A simple open-source method for highly multiplexed imaging of single cells in tissues and tumours
=================================================================================================

* Jia-Ren Lin
* Benjamin Izar
* Shaolin Mei
* Shu Wang
* Parin Shah
* Peter K Sorger

## ABSTRACT

Intratumoural heterogeneity strongly influences the development and progression of cancer as well as responsiveness and resistance to therapy. To improve our ability to measure and analyze such heterogeneity we have developed an open source method for fluorescence imaging of up to 60 protein antigens at subcellular resolution using formalin-fixed, paraffin-embedded (FFPE) tissue samples mounted on glass slides, the most widely used specimens for the diagnosis of cancer and other diseases. As described here, tissue-based cyclic immunofluorescence (t-CyCIF) creates high-dimensional imaging data through successive acquisition of four-color images and requires no specialized instruments or reagents. We apply t-CyCIF to 14 cancer and healthy tissue types and quantify the extent of cell to cell variability in signal transduction cascades, tumor antigens and stromal markers. By imaging immune cell lineage markers we enumerate classes of tumour-infiltrating lymphocytes (TILs) and their spatial relationships to the tumor microenvironment (TME). The simplicity and adaptability of t-CyCIF makes it a powerful method for pre-clinical and clinical research and a natural complement to single-cell genomics.

## INTRODUCTION

Advances in DNA and RNA profiling have dramatically improved our understanding of oncogenesis and propelled the development of targeted anti-cancer drugs1. Sequence data are particularly useful when an oncogenic driver is both a drug target and a biomarker of drug response, *BRAFV600E* in melanoma2 or *BCR-ABL*3 in chronic myelogenous leukemia, for example. However, in the case of drugs that act through cell non-autonomous mechanisms, such as immune checkpoint inhibitors (ICIs), tumour-drug interactions must be studied in the context of a multi-cellular environment that includes both cancer and non-malignant stromal and infiltrating immune cells. Multiple studies have established that these aspects of the tumor microenvironment strongly influence the initiation, progression and metastasis of cancer4 and the magnitude of responsiveness or resistance to therapy5.

Single cell transcript profiling provides a means to dissect the tumour ecosystems and quantify cell types and states6. However, single-cell sequencing usually requires disaggregation of tissues, resulting in loss of spatial context6,7. Tissue imaging preserves this context but as currently performed, has relatively low dimensionality, particularly in the case of slides carrying formalin-fixed, paraffin-embedded (FFPE) tissue slices. These are among the most commonly acquired clinical samples and are routinely used to diagnose disease following staining with Haemotoxylin and Eosin (H&E). The potential for immunohistochemistry (IHC) of such samples to aid in diagnosis and prioritization of therapy is well established8 but IHC is primarily a single channel method: imaging multiple antigens typically involves sequential tissue slices or harsh stripping protocols (although limited multiplexing is possible using IHC and bright-field imaging9). Antibody detection by formation of a brown diaminobenzidine (DAB) or similar precipitate is also less quantitative than fluorescence10. The limitations of IHC are particularly acute in immuno-oncology11 in which it is necessary to quantify multiple immune cell types (regulatory and cytotoxic T cells for example) in parallel with expression of multiple tumor antigens and immune checkpoints, such as PD-1/PD-L1 and CTLA-4.

A variety of multiplexed approaches to analyzing tissues that do not involve conventional microscopy have been developed with the goal of simultaneously assaying cell identity, state, and morphology12–16. For example, FISSEQ17 enables genome-scale RNA profiling of tissues at single-cell resolution and multiplexed ion beam imaging (MIBI) and imaging mass cytometry achieve a high degree of multiplexing and excellent signal to noise ratios using metals as labels and mass spectrometry as a detection modal ity12,18. Despite the potential of these new methods, they require specialized instrumentation and reagents, one reason that the great majority of translational and clinical studies still rely on H&E or single-channel IHC staining. There remains a substantial need for highly multiplexed methods that (i) minimize the requirement for specialized instruments and costly, proprietary reagents, (ii) work with conventionally prepared FFPE tissue samples (iii) enable imaging of 20-60 antigens at subcellular resolution across a wide range of cell and tumour types (iv) collect data with sufficient throughput that large specimens (several square centimeters) can be imaged and analyzed (v) generate high resolution data typical of optical microscopy and (vi) allow investigators to customize the antibody mix to specific questions or tissue types. Among these requirements the last is particularly critical: at the current, early stage of development of high dimensional histology, it is essential that pathologists be able to test the widest possible range of antibodies and antigens in search of those with the greatest diagnostic and predictive value.

This paper describes an open-source method for highly multiplexed fluorescence imaging of tissues, tissue-based cyclic immunofluorescence (t-CyCIF), that significantly extends a method we previously described for tissue culture cells19. t-CyCIF assembles images of FFPE tissue slices stained with up to 60 different fluorescent antibodies via successive rounds of 4-channel imaging. t-CyCIF uses widely available reagents, conventional slide scanners, automated slide processors and freely available protocols to create a method that is easy to implement in any research or clinical laboratory. We believe that high dimensional imaging methods by t-CyCIF will become a powerful complement to single cell genomics, enabling routine analysis of the phenotypic geography of cancer at single-cell resolution.

## RESULTS

### t-CyCIF enables multiplexed imaging of FFPE tissue and tumor specimens at sub-cellular resolution

In t-CyCIF, multiplexing is achieved using an iterative process (a cycle) performed on a slice cut from a block of FFPE tissue. We have developed t-CyCIF staining conditions for antibodies targeting ~140 different proteins including immune lineage makers, signaling proteins and phosphorylated kinases and transcription factors (some of which are drug targets), markers of cell state including cell cycle stage, quiescence and apoptosis etc. (Table 1). In the implementation described here, each cycle involves four sequential steps (Figure 1A): (i) staining with fluorophore-conjugated antibodies against different protein antigens; we currently use antibodies conjugated to Alexa 488, 555 and 647 (ii) staining with Hoechst 33342 to mark nuclei (iii) four-channel imaging at low and high magnification (10X and 40X objectives) (iv) fluorophore oxidation using hydrogen peroxide, high pH and UV light followed by a wash step. To reduce the level of auto-fluorescence and minimize non-specific staining, we perform a pre-staining cycle prior to any incubation with primary antibodies (Figure S1A); pre-staining involves incubation with secondary antibodies alone followed by fluorophore oxidation (Figure 1B, Figure S1B-D). The current protocol has been optimized for samples prepared in the standard manner for pathologic diagnosis of cancer (4-5 μm thick FFPE slices mounted on a glass slide). We find that incubation in oxidation solution for 15 min is often adequate for Alexa 555 and 647 conjugated antibodies but reduction of Alexa 488 fluorescence to background levels typically requires 60 min (Figure 1C-E, Figure S1E-G); we routinely perform 60-minute oxidation reactions. 

![Figure 1.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F1.medium.gif)

[Figure 1.](http://biorxiv.org/content/early/2017/06/19/151738/F1)

Figure 1. Steps in the t-CyCIF process and their properties
**(A)** Schematic of the cyclic process whereby t-CyCIF images are assembled via multiple rounds of four-color imaging. **(B)** Image of human tonsil prior to pre-staining and then over the course of three rounds of t-CyCIF. Arrows shows the position of an object that fluoresces in the green channel (use for Alexa-488 imaging) and becomes progressively less intense with cycle number; the phenomenon of decreasing background signal and increasing signal-to-noise ratio as cycle number increases is common. **(C)** Images of tonsil tissue stained with Alexa 488 conjugated to anti-PCNA antibodies and then subjected to fluorophore inactivation conditions (a high pH hydrogen peroxide solution and light) for 0-60 min. **(D)** Distribution of pixel intensities in the sample in panel C. **(E)** Effect of bleaching duration (0-60 minutes) and fluorophore inactivation for tonsil tissue stained with antibodies conjugated to Alexa 488, 570 or 647. **(F)** Impact of cycle number on antigenicity as evaluated using tonsil tissue stained with Alexa 488 conjugated anti-PCNA antibodies. Because primary antibodies remain bound to antigen after fluorophore inactivation, it is not possible to assay the same field repeatedly with the same antibody; thus, different fields are show for each cycle. (**G**) Distribution of single-cell intensities from images in panel (F). Each distribution represents ~2 x 104 cells. **(H)** Impact of t-CyCIF cycle number on tissue integrity as measured by the fraction of cells detected in the Hoechst channel in successive cycles; error bars represent the standard error of the mean (S.E.M) for different fields.

View this table:
[Table 1.](http://biorxiv.org/content/early/2017/06/19/151738/T1)

Table 1. 
List of antibodies tested and validated for t-CyCIF.

Virtually all tissue samples we have examined can be successfully subjected to 8-20 t-CyCIF cycles, yielding data on the spatial distributions of 24-60 different antigens plus nuclear morphology. The primary requirement appears to be good cellularity: samples in which cells are very sparse tend to be too fragile for repeated imaging, We achieve subcellular resolution using a fluorescence slide scanner (in our studies a RareCyte CyteFinder with 10X 0.3 NA objective and field of view of 1.6 x1.4 mm and a 40X 0.6 NA objective and field of view of 0.42 x 0.35 mm). Immunogenicity does not fall appreciably with cycle number (Figure 1F-G) and the signal-to-noise ratio can actually increase due to a reduction in auto-fluorescence as cycle number increases (Figure S1H-J)20. The primary limitation in the number of cycles is tissue integrity: some tissues are physically more robust and can undergo more staining and washing procedures than others (Figure 1H). To date the highest cycle number has been obtained with normal tonsil and skin and with glioblastoma multiforme (GBM), pancreatic cancer and melanoma (Figure S2 and S3). Cell morphology is preserved through multiple cycles: for example, in a t-CyCIF image of tonsil tissue (Figure 2A, Table S1), we can distinguish membrane staining of anti-CD3 and CD8 in Cycle 2, and staining of the nuclear lamina, and nuclear exclusion of NF-κB in Cycle 6 (Figure 2B). 

![Figure 2.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F2.medium.gif)

[Figure 2.](http://biorxiv.org/content/early/2017/06/19/151738/F2)

Figure 2. Ten-cycle t-CyCIF of human tonsil tissue.
**(A)** A selected image across ten cycles stained with antibodies against the antigens indicated; see Table S1 for a list of antibodies used. **(B)** Images of selected channels and cycles emphasizing sub-cellular features.

Multiplexed immunofluorescence enables high resolution imaging of large samples. Figure 3 shows a ~2 x 1.5 cm pancreatic ductal adenocarcinoma (PDAC) subjected to 8 rounds of t-CyCIF (Figure 3A, Table S2). The image comprises ~140 10X fields stitched together to reconstruct the full specimen. Differences in subcellular distribution are evident for many markers, but for simplicity, in this paper we only analyze intensity values integrated over a whole cell. Images were segmented using a conventional watershed algorithm and total fluorescent signal was calculated for each cell and antibody stain (Figure 3B-C; see Online Methods for a discussion of caveats). This yielded ~1.5 x 105 single cells each with 25 intensity values. When we analyzed the levels of pERKT202/Y204 (henceforth pERK, the phosphorylated, active form of the kinase) on a cell by cell basis we found that they were highly correlated with the levels of an activating phosphorylation of the downstream kinase pS6S235/S236 (r = 0.81). Similarly, β-catenin levels (a measure of canonical WNT pathway signaling) were highly correlated with E-cadherin and Keratin levels, whereas Vimentin and VEGFR2 receptor levels were anti-correlated (Figure 3D), recapitulating the known dichotomy between epithelial and mesenchymal cell states in normal and diseased tissues 

![Figure 3.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F3.medium.gif)

[Figure 3.](http://biorxiv.org/content/early/2017/06/19/151738/F3)

Figure 3. Eight-cycle t-CyCIF images of human tumor samples.
**(A)** t-CyCIF image of pancreatic adenocarcinoma. On the left, the entire tumour, comprising 143 stitched 10X fields of view is shown. On the right, a representative field is shown through all 8 t-CyCIF rounds. **(B)** Representative field at high magnification showing the spatial distribution of PCNA, β-catenin, Ki67 and pERK-positive cells. **(C)** Representation of t-CyCIF image in panel (B) following image segmentation with each dot denoting the centroid of a single cell and the color representing the intensity of a particular antibody stain. **(D)** Quantitative single-cell signal intensities of 24 proteins (rows) measured in ~4 x 103 cells (columns) from panel (C). The degree of correlation of each measured protein with E-cadherin (at a single-cell level) is shown numerically; proteins highlighted in red are further analyzed in panel (E). **(E)** t-stochastic neighbor embedding (t-SNE) of cells analyzed in panel (D) with intensity measurements for selected proteins. Circled regions represent t-SNE domains in which the relationship between pERK levels (a measure of MAPK signaling) and β-catenin levels (a measure WNT signaling) are, on average, negatively (a) or positively (b) correlated or uncorrelated (c). **(F)** t-CyCIF of a clear cell renal cancer subjected to 12-cycle t-CyCIF. Regions high in α-smooth muscle actin (α-SMA) correspond to stromal components of the tumors, those low in α-SMA represent regions enriched for malignant cells. **(G)** The same tumor region as in (F) following image segmentation. Each dot denotes the centroid of a single cell. Cells staining positively for α-SMA or the T cell marker CD8 are highlighted in red or green, respectively, and other cells are represented only by nuclear staining (blue). **(H)** Fraction of cells (relative to all cells) with positive staining for immune markers depending on the tumor region (α-SMA high vs. α-SMA low).

The WNT pathway is frequently activated in PDAC and is important for tumourigenesis of multiple gastrointestinal tumours21. Approximately 90% of sporadic PDACs also harbor driver mutations in KRAS, activating the MAPK pathway and promoting tumourigenesis22. Studies comparing these pathways have come to different conclusions with respect to their relationship: some studies show concordant activation of MAPK and WNT signaling and others argue for exclusive activation of one pathway or the other 23. Using t-Stochastic Neighbor Embedding (t-SNE), which clusters cells in 2D based on their proximity in the 25-dimensional space of image intensity data, we identified multiple sub-populations within the same tumor sample representing negative, positive or no correlation between pERK and β-catenin levels (marked with labels “a”, ‘b” or “c”, respectively in Figure 3E). In PDACs malignant cells can be distinguished from stromal cells, to a first approximation, by high proliferative index, which we assessed by we measured by staining for Ki-67 and PCNA24. When we gated for cells that were both Ki67high and PCNA high cells and thus likely to be malignant, we again failed to find a fixed relationship between pERK and β-catenin levels. While we cannot exclude the possibility of phospho-epitope loss during sample preparation, it appears that the full range of possible relationships between the MAPK and WNT signaling pathways described in the literature can be found within a tumor from a single patient. This illustrates the impact of tumor heterogeneity on the activities of key signal transduction pathways.

### Multiplex imaging of tumour architecture and immune infiltrates

Immuno-oncology drugs, including ICIs that target CTLA-4 and PD-1/PD-L1 are rapidly changing the therapeutic possibilities for traditionally difficult-to-treat cancers, such melanoma, renal and lung cancers, but responses are still highly variable across and within cancer types. Expression of PD-L1 correlates with responsiveness to ICIs such as pembrolizumab and nivolumab25 but the negative predictive value of PD-L1 alone is insufficient to stratify patient populations26. Imaging for multiple markers such as PD-1, PD-L1, CD4 and CD8 using IHC on sequentially cut tumour slices appears to represent a superior approach to identifying ICI-responsive metastatic melanomas5. As a first step in developing multiplexed CyCIF immune biomarkers we developed staining conditions for antibodies against CD45, CD3, CD8, CD45RO, FOXP3, PD-1 and PD-L1 (Table 1). In a typical FFPE section from clear-cell renal cell carcinoma (Figure 3F) we could distinguish a domain rich in non-malignant stroma, which stained strongly for the alpha isoform of smooth muscle actin (α-SMA) and one enriched in tumour cells and low in α-SMA. To examine the spatial distribution of cytotoxic CD8+ T cells within this tumor we stained for several immune markers, including CD8 (Figure 3G), CD3, PD-1 and PD-L1. CD3+ CD8+ TILs were 4-fold enriched in the tumour-domain (Figure 3H). The subset of CD3+ CD8+ PD-1+ cells that represents a population of putatively exhausted T cells was 18-fold enriched in the tumor-domain (Figure 3H). Moreover, PD-1 and PD-L1 positive cells were 13 to 20-fold more prevalent in the tumour-rich domain (yellow bars, α-SMA low domain) as compared to the tumor stroma (blue bars, α-SMA high domain). Together, these data suggest that PD-1/PD-L1 interactions occur predominantly within tumour rich domains of kidney cancer and show the potential of t-CyCIF to quantify key features of TILs in combination with their positions in a tumour. Showing that such features constitute true biomarkers will, of course, required additional validation with clinical cohorts.

### Analysis of diverse tumour types and grades using CyCIF of tissue-microarrays (TMA)

Tissue microarrays (TMAs) provide a means to analyze a large number of tissues and tumour samples simultaneously, and are frequently used to study patient biopsies collected in clinical trials. We applied 8 cycle t-CyCIF to TMAs containing 39 individual biopsies from 13 healthy tissues as well as low and high-grade tumours for 13 type of cancer (Figure 4A, Figure S4, Table S2 for antibodies used, Table S3 for TMA details and naming conventions) and then performed t-SNE on single cell intensity data (Figure 4B). The great majority of TMA samples mapped to one or a few discrete locations in the t-SNE projection (compare normal kidney tissue -KI1, low grade tumours -KI2, and high grade tumours – KI3; Figure 4C), while other tumours, such as ovarian cancer, showed a scattered pattern in the t-SNE projection (Figure 4D). Overall, there was no separation between normal tissue and tumours regardless of grade (Figure 4E). In a number of cases, high grade cancers from multiple different tissues of origin co-clustered, implying that transformed morphologies and cell states were closely related. For example, while healthy and low grade pancreatic and stomach cancer occupied distinct t-SNE domains, high grade pancreatic and stomach cancers were intermingled and could not be readily distinguished (Figure 4F), recapitulating the known difficulty in distinguishing high grade gastrointestinal tumours of diverse origin by histophathology.27 Nonetheless, t-CyCIF might represent a means to identify discriminating biomarkers by efficiently sorting through large numbers of alternative antigens, particularly those informed by known genetic features of each disease. Overall we conclude that t-CyCIF can be used on a wide range of normal tissues and tumor types to quantify inter- and intra-tumour heterogeneity. 

![Figure 4.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F4.medium.gif)

[Figure 4.](http://biorxiv.org/content/early/2017/06/19/151738/F4)

Figure 4. Eight-cycle t-CyCIF of a tissue microarray (TMA) including 13 normal and tumor tissue types.
The TMA carried 13 normal tissue types, and corresponding high and low grade tumors, for a total of 39 specimens (Table S2-3). **(A)** Selected images of different tissues illustrating the quality of t-CyCIF images (additional examples shown in Figure S3. A full gallery of staining for all samples from this TMA is available online ([https://omero.hms.harvard.edu/webclient/?show=dataset-2037](http://https://omero.hms.harvard.edu/webclient/?show=dataset-2037)). **(B)** t-SNE plot of single-cell intensity data derived from all 39 samples; data were analyzed using the CYT package (see materials and methods). Tissues of origin and corresponding malignant lesions were labeled as follows: BL, bladder cancer; BR, breast cancer CO, Colorectal adenocarcinoma, KI, clear cell renal cancer, LI, hepatocellular carcinoma, LU, lung adenocarcinoma, LY, lymphoma, OV, high-grade serous adenocarcinoma of the ovary, PA, pancreatic ductal adenocarcinoma, PR, prostate adenocarcinoma, UT, uterine cancer, SK, skin cancer (melanoma), ST, stomach (gastric) cancer. Numbers refer to sample type; “1” to normal tissue, “2” to -grade tumors and “3” to high grade tumors. **(C)** Detail from panel B of normal kidney tissue (KI1) a low-grade tumor (KI2) and a high-grade tumor (KI3) **(D)** Detail from panel B of normal ovary (OV1) low-grade tumor (OV2) and high-grade tumor (OV3). **(E)** t-SNE plot from Panel B coded to show the distributions of all normal, low grade and high grade tumors. **(F)** tSNE clustering of normal pancreas (PA1) and pancreatic cancers (low grade, PA2, and high grade, PA3) and normal stomach (ST1) and gastric cancers (ST2 and ST3, respectively) showing intermingling of high grade tumors.

### Quantitative analysis reveals global and regional heterogeneity and multiple histologic subtypes within the same tumour in glioblastoma multiforme (GBM)

Data from single-cell genomics has revealed the extent of intra-tumour heterogeneity28 but our understanding of this phenomenon would benefit greatly from spatially-resolved data12. To study this using t-CyCIF we performed 8 cycle imaging on glioblastoma multiforme, a highly aggressive and genetically heterogeneous29 brain cancer that is classified into four histologic subtypes30. We imaged a 2.5 cm x 1.8 mm resected tumor sample for markers of neural development, cell cycle state and signal transduction state (Figure 5A-B, Table S5). Phenotypic heterogeneity at the level of single tumor cells was assessed at three spatial scales corresponding to: (i) 1.6 x 1.4 mm fields of view (252 total) each of which comprised 103 to 104 cells (ii) seven macroscopic regions of ~104 to 105 cells each, corresponding roughly to tumour lobes and (iii) the whole tumour comprising ~106 cells. To quantify local heterogeneity we computed the informational entropy on a-per-channel basis for 103 randomly selected cells in each field (Figure 5C, see online Methods for details). In this setting, informational entropy is a measure of cell-to-cell heterogeneity on a mesoscale corresponding to 10-30 cell diameters. For a marker such as EGFR, which is a driving oncogene in GBM, informational entropy was high in some areas (Figure 5C; red dots) and low in others (blue dots). Areas with high entropy in EGFR abundance did not co-correlate with areas that were most variable with respect to a downstream signaling protein such as pERK. Thus, the extent of local heterogeneity varied with the region of the tumor and the marker being assayed. 

![Figure 5.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F5.medium.gif)

[Figure 5.](http://biorxiv.org/content/early/2017/06/19/151738/F5)

Figure 5. Molecular heterogeneity in a single GBM tumor.
**(A)** Representative low magnification image of a GBM specimen generated from 221 stitched 10X frames; the sample was subjected to 10 rounds of t-CyCIF using antibodies listed in Table S4 **(B)** Magnification of frame 152 (whose position is marked with a white box in panel A) showing staining of pERK, pRB and EGFR; lower panel shows a further 4-fold increase in magnification to allow single cells to be identified. **(C)** Normalized Shannon entropy of each of 221 fields of view to determine the extent of variability in signal intensity for 1000 cells randomly selected from that field for each of the antibodies shown. The size of the circles denotes the number of cells in the field and the color represents the value of the normalized Shannon entropy (data are shown only for those fields having more than 1,000 cells; see Online Methods for details).

Unsupervised clustering using expectation–maximization Gaussian mixture (EMGM) modeling on all cells in the tumour yielded eight distinct clusters, four of which in aggregate encompassed 85% of the cells (Figure 6A). Among these, cluster one had high EGFR levels, cluster two had high NGFR and Ki67 levels and cluster six had high levels of vimentin; cluster five was characterized by high keratin and pERK levels. The presence of four highly populated t-CyCIF clusters is consistent with data from single-cell RNA-sequencing of ~400 cells from five GBMs7. Three of the t-CyCIF clusters have properties reminiscent of classical (cluster 1), pro-neural (cluster 2) and mesenchymal (cluster 6) histological subtypes, but additional work will be required to confirm such assignments. To study the relationship between phenotypic diversity and tumor architecture, we mapped each cell to an EMGM cluster (denoted by color). Extensive intermixing was observed at the level of fields of view and overall tumor domains (Figure 6B). For example, field of view 147 was highly enriched for cells corresponding to cluster 5 (yellow), but a higher-magnification view revealed extensive intermixing of four other cluster types on a scale of ~3-5 cell diameters (Figure 6C). At the level of larger, macroscopic regions, the fraction of cells from each cluster also varied dramatically (Figure 6D, Figure S5). These findings have several implications. First, they suggest that GBM is a phenotypically heterogeneous on a spatial scale of 5-1000 cell diameters and that cells corresponding to distinct t-CyCIF clusters are often found in the vicinity of each other. Second, sampling a small region of a large tumour has the potential to substantially misrepresent the proportion and distribution of tumour subtypes, with implications for prognosis and therapy. Similar concepts likely apply to other tumor types with high genetic heterogeneity, such as metastatic melanoma, as recently indicated by single-cell genomic analyses6, and are therefore relevant to diagnostic and therapeutic challenges arising from tumor heterogeneity. 

![Figure 6.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F6.medium.gif)

[Figure 6.](http://biorxiv.org/content/early/2017/06/19/151738/F6)

Figure 6. Spatial distribution of molecular phenotypes in a single GBM.
**(A)** Intensity values from the tumor in Figure 5 were clustered using expected-maximization with Gaussian mixtures (EMGM), yielding eight clusters, of which four clusters accounted for the majority of cells. The number of cells in each cluster is shown as a percentage of all cells in the tumor. **(B)** EMGM clusters (in color code) mapped back to singles cells and their positions in the tumor. The coordinate system is the same as in Figure 5, Panel A. The positions of seven macroscopic regions (R1-R7) representing distinct lobes of the tumour are also shown. **(C)** Magnified view of Frame 147 from region R5 with EMGM cluster assignment for each cell in the frame shown as a dot. **(D)** The proportional representation of EMGM clusters in each tumor region as defined in panel B.

## DISCUSSION

The spatial heterogeneity of solid tumours poses a scientific, diagnostic and therapeutic challenge that is not sufficiently addressed using current methods. We have developed a simple, public-domain approach for quantitative assessment of 20-60 protein antigens in ~5-10μm thick FFPE tissue slices, which represent the norm for diagnosis of human disease and study of mouse models. We describe several applications of t-CyCIF in studying oncogenic signaling, tumour heterogeneity and immune cell-tumour interaction, none of which requires specialized equipment (beyond a slide scanner) or proprietary reagents. t-CyCIF is not as technically sophisticated as FISSEQ17, MIBI18 or tissue-based mass cytometrfy12, but we regard simplicity as a primary virtue: we have taught t-CyCIF to several other research groups and are confident that it can be readily adopted by many clinical and translational research laboratories. Cyclic immunofluorescence also appears to be substantially higher in throughput than non-optical methods, particularly when multiple slides are processed in parallel, and considerable opportunity exists for further improvement, for example, by switching from four to six channel imaging per cycle. Good linearity and resolution (~ 400 nm laterally in the current work, but potentially better with higher NA optics or super-resolution microscopes) are additional advantages of direct fluorescence imaging as compared to methods that rely on enzymatic amplification, laser ablation or mechanical picking of tissues. However, some signals, particularly those associated with phospho-epitopes can be very dim and amplification or use of very bright fluorophores such as quantum dots will be required to image them. We therefore continue to optimize t-CyCIF and will make updates available via our website.

In an initial test of t-CyCIF we quantified the relationship between WNT and MAPK-signaling in PDAC. Prior studies performed on tumours or on populations of cells under different conditions have reported conflicting results as to whether or not WNT signaling positively regulates MAPK signaling31. Our analysis of PDAC suggests that the activities of the WNT and MAPK signaling cascades can by uncorrelated, positively correlated or negatively correlated within different regions of a single tumor. Thus, what appears to be a set of conflicting findings most likely represents heterogeneity arising from differences in microenvironment, genotype or both. Such data adds new insight into our understanding of disease mechanism but variability may complicated the use of MEK-inhibitors in PDAC.32

In a second test of t-CyCIF we studied within-tumor heterogeneity in GBM, a cancer with multiple histological subtypes whose differing properties impact prognosis and therapy.30,33 Clustering antigen abundance data from t-CyCIF images also reveals multiple phenotypic classes within a single tumor. We have not yet established the link between t-CyCIF clusters and known histological subtypes but our results show that cells with very different characteristics are intermingled at multiple spatial scales with no evidence of recurrent patterns of heterogeneity. In regions of the GBM we have studied in detail, heterogeneity on a scale of 10-100 cell diameters is as great as it is between distinct lobes. The proportion of cells from different clusters also varies dramatically from one lobe to the next. Variation on this spatial scale is likely to impact the interpretation of small core needle biopsies.

In a third test we characterized tumour-immune cell interactions. Immune checkpoint inhibitors (ICI) produce durable responses in a portion of patients but identifying potential responders and non-responders remains a challenge. Conventional single channel IHC on checkpoint proteins and ligands lacks sufficient positive and negative predictive value to stratify therapy or justify withholding ICIs in favor of small molecule therapy. Multivariate predictors based on multiple markers such as CD3, CD4, CD8, PD-1 etc. are likely to be more effective, but still underperform as patient stratification approaches5. This likely arises because cell types other than CD8+ TILs, including malignant, stromal, and myeloid-derived cells affect responses to ICIs. t-CyCIF represents a simple method for simultaneously assessing up to 60 predictors, including several processes and mechanisms, such as angiogenesis regulators, DNA damage, tyrosine kinase expression, proliferation, and others (Figure S3). In the current work, we perform a simple analysis showing that TILs can be subtyped, analyzed for PD-1 levels and proximity to PD-L1 ligand at a single cell level. Next steps include validating more antibodies and developing an efficient means to relate staining patterns to immune cell classes that have been defined primarily by flow cytometry.

Highly multiplexed histology is still in an early stage of development and better methods for segmenting cells, quantifying fluorescence intensities and analyzing the resulting data are required. With better data analysis methods, cell-to-cell heterogeneity in t-CyCIF images should enable reconstruction of signaling network topologies relevant to different regions of a single tumor12,34 based on the observation that proteins naturally and randomly fluctuate in abundance and activity from one cell to the next. When these fluctuations are highly correlated, they are likely to reflect causal associations35. Additional work will also be required to determine which types of heterogeneity are most significant for therapeutic response and disease progression: some cell-to-cell differences observed by fixed cell imaging arise from time-dependent fluctuation. However, we have observe cell-to-cell heterogeneity not only among proteins known to change over the course of a single cell cycle but also among long-lived proteins. Finally, validation of t-CyCIF-based biomarkers will require extensive testing in patient cohorts. It is important to note in this context that the current study describes a technology and its potential applications to histopathology, not actual diagnostic biomarkers. The associations we describe might not prove statistically significant when tested on larger, well-controlled sets of samples.

In conclusion, the t-CyCIF approach to multi-parametric imaging is robust, simple and applicable to many types of tumours and tissues; as an open platform, it allows investigators to mix and match antibodies depending on the requirements of a specific type of sample. In the long run we expect t-CyCIF to be complementary to, and used in parallel with more sophisticated protein and RNA imaging methods that may have greater sensitivity or channel capacity, although it seems probable that direct imaging will always have better resolution and speed than laser ablation or mechanical picking. A particularly important future development will be cross-referencing tumor cell types identified by single-cell genomics with those identified by multiplexed imaging, making it possible to precisely define the genetic geography of human cancer.

## Author contributions

JR, BI and PKS conceived the study. JR, BI, SM and PS performed experiments. JR, BI, SW and PKS performed analyses. BI, JR and PKS wrote the manuscript.

## Competing financial interests

PKS is a member of the Board of Directors of RareCyte Inc., which manufactures the slide scanner used in this study, and co-founder of Glencoe Software, which contributes to and supports open-source OME/OMERO image informatics software. Other authors have no competing financial interests to disclose.

## ONLINE METHODS

We continue to improve the methods described here; periodic updates can be found at our web site [http://lincs.hms.harvard.edu/resources/](http://lincs.hms.harvard.edu/resources/). A video illustrating the t-CyCIF approach can be found at [https://youtu.be/fInnargF2fs](http://https://youtu.be/fInnargF2fs).

### Patients and specimens

Tumor tissue and FFPE specimens were collected from patients under IRB-approved protocols (DFCI 11-104) at Dana-Farber Cancer Institute/Brigham and Women’s Hospital, Boston, Massachusetts. Tonsil samples were purchased from American MasterTech (CST0224P). Tissue microarrays (TMA) were obtained from Protein Biotechnologies (TMA-1207).

### Reagents and antibodies

All conjugated and unconjugated primary antibodies used in this study are listed in Table 1. Indirect immunofluorescence was performed using secondary antibodies conjugated with Alexa-647 anti-Mouse (Invitrogen, Cat. A-21236), Alexa-555 anti-Rat (Invitrogen, Cat. A-21434) and Alexa-488 anti-Rabbit (Invitrogen, Cat. A-11034). 10 mg/ml Hoechst 33342 stock solution was purchased from Life Technologies (Cat. H3570). 20x PBS was purchased from Santa Cruz Biotechnology (Cat. SC-362299). 30% hydrogen peroxide solution was purchased from Sigma-Aldrich (Cat. 216763). PBS-based Odyssey blocking buffer was purchased from LI-COR (Cat. 927-40150). All reagents for the Leica BOND RX were purchased from Leica Microsystems.

### Pre-processing and pre-staining tissues for t-CyCIF

#### Automated dewaxing, rehydration and pre-staining

Pre-processing of FFPE tissue and tumor slices mounted on slides was performed on a Leica BOND RX automated stained using the following protocol:

View this table:
[Table2](http://biorxiv.org/content/early/2017/06/19/151738/T2)

#### Manual dewaxing, rehydration and pre-staining

In our experience dewaxing, rehydration and pre-staining can also be performed manually with similar results. For manual pre-processing, FFPE slides were first incubated in a 60°C oven for 30 minutes. To completely remove paraffin, slides were placed in a glass slide rack were then immediately immersed in Xylene in a glass staining dish (Wheaton 900200) for 5 min and subsequently transferred to a another dish containing fresh Xylene for 5 min. Rehydration was achieved by sequentially immersing slides, for 3 min each, in staining dishes containing 100% ethanol, 90% ethanol, 70% ethanol, 50% ethanol, 30% ethanol, and then in two successive 1xPBS solutions. Following rehydration, slides were placed in a 1000 ml beaker filled with 500 ml citric acid, pH 6.0, for antigen retrieval. The beaker containing slides and citric acid buffer was microwaved at low power until the solution was at a boiling point and maintained at that temperature for 10 min. After cooling to room temperature, slides were washed 3 times with 1xPBS in vertical staining jars followed by blocking with Odyssey blocking buffer. Buffer was applied to slides as a 250-500 μl droplet for 30 mins at room temperature; evaporation was minimized by using moist in a slide moisture chamber (Scientific Device Laboratory, 197-BL). Slides were then pre-stained by incubation with diluted secondary antibodies (listed above) for 60 minutes, followed by washing 3 times with 1x PBS. Finally, slides were incubated with Hoechst 33342 (2 μg/ml) in 250-500 μl Odyssey blocking buffer for 30 min. in a moisture chamber and washed 3 times with 1xPBS in vertical staining jars.

### Cyclic immunofluorescence with primary antibodies and Hoechst 33342

All primary antibodies (fluorophore-conjugated and unconjugated) were diluted in Odyssey blocking buffer. Slides carrying dewaxed, pre-stained tissues were then stained in a moisture chamber by dropping the diluted primary or fluorophore-conjugated antibody (250-500 μl per slides) on tissue followed by incubation at 4°C for ~12 hr. Slides were washed four times in 1x PBS by dipping in a series of vertical staining jars. For indirect immunofluorescence, slides were incubated in diluted secondary antibodies in a moisture chamber for 1 hr at room temperature followed by four washes with 1x PBS. Slides were incubated in Hoechst 33342 at 2 μg/ml in Odyssey blocking buffer for 15 min at room temperature, followed by four washes in 1x PBS. Stained slides were mounted prior to image acquisition (see the Mounting section below).

### Primary antibodies

For t-CyCIF, we selected commercial antibodies previously validated by their manufacturers for use in immunofluorescence, immunocytochemistry or immunohistochemistry (IF, ICC or IHC). When possible, we checked antibodies on reference tissue known to express the target antigen, such as immune cells in tonsil tissue or tumor-specific markers in tissue microarrays. The staining patterns for antibodies with favorable signal-to-noise ratios were compared to those previously reported for that antigen by conventional antibodies. An updated list of all antibodies tested to date can be found at the HMS LINCS website ([http://lincs.hms.harvard.edu/resources/](http://lincs.hms.harvard.edu/resources/)). The extent of validation is quantified on a level between 0 and 2: “Level 0” represents antibodies for which staining was not detected using tissues for which the antigen is thought to be present based on published data; “Level 1” represents the expected pattern of positive staining in a limited number of tissues types (e.g. CD4 antibody in tonsil tissue alone); “Level 2” represents the expected pattern of positive staining in all tissues or tumor types been tested (N>= 3). Higher levels will be assigned in the future to antibodies that have undergone extensive validation; for example, side-by side comparison of against an established IHC positive control.

### Mounting & de-coverslipping

Immediately prior to imaging slides were mounted with 1x PBS or, if imaging was expected to take longer than 30 minutes (this occurs in the case of samples larger than 4 cm2 corresponding to about 200 fields of view with a 10X objective) with 1x PBS containing 10% Glycerol. Slides were covered using 24 x 60mm No. 1 coverslips (VWR 48393-106) to prevent evaporation and to facilitate susbequent decoverslipping via gravity. Following image acquisition, slides were placed in a vertical staining jar containing 1x PBS for at least 15 min. Coverslips were released from slides (and the tissue sample) via gravity as the slides were slowly drawn out of the staining jar.

### Fluorophore inactivation

After imaging, fluorophores were inactivated by placing slides horizontally in 4.5% H2O2 and 24 mM NaOH made up in PBS for 1 hour at RT in the presence of white light. Following fluorophore inactivation, slides were washed 4 times with 1x PBS by dipping them in a series of vertical staining jars to remove residual inactivation solution.

### Image acquisition

Stained slides from each round of CycIF were imaged with a CyteFinder slide scanning fluorescence microscope (RareCyte Inc. Seattle WA) using either a 10X (NA=0.3) or 40X long-working distance objective (NA = 0.6). Imager5 software (RareCyte Inc.) was used to sequentially scan the region of interest in 4 fluorescence channels. These channels are referred to by the manufacturer as: (i) a DAPI channel with an excitation filter having a peak of 390 nm and half-width of 18nm and an emission filter with a peak of 435nm and half-width of 48nm; (ii) FITC channel having a 475nm/28nm excitation filter and 525nm/48nm emission filter (iii); Cy3 channel having a 542nm/27nm excitation filter and 597nm/45nm emission filter and (iv); Cy5 channel having a 632nm/22nm excitation filter and 679nm/34nm emission filter. Imaging was performed with a 2x2 binning strategy to increase sensitivity, reducing exposure time and photo bleaching. We have tested slide scanners from several other manufacturers (e.g. a Leica Aperio Digital Pathology Slide Scanner, GE IN-Cell Analyzer 6000 and GE Cytell Cell Imaging System) and found that they too can be used to acquire images from t-CyCIF samples. Slides can also be analyzed on conventional microscopes, but the field of view is typically smaller, and an automated stage is required for accurate stitching of individual fields of view into a complete image of a tissue.

### Image processing

Quantitative analysis of tissue images is a complex, in large part because cells are close packed. Background can be uneven across large images and signal to noise ratios relatively low, particularly in the case of anti-phospho-protein antibodies. We have only started to tackle these issues in the case of high dimensional t-CyCIF data and users of the method are encouraged to thoroughly research image processing methods themselves. We expect to update methods and algorithms described here on a regular basis and users are encouraged to check our Web site for additional information.

#### Background subtraction and image registration

Background subtraction was performed using the previously established rolling ball algorithm (with a 50-pixel radius) in ImageJ. Adjacent background-subtracted images from the same sample were then registered to each using an ImageJ script as described previously1. In brief, DAPI images from each cycle were used to generate reference coordinates by Rigid-body transformation. To generate virtual hyper-stacked images, the transformed coordinates were applied to images from four channel imaging of each CyCIF cycle.

#### Single-cell segmentation & quantification

To obtain intensity values for single cells, images were segmented using a previously described2 conventional waterfall algorithm based on nuclear staining by Hoechst 33342. Images were binarized in the Hoechst channel and then converted into regions of interest (ROIs) for each cell. The Watershed algorithm in ImageJ was then applied to enlarge ROIs (by 3 pixels in the case of 10 x images) and encompass a significant portion of the cytoplasm and membrane for each cell. ROIs were then used to compute intensity values from all channels. All scripts can be found in our Github repository ([https://github.com/sorgerlab/cycif](http://https://github.com/sorgerlab/cycif)).

### High-dimensional single-cell analysis by t-SNE and EMGM

Raw intensity data generated from registered and segmented images was imported into Matlab and converted to comma separated value (csv) files. The viSNE implementation of t-SNE and EMGM algorithms from the CYT single-cell analysis package were obtained from the Pe’er laboratory at Columbia University3. Intensity-based measurements (such as flow cytometry or imaging cytometry) of protein expression have approximately log-normal distribution4, hence, t-CyCIF raw intensity values were first transformed in log or in inverse hyperbolic sine (*asinh*) using the default Matlab function or the *CYT* package3, respectively. Between-sample variation was then normalized on a per-channel basis by aligning intensity measurements that encompass values between 1st and the 99th percentile (using the *CYT* package). Data files were aggregated and used to generate viSNE plots. All viSNE/t-SNE analyses used the following settings: perplexity -30, epsilon =500, lie factor =4 for initial 100 iterations and lie factor -1 for remaining iterations. For EMGM clustering, k = 8 was used for dividing samples into groups. Intensity values from all antibody channels (plus area and Hoechst intensity) were used for unsupervised clustering.

### Calculating Shannon entropy values

Images were divided into regular grids and 1000 cells from each region used to calculate the non-parametric Shannon entropy as follows: ![Formula][1]</img>  where E1(s) is the Shannon entropy of signal **s**; si is the per-pixel intensity of signal **s** at a given point. Normalized Shannon entropy as calculated as Enormalized = Eregion / Esample

### Data availability

Intensity data used to generate figures 1D-E, 1G-H, 3D-E, 3G-H, 4B-F, 5C, 6A-6D is available in supplementary materials and can be downloaded from the HMS LINCS website ([http://lincs.hms.harvard.edu/lin-tbd-2017/](http://lincs.hms.harvard.edu/lin-tbd-2017/)

### Code availability

All ImageJ & Matlab scripts used in this study are available at the Sorgerlab GitHub Repo ([https://github.com/sorgerlab/cycif](http://https://github.com/sorgerlab/cycif)).

### Image availability

All images can be accessed through HMS LINCS page ([http://lincs.hms.harvard.edu/lin-tbd-2017/](http://lincs.hms.harvard.edu/lin-tbd-2017/)). Note that publically available OMERO Web clients do not yet support viewing of very large stitched images and only representative image fields are shown. We are currently working on software that will enable the analysis and distribution of very large images via an OMERO Web client.

Specific image sets can be accessed as follows:

For Tonsil composite images (Figure 2 & Table S1):

[https://omero.hms.harvard.edu/webclient/?show=dataset-2038](http://https://omero.hms.harvard.edu/webclient/?show=dataset-2038)

For PDAC composite images (Figure 3 & Table S2):

[https://omero.hms.harvard.edu/webclient/?show=dataset-2039](http://https://omero.hms.harvard.edu/webclient/?show=dataset-2039)

For RCC composite images (Figure 3 & Table S4):

[https://omero.hms.harvard.edu/webclient/?show=dataset-2040](http://https://omero.hms.harvard.edu/webclient/?show=dataset-2040)

For TMA panels (Figure 3& Table S2/S3):

[https://omero.hms.harvard.edu/webclient/?show=dataset-2037](http://https://omero.hms.harvard.edu/webclient/?show=dataset-2037)

For GBM composites images (Figure 5, Figure 6 & Table S5)

[https://omero.hms.harvard.edu/webclient/?show=dataset-2041](http://https://omero.hms.harvard.edu/webclient/?show=dataset-2041)

![Supplementary Figure S1.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F7.medium.gif)

[Supplementary Figure S1.](http://biorxiv.org/content/early/2017/06/19/151738/F7)

Supplementary Figure S1. Technical aspects of the t-CyCIF process (Covering data in Figures 1 and 2).
**(A)** Schematic of pre-processing steps for t-CyCIF. **(B-D)** Evaluation of auto-fluorescence and fluorophore inactivation. Tonsil tissue was stained with Hoechst 33342 and imaged in all four channels before and after 3 cycles of fluorophore inactivation. **(E and F)**. Single-cell quantification of fluorophore inactivation in tonsil tissue stained with antibodies indicated. **(H-J).** The effect of repeated rounds of fluorophore inactivation on antigenicity. Tonsil tissue was stained with Ki67-Alexa647 (colored red in **H**) and CD11c-Alexa 555 antibodies (colored green in **H**). The intensities were measured after 1, 2, 3, and 4 cycles of fluorophore inactivation. **(I and J)** Distributions of single-cell fluorescence intensities measured in (H) from different cycles are shown.


![Supplementary Figure S2.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F8.medium.gif)

[Supplementary Figure S2.](http://biorxiv.org/content/early/2017/06/19/151738/F8)

Supplementary Figure S2. Retention of tissue ultra-structure during t-CyCIF (Covering data in Figure 2).
**(A)** Normal skin imaged with 5-cycle t-CyCIF with as indicated. **(B)** Melanoma samples imaged via 14 cycle t-CyCIF showing the (pseudo-colored) Hoechst channel only as a means to judge tissue integrity. Cycle 0 denotes the pre-staining cycle. **(C)** Higher magnification view of the sample images of (B).

![Supplementary Figure S3.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F9.medium.gif)

[Supplementary Figure S3.](http://biorxiv.org/content/early/2017/06/19/151738/F9)

Supplementary Figure S3. 60-marker t-CyCIF in a melanoma tumor.
**(A)** Gallery of pseudo-colored cycles of a melanoma specimen that underwent 20 cycles of t-CyCIF. **(B)** Representative cycles 0, 5, 10 and 20 of the specimen from panel (A) highlighting markers of various biological processes, such as cancer cell autonomous expression of angiogenesis receptor VEGFR2, DNA damage (i.e. γH2xa), receptor tyrosine kinase (i.e. EGFR) and cell cycle markers (i.e. PCNA).

![Supplementary Figure S4.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F10.medium.gif)

[Supplementary Figure S4.](http://biorxiv.org/content/early/2017/06/19/151738/F10)

Supplementary Figure S4. Imaging of tissue microarrays by t-CyCIF (covering data in Figure 4).
Gallery of additional representative images of 8-cyle t-CyCIF of biopsies in a tissue microarray.

![Supplementary Figure S5.](http://biorxiv.org/https://www.biorxiv.org/content/biorxiv/early/2017/06/19/151738/F11.medium.gif)

[Supplementary Figure S5.](http://biorxiv.org/content/early/2017/06/19/151738/F11)

Supplementary Figure S5. Cell-to-cell heterogeneity in a GBM tumor (covering data in Figure 6)
**(A)** EMGM clusters (as defined and color-coded in Figure 6) frame 100 in tumor region R5 mapped onto the positions of single cells. The percentage of cells in each cluster is shown. **(B).** Representative image of the same frame for cytokeratin, c-Jun and EGFR staining.

View this table:
[Table S1.](http://biorxiv.org/content/early/2017/06/19/151738/T3)

Table S1. 
List of antibodies used to stain tonsil tissue presented in Figure 2

View this table:
[Table S2.](http://biorxiv.org/content/early/2017/06/19/151738/T4)

Table S2. 
List of antibodies used to stain PDAC sample presented in Figure 3 and TMA in Figure 4.

View this table:
[Table S3.](http://biorxiv.org/content/early/2017/06/19/151738/T5)

Table S3. 
Description and naming convention of TMA used in Figure 4.

View this table:
[Table S4](http://biorxiv.org/content/early/2017/06/19/151738/T6)

Table S4 
List of antibodies used to stain RCC tissue presented in Figure 3

View this table:
[Table S5.](http://biorxiv.org/content/early/2017/06/19/151738/T7)

Table S5. 
List of antibodies used to stain GBM tissue presented in Figure 5 and Figure 6.

## Acknowledgements

This work was funded by P50GM107618 and U54HL127365 to PKS. This work was supported by a DF/HCC GI SPORE Developmental Research Project Award and the DFCI Claudia Adams Barr Program for Innovative Cancer Research Award to BI. We thank Brian Wolpin and Christine Lian for providing tissue slides, Zoltan Maliga for technical support and Levi Garraway, and members of Ludwig Center for Cancer Research at Harvard for fruitful discussions.

*   Received June 18, 2017.
*   Revision received June 18, 2017.
*   Accepted June 19, 2017.


*   © 2017, Posted by Cold Spring Harbor Laboratory

This pre-print is available under a Creative Commons License (Attribution-NonCommercial-NoDerivs 4.0 International), CC BY-NC-ND 4.0, as described at [http://creativecommons.org/licenses/by-nc-nd/4.0/](http://creativecommons.org/licenses/by-nc-nd/4.0/)

## REFERENCES

1.  1.Garraway, L. A. & Lander, E. S. Lessons from the cancer genome. Cell 153, 17–37 (2013).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/j.cell.2013.03.002&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=23540688&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000316853700004&link_type=ISI) 

2.  2.Chapman, P. B. et al. Improved Survival with Vemurafenib in Melanoma with BRAF V600E Mutation. N. Engl. J. Med. 364, 2507–2516 (2011).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1056/NEJMoa1103782&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=21639808&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000292199800006&link_type=ISI) 

3.  3.Druker, B. J. & Lydon, N. B. Lessons learned from the development of an Abl tyrosine kinase inhibitor for chronic myelogenous leukemia. J. Clin. Invest. 105, 3–7 (2000).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1172/JCI9083&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=10619854&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000084579300001&link_type=ISI) 

4.  4.Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646–674 (2011).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/j.cell.2011.02.013&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=21376230&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000288007100007&link_type=ISI) 

5.  5.Tumeh, P. C. et al. PD-1 blockade induces responses by inhibiting adaptive immune resistance. Nature 515, 568–571 (2014).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/nature13954&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=25428505&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000346247600054&link_type=ISI) 

6.  6.Tirosh, I. et al. Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352, 189–196 (2016).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEyOiIzNTIvNjI4Mi8xODkiO3M6NDoiYXRvbSI7czozNzoiL2Jpb3J4aXYvZWFybHkvMjAxNy8wNi8xOS8xNTE3MzguYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

7.  7.Patel, A. P. et al. Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344, 1396–1401 (2014).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEzOiIzNDQvNjE5MC8xMzk2IjtzOjQ6ImF0b20iO3M6Mzc6Ii9iaW9yeGl2L2Vhcmx5LzIwMTcvMDYvMTkvMTUxNzM4LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

8.  8.Bodenmiller, B. Multiplexed Epitope-Based Tissue Imaging for Discovery and Healthcare Applications. Cell Syst. 2, 225–238 (2016).
    
    

9.  9.Stack, E. C., Wang, C., Roman, K. A. & Hoyt, C. C. Multiplexed immunohistochemistry, imaging, and quantitation: A review, with an assessment of Tyramide signal amplification, multispectral imaging and multiplex analysis. Methods 70, 46–58 (2014).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/j.ymeth.2014.08.016&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=25242720&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

10. 10.Rimm, D. L. What brown cannot do for you. Nat. Biotechnol. 24, 914–916 (2006).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/nbt0806-914&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=16900128&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

11. 11.Postow, M. A., Callahan, M. K. & Wolchok, J. D. Immune Checkpoint Blockade in Cancer Therapy. J. Clin. Oncol. Off. J. Am. Soc. Clin. Oncol. 33, 1974–1982 (2015).
    
    

12. 12.Giesen, C. et al. Highly multiplexed imaging of tumor tissues with subcellular resolution by mass cytometry. Nat. Methods 11, 417–422 (2014).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/nmeth.2869&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=24584193&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000333749900024&link_type=ISI) 

13. 13.Gerdes, M. J. et al. Highly multiplexed single-cell analysis of formalin-fixed, paraffin-embedded cancer tissue. Proc. Natl. Acad. Sci. U. S. A. 110, 11982–11987 (2013).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NDoicG5hcyI7czo1OiJyZXNpZCI7czoxMjoiMTEwLzI5LzExOTgyIjtzOjQ6ImF0b20iO3M6Mzc6Ii9iaW9yeGl2L2Vhcmx5LzIwMTcvMDYvMTkvMTUxNzM4LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

14. 14.Micheva, K. D. & Smith, S. J. Array tomography: a new tool for imaging the molecular architecture and ultrastructure of neural circuits. Neuron 55, 25–36 (2007).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/j.neuron.2007.06.014&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=17610815&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000248010700006&link_type=ISI) 

15. 15.Remark, R. et al. In-depth tissue profiling using multiplexed immunohistochemical consecutive staining on single slide. Sci. Immunol. 1, aaf6925-aaf6925 (2016).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6MTA6ImltbXVub2xvZ3kiO3M6NToicmVzaWQiO3M6MTE6IjEvMS9hYWY2OTI1IjtzOjQ6ImF0b20iO3M6Mzc6Ii9iaW9yeGl2L2Vhcmx5LzIwMTcvMDYvMTkvMTUxNzM4LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

16. 16.Gerner, M. Y., Kastenmuller, W., Ifrim, I., Kabat, J. & Germain, R. N. Histo-cytometry: a method for highly multiplex quantitative tissue imaging analysis applied to dendritic cell subset microanatomy in lymph nodes. Immunity 37, 364–376 (2012).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/j.immuni.2012.07.011&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=22863836&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000308261800019&link_type=ISI) 

17. 17.Lee, J. H. et al. Highly multiplexed subcellular RNA sequencing in situ. Science 343, 1360–1363 (2014).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEzOiIzNDMvNjE3Ny8xMzYwIjtzOjQ6ImF0b20iO3M6Mzc6Ii9iaW9yeGl2L2Vhcmx5LzIwMTcvMDYvMTkvMTUxNzM4LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

18. 18.Angelo, M. et al. Multiplexed ion beam imaging of human breast tumors. Nat. Med. 20, 436–442 (2014).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/nm.3488&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=24584119&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

19. 19.Lin, J.-R., Fallahi-Sichani, M. & Sorger, P. K. Highly multiplexed imaging of single cells using a high-throughput cyclic immunofluorescence method. Nat. Commun. 6, 8390 (2015).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/ncomms9390&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=26399630&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

20. 20.Lin, J.-R., Fallahi-Sichani, M., Chen, J.-Y. & Sorger, P. K. Cyclic Immunofluorescence (CycIF), A Highly Multiplexed Method for Single-cell Imaging. Curr. Protoc. Chem. Biol. 8, 251–264 (2016).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1002/cpch.14&link_type=DOI) 

21. 21.Jones, S. et al. Core Signaling Pathways in Human Pancreatic Cancers Revealed by Global Genomic Analyses. Science 321, 1801–1806 (2008).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEzOiIzMjEvNTg5Ny8xODAxIjtzOjQ6ImF0b20iO3M6Mzc6Ii9iaW9yeGl2L2Vhcmx5LzIwMTcvMDYvMTkvMTUxNzM4LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

22. 22.Vogelstein, B. et al. Cancer Genome Landscapes. Science 339, 1546–1558 (2013).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjEzOiIzMzkvNjEyNy8xNTQ2IjtzOjQ6ImF0b20iO3M6Mzc6Ii9iaW9yeGl2L2Vhcmx5LzIwMTcvMDYvMTkvMTUxNzM4LmF0b20iO31zOjg6ImZyYWdtZW50IjtzOjA6IiI7fQ==) 

23. 23.Jeong, W.-J. et al. Ras stabilization through aberrant activation of Wnt/β-catenin signaling promotes intestinal tumorigenesis. Sci. Signal. 5, ra30 (2012).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6ODoic2lndHJhbnMiO3M6NToicmVzaWQiO3M6MTA6IjUvMjE5L3JhMzAiO3M6NDoiYXRvbSI7czozNzoiL2Jpb3J4aXYvZWFybHkvMjAxNy8wNi8xOS8xNTE3MzguYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9) 

24. 24.Bologna-Molina, R., Mosqueda-Taylor, A., Molina-Frechero, N., Mori-Estevez, A. D. & Sánchez-Acuña, G. Comparison of the value of PCNA and Ki-67 as markers of cell proliferation in ameloblastic tumor. Med. Oral Patol. Oral Cir. Bucal 18, e174–e179 (2013).
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=23229269&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

25. 25.Mahoney, K. M. & Atkins, M. B. Prognostic and predictive markers for the new immunotherapies. Oncol. Williston Park N 28 Suppl 3, 39–48 (2014).
    
    

26. 26.Sharma, P. & Allison, J. P. The future of immune checkpoint therapy. Science 348, 56–61 (2015).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6Mzoic2NpIjtzOjU6InJlc2lkIjtzOjExOiIzNDgvNjIzMC81NiI7czo0OiJhdG9tIjtzOjM3OiIvYmlvcnhpdi9lYXJseS8yMDE3LzA2LzE5LzE1MTczOC5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

27. 27.Varadhachary, G. R. & Raber, M. N. Cancer of Unknown Primary Site. N. Engl. J. Med. 371, 757–765 (2014).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1056/NEJMra1303917&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=25140961&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

28. 28.Turner, N. C. & Reis-Filho, J. S. Genetic heterogeneity and cancer drug resistance. Lancet Oncol. 13, e178–e185 (2012).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/S1470-2045(11)70335-7&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=22469128&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000302287000023&link_type=ISI) 

29. 29.Brennan, C. W. et al. The somatic genomic landscape of glioblastoma. Cell 155, 462–477 (2013).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/j.cell.2013.09.034&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=24120142&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000325719800021&link_type=ISI) 

30. 30.Olar, A. & Aldape, K. D. Using the molecular classification of glioblastoma to inform personalized treatment. J. Pathol. 232, 165–177 (2014).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1002/path.4282&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=24114756&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

31. 31.Ilmer, M. et al. RSPO2 Enhances Canonical Wnt Signaling to Confer Stemness-Associated Traits to Susceptible Pancreatic Cancer Cells. Cancer Res. 75, 1883–1896 (2015).
    
    [Abstract/FREE Full Text](http://biorxiv.org/lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6NDoiQUJTVCI7czoxMToiam91cm5hbENvZGUiO3M6NjoiY2FucmVzIjtzOjU6InJlc2lkIjtzOjk6Ijc1LzkvMTg4MyI7czo0OiJhdG9tIjtzOjM3OiIvYmlvcnhpdi9lYXJseS8yMDE3LzA2LzE5LzE1MTczOC5hdG9tIjt9czo4OiJmcmFnbWVudCI7czowOiIiO30=) 

32. 32.Zeitouni, D., Pylayeva-Gupta, Y., Der, C. J. & Bryant, K. L. KRAS Mutant Pancreatic Cancer: No Lone Path to an Effective Treatment. Cancers 8, (2016).
    
    

33. 33.Phillips, H. S. et al. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell 9, 157–173 (2006).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1016/j.ccr.2006.02.019&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=16530701&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 
    
    [Web of Science](http://biorxiv.org/lookup/external-ref?access_num=000236378500005&link_type=ISI) 

34. 34.Sachs, K., Gifford, D., Jaakkola, T., Sorger, P. & Lauffenburger, D. A. Bayesian network approach to cell signaling pathway modeling. Sci. STKE Signal Transduct. Knowl. Environ. 2002, pe38 (2002).
    
    

35. 35.Vilela, M. & Danuser, G. What’s wrong with correlative experiments? Nat. Cell Biol. 13, 1011 (2011).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/ncb2325&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=21892139&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

## REFERENCES

1.  1.Lin, J.-R., Fallahi-Sichani, M. & Sorger, P. K. Highly multiplexed imaging of single cells using a high-throughput cyclic immunofluorescence method. Nat. Commun. 6, 8390 (2015).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/ncomms9390&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=26399630&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

2.  2.Carpenter, A. E. et al. CellProfiler: image analysis software for identifying and quantifying cell phenotypes. Genome Biol. 7, R100 (2006).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1186/gb-2006-7-10-r100&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=17076895&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

3.  3.Amir, E. D. et al. viSNE enables visualization of high dimensional single-cell data and reveals phenotypic heterogeneity of leukemia. Nat. Biotechnol. 31, 545–552 (2013).
    
    [CrossRef](http://biorxiv.org/lookup/external-ref?access_num=10.1038/nbt.2594&link_type=DOI) 
    
    [PubMed](http://biorxiv.org/lookup/external-ref?access_num=23685480&link_type=MED&atom=%2Fbiorxiv%2Fearly%2F2017%2F06%2F19%2F151738.atom) 

4.  4.Bagwell, C. B. Hyperlog-a flexible log-like transform for negative, zero, and positive valued data. Cytom. Part J. Int. Soc. Anal. Cytol. 64, 34–42 (2005).

 [1]: /embed/graphic-13.gif