Overview of multiplex immunohistochemistry/immunofluorescence techniques in the era of cancer immunotherapy

3.3.1 Imaging mass cytometry (IMC)

One of the more commercialized and accessible multiplex tissue imaging techniques is imaging mass cytometry (IMC; Fluidigm, South San Francisco, California, USA; Figure 1B). This method combines high-resolution laser ablation with mass cytometry for the simultaneous evaluation of more than 100 biomarkers (although this is theoretical and based on the availability of additional isotopes which makes the actual value > 40 biomarkers) from tissue sections labelled with metal-tagged antibodies as shown in Figure 1B, with single-cell and spatial resolution (Figure 3) [16]. The number of markers that can be examined is the main benefit of IMC but certain constraints remain. For example, the function of the selected antibodies must be thoroughly validated for each IMC run. In addition, the acquisition speed constrains the area that can be imaged by IMC. To overcome this caveat, immunofluorescence is used to select specific regions of a slide for IMC analysis. However, the number of slides that can be imaged also remains limited - which begs the question if the quantity of biomarkers that can be evaluated by IMC is really of true value. Cell segmentation remains the most challenging process in multiplexed image analysis because cell borders are not always clearly defined and individual pixels may contain information from more than one cell [65]. Therefore, tissue imaging data tends to be noisier than the data obtained using suspension analysis methods.

As with other platforms, there are also constraints associated with the IMC platform itself. IMC has lower sensitivity than traditional IF as it directly measures antibody abundance and lacks signal amplification or options to increase “exposure time”. This is easily achievable with fluorescence-based imaging technologies. Furthermore, due to limited precision of the laser spot, the x-y resolution of IMC is set at 1 μm, which is relatively low compared to other technologies. For example, mIHC/IF, which will be discussed below, offers 0.5 and 0.25 μm resolution [3-14]. This should be sufficient for cell-level analysis since the average cell size is ∼10 μm in diameter. However, at this resolution, it is difficult to perform subcellular analysis. In addition, the acquisition process is time-consuming as seen in Table 2. For a 1000 μm² region of interest (ROI), ablation takes ∼2 h. This slow rate of image acquisition impairs system throughput and may introduce batch effects due to hourly and daily instrument drift. The IMC platform does not provide options to capture colorimetric images, such as hematoxylin and eosin images, and this issue can be only bypassed by examination of consecutive sections before or after IMC. IMC is also disruptive to the tissue as it is ablated and thus, not reusable. However, similar to other commercial multiplex techniques, IMC also offers a quantitative option using two free-access software packages (Table 1), namely, CellProfiler [66, 67] and histoCAT [68]. These packages can perform cell segmentation and quantitative image analysis. Lastly, just like all the hyperplexed mIHC/IF system, there may be the problem of steric hindrance and non-specific binding between antibodies and epitopes. Hence, proper staining controls need to be in place to safeguard the quality of the staining and imaging.

3.3.2 MIBI

A rather new alternative is Multiplexed Ion Beam Imaging (MIBI, Figure 1B) by IONpath (www.ionpath.com) where tissues are stained with up to 40-100 metal-labelled antibodies that are ionized by high-energy beams to generate secondary ions which are then detected by an imaging mass spectrometer over a five-log dynamic range [15, 69]. Similar to DISCOVERY ULTRA, it is compatible with formalin-fixed paraffin-embedded (FFPE) tissues which are the most common type of specimens in clinical settings [70]. However, as shown in Table 1, it can also perform imaging and analyzing of tissue samples with MIBItracker Software (IONpath, Menlo Park, California, USA), something that other platforms do not provide as a bundled analytic tool. Its ability to rescan slides at multiple resolutions helps to make sub-cellular imaging possible with MIBIscope Software (IONpath, Menlo Park, California, USA). Hence, it not only provides information about the degree of immune cells infiltration, spatial information about cell morphology, and localization at light microscopy resolution but also measures high and low protein levels in individual cells like PD-1 and FOXP3 which are often missed. This makes MIBI potentially very useful at identifying heterogeneity in the composition of tumors. MIBI can be quite costly as illustrated in Table 1 [71], and as more markers are added, longer acquisition times may be required. Similar to IMC, steric hindrance and non-specific binding between antibodies and epitopes are still a challenge whereby proper controls are needed for the staining and imaging.

3.3.3 Fluorescence-based mIHC/IF

3.5.1 CO detection by indEXing (CODEX)

CODEX is a multiparametric imaging technology commercialized by Akoya Biosciences, Inc. (www.akoyabio.com). Unlike conventional IF and IHC which are limited to measuring few parameters simultaneously and single-cell technologies like mass cytometry that have limited spatial information, CODEX uses high-throughput technology to detect more than 50 biomarkers simultaneously in a single tissue sample. In addition, it provides information about the relative abundance and expression of these biomarkers at a spatial level [22].

By using antibodies conjugated to barcodes comprised of unique oligonucleotide sequences, the CODEX assay can target specific barcodes using a dye-labelled Reporter (fluorophore) for highly specific detection. Cycles of labelling, imaging and removing reporters are performed by a fully automated fluidics system and images are collected and compiled across all the cycles to achieve single-cell resolution [83]. The platform is compatible with the existing three-colored fluorescence microscope, enabling the conversion of a simple fluorescence microscope into a tool for multiparametric imaging and cytometry [84].

CODEX antibodies have demonstrated comparable staining patterns when compared to other IF and IHC grade dye-conjugated antibodies [84, 85]. The compatibility of the system with various tissue types including spleen, tonsil, lymph node and FFPE melanoma has been demonstrated [85]. It has also successfully achieved deep and previously uncharacterized immune profiling of the mouse splenic architecture by comparing normal murine spleens to those with autoimmune disease [86].

Although a study reported minimal steric hindrance when large panels of CODEX antibodies are used for staining, it was found that there are some cases where the signal might be lower [22]. Hence, there is still a need for the users to perform proper controls for specific antibodies and combinations of interest.

3.5.2 Digital Spatial Profiling (DSP)

DSP is a method of high-plex spatial profiling of proteins and RNA using oligonucleotide detection technologies as shown as Figure 1D. In this method, oligonucleotide barcodes are conjugated to antibodies using a photocleavable UV light-sensitive linker. UV light is then used to decouple the high-plex oligo tags from the antibody and retrieved from the surface of the tissue, enabling the sample to be reused. The oligo barcodes undergo quantitative analysis and are mapped back to tissue location to allow spatial profiling at the defined region of interest. It is also a high-throughput technology that can analyze up to 16-20 sections per day [21].

DSP has been successfully used to characterize the tumor profile of patients with melanoma on checkpoint blockade therapy and has demonstrated a correlation between baseline immune infiltration and treatment response [87, 88]. A validation study [89] also showed a robust detection of high abundance protein and RNA targets. When compared with IHC, the dynamic range of DSP was found to be significantly greater [89]. DSP has also shown high concordance with quantitative immunofluorescence (QIF) and has been validated based on regression and outcome assessment [90].

However, there are some limitations of the DSP platform as shown in Table 1. Profiling every cell at single-cell resolution using DSP may be impractical and cost-prohibitive. Thus, a general advantage of imaging-based methods compared to DSP is the ability to get multiplexed information on each cell in the tissue slice rather than high-plex information of regions of interest as in DSP [91]. In other words, the DSP can only be visualized based on its ‘morphology kit’ of up to 3 markers, including cytokeratin and CD45. The multiplexing of that technique (up to 60 markers) does not provide any visualization. Hence, it only provides a 60-marker snapshot of the selected regions of interest based on the visualization of the ‘morphology kit’, providing limited information on the spatial distribution of cells. Despite this, the protein plus gene expression data harvested from the same regions of interest is attractive to most of the researchers.

3.5.3 Ultivue's InSituPlex

Ultivue's InSituPlex technology is a method that detects a unique DNA barcode conjugated to the primary antibodies. This enables a higher order of multiplexing without damaging the tissue sample. The tissue sample is first incubated with barcoded antibodies. Then, it undergoes a single amplification step that increases the ratio of barcodes per antibody to enable more complementary probe strands to bind. Lastly, complementary fluorescent DNA probes are added which bind to the targets, enabling imaging of the sample.

InSituPlex is particularly promising in the field of immuno-oncology research. A separate study using 15 different tumor markers on an FFPE tissue sample was able to yield high-dimensional images and successfully spatially profile the different sub-populations of immune cells (Table 1) [92]. In another study involving 5 samples from 4 tumor types (breast, lung, colon, melanoma), InSituPlex technology was able to provide analysis on the different phenotypic subtypes of immune cells, proliferative indexes and cell densities [93]. This is all achieved while maintaining the integrity of the tissue samples. Moreover, Insituplex can be adopted easily in laboratories with the usual fluorescent microscope (Figure 4), eliminating the need for any other more costly platforms (Table 1). It is also potentially much more reproducible than other multiplexing techniques available, with a coefficient of variation (CV) of less than 10% as compared to that of other IHC based assays which have a CV of less than 15%, making it suitable for translational applications in the future [94].

3.6.1 HALO

HALO (Indica Labs, Albuquerque, New Mexico, USA, ; https://www.indicalab.com/halo/) is an image analysis platform for quantitative tissue analysis in digital pathology, mainly used for analysis in oncology, toxicology, metabolism, neuroscience, and more [90-94]. It reports the morphology and multiplexed expression data of each and every cell across the entire tissue sections. Its ease of usage comes from maintaining an interactive link between cell data and cell image. Apart from being able to see the analysis outputs for a specific cell, HALO's sorting and filtering tools also allow the user to evaluate millions of other cells in comparison with corresponding cell populations. As such, it can locate cells with the highest intensity of a selected biomarker and can be used in segmenting tumor and stroma among other uses. HALO comes with a myriad of add-ons such as the Tissue Classifier to automate tissue segmentation, FISH-IF Quantification to contextualize protein and gene expression profile of every cell, Spatial Analysis to identify relative spatial distribution of cells, Tissue Microarray Add-on for batch analysis of whole slide TMA images, and Serial Section Analysis to analyze serial section(s) stained for additional IHC markers. These add-ons are also compatible with a broad spectrum of image and digital slide formats, namely JPG, TIF, ND2, MRXS, QPTIFF, component TIFF, VSI, NDPI, NDPIS, SVS, AFI, CZI, SCN, LIF, and BIF. More importantly, HALO itself is compatible and frequently used with Vectra and Insituplex as the recommended imaging platform [87, 88, 95]. While quantitative biomarker analysis platforms like HALO might accurately inform patient selection for clinicians to better predict response to treatment, the potential clinical translation of the software is shown in Table 2; HALO is limited by its cost and has yet to obtain accreditation for In Vitro Diagnostic (IVD) medical device. Up to date, it is mainly used for research purposes and by Contract Research Organisations (CROs) where it is used to meet their pre-clinical high-volume image analysis needs.

3.6.2 Oncotopix

As shown in Table 2, the other commercially available image analysis software for quantitative digital pathology is Oncotopix (Visiopharm, Hoersholm, Denmark; https://www.visiopharm.com/) which is claimed to have been designed for cancer diagnostics but is also used for research and pharmacological purposes [96, 97]. It uses image analysis, machine learning, and artificial intelligence to analyze, manage and report massive data sets, connect to laboratories and their information management system, and compatible with all image/slide formats. Unlike other software in the market, it has been applied in multiple clinical uses as an in vitro diagnostic (IVD) approved device such as in breast and lung cancer detection [98, 80]. For breast cancer, it is essential to assess for estrogen receptor (ER), progesterone receptor (PR), Ki-67 and human epidermal growth factor receptor 2 (HER2) invasive cancers and to exclude non-invasive structures [89, 99-101]. However, the exclusion of non-invasive tumor manually can be challenging and labor-intensive, especially in cancers with a diffuse growth pattern. The Invasive Tumor Detection APP under Oncotopix automatically detects the invasive tumor. The software combines with a full biomarker analysis workflow, to allow automatic hot spot detection, accurate biomarker quantification, and visualization of tumor heterogeneity. This is based on a physical double stain of p63⁺CK7/19⁻- invasive tumor components are identified and are labelled as regions positive for CK, whereas non-invasive tumor components (regions positive for both CK and p63) are excluded. Similarly, for lung cancer, the tissue is first categorized into tumor and stroma using cytokeratin staining. It then undergoes tissue segmentation where individual cells are identified by the presence of DNA (DAPI) and membrane proteins. The phenotyping algorithm then scans multiple images, automatically identifies cellular phenotypes, and categorizes them accordingly. Using Phenotypic Matrix, Phenotypic Profile and/or tSNE plots, the phenotypes can be summed up and further visualized for a better understanding of the data. This APP is made to work with amplified fluorescence signals (unmixed with multispectral imaging) where each stain can be spectrally isolated from background autofluorescence and spectrally overlapping fluorophores.

3.6.3 HistoCAT

HistoCAT (Fluidigm, South San Francisco, California, USA) is an interactive free-access computational platform that makes quantitative analysis of highly multiplexed, single-cell-resolved tissues [95, 96]. HistoCAT combines high-dimensional image visualization, analysis methods for cell phenotype characterization, and novel algorithms for the comprehensive study of cell-to-cell interactions and the social networks of cells within complex tissues [95].

HistoCAT operates by first extracting data of individual cells from images, including the quantity of all measured markers for a cell and within the area of interest, morphological features like cell size and shape, and features of the cell's environment such as cell neighbors and crowding. This data is synthesized and represented in a flow cytometry standard format (.fcs) file for further analysis using the HistoCAT or other compatible analytic software. HistoCAT promotes the idea of “Round-trip” analyses, by comparing the single-cell phenotypes and their interactions of a specific area of an image to the entire dataset, and back to the visualization of unique cells on images which would help users define and understand important cell populations and their spatial context within tissues. HistoCAT also possess a novel algorithm which can detect proximate cell-to-cell interactions that are more frequently detected than expected by chance and identify the more significant interactions and unique cell environments across entire datasets and cohorts. As demonstrated in Table 1 and 2, HistoCAT is also one of the recommended image analysis software for images processed through imaging mass cytometry, as mentioned above [97, 98], where more than 40 unique metal-isotope-labeled antibodies can be detected simultaneously at a resolution comparable to fluorescence microscopy.

The software is, however, limited by its ability to perform tissue segmentation, where it is coupled with other cell segmentation software such as the CellProfiler (https://Cellprofiler.org). Besides, a small percentage of cells will overlap in 5-μm thick tissue sections, leading to only a portion of a cell being present within the analyzed section. Small differences in segmentation between adjacent cells can thus result in a small spillover of signal between neighboring cells. HistoCAT addresses this by allowing for cell boundaries to be defined using a combination of different markers to improve the accuracy of segmentation masks [95]. Overall, HistoCAT is still in the phase of constant upgrade and maintenance with the latest version uploaded in August 2019 by Bodenmiller Labs.

3.6.4 QuPath

QuPath (https://qupath.github.io) is another high-throughput, free access image analysis software for whole slide image analysis with robust batch-processing and scripting functionalities [99]. It is capable of handling large whole slide imaging data without needing to crop or down-sample the image data into a manageable size for subsequent analysis. It also serves as a platform for researchers to develop and share novel algorithms and workflows to analyze multiplexed tissue images as shown in Table 2. A common workflow of analysis in QuPath for a tissue microarray (TMA) sample starts with creating a multi-slide project with automated TMA dearraying and stain estimation. This is followed by the single-cell analysis which includes cell detection, feature computation and trainable cell classification. The data is then integrated and further analyzed through batch processing and survival analysis. QuPath also offers supplementary functionalities such as support for whole face tissue sections and fluorescence image analysis, data transfer with existing software (e.g. ImageJ and MATLAB), scriptable data mining, and rapid processing and export of spatial, morphological and intensity-based data. There is also good interobserver reproducibility of bioimage data using QuPath in the TMA setting [100].

The technology is, however, limited by its ability to perform digital scoring of biomarkers with more complex patterns of staining, such as mismatch repair proteins or immune checkpoint inhibitors like PD-L1. This arises from the nature of staining multiple cell populations (tumor and immune) and the distinction between tumor and immune cell staining being difficult due to a patchy pattern of staining [100]. Several studies have illustrated the potential of QuPath in TMA biomarker scoring [99, 101, 102], and it is also currently actively utilized by researchers to analyze mIHC/IF samples in the study of cancer immunotherapy [103]. As a whole, QuPath is still a work in progress, has only become publicly available in 2016 and is now being constantly updated with new features by its owner, P. Bankhead.

3.6.5 3-dimension (3D) Imaging

An emerging alternative to the 2-dimension (2D) imaging techniques mentioned previously would be 3D imaging which allows for a more detailed reconstruction of the molecular properties in a 3D tissue sample. Intravital microscopy was the earliest tool used in 3D imaging. It is a technique that enables visualization of the cellular processes in real-time and has been used extensively to study dynamic processes in live animals [104]. However, the imaging volume is limited, and a significant portion of imaged tissues cannot be analyzed. This was followed by the development of ‘Histo-cytometry’ which involves staining a sample with up to 14 antibodies at one time and performing quantitative analysis whereby each cell has defined parameters which can be visualized on the tissue sample. ‘Histo-cytometry’ demonstrated the ability to provide cellular subset discrimination of CD11c⁺ dendritic cells, CD8, CD4 and B cells that was comparable to the findings from flow cytometric analysis of the dissociated cells[105]. Following this, new tools such as Clearing-enhanced 3D (Ce3D) imaging have emerged. Ce3D imaging has superior scanning depth and allows for adequate visualization of large tissue volumes, while preserving the capacity for multiplex antibody staining. Furthermore, image analysis can be done using advanced platforms such as volumetric histocytometry to quantitatively characterize the cells with respect to their spatial positioning, phenotypic and functional markers [106]. Ce3D can also be used to examine tissue samples in any plane of interest. This is particularly important as it can reduce sampling errors in samples that have an uneven and asymmetrical distribution of cells, such as in a tumor with immune cell infiltrate [107]. Ce3D has also been used successfully in the visualization of cellular anatomical relationships such as in dense dendritic cell aggregates in the thymic medullary regions [106]. However, Ce3D has some limitations. There may be problems with antibody penetration in thick tissues. For example, in brain tissues that are rich in lipids, longer incubation times of up to 2 weeks are required to penetrate 1 mm of the tissue [107]. While 3D imaging is not as widely accessible to most labs and regions of the world as other 2D imaging techniques, it still remains an exciting area of ongoing development with great potential for application in the era of cancer immunotherapy.