaping immune-mediated cytotoxicity.12 When blocking either PD-1 or PD-L1 with ICI, T cells regain their functionality and cancer cells can be eliminated.18 19 Despite improved survival rates, the efficacy of ICI therapy targeting PD-1 varies greatly among patients and cancer diagnoses. The mechanism causing failure of ICI therapy remains largely unknown.

Galectin 3 (Gal-3) is a member of the galectin family, comprising small lectins with well-conserved carbohydrate recognition domains (CRDs) for β-galactosidases. Gal-3 can bind glycosylated sites on proteins. Gal-3 is the sole member of the chimera-type galectin, and it is structurally unique among galectins, consisting of only one CRD linked to an N-terminal peptide domain.20 Gal-3 can multimerize, potentially resulting in large structures found both in a soluble form in plasma and bound to cellular receptors.21 22 In cancer, Gal-3 has suggestive immunosuppressive effects when present in the tumor microenvironment.23

PD-1 and its ligands are modified by post-translational glycosylation, enabling specific binding of Gal-3.24 We have previously shown that Gal-3 can bind these glycosylated sites on the PD-1 receptor.25 In addition, Gal-3 presence in tumor is associated with a poor prognosis for patients with non-small cell lung carcinoma receiving treatment with pembrolizumab, for reasons unknown.26

We hypothesized that Gal-3 can interfere with the binding between PD-1 and pembrolizumab. We also investigated the potential of sPD-1, and plasma Gal-3 as biomarkers in 40 patients with metastatic melanoma treated with pembrolizumab.

Materials and methodsPatient material

Patients with metastatic MM (n=40) were included in this study. At baseline, pembrolizumab treatment was initiated. Inclusion criteria for patients were: age ≥18, inoperable stage III or stage IV MM, and performance status 0–1. The exclusion criteria were pregnancy and breastfeeding. Blood samples were obtained at treatment initiation (baseline) and after 3 and 6 weeks of treatment. C reactive protein (CRP), l-lactate dehydrogenase, white blood cells, neutrophils, lymphocytes, platelets, hemoglobin, cortisol, albumin, alkaline phosphatase, alanine transaminase, aspartate transaminase, creatinine, potassium, sodium, glucose, bilirubin, thyroxin, thyrotropin and vitamin D3 levels were all quantified at baseline. Selected parameters are presented in table 1. Follow-up was 3 years where survival and progression-free survival were recorded (table 1). Endpoints were progression-free survival and overall survival. Patients included were in average 70 years old with a 30/70 female/male ratio. All patients were naïve to treatment and received pembrolizumab as first line of therapy. 25/40 patients had disease progression during the follow-up. Diagnostic biopsies from 19 patients were available and analyzed. The diagnostic biopsy was from the metastatic site and performed prior to treatment initiation.

Table 1

Patient characteristics

Healthy controls (HC) were gender matched (30/70 female/male) but slightly younger than the MM population (age: 63 vs 70, p=0.02). Blood from patients and HC was collected in EDTA tubes and plasma was separated by centrifugation and stored at −80°C until use. Peripheral blood mononuclear cells (PBMC) were collected by Ficoll density gradient centrifugation and cryo-preserved at −150°C until use.

ELISA

Plasma levels of sPD-1 (RnD systems, cat: DY1086), sPD-L2 (RnD systems, cat: DY1224), sPD-L1 (Abcam, cat: ab277712), and Gal-3 (RnD systems, cat: DGAL30) were measured by ELISA. Samples were investigated in duplicates, and cut-off calculated as two times the SD of the blank. Manufacturers protocol was followed, except for addition of 30 µg/mL bovine, goat and mouse IgG in the PD-1 and PD-L2 ELISA, to block binding of heterophilic antibodies.

Immunohistochemistry staining for Gal-3

Staining was done at the core facility at the Department of Pathology, Aarhus University Hospital, Denmark. In brief, formalin fixed and paraffin embedded sections were cut in 3 µm sections, mounted on SuperFrost Plus slides and dried for 1 hour at 60°C. Immunohistochemistry (IHC) staining was performed on the Benchmark Ultra (Ventana, ROCHE). Heat-induced epitope retrieval was performed using CC1 buffer (high pH) for 32 min at 95°C. Sections were blocked for endogenous peroxidase and afterwards incubated for 12 min at 37°C with the primary antihuman Gal-3—mouse monoclonal antibody (clone: 9C4 RTU, Ventana, ROCHE). The antigen binding was visualized using the Optiview DAB IHC detection system.

Surface plasmon resonance

All surface plasmon resonance (SPR) experiments were performed on a Biacore 3000 (Cytiva) instrument, running at 25°C and a data collection rate of 1 Hz. The running buffer was 10 mM Hepes pH 7.5, 150 mM NaCl, 2 mM CaCl2 and 0.05% Tween-20. PD-1 (rhPD-1, R&D systems, 1086-PD) and PD-L1 (rhPD-L1, R&D systems, 156-B7-100) were both recombinant Fc-fused proteins. Gal-3 (kindly provided by Professor Hakon Leffler) was recombinantly produced in E. coli as described by Salomonsson et al.27 For evaluation of the effect of Gal-3 on the binding between PD-1 and pembrolizumab (MSD) and nivolumab (Bristol-Meyers) and PD-L1 interaction with atezolizumab (Roche) and durvalumab (Astra Zeneca), the antibodies were captured in the active flow cell of a protein G-coupled CM5 chip. A similar in-line protein G-coupled flow cell was used for referencing. Next, the analyte PD-1/PD-L1 (1 µg/mL), preincubated with a titration series of 100–6400 nM of Gal-3, was injected in both flow cells for 180 s followed by a dissociation phase of 120 s. To avoid binding between the Fc-fused PD-1/PD-L1 and the protein G surface, the preincubated analyte-mix also contained 500 nM of protein G. At the end of each cycle, both flow cells were regenerated with a 30 s injection of 10 mM glycine pH 1.5.

Data were further referenced by subtraction of a blank (buffer) run. The resulting double-referenced sensorgrams were used to generate steady state plots, in which steady state was defined as the averaged response 30 s after the ended analyte injection (online supplemental figure S1). At this time point, Gal-3 is presumed to have dissociated so that the remaining response can be attributed to PD-1/PD-L1 binding to their respective antibodies.

Electron microscopy

The rhPD-1 (RnD, 1086-PD) and Gal-3 (RnD, cat: 8259-GA) samples were mixed, and the complex was purified by size-exclusion chromatography on an Äkta Pure microbore system using a Superdex200 3.2/300 column (Cytiva Life Sciences) and PBS buffer. 1.2/1.3 300 mesh UltrAuFoil grids (Quantifoil) were glow discharged for 45 s at 15 mA using a Quorum GloQube Plus (Quorom Technologies). 3 μL of the rhPD-1:Gal-3 complex (~0.1 mg/mL) was applied to grids, blotted for 6 s, and plunge-frozen in liquid ethane using a Vitrobot Mark IV plunge freezer using a blot force of zero (Thermo Fisher Scientific) maintained at 4°C and 99% humidity. Vitrified grids were stored in liquid nitrogen until imaging.

Data collection was performed on a Titan Krios G3i (Thermo Fisher Scientific) at the Danish National cryo-EM Facility—EMBION (Aarhus, Denmark) operated at 300 kV with a Bioquantum/K3 setup (Ametek/Gatan) automated with the EPU2.7 software. A nominal magnification of 130,000× (pixel size of 0.647 Å) and an under-focus range between 0.5 and 1.5 µm was used for data collection. An energy filter slit width of 20 eV was used. Exposures were collected as movies of 52 dose fractions with an exposure rate of ~17.5 electrons pixel−1 s−1 and a total exposure of ~58 electrons Å−2. A total of 5500 movies were collected for the rhPD-1:Gal-3 sample. Out of 5500 movies collected on the rhPD-1:Gal-3 sample, 4893 raw movies were used based on curation on ice thickness, CTF fit resolution and total motion. Image processing was performed using cryoSPARC V.4.5.1 software28 (Structura Biotechnology). Patch motion correction was performed with Alignparts algorithm29 and patch CTF estimation using L-BFGS algorithm30 as used in the cryoSPARC implementation. A template picker was used to select 314,153 particle images based on internally generated volumes from selected 2D classes based on an initial blob picking round. The particles were analyzed by 2D classification. This revealed the presence of oligomeric species. The rhPD-1:Gal-3 complex was expected to form a 2:2 complex based on the Fc-fusion to PD-1 in rhPD-1. It was not possible to visualize the entire complex as judged from both 2D classes and 3D ab initio reconstructions. Volume classes of Fc dimers and a 2:2 complex of PD-1:Gal-3 could be identified but only low-resolution maps were obtained. Because of the apparent high flexibility an approach focusing on the 1:1 PD-1:Gal-3 part of the 2:2 complex was used. Ab initio reconstructions with multiple classes resulted in a volume class based on 15 025 particles as the best map of the PD-1:Gal-3 complex. Further refinements did not result in improved maps, probably because of the low number of particles from the low concentrations of sample and high flexibility of the particle. Crystal structures of Gal-1 (3ZSJ) and PD-1 (7CU5) were docked manually into the best volume class followed by automatic fitting using the “Fit in Map” algorithm in UCSF ChimeraX.31

Flow cytometry

PBMCs from metastatic MM patients at baseline (n=11) and after 6 weeks of treatment (n=4) were stained both as naïve cells and after stimulation for 24 hours with anti-CD3 (BD, cat: 555329)/ /CD28 (BD, cat: 555725). Staining was done for: Tigit BV421 (BD, cat: 741182), CD4 BV605 (Biolegend, cat: 317437), CD69 BV650 (BD, cat: 563835), Lag3 BV711 (Biolegend, cat: 369319), CD8 BV785 (Biolegend, cat: 344740), PD1 AF488 (Biolegend, cat: 329936), Gal3 PE (Biolegend, cat: 126706), CTLA4 Pe-Cy5.5 (Biolegend, cat: 349928), CD3 PeCy7 (Biolegend, cat: 300420), CD45RO APC (Biolegend, cat: 304210), CD25 AF700 (BD, cat: 561398), and L/D nIR (ThermoFisher, cat: L34975). Antibodies were used in the recommended concentration. Unspecific binding was blocked using mouse IgG (40 µg/mL) (Jackson ImmunoResearch, cat: 015-000-003). Compensation was done on beads (Ultra comp/Arc amine reactive compensation beads, ThermoFisher cat: 01-3333-41/A10628). Gating was done on single cells and live cells using fluorescence minus one (FMOs) as controls (online supplemental figure S2).

In vitro cell cultures

Plasma from HC (five donors) were pooled to obtain Gal-3-containing plasma. CD4+ T cells were collected from HC using magnetic beads and negative selection (StemCell, cat:17952) (n=5). CD4+ T cells were stimulated with anti-CD3/anti-CD28 18 hours. Cells were incubated with 20% plasma in RPMI (Gibco, cat: 12633020) media with 10 µg/mL BFA (ThermoFisher, cat: 00-4506-51). After 30 min incubation cells were added to wells containing rhPD-L1, 1 µg/mL (RnD systems, cat: 156-B7) or PBS, incubated for 4 hours and intracellular flow was performed staining for OX40 APC (eBioscience, cat: 17-1342-82), Gal-3 PE (Biolegend, cat: 126706), Live Dead nIR (ThermoFisher, cat: L34975), PD-1 AF488 (Biolegend, 329936). Permeabilization was done using True Nuclear Factor Perm kit following the protocol from the manufacturer (Biolegend, cat: 424401). Intracellular staining was done for: tumor necrosis factor-alpha PeCy5.5 (cat: 560679), interleukin 2 (IL-2) PeCy7 (Biolegend, cat: 500326), IFN-γ AF700 (Biolegend, cat: 502520). Gating was done on single-live cells, using FMOs as controls. Gating on intracellular cytokines was compared between the Gal-3+ and Gal-3− population using the same gate (online supplemental figure S2). T-SNE dimensionality reduction was performed using FlowJo with the following parameters: 50 000 single, live, events, clustering channels Gal-3, PD-1, IFN-γ, IL-2, TNFα, OX40.

In vitro effects of pembrolizumab

PBMCs from HC were pre-stimulated with anti-CD3/anti-CD28 for 24 hours (n=5). Cells were subsequently incubated with pembrolizumab (5 µg/mL) (Merck), rhGal-3 (1 µg/mL) (RnD, cat: 8259-GA), or combinations thereof. After 48 hours in culture, the media was removed, and IL-2 was measured in the supernatant (Invitrogen, cat: BMS221-2).

Statistics

Data analysis was performed in GraphPad Prism (GraphPad). Data were investigated for normal distribution, and when fitted, t-test were used to compare timepoints. When data did not follow a normal distribution, differences were analyzed using Mann-Whitney U test. When applicable, paired data analysis was used. Difference between multiple data point of paired values were examined by a mixed-effects analysis. Kaplan-Meier plot was used to investigate the association between survival and progression-free survival. Correlations were investigated by Spearman’s correlation and subsequently linear regression analysis. Data from biopsies were analyzed using Fisher’s exact test.

ResultsGal-3 blocks the binding between PD-1 and pembrolizumab

We have previously demonstrated the binding between Gal-3 and PD-1 as well as PD-L1.25 We, therefore, hypothesized that Gal-3 binding to PD-1 could interfere with the binding between PD-1 and pembrolizumab, or PD-L1 and azeterolizumab. To test this hypothesis, we performed SPR studies in which rhPD-1 was injected over a surface with immobilized pembrolizumab. As expected, in the absence of Gal-3, a strong binding was observed between PD-1 and pembrolizumab. Adding Gal-3 in increasing concentration compromised this binding in a dose-dependent manner. At a concentration of 6.4 µM, equivalent to 166 µg/mL, Gal-3 almost completely blocked the binding between PD-1 and pembrolizumab (figure 1A). A similar curve was observed for PD-1, nivolumab and Gal-3 (figure 1B). When investigating the effect of Gal-3 on the binding between PD-L1 and atezolizumab or durvalumab, only a slight decrease in the binding between the antibody and rhPD-L1 was observed (figure 1C,D). The binding between rhPD-1 and rhPD-L1 was also investigated with increasing concentrations of Gal-3. However, due to the ability of Gal-3 to multimerize and bind to both rhPD-1 and rhPD-L1 it was not possible to determine a steady state (data not shown). These data suggest that especially the interaction between PD-1 and its antibodies will be compromised by high concentrations of Gal-3 in the microenvironment.

Figure 1

Effect of galectine-3 (Gal-3) on binding between programmed cell death 1 (PD-1)/programmed death-ligand 1 (PD-L1) and anti-PD-1/PD-L1 antibodies. (A) Surface plasmon resonance sensorgrams of the binding between Gal-3:PD-1 and antibodies; pembrolizumab, (B) nivolumab, and Gal-3:PD-L1 binding to antibodies; (C) atezolizumab and (D) durvalumab. Preincubated rhPD-1/PD-L1 and Gal-3 (100–6400 nM) were injected over immobilized antibodies.

Gal-3 binding to PD-1 suggests a steric interference with the pembrolizumab binding site

We next continued to visualize the binding between rhPD-1 and Gal-3 using cryogenic electron microscopy (cryo-EM). The rhPD-1-Gal-3 complex was purified using size exclusion chromatography and the peak fraction was used for preparing grids for cryo-EM. The data collected indicated a flexible structure and it was only possible to obtain a low-resolution (~7 Å) map of the complex (figure 2A,B). Known structures of PD-1 and Gal-3 were docked into the map and in the best configuration obtained. The docked structures indicated PD-1 and Gal-3 to interact through a glycosylation at Asparagine58 of PD-1, where the glycosylation on PD-1 was located in close proximity to the carbohydrate binding site of Gal-3 (figure 2C,D). The model of the PD-1:Gal-3 complex obtained from docking into the cryo-EM map was next compared with the crystal structure of PD-1 and pembrolizumab. This indicated that binding of Gal-3 to PD-1 would sterically block the binding of pembrolizumab to PD-1 because of overlapping binding sites (figure 2E,F), which is in accordance with our binding data.

Figure 2

Cryo-EM visualization of the binding between PD-1, Gal-3 and pembrolizumab. (A and B) Orthogonal views of cryo-EM maps without docked structures. (C and D) Orthogonal views of docked crystal structures of Gal-3 (magenta, 3ZSJ) and PD-1 (slate, 7CU5) into the cryo-EM map (A and B). (E) Cartoon representation of crystal structure (5B8C) of pembrolizumab (brown) bound to PD-1 (slate). (F) Overlay of pembrolizumab:PD-1 crystal structure (5B8C) with model of the PD-1:Gal-3 complex showing significant overlap in the pembrolizumab and Gal-3 binding sites on PD-1 resulting in steric hindrance by PD-1 bound Gal-3 on pembrolizumab binding to PD-1. cryo-EM, cryogenic electron microscopy; Gal-3, galectine-3; PD-1, programmed cell death 1.

Having confirmed the binding between rhPD-1 and Gal-3, and further having visualized the potential binding site overlapping with the binding site of pembrolizumab, we continued to investigate the interplay between PD-1, pembrolizumab and Gal-3 in vitro cultures.

High cellular expression of surface Gal-3 downregulates T cell activity and is not reversed by pembrolizumab

Using an in vitro model to mimic a microenvironment, we aimed to investigate the effects of Gal-3. To support the translation of this model, we added pooled normal human plasma containing Gal-3 but with low levels of sPD-1 to sorted CD4+ T cells and subsequently investigated T cell activity. The concentration of PD-1 and Gal-3 in the added, pooled plasma was not determined, but levels were expected to be comparable to levels established in HC plasma and as previously investigated.32 After plasma incubation, surface Gal-3 expression increased significantly from 2.5% (0.78%–4.4%) in untreated cells to 17.7% (9.1%–26%) (p<0.01) in cells treated with plasma (figure 3A). Anti-CD3/CD28 stimulated CD4+ T cells were cultured in the presence of rhPD-L1 to ensure a binding partner for the PD-1 receptor. In these cells, intracellular IFN-γ and intracellular IL-2 were significantly decreased in the Gal-3 positive cells, compared with the Gal-3 negative cells (figure 3B). This was further visualized by t-SNE, where the Gal-3high population expressed low amount of the pro-inflammatory T cell cytokines, IL-2 and IFN-γ, but preserved high expression of PD-1. By contrast, the expression of the pro-inflammatory receptor OX40 was low on this T cell population (figure 3C). The setup was repeated without, rhPD-L1 present in the culture. No difference was observed on IL-2 production, although OX40 expression was lower, with PD-L1 present in the system (online supplemental figure S3). Having observed the immune modulatory effects of Gal-3 from plasma, we continued to evaluate the effects of adding a high concentration of rhGal-3 (1 µg/mL) to cell cultures, and we investigated if cytokine synthesis could be reversed by the addition of pembrolizumab. In line with Gal-3 from plasma, addition of a rhGal-3 (1 ug/mL) to stimulated PBMCs significantly decreased IL-2 production (162 pg/mL (117–217 pg/mL) to 76 pg/mL (25–129 pg/mL), p=0.003), but combining rhGal-3 and pembrolizumab did not rescue IL-2 production when comparing to pembrolizumab only (142 pg/mL (93–199 pg/mL) vs 75 pg/mL (25–115 pg/mL), p=0.002) (figure 3D).

Figure 3

In vitro experiments investigating the effects of Gal-3 in the microenvironment. (A) CD4+ T cells cultured with plasma from healthy controls have increased expression of Gal-3. (B) CD4+ T cells with increased surface Gal-3 produce less IFN-γ and IL-2 on stimulation with anti CD3/CD28. (C) tSNE plot visualizing the coexpression of OX40, IFN-γ, PD-1, Gal-3, IL-2 and TNF-α. Cells incubated with plasma (top) have an increased expression of Gal-3 and a corresponding decreased expression of IL-2 and IFN-γ on the same cellular subsets. Cells not incubated with plasma are represented in the bottom panel. (D) In vitro stimulation of peripheral blood mononuclear cells from healthy controls with pembrolizumab (5 µg/mL) in the presence or absence of Gal-3 (1 µg/mL). IL-2 production in the supernatant, measured by ELISA decrease in the presence of Gal-3 and cannot be rescued by pembrolizumab. **p<0.01, ***p<0.001. Gal-3, galectin 3; IFN, interferon; IL, interleukin; PD-1, programmed cell death 1; TNF-α, tumor necrosis factor-alpha.

This confirmed that T cells binding Gal-3 on their surface, decrease their production of pro-inflammatory cytokines, and that pembrolizumab did not reverse the decreased T cell activity when Gal-3 is present in a high concentration. Having observed the structural data as well as the in vitro experiments, we continued to investigate the interplay between Gal-3 and pembrolizumab in a cohort of patients with metastatic MM.

Pembrolizumab treatment results in increased plasma sPD-1, but not plasma Gal-3

In the search for predictive biomarkers, we measured plasma levels of sPD-1, its ligands, and Gal-3. At baseline, levels of sPD-1 did not differ from HC. After 3 weeks of treatment with pembrolizumab, sPD-1 levels increased significantly, but did not change further after 6 weeks of treatment (figure 4A). Plasma levels of the two PD-1 ligands remained unchanged over the 6-week treatment period but were increased compared with levels in HC (figure 4B,C). Likewise, plasma levels of Gal-3 did not change over the 6 week treatment period, and levels were also increased compared with levels in HC (figure 4D). For all measured plasma proteins, we observed intercorrelation between the three timepoints (data not shown).

Figure 4

Investigation of potential biomarkers in plasma. (A) Plasma levels of sPD-1 measured by ELISA before treatment initiation (0) and after 3 and 6 weeks (n=40), compared with HC (n=20). (B) Plasma levels of sPD-L1 measured by ELISA before treatment initiation (0) and after 3 and 6 weeks (n=40), compared with HC (n=20). (C) Plasma levels of sPD-L2 measured by ELISA before treatment initiation (0) and after 3 and 6 weeks (n=40), compared with HC (n=20). (D) Plasma levels of Gal-3 measured by ELISA before treatment initiation (0) and after 3 and 6 weeks (n=40), compared with HC (n=20). (E) Baseline plasma Gal-3 levels in disease progression “yes” (n=25), “no” (n=15) (left) and corresponding Kaplan-Meyer curve displaying progression-free survival (right). High Gal-3 levels (above median) are associated with a better prognosis. (F) Three weeks plasma sPD-1 levels in disease progression “yes” (n=25), “no” (n=15) (left) and corresponding Kaplan-Meyer curve displaying progression-free survival (right). Low sPD-1 levels (below median) are associated with a better prognosis, however, not significant when evaluated by Kaplan-Meyers analysis. *p<0.05, **p< 0.01, ***p<0.001. Gal-3, galectin 3; HC, healthy controls; sPD-1, soluble programmed cell death 1.

Plasma sPD-1 levels correlate with markers of inflammation and plasma Gal-3 levels correlate with longer progression-free survival in the study period

Next, we investigated for correlation with biochemical markers recorded at baseline. Using Spearman correlation, we observed a correlation between baseline CRP and sPD-1 after 6 weeks. We also observed an inverse correlation with albumin, suggesting that a higher degree of inflammation at treatment initiation is associated with a more significant increase in sPD-1 levels. We made a similar observation for sPD-L1, which also correlated to CRP and neutrophilic count. sPD-L2 did not correlate to clinical parameters. Plasma levels of Gal-3 correlated with creatinine and inversely with platelets (table 2).

Table 2

Soluble plasma proteins and correlation with clinical data

We progressed to examine for correlations between plasma levels of the proteins and progression-free survival or death. We found no associations to death for any of the proteins measured in plasma. Plasma Gal-3 at baseline correlated with time to progression (table 2). We observed baseline plasma Gal-3 levels to be significantly higher in patients with a longer progression free survival after pembrolizumab treatment. Stratifying patients by plasma Gal-3 levels above and below the median (9.6 ng/mL), confirmed high plasma Gal-3 levels to be associated with longer progression-free survival (figure 4E). Plasma levels of sPD-1 after 3 weeks of pembrolizumab treatment correlated positively with progression-free survival, supporting that increased sPD-1 levels 3 weeks after pembrolizumab treatment were associated with a decreased progression-free survival (table 2). We stratified patients by progression “yes” or “no” and accordingly observed significantly higher sPD-1 levels in patients with disease progression. Next, we stratified patients by sPD-1 levels above and below the median value (4.2 ng/mL) but observed no significant difference when we evaluated the probability of disease progression in the two groups (figure 4F). PD-L1 and PD-L2 levels did not correlate with disease progression (data not shown).

Gal-3 is expressed in MM metastatic biopsies

We next evaluated the biopsies from the metastatic site. 18/19 biopsies were positive for PD-L1 (expression >1%). MM metastatic biopsies all expressed Gal-3. The expression of Gal-3 was categorized into 1–4, with 4 being the maximum expression. 4/19 biopsies had an expression of 1 (figure 5A), 5/19 biopsies had an expression of 2 (figure 5B), 8/19 biopsies had an expression of 3 (figure 5C) and 2/19 biopsies had an expression of 4 (figure 5D). The grading of Gal-3 expression was subsequently quantified into low (expressions 1 and 2) and high (expressions 3 and 4). We observed a numerical difference between disease progression in the two groups, as 5/9 (56%) patients had progressive disease in the low group, and 8/10 (80%) patients in the high group had progressive disease. We observed no statistically significant difference in progression-free survival between the two groups (figure 5E).

Figure 5

Expression of galectin 3 (Gal-3) at the site of diagnostic biopsy. (A–D) Gal-3 expression at the site of biopsy, graded 1–4 (left to right) as visualized. (E) Number of patients with or without progressive disease in hi/lo Gal-3 expression (left), right visualized by Kaplan-Meyers analysis. Fisher’s exact test; p=0.3, (F) expression of programmed cell death 1 (PD-1) on CD4+ and CD8+ T cells before and 3 weeks after treatment with pembrolizumab. PD-1 expression decreased significantly with treatment. (G) Evaluating activated T cells at baseline in patients with or without progressive disease. CD8+CD69% cells are close-to-significantly higher in patients without disease progression. (H) %CD4+PD-1+ (left) and CD4+Gal-3+ cells (right) in patients with or without progression. No difference was observed. (I) Peripheral PD-1 expression in CD4+ +T cells from patients with hi/low Gal-3 grade tumor. (J) Peripheral Gal-3 expression in CD4+ +T cells from patients with hi/low-grade tumor. *p<0.05.

PD-1 and Gal-3 are expressed by peripheral T cells in metastatic MM patients

By flow cytometry, we investigated the cellular expression of PD-1, Gal-3, other checkpoint receptors, and activation markers. CD25, CD69, CD45Ro, Gal-3, and the checkpoint receptors CTLA4, LAG3 TIM3, and Tigit did not change in response to treatment with pembrolizumab (online supplemental figure S4), whereas PD-1 expression decreased on both CD4 and CD8 T cells after 3 weeks of treatment (figure 5F). We next evaluated cells from patients with progression-free survival and with disease progression. First, evaluating the percentage of activated T cells, we observed a slight increase in CD8+CD69+ T cells in patients with progression-free survival (p=0.06) (figure 5G). Next, we evaluated the difference in PD-1 and Gal-3 on CD4 T cells, but no difference was observed (figure 5H). We also investigated for association with hi/lo Gal-3 grade tumor. No difference in T cell activation markers was observed (online supplemental figure S4), however, we observed increased PD-1 expression on CD4+ T cells (figure 5I), which adds to the less favorable prognosis of patients with high tumor Gal-3. We observed no difference when evaluating peripheral Gal-3 expression (figure 5J). No difference was observed when evaluating CD8+ T cells (online supplemental figure S4). Expression of other checkpoint receptors did not differ between the two groups (data not shown). These data demonstrate a slightly more activated T cell profile from patients with progression-free survival.

Discussion

Targeting the PD-1 receptor in metastatic MM has resulted in long-term survival for close to 40% of patients. This is a tremendous change for these patients. However, many patients still have progressive disease,33 which calls for an ongoing search for biomarkers as well as optimizing the treatment by combination therapy.34 Targeting Gal-3 in combination with PD-1 has proven efficacy in both preclinical models as well as clinical studies.35–37 The interaction between PD-1 and Gal-3 is established,25 38 39 but how this interferes with pembrolizumab remains to be fully elucidated. We use both SPR and cryo-EM to demonstrate that Gal-3 can sterically block the binding between PD-1 and pembrolizumab. We first used SPR to demonstrate the absent binding between PD-1 and pembrolizumab in the presence of high concentrations of Gal-3. The concentration needed for complete inhibition of the binding between PD-1 and pembrolizumab is significantly higher than what we measure in plasma and the concentration we use in our in vitro experiments. Yet, we do observe a clear effect of adding rhGal-3 to our in vitro cultures. This could be explained by a longer exposure time and thereby increased interaction between the proteins in the cultures. Increased multimerization of Gal-3 in the cultures could also add to the explanation.38 This suggests, that for Gal-3 to block the binding between PD-1 and pembrolizumab, the presence of Gal-3 needs to be continuous and sustained in a local environment. This supports that especially in the microenvironment, Gal-3 can block the interaction between PD-1 and pembrolizumab, this being in accordance with high metastatic site Gal-3 associated with a poor prognosis also in non-small cell lung cancer.26 In line with this, in our cohort, 80% of patients with high metastatic site Gal-3 experienced disease progression, whereas only 56% of patients with low metastatic site Gal-3 experienced progression. Because our study is challenged by a limited number of patients, and especially a low number of biopsies, we only show a numerical difference to support our SPR data, demonstrating that a high level of Gal-3 can inhibit the PD-1 and pembrolizumab interaction. In addition, we also observed a significantly higher expression of PD-1 on CD4+T cells from patients with high metastatic Gal-3. This may add further to the less favorable effects of pembrolizumab treatment for patients with high metastatic site Gal-3. We continued to visualize the interaction between PD-1, pembrolizumab and Gal-3 using cryo-EM. Despite a sub-optimal resolution, we were able to predict the structure for both Gal-3 and PD-1, and to visualize their binding site. We finally demonstrated the binding site for Gal-3 to be sterically overlapping with the pembrolizumab binding site, again supporting that Gal-3 actively blocks the binding between PD-1 and pembrolizumab in the closed microenvironment. These data are supported by previous studies, suggesting an interaction between PD-1, Gal-3 and pembrolizumab.35 We continued to investigate the functional implications of Gal-3 in the microenvironment and provided supportive evidence that a high concentration of Gal-3 resulted in reduced cytokine production from T cells, which could not be rescued by pembrolizumab. These findings demonstrate that Gal-3 has direct immunosuppressive effects as previously suggested,40 41 and these effects cannot be reversed by pembrolizumab, when the concentration of Gal-3 is high.

Both sPD-1 and sPD-L1 have been suggested as biomarkers for ICI therapy, where high levels of sPD-L1 are associated with disease progression.42–44 We did not find an association between sPD-L1 levels and disease progression; however, we did observe a more progressive disease in patients with high sPD-1 3 weeks after treatment initiation. Gal-3 in plasma has previously been demonstrated to be increased in patients with breast cancer45 and it has been suggested that high plasma Gal-3 is associated with a better response to chemotherapy.46 In line with this, we also demonstrated high plasma Gal-3 to be associated with longer progression-free survival. These data were unexpected and in contrast to what we and others report from the tumor microenvironment, where high levels of Gal-3 are associated with a poor prognosis.26 Several hypotheses could explain this contradiction. High Gal-3 in plasma could represent a high immune activity,47 and thereby be associated with a more favorable prognosis. Because plasma is a systemic compartment the risk of blocking the interaction between PD-1 and pembrolizumab becomes less significant. In the microenvironment, both cells, antibodies and Gal-3 are more closely localized, and the interactions occur over longer periods of time. This could favor both the immune suppressant effect of Gal-3 and the blocking of the PD-1/pembrolizumab interaction. Finally, other structural components in the microenvironment could favor multimerization of Gal-3.48 In the microenvironment the association with myeloid derived cells, including macrophages is closer and it is demonstrated that Gal-3 can drive the differentiation of M2 macrophages, not favorable for tumor clearance.49 Collectively, this highlights the difference between the immune regulation in the tumor microenvironment and the peripheral circulation. The effects are, however, also demonstrated by our SPR and in vitro data, suggesting that the interaction between PD-1 and pembrolizumab is inhibited only when Gal-3 is present in a high concentration.

Because immune activation is crucial for tumor clearance, we finally demonstrated that more activated CD8 T cells, were present in patients with a better prognosis. Gal-3 levels on the cells in the peripheral blood, did however not differ, again supporting the effects of Gal-3 to be more pronounced in the metastatic microenvironment.