Studying tumor microenvironment may identify prognostic or predictive biomarkers of clinical response to immunotherapy. Presented here, is an innovative method based on in situ fluorescence multispectral imaging to analyze and count automatically various subpopulations of CD8+ T cells. This reproducible and reliable technique is suitable for large cohort analyses.
Immune cells are important components of the tumor microenvironment and influence tumor growth and evolution at all stages of carcinogenesis. Notably, it is now well established that the immune infiltrate in human tumors can correlate with prognosis and response to therapy. The analysis of the immune infiltrate in the tumor microenvironment has become a major challenge for the classification of patients and the response to treatment.
The co-expression of inhibitory receptors such as Program Cell Death Protein 1 (PD1; also known as CD279), Cytotoxic T Lymphocyte Associated Protein 4 (CTLA-4), T-Cell Immunoglobulin and Mucin Containing Protein-3 (Tim-3; also known as CD366), and Lymphocyte Activation Gene 3 (Lag-3; also known as CD223), is a hallmark of T cell exhaustion. We developed a multiparametric in situ immunofluorescence staining to identify and quantify at the cellular level the co-expression of these inhibitory receptors. On a retrospective series of frozen tissue of renal cell carcinomas (RCC), using a fluorescence multispectral imaging technology coupled with an image analysis software, it was found that co-expression of PD-1 and Tim-3 on tumor infiltrating CD8+ T cells is correlated with a poor prognosis in RCC. To our knowledge, this represents the first study demonstrating that this automated multiplex in situ technology may have some clinical relevance.
In the past few years, immunotherapy has emerged as a very promising treatment for many types of cancers, including RCC. Particularly, immunotherapy based on the inhibition of inhibitory checkpoints like PD-1 and CTLA-4 has been reported to be clinically effective1,2,3,4,5,6. Monoclonal antibodies against CTLA-4, PD-1, or Program Death Ligand 1 (PD-L1) are already approved in several cancers and lead to long lasting clinical responses in more than 20% of patients7. Nevertheless, not all patients are responders, the cost of the treatment is high, and these treatments are toxic, leading to potential serious autoimmune-like side effects. Therefore, the current challenge is to identify predictive markers to those new immunotherapies. The rate of mutations in the tumor, the expression of PD-L1, or the levels of intratumoral CD8+ T cell infiltration have been reported to correlate with clinical response. However, this association is still too weak to recommend the use of these clinical biomarkers in clinical practice except for the companion test for PD-L1 before the administration of Pembrolizumab in non-small cell lung cancer (NSCLC) patients8,9,1011,12. It has been demonstrated that the co-expression of many inhibitory receptors like PD-1, Tim-3, Lag-3, and CTLA-4, induces a cell exhaustion phenotype and resistance to therapy13,14,15. Since peripheral blood is not representative of the tumor microenvironment, it is of high interest to analyze the phenotypic features of the cells in situ. PD-1 and Tim-3 co-expressing T cells are known to be functionally impaired cells in several contexts13,16,17. In this study, the prognostic impact of the co-expression of the two inhibitory receptors PD-1 and Tim-3 on CD8+ T cells was assessed.
Up until now, studying the co-expression of multiple markers on tumor infiltrating lymphocytes (TILs) has mainly been performed by flow cytometry analysis, making it necessary to work on fresh tumors and therefore precluding retrospective analyses. With conventional in situ staining, only one staining at a time can be performed, and the characterization of the cell type that co-expresses the markers is not possible. For example, PD-L1 is expressed by many cell types of the tumoral microenvironment, making it difficult to define by conventional immunohistochemistry analysis which cells expressing PD-L1 are the more relevant for correlative studies. In this work, we developed an innovative in situ multiparametric immunofluorescence method with computer-counting to correlate the co-expression of PD-1 and Tim-3 by tumor infiltrating CD8+ T-cells with clinical outcomes in RCC. This technique has several advantages, including the possibility to analyze at a single cell level and multiple markers at the same time using a multispectral camera that can capture restricted intervals >10 nm through liquid crystal filters18. Moreover, the procedure is automatic which enables an inter-operator reproducibility and a shortened analysis compared to manual techniques19. In the cancer field, several studies reported convincing multiple stainings of immune molecules like PD-1, PD-L1, and CD8 in Merkel-cell carcinoma, lung cancer, and head and neck cancer20,21,22,23. The automated cell count is possible with training by the user (phenotyping step). The fluorescence is measured in the different cell compartments (nuclei, cytoplasm, and membrane).
Here, different subsets of tumor-infiltrating CD8+ T cells expressing PD-1 and/or Tim-3 in a large cohort of RCC were counted and the results were correlated with clinical gravity scores and survival parameters. It was also possible to analyze membrane fluorescence intensity at the cellular resolution with Mean Fluorescence Intensity (MFI) data like in cytometry. As far as we know, this represents the first study reporting prognostic results using this multispectral imaging based count technique.
This study was conducted in accordance with the Declaration of Helsinki and was approved by the local ethics committee (CPP Ile de France nr. 2012-05-04). Informed consent was obtained from the participant included in the cohorts.
1. Tissue Material
2. In Situ Immunofluorescence Staining of TILs
3. Fluorescence Analysis and Automated Cell Count
Using the general protocol described above, we aimed to quantify intratumoral CD8+ T cells co-expressing the inhibitory receptors PD-1 and Tim-3 in frozen tissues from patients with RCC, and to correlate the results with clinical outcomes25.
Optimization of the CD8/PD-1/Tim-3 Staining:
Different species of antibodies (mouse, rabbit and goat, see Table of Material) were chosen in order to avoid cross reaction between the different antibodies. The protocol was validated on a tonsil, which is an organized lymphoid tissue. PD-1 staining could be found in the germinal centers (Figure 1), whereas CD8 and Tim-3 staining are present surrounding the germinal centers. The observation of this typical staining validated this protocol. The staining was also validated on the tissue of interest (kidney).
The absence of signal of the primary antibody incubated with their non-corresponding secondary antibody (data not shown) was confirmed. To analyze accurately the fluorescence staining and to make specific spectral libraries, individual slides were singly stained with each marker (CD8, PD-1, Tim-3, or DAPI) in the tissue of interest (kidney). A non-stained slide in the same conditions is necessary to extrapolate the autofluorescence of the tissue. The staining was analyzed by the automated microscope, which integrated the spectrums of the different fluorophores; each marker was well-differentiated and no overlap of markers was observed (Figure 3).
Quantitative and Qualitative Characteristics of the CD8+ Infiltrate in 87 RCC Patients:
Results showed that approximatively half of CD8+ T cells express PD-1 (mean 53.9%; SE: 30.49%) and about one third were double positive PD-1+ Tim-3+ (mean 38.16%; SE: 28.11%). Tim-3 expression without PD-1 expression was detected in less than 3% of CD8+ T cells (data not shown). The mean numbers of CD8+ T cell subpopulations were: total CD8+ T cells 116.5 (SE: 216), PD-1+ CD8+T cells 89.32 (SE: 191.8), PD-1+ Tim-3+ 66.9 (SE: 143), and PD-1+ Tim-3– CD8+ T cells 22.38 (SE: 57.5) per field. The total number of intratumoral CD8+ T cells was positively correlated with the percentage of PD-1+ Tim-3+ on CD8+ T cells.
Functional Characterization of the CD8+ T Cells Depending on Their Phenotype PD-1 and Tim-3:
PD-1 expression level is known to be correlated with T cell exhaustion so we next aimed to measure PD-1 expression on CD8+ T cells according to Tim-3 expression status. As the mean fluorescence intensity of the different fluorochromes is reported at the cellular level, a small series of patients were manually analyzed for the cellular data: results showed a correlation of PD-1 membrane MFI with the cell subtype. PD-1 fluorescent intensity was measured (Figure 6) at the cellular level on PD-1+ Tim-3+ versus PD-1+ Tim-3– CD8+ T cells (A panel) and at the individual patient level after integrating these various cell signals on a tissue section. In a series of 9 patients, a comparison with Wilcoxon paired test found that PD-1 MFI is higher when Tim-3 is co-expressed, reinforcing the functional relevance of Tim-3 expression (B panel).
Expression of PD-1 and Tim-3 Ligands at the Surface of the Tumor Cells from Multiple Co-staining:
As T cell exhaustion is mediated through receptor/ligand interaction, we assessed the expression of PD-1 and Tim-3 ligands (PD-L1 and Galectin-9 (Gal-9), respectively) on renal tumor cells identified by pan-Keratin staining. Positivity is defined as a cut-off of 10% of tumor cells expressing the marker: (i) 58 out of 87 patients (66%) were positive for PD-L1; (ii) all the 15 tumors analyzed were positive for Gal-9 (Figure 7A, B).
Clinical Impact of the Co-expression of Tim-3+ PD-1+ on CD8+ T Cells:
Tumor Node Metastasis (TNM) scoring, Fuhrman grade, tumor size, and UISS score are histological characteristics (combined with clinical score for UISS) used to define the prognosis value of primary RCC. A positive correlation was observed between the number and/or percentage of tumor-infiltrating CD8+ T cells expressing PD-1 and Tim-3 (but not expressing only PD-1 without Tim-3) with all these parameters (Table 1). In line with this more pejorative phenotype, RCC patients with CD8+ T cells co-expressing PD-1 and Tim-3 above the median percentage (34.7) were more likely to relapse (p = 0.046; HR 2.9; 95% CI: 1.02-8.21). This correlation was not observed with the percentage of PD-1 on CD8+ T cells independently of Tim-3 status (Table 2). A correlation was also demonstrated between the percentage of CD8+ T cells co-expressing PD-1 and Tim-3 and the 36-month overall survival rate (data not shown). We also validated the triple immunostaining on tonsil paraffin tissues (Figure 1).
Figure 1: CD8-PD-1-Tim-3 immunostaining on a tonsil paraffin-embedded tissue. Paraffin embedded tissue sections instead of cryopreserved sections were selected for triple immunostaining. After dewaxing and rehydration, the same protocol of staining previously developed for cryopreserved sections was applied. CD8+ T cells are located outside the germinal center and non-CD8+ T cells expressing PD-1 are clustered in the germinal center. The microscope magnification is 20x and the scale bar represents 20 µm. Please click here to view a larger version of this figure.
Figure 2: Automated analysis of renal cell carcinoma tissue specimens. After tissue recognition in bright field image (A, scale bar 1 mm), 4x microscope magnification low resolution imaging in RGB (red blue green conventional fluorescence) is used to select the fields (B, scale bar 500 µm. The full red square represents a field. A high-resolution image (multispectral cube fluorescence image specific to this technology) with at 20x magnification performed by the multispectral microscope is shown (C, scale bar 100 µm). Please click here to view a larger version of this figure.
Figure 3: Multispectral microscopy and software coupled analysis for the generation of a composite image of a renal cell carcinoma (20x microscope magnification). The high resolution raw image (A) is a fluorescent image of the field after multispectral scanning by fluorescence filter cubes. Blue signal is AF488, Red is Cyanine 5, and green is Cyanin3. The merge shows different colors for example, yellow is a merge of the three fluorophores. This image is incorporated in the image analysis software. Spectral library integration allows an accurate spectrum separation of each fluorophore and autofluorescence leading to a fluorescent composite image (B). Representative colors are chosen by the operator: blue illustrates Cyanine 5 (CD8), Red Cyanine 3 (PD-1), Green AF488 (Tim-3). Merging mix color such as purple is used for CD8-PD-1 double positive. The scale bar represents 20 µm. Please click here to view a larger version of this figure.
Figure 4: PD-1 and Tim-3 expression on tumor-infiltrating CD8+ T cells on a clear cell renal cell carcinoma specimen analyzed by fluorescence multispectral imaging technology (20x microscope magnification). (A) CD8 (blue), PD-1 (red), and Tim-3 (green) were stained on frozen tissue sections derived from RCC patients. Colocalization of these three markers can be detected by merging the mono-staining pictures. Isotype controls were performed in each experiment. The scale bar represents 20 µm. Yellow boxes identify co-stained cells: a Tim-3– PD-1+ CD8 cell and a Tim-3+ PD-1+ CD8 cell. (B) Triple co-staining for CD8, PD-1, and Tim-3 (merged) is shown on the left with the box indicating that green is CD8+ T cells co-expressing PD-1 and/or Tim-3. For automated counting with the image analysis software, the identification of the cells relies on the presence of a nuclear staining (DAPI); the cell segmentation is also based on morphometric characteristics represented as red boundaries (middle panel). An algorithm trained by the user until concordance with visual count is performed (right), and after automated phenotyping, identifies green dots as CD8+ T cells co-expressing PD-1 and Tim-3, red dots as CD8+ T cells expressing PD-1 without Tim-3, and blue dots as CD8+ T cells not expressing PD-1 or Tim-3. The scale bar represents 20 µm. Yellow box shows PD-1 and/or Tim-3 expressing CD8+ T cells. Please click here to view a larger version of this figure.
Figure 5: Computing the raw data of multispectral analysis. Data of the count of each cell category present in each field (i.e., 5 fields/patient, i.e., 5 .txt files) are collected after image analysis. An R script combines the 5 fields data, only considering cells phenotyped with a confidence interval > 55%. The microscope magnification is 20x and the scale bar represents 100 µm. Please click here to view a larger version of this figure.
Figure 6: PD-1+ Tim-3+ CD8+ T cells express higher levels of PD-1 compared to PD-1+ Tim-3– CD8+ T cells in renal cell carcinoma. The intensity of PD-1 is detected at the cellular level (left panel) by in situ immunofluorescence analysis on PD-1+ Tim-3+ and PD-1+ Tim-3– CD8+ T cells. At the individual level (middle panel): results are shown for the whole CD8+ T cell subsets PD-1+ Tim-3+ versus PD-1+ Tim-3– present on a tissue section (i.e., one patient; Mann Whitney test). Comparative analysis of the mean PD-1 intensity was measured in a series of 9 patients (right panel) (Wilcoxon test, p-values lower than 0.05 were considered as significant). The microscope magnification is 20x and the scale bar represents 10 µm. Please click here to view a larger version of this figure.
Figure 7: PD-L1 and Gal-9 i.e. PD-1 and Tim-3 ligands staining on renal cell carcinoma. (A) Tumor cells are identified by pan-Keratin staining. Co-staining of PD-L1 and pan-keratin (A panel) identifies the expression of PD-L1 on tumor cells in 58 patients out of 87 analyzed. (B) Galectin-9 was expressed in all of the 15 tumors from RCC patients analyzed. Positivity was considered if at least 10% of cells were stained for the marker analyzed. Isotype controls were included for each staining. Yellow boxes represent positive tumor (pan-Keratin+) cells for PD-L1 (panel A) or Galectin-9 (panel B). The microscope magnification is 20x and the scale bar represents 20 µm. Please click here to view a larger version of this figure.
%/CD8+ T cells | PD-1+ | PD1+ Tim-3+ | PD-1+ Tim-3– |
TNM | 0.04 | 0.003 | 0.22 |
Furhman | 0.01 | 0.004 | 0.74 |
Tumor Size | 0.08 | 0.01 | 0.37 |
UISS | 0.01 | 0.01 | 0.63 |
Table 1: Correlation between the expression of PD-1 alone or combined with Tim-3 on CD8+ T cells and clinical prognostic parameters of RCC patients: p-values. The percentage of PD-1+, PD-1+ Tim-3+ or PD-1+ Tim-3– on CD8+ T cells selected as a continuous variable measured by in situ immunofluorescence in the cohort of 87 patients was correlated with various clinical parameters defined as a binary (TNM, Fuhrman grade, UISS score) or a continuous variable (tumor size). TNM was divided into two groups: localized disease (pT1 and pT2) and advanced disease (pT3, pT4, N+, or M+). Fuhrman grade was defined as low (grade I or II) and high (grade III or IV) and the UISS score was divided into 3 classes (0, 1, and 2). P-values indicating a significant correlation are shown in bold.
CD8+ T cell subset/CD8 | PD-1+ Tim-3+ | PD-1+ Tim-3– |
Median (%) | 34.7 | 8.6 |
Correlation with PFS | P = 0.046 | P = 0.8 |
Table 2: Correlation between PD-1 and Tim-3 co-expression on CD8+ T cells and progression free survival. Two groups of RCC patients (n = 87) depending on the percentage of PD-1 without Tim-3 (right), PD-1 and Tim-3 (Left) relative to the median were individualized. The correlation with the PFS was made with the median selected as a cut-off. P-values indicating a significant correlation are shown in bold.
Supplemental Table 1: Buffer composition of TBS and TBST. TBS is used for antibody mixes and dilutions. For washes, all steps are performed with TBS except for the washes following primary antibodies, where TBST is recommended. Please click here to download this file.
Supplemental Table 2: Detailed antibodies mix composition for CD8-PD-1-Tim-3 immunofluorescence staining. For each primary, secondary, and tertiary antibody mixes, the dilution to perform in the corresponding volume of TBS is given for a total volume of 1 mL. For one slide, the necessary total reaction volume is 100 µL for a tissue area of 2 cm2. Please click here to download this file.
Modifications and Troubleshooting:
The tissue quality is an important parameter; it can be easily checked by hematoxylin and eosin coloration.
One advantage of the technique is the possibility of multiplexed stainings, but to avoid fluorophore spillover, it recommended to choose emission wavelengths with delta of 10 nm minimum. Here, because we had 4 stainings including DAPI, we chose to spread out the wavelengths (DAPI: 460 nm, AF488: 519 nm, Cyanine 3: 570 nm, Cyanine 5: 670 nm). The risk of overlapping signals from various fluorophores is avoided by using the software to compute the pure spectrum of a fluorophore from a mixed emission signal, combined with an automated image analysis. The mono-stained slides also confirm the absence of overlap between fluorophores and act as a compensation thanks to the spectral library.
Sometimes positivity for one marker in one cell can be difficult to determine if the staining/fluorescence is weak. The negative control included in each experiment is a good way to determine a threshold of positivity.
In the coupled software used to analyze images, a scoring step for determining the positivity of the cell exists but it only can analyze two markers simultaneously. Because we can have triple staining on the same cell (CD8, PD-1, and Tim-3), we chose the phenotyping step.
The phenotyping step can be laborious, particularly if the type of staining and background is variable from one sample to another.
Limitations of the Technique:
Even if the number of co-stainings is an advantage compared to conventional fluorescence, it is not possible to analyze many colors as in flow cytometry. Especially when the colocalization of multiple markers is studied, be aware for sensitivity that is lower compared with confocal microscopy (for a 20x microscope magnification, approximately 0.5 µm versus 0.1 µm per pixel for confocal, depending on the brand) and check for fluorophore spillover (refer to Critical Steps Within the Protocol, below).
A large limitation of the technique is the raw data format. The raw data are .txt files which might be tricky to use. As the software gives one .txt file per image with CI for the phenotype, the manual treatment of each of the 5 files per patient of a huge series is very laborious. A bioinformatics specialist is required to compute all the data in a statistical software.
Significance with Respect to Existing Methods:
Multiplexed staining analysis is easily possible. Compared to flow cytometry, which analyzes fresh samples, this technique allows a rapid counting in a reproducible method of the different cell subsets present in the tumor microenvironment, in large cohorts of frozen samples with an automated procedure26,27. We could automatically count infiltrating CD8+ T cells expressing PD-1 and co-expressing Tim-3 or not in the setting of RCC. The automatic count avoids several biases like eye fatigue and non-reproducible count, and is not time consuming.
As the large cohort studied was on frozen sample, we had retrospective data so we could link the absolute number or percentage of the CD8+ T cell subsets with biological parameters and clinical outcome. As there is an important heterogeneity of the immune infiltration from one patient to another28, this represents an advantage compared to fluorocytometry that reports only MFI and percentage but not the infiltration grade.
The fluorescence intensity is quantitatively reported for each cell, and in the different cell compartments (membrane/cytoplasm/nuclei), we could evaluate the level of PD-1 staining by MFI at the membrane. This represents another interesting tool that is similar to flow cytometry. We found that the PD-1+ CD8+ T cells expressing Tim-3 had higher expression of PD-1 and were associated with a poor prognosis (impact on progression free survival). To our knowledge, it is the first study using this method that shows a clinical significance of a specific cell subset.
Future Applications:
The field of application is immense. We also highlighted in lung cancer tumor's microenvironment another specific cell subset: the tissue resident memory T cells called TRM (CD103+ CD49a+), and they are correlated with a better overall survival21,29,30. In another work using this technique, our team found that PD-L1 was overexpressed on tumor cells in a subtype of lung cancer with an anaplastic lymphoma kinase (ALK) rearrangement, and this expression was correlated with CD8+ T infiltrate10. This technique is suitable for the evaluation of critical biological parameters in several contexts and can represent a valuable tool for biomarker discovery. Since the xy coordinates are given, it is one of the first tools to study the topography and cell/cell interactions with ease31.
Critical Steps Within the Protocol:
One of the critical steps is the optimization of the co-staining, which can be time-consuming. The choice of a good antibody is crucial, for instance multiple clones have been tested for Tim-3 in the present case. Furthermore, different fluorophores for the same marker can give different results so it is important to try several combinations. Examining the staining specificity with antibodies that should not react together is recommended. Despite the possibility to analyze several fluorophores at the same time, checking for spillover with the mono-stained slides is recommended. Another crucial step is the phenotyping step that relies on manual training to perform carefully.
In total, the use of this multiplexed technique by an adept operator represents an elegant tool to analyze multiple cell subsets of the local immune infiltrate, which is of importance since the peripheral blood analysis cannot be representative of the local tumor environment.
The authors have nothing to disclose.
This work was supported by grants from Institut National du Cancer (INCA) (ET), Ligue contre le Cancer (ET), Université Sorbonne Paris Cité (ET), ANR (Selectimmunco) (ET), Labex Immuno-Oncology (ET), SIRIC CARPEM (CG, ET). EdG was funded by a fellowship of Fondation ARC. EV and CD were funded by a fellowship of APHP (bourse année recherche). ChB is funded by a fellowship of Université Sorbonne Paris Cité (contrat doctoral). The authors thank Bristol Myers Squibb for their funding in this project. The authors thank the department of Pathology of Hopital Européen Georges Pompidou and Necker (Laurianne Chambolle, Elodie Michel and Gisèle Legall). The authors thank the Histology platform of PARCC, Hopital Européen Georges Pompidou (Corinne Lesaffre).
Vectra 3 Automated Quantitative Pathology Imaging | Perkin Elmer | CLS142338 | |
inForm cell analysis 2.1. | Perkin Elmer | CLS135781 | |
R software | https://www.r-project.org | ||
Dakopen delimiting pen | Dako | S2002 | |
Tris Buffer Salin TBS Tablets | Takara, Bio Inc. | TAKT9141Z | pH7.6 100 tablet |
Tris Buffer Salin Tween 20 TBS(+Tween20) | Takara, Bio Inc. | TAKT9142Z | pH 7.6 100 tablets |
Biotin blocking system | Dako | X0590 | Avidin 0.1% and Biotin 0.01% |
normal donkey serum | Jackson Immunoresearch | 017-000-001 | 5% vol./vol. concentration |
Fluoroshield with DAPI | Sigma-aldrich | F6057 | 1.5 µg/mL concentration |
Knittel glass coverslip | Knittel Gläser, | 100039 | 24×60 mm 100 cover slips |
Rabbit anti-CD8 Clone P17-V | novus | NBP1-79055 | use at 4µg/mL |
Mouse anti-PD-1 Clone NAT | abcam | ab52587 | use at 2 µg/mL |
Goat anti-Tim-3 | R&D | AF2365 | use at 3 µg/mL |
Rabbit anti-PD-L1 Clone SP142 | Roche | 7309457001 | use at 1 µg/mL |
Mouse AF647 labeled pan- Keratin Clone C11 | Cell Signalling | 4528 | use at 0.5 µg/mL |
Goat anti-human gal9 | R&D | AF2045 | use at 0.3 µg/mL |
Cyan 5 conjugated donkey anti-rabbit | Jackson Immunoresearch | 711-175-152 | use at 5 µg/mL |
Biotinylated F(ab’2) donkey anti-mouse IgG | Jackson Immunoresearch | 715-066-150 | use at 3 µg/mL |
Alexa Fluor488 conjugated donkey anti-goat IgG | abcam | ab150133 | use at 5 µg/mL |
Cy3 labeled streptavidin | Amersham | PA43001 | use at 3 µg/mL |
negative control mouse IgG1 | Dako | X0931 | use at 2 µg/mL |
IgG from goat serum | Sigma-aldrich | I5256 | use at 3 µg/mL |
IgG from rabbit serum | Sigma-aldrich | I5006 | use at 4µg/mL |