Sequential Chromogenic IHC: Spatial Analysis of Lymph Nodes Identifies Contact Interactions between Plasmacytoid Dendritic Cells and Plasmablasts

The clinical success of immunotherapeutic approaches for patients with cancer warrants a better understanding of immune responses in cancer, to increase patient outcomes and decrease adverse reactions. Biomarker discovery and assessment is a key tool in modern immuno-oncology practice (1). Defining which biomarkers are useful as prognostic and predictive indicators of disease course and therapeutic outcome is not only intrinsically challenging due to the complexity of the immune system, but also because no single marker has been validated to correlate with clinical response (2, 3). The need to quantify multiple biomarkers is especially important in the context of immune responses to cancer, which summons both adaptive and innate immune cells to both the tumor microenvironment and sentinel lymph nodes (4). Recent advances in our understanding of the interplay between cancer and the immune system highlight the importance of the spatial organization of immune cells, adding another layer of complexity to the problem (5–9). Thus, innovative wet and dry lab techniques to tackle these challenges are urgently needed.

Sequential chromogenic IHC (scIHC) is a novel experimental and analytic approach that allows researchers to assess multiple (>10) biomarkers on a single tissue section. This technique promises to vertically innovate biomarker discovery and assessment of disease states (10). Contrary to multiplex IHC, in which two or more biomarkers are detected at the same time, biomarkers in a sequential IHC panel are tested one at a time (11). Hence, the cells present on the slide can be sampled for tens of biomarkers, such as it is common practice in flow cytometry, with the added advantage of retaining the relative location to each other (10). Once data are collected, the coexpression of various markers within the same cell and its spatial relations with nearby immune and tumor cells can be quantified (12). These features grant scIHC a spot in the family of approaches commonly referred to as “tissue cytometry” (13). Compared with other tissue cytometry approaches, the technique described here is not limited to 5–10 biomarkers (such as MICSSS and SIMPLE, reviewed in ref. 13). Fluorescence-based protocols (such as cIF and MxIF) can be somewhat less time-intensive than scIHC, but are not integrated in pathology labs and some methods (like CODEX and NanoString) require special devices and reagents (13). The consumables employed during scIHC are commercially available and widely used in clinical pathology laboratories, which makes scIHC a simple and affordable technique as compared with other equivalent approaches (14). To benefit patients, the transition into the clinic will require the reduction of large, research-grade biomarker panels down to the minimum number of biomarkers strictly necessary for clinical decision making.

Our group recently employed scIHC to quantify the tumor immune microenvironment of sinonasal squamous cell carcinoma (SNSCC). We identified potential prognostic biomarkers in a cohort of 38 patients by showing increased T-cell populations and decreased myeloid cell populations in patients with SNSCC without recurrent disease, as compared with recurring patients (15). Although the characterization of the tumor immune microenvironment can inform treatment decisions and provide useful prognosticators (5, 16–19), the unpredictable nature of the tumor microenvironment can represent a significant obstacle to reproducibly and accurately assessing immune responses (20–22). Moreover, polyclonal antibodies cannot generally be employed to assess the tumor immune microenvironment, likely due to the presence of aberrant cells carrying different glycosylations and oxidized epitopes that greatly increase background signal (personal observations). On the other hand, regional lymph nodes are highly structured organs that are inhabited by normal immune cells responsible for key immune reactions to tumor antigens (23). In addition, immune cell location within highly compartmentalized lymph nodes can be indicative of function (24, 25). Finally, sentinel node biopsy and lymph node dissection are standard of care in patients with early disease (T1-2) and with evident nodal disease at the time of diagnosis, respectively (26–28). For these reasons, analysis of patient lymph nodes represents a promising alternative to inform treatment decisions and provide important prognostic indications.

To develop a novel scIHC panel, it is not enough to simply add single IHC steps back-to-back. Several considerations into reagent selection and compatibility need to be assessed to avoid getting into a daunting task riddled with unexpected complications. Here, we examine major steps from classical IHC protocols and highlight the impact they have on the scIHC procedure. As a case study, we describe the development of a novel scIHC panel, with special focus on staining order, on how to avoid signal carryover and secondary reagent cross-reactivity, and on comparison between alternative procedures. Findings for each antibody tested are provided in an external resource to guide development of other customized panels. We use this scIHC panel to quantify immune parameters in regional lymph nodes from patients diagnosed with head and neck cancer (HNC; Supplementary Table S1), that we analyzed by developing a novel bioinformatics pipeline. Together, our findings provide a systematic and practical approach to develop novel biomarker panels for sequential histologic studies that will contribute to improve our understanding of immune responses in patients with cancer.

Formalin-fixed paraffin-embedded (FFPE) tissue sections (5 μm) of N = 10 tumor-free N0 regional lymph nodes from anonymized patients diagnosed with HPV+ or HPV– squamous cell carcinoma of the head and neck were obtained from the Oregon Health & Science University (OHSU) Knight Cancer Institute Biolibrary [Institutional Review Board (IRB) #4918; Supplementary Table S1] in accordance with full ethical approval by the OHSU Institutional Review Board. OHSU Knight Cancer BioLibrary has complied with all applicable laws, rules, and regulations relating to the collection and/or use of the materials and it has obtained all necessary approvals, including informed consent or waiver of consent under an IRB-approved protocol, in writing, for the collection and use of materials.

Our protocol was guided by the seminal work of Tsujikawa and colleagues (11, 29) with some modifications. FFPE patient tissue sections were baked for 60°C for 60 minutes total, with a 5-minute incubation at room temperature after the first 30 minutes. These slides were then deparaffinized by submerging them in xylene for 5 minutes, and then repeated again for another 5 minutes. Slides were then rehydrated by incubating in serially graded ethanol, with a final incubation in distilled water. An initial counterstain of hematoxylin (Dako, S3301) was performed for 1 minute, with washes in distilled water, and placed in TBS with 2% Tween, pH 7.4 (Boston BioProducts, IBB-180X). Slides were coverslipped (Corning, 2975246) and imaged using an Aperio ImageScope AT (Leica Biosystems) at 20X magnification. Coverslips were removed by gentle agitation of the slides in TBS with 2% Tween (TBST). Slides were then subjected to heat-mediated antigen retrieval by placing slides in boiling citrate buffer with a pH of 6.0 (Abcam, ab93678) and steamed for 30 minutes (100°C). Slides were allowed to cool to room temperature, washed with distilled water, and placed in TBST. Endogenous peroxidase activity was blocked by incubating slides with Dako dual endogenous peroxidase block (Dako, S2003) for 10 minutes at room temperate. Slides were washed with distilled water, placed in TBST for 1 minute. Additional protein blocking was performed by incubating slides in 1X PBS (Corning, 21-040-CM) containing 5% horse serum and 2.5% BSA for 10 minutes at room temperature. Slides experienced primary antibodies at saturating dilutions, with incubation times and temperatures previously optimized during testing (Supplementary Table S2). After primary antibody incubation, slides were washed in TBST and incubated with an appropriate F(ab’) fragment–specific secondary antibody–labeled polymer conjugated to horseradish peroxidase (HRP) for 30 minutes at room temperature (Supplementary Table S3). After primary and secondary antibody detections steps, signal was visualized using an alcohol-soluble peroxidase substrate 3-amino-9-ethylcarbazole (AEC; Vector Labs, SK-4200), followed by whole-slide digital scanning. After imaging, AEC was removed using graded ethanol incubations, briefly washed in distilled water, and placed in TBST. Secondary HRP signal was either inactivated by performing two blocking steps of Dako dual endogenous peroxidase block (Dako, S2002) for 10 minutes at room temperature (each), in addition to a protein blocking step to allow for another primary antibody produced in a different species to be applied as a different round, or antibodies were stripped in heated citrate buffer (pH 6.0) to begin a new cycle.

In the last cycle of the scIHC panel, a Tris-Ethylene diamine tetra-acetic acid (EDTA) antigen retrieval buffer, pH 9.0 (Abcam, ab93684) was required for efficient signal detection. Antibodies from the previous cycle required stripping in a Citrate buffer, as mentioned above, before “conditioning” the tissue with the EDTA-based buffer. This “conditioning” was also heat-mediated, and steamed for 30 minutes. The protocol for staining and visualization was performed as in previous cycles, including both primary and secondary antibodies. Antibodies that require a Tris-EDTA antigen retrieval can also be placed in rounds at the first cycle(s) if successive heat treatment adversely affects the tissue integrity on the slide. Changing antigen retrieval buffers within and experiment should be avoided; however, this method can be utilized if that is not possible, as was the case for this experiment. At the end of the last cycle, AEC was removed from the slides using ethanol gradient washes with, a final counterstain with hematoxylin was performed, and slides were imaged.

In addition to determining an antibody's saturating dilution and incubation parameters, it is necessary to identify its appropriate location within the panel. Ideally, an antibody of interest would be tested in each round of every cycle; however due to budgetary and temporal constraints, this may not be possible. To overcome such constraints, we created and utilized a detailed database (AirTable Inc.) to track all tests in an effort to optimize the development of the scIHC panel (Supplementary Table S4). As tests and experimental data are generated, the database is continually updated to streamline the creation of future scIHC panels, and currently has >300 entries. The database can be accessed at https://airtable.com/shrJkmkXLOik5HlRc.

All slides were kept in TBST with 0.02% sodium azide and used as a repository of tissue to test at different cycles. A separate database was used to track slides, their tissue type, their, in addition to the cycle and round number. If an antibody was tested and determined to work at cycle 1 and again at cycle 5, the authors assumed this antibody would also work in cycles 2–4. Moreover, if a specific antibody did not work at a given cycle, it was assumed it would also not work in subsequent cycles.

For efficient, parallel handling and processing of the large scIHC images, the data were first chunked into a multiresolution, multichannel format of small image squares of 512×512 pixels, compatible with the open-source multi-channel volume image annotation tool KNOSSOS (https://knossos.app; ref. 30). KNOSSOS has originally been developed for serial section three-dimensional (3D) electron microscopy data, but it can also be used with bidimensional data, which is a 3D image with just one slice. Subsequently, each IHC image was registered to the HEM image. Alignment between all the biomarker images for each section is necessary because the scIHC procedure causes shifts and micro-deformations in the tissue. Our computational methods correct for both local distortions and macro-shifts of the tissues on the slide (Supplementary Fig. S1). After a global rigid alignment at downsampled resolution, local nonlinear distortions were corrected chunkwise as follows. First, correspondence point candidates were automatically determined in the IHC images and the HEM image using a customized normalized cross-correlation procedure (31). Next, correspondence point candidates with locally nonconsistent distortion directions and amplitudes were pruned and a distortion field was fitted to the remaining set of correspondence points using fast natural neighbor interpolation. An example of the alignment outcome is provided in Supplementary Fig. S2. To ignore low-quality data, including out of focus or damaged areas, as well as cells from nearby tissues that may express some of the biomarkers used but that do not belong to the analysis (e.g., CD141 is also expressed in endothelial cells in fat tissue), we trained the computational network to detect and mask artifacts like folds, out-of-focus areas, red blood cells within vessels and non-lymph node tissue (Supplementary Fig. S3). Cell outlines were segmented in the hematoxylin image (before cycle 1) using a parallelized Cellpose-based segmentation workflow (32). To maximize the segmentation accuracy, the network was iteratively retrained with targeted human generated ground truth. Cell segmentation accurately detected most nuclei irrespective of their shapes, including round lymphocyte nuclei and elongated macrophage nuclei (Supplementary Movie S1). Regions of interest have been determined automatically using a tailored U-net–based artifact and tissue segmentation workflow. Within the regions of interest, each segmented cell was assigned a unique ID and the average intensity of the magenta color channel within the segmented cell was calculated for all registered IHC images. The final output of the pipeline was a list of unique cell IDs with x/y coordinates and average intensities for each of the scIHC markers, stored in FCS format for import in Cytobank (community version, BD), a web-based flow cytometry data analysis application. Subsets of interest (see below) were gated in Cytobank based on biomarker intensities. Each gated cell subset was then exported as a list of cells defined by x/y coordinates. These lists were visualized in KNOSSOS for visual inspection of the results (cell segmentation and biomarker presence). For spatial analysis, each cell within the plasmablast subset (defined by CD138 expression) was assigned the median Euclidean distance to the nearest 5 cells from the following subsets: plasmacytoid dendritic cells (pDC: BDCA2), type-1 and type-2 conventional dendritic cells (cDC1: CD141⁺CD1c⁻; cDC2: CD141⁻CD1c⁺), sinusoidal macrophages (CD169), non-sinusoidal macrophages (CD169⁻CD68⁺), natural killer (NK) cells (CD56), and granulocytes (CD66b). Frequency distributions of nearest neighbors were plotted with Graphpad Prism (version 9).

Image cytometry data can be accessed at https://community.cytobank.org/cytobank/experiments/110638. Optimization data on primary antibodies can be accessed at https://airtable.com/shrJkmkXLOik5HlRc. Original image scans in SVS format available upon request.

An overview of the scIHC protocol is depicted in Fig. 1, along with the indication of which steps of the protocol were improved, as described below.

Once a list of biomarkers of interest is defined, the first step in designing a scIHC panel is testing the individual antibodies against those biomarkers. The number of cycles needed is determined by the highest number of primary antibodies from the same species: for example, if a list of 20 primary antibodies contains 8 made in mouse, 10 in rabbit and 2 in rat, 10 cycles will be needed. To simplify development of new panels and avoid testing of primary antibodies at each stripping cycle, we collected all our testing into a detailed database for other investigators to use (see Materials and Methods). The need to identify the latest cycle at which the primary antibodies still work comes from the fact that multiple stripping cycles likely affect epitope conformation and subsequent ability of the primary antibody to detect its target. Some antigens are preserved across many cycles (see below), while others are more sensitive to the relatively harsh condition necessary for stripping primary antibodies from the previous cycle. Luckily, multiple antibodies can be assigned to the same order position (i.e., the same stripping cycle) if they can be placed in different rounds within that stripping cycle (i.e., they are from different species), which greatly simplifies primary antibody order assignment. Antigens with epitopes sensitive to stripping should be prioritized in earlier cycles (14). Thus, determining the staining order is crucial to identify each primary antibody's order within the panel. To this end, practice slides should be reused as the various primary antibodies are tested to identify ideal staining conditions, including incubation time/temperature and dilution factor. This way, information on staining efficiency in several different cycles can be obtained. For example, a mouse anti-human CD66b (see Supplementary Table S2) was determined to work at a dilution of 1:600 for 30 minutes at room temperature at cycle 1 (Fig. 2A). The slide was stripped and used to identify ideal staining conditions for a rabbit anti-human CD141, and subsequently stripped again for assessment of other primary antibodies. The same mouse anti-human CD66b was then retested at cycle 9 under the identified optimal conditions (Supplementary Table S2) but no staining was observable (Fig. 2B). Similarly, the rabbit anti-human CD141 was determined to work at a dilution of 1:100 overnight at 4°C in cycle 2 (Fig. 2C), but not at cycle 9 (Fig. 2D). Stripping resistance of a biomarker can be defined as the ability of one of its epitopes to restore its conformation for binding to a given antibody. Biomarkers resistant to multiple stripping cycles can be found. For example, a mouse anti-human CD68 and a rabbit anti-human BCL6 antibodies worked well in cycle 11 (see optimization database on Airtable). All data from these antibody tests were collected into a detailed database (see Materials and Methods), in a searchable format, that allows to determine a tentative order in which antibodies could be placed into a new panel, without the need to systematically assess each antibody at every cycle. Collecting this information in a database is useful for determining cycle placement when developing new panels, such that testing could be tailored based on the data already available, thereby greatly reducing the amount of work necessary.

In addition to optimization of staining conditions for primary antibodies to be used in a scIHC panel, it is also necessary to have reliable secondary reagents for efficient biomarker detection. To maximize the number of biomarkers within each stripping cycle (see above), we expanded the repertoire of anti-human primary antibodies to include less common species, such as goat, rat, and sheep (mouse and rabbit anti-human primary antibodies arguably are the most common). These primary antibodies are used sequentially within a cycle. To match these additional primary antibodies with their corresponding secondary reagents, we compared traditional HRP-conjugated antibodies (IgG) with HRP polymers (Fab’). Although polymeric secondary antibodies are widely used in IHC, a formal comparison with their monomeric counterparts in scIHC has not been assessed. These tests are especially important given the increased potential for cross-reactivity of secondary polymeric reagents used sequentially, as we discuss in Section 3.4. Even after optimizing dilutions and incubation times, we found that detection using an HRP-conjugated IgG showed dimmer staining as compared with an HRP polymer, independent of biomarker identity and while maintaining the same primary antibody conditions (Fig. 3). The secondary HRP polymers not only showed strong staining in early cycles, they also demonstrated better signaling in later cycles. For example, CD11c detection using the traditional secondary HRP-conjugated IgG at cycle 1 (Fig. 3A) was less intense than a secondary HRP polymer using the same primary antibody conditions at cycle 4 (Fig. 3B). The difference was more pronounced when observing Tbet signaling, with the secondary HRP-conjugated IgG showing dim stain at cycle 3 (Fig. 3C), but the secondary HRP polymer showing a much stronger signal at cycle 11 (Fig. 3D). In conclusion, HRP polymers outperform traditional HRP-conjugated IgG not only because they are ready to use and save time, but also because these reagents provide increased signal intensity and reduced background.

Once each primary antibody has been assigned to a stripping cycle, preliminary testing of the whole scIHC panel can start. During this final testing phase, special consideration should be paid to signal carryover within each stripping cycle. We define carryover as signal from the previous round appearing in the current round. This can happen because primary antibodies and secondary reagents are consecutively added to the section during each round, without stripping (Fig. 1). Thus, secondary reagents may form chromogenic signal where previously used primary antibodies are bound. As a result, the antibody signal cannot be trusted because it would be a mix of multiple markers. It is important to eliminate carryover in scIHC to ensure accurate data acquisition and analysis. To detect carryover, we routinely checked for similarities in signal patterns between the current and the previous rounds. During testing of patient samples, we had an interesting case of selective carryover in some patient samples but not others which allowed us to investigate the issue in detail (Fig. 4). More specifically, a goat anti-human CD20 was used in the second round. Appropriate signal was detected in all patient samples (see Fig. 4A and E). A mouse anti-human CD66b antibody was used in the third round. Most patient samples showed identical CD20-like signal as the previous round that we interpreted as carryover (Fig. 4B); however, 1 patient had the expected CD66b staining (Fig. 4F). This created an opportunity to test conditions that eliminate carryover, and recover accurate signal for data acquisition. We reasoned that carryover is due to insufficient blocking of peroxidase from the previous round, which involved a very abundant marker, CD20. To determine whether extra peroxidase blocking would aid in avoiding signal carryover of goat anti-human CD20 into the round with mouse anti-human CD66b, AEC from all patient slides was removed and endogenous peroxidase activity was blocked twice. To confirm absence of peroxidase activity, we applied AEC to the slides immediately after the double blocking step above and imaged the slides. After confirming complete absence of peroxidase activity (Fig. 4C and G), we proceeded with staining for CD66b and its secondary reagent. Upon AEC visualization, we observed the expected signal—albeit somewhat dimmer, in the patient sample that did not have carryover (Fig. 4H), suggesting that double peroxidase treatment does not prevent detection of biomarkers. Importantly, we observed a different staining pattern that corresponded to CD66b signaling in patient samples that previously contained carryover (Fig. 4D), suggesting that, at least for abundant markers like CD20 in lymph nodes, double peroxidase blocking is an effective strategy to prevent carryover.

The use of primary antibodies from different species allows greater flexibility in assigning an order to each of them in the panel by placing multiple biomarkers in the same stripping cycle. Because of the variety of species we used in each cycle, we were primarily concerned with cross-reactivity between secondary reagents. Although methods to avoid cross-reactivity between secondary antibodies in multicolor IHC approaches are not new, secondary reagent cross-reactivity can happen when multiple sequential rounds of IHC are performed. Given the novelty in the sequential nature of the technique, how the secondary reagents may cross-react has not been studied in detail. This could happen when secondary reagents from two different rounds within a stripping cycle recognize each other, leading to a binding strong enough to localize the active peroxidase where the inactivated one from the previous round is. The end result is signal formation from both rounds (Fig. 5A). For example, when we started including goat anti-human primary antibodies in our lymph node scIHC panel, we detected them with a rabbit anti-goat secondary HRP polymer (Supplementary Table S3). If a rabbit and goat primary antibodies were to be used within the same stripping cycle, then the two secondary HRP polymers would bind to each other, leading to cross-reactivity (Fig. 5B). Depending on the markers involved, this issue can be extremely difficult to realize and often remains unnoticed, leading to confounding results and even when noticed, it limits the choice of species that can be used within a stripping cycle (in the example above, either rabbit or goat), leading to bigger complications in defining the biomarker order for the panel. To avoid this, whenever a validated primary antibody made in goat is added to a cycle, a horse anti-rabbit secondary reagent should be used in that same cycle in place of the goat anti-rabbit secondary HRP polymer. This avoids cross-reactivity between secondary reagents (Supplementary Fig. S4). Of note, cross-reactivity does not happen when only one of the two secondary reagents can recognize the other: for example, a rabbit anti-goat and a goat anti-rat secondary HRP polymers (Fig. 5C). We have summarized these cross-reactivity concerns and additional cost-effective analysis in a “decision table” to determine antibody order within a stripping cycle (Table 1; Supplementary Fig. S3).

Antibody stripping is the process of removing bound primary antibodies from the tissue. It is a necessary step in scIHC, not only because it allows to assess additional biomarkers, but also because the treatment to strip primary antibodies also unmasks antigenic epitopes for better detection. However, it is important to distinguish the two procedures because they have different purposes. Importantly, selecting the appropriate antigen retrieval technique depends on a variety of variables such as the method and duration of fixation, the type of tissue, the target antigen and the antibody used (33). Most of the antibodies we selected performed optimally using a citrate-based antigen retrieval buffer, pH 6.0. However, we had one antibody, BCL6, which did not perform well in acidic antigen retrieval conditions. Instead, this antibody performed optimally in a Tris-EDTA antigen retrieval buffer, pH 9.0. To assess whether this buffer can efficiently strip primary antibodies from previous cycle (which include CD68), we imaged the section after AEC treatment (without restaining with primary/secondary antibodies). We observed carryover from the previous cycle (Fig. 6A and B). To properly strip primary antibodies from previous cycle and unmask BCL6 epitopes, we stripped the slides with citrate buffer, allowed to cool to room temperature, and then performed antigen retrieval with Tris-EDTA buffer. This procedure was necessary to detect the appropriate signal of BCL6 without the carryover of CD68 from the previous cycle (Fig. 6C). The complete absence of carryover between sequential rounds of staining is demonstrated by: (i) CD68⁺ cells from panel A can be easily distinguished from BCL6+ cells found in their vicinity, in panel C, because their morphologies and staining pattern are very different; (ii) CD68⁺ cells are not stained in panel C; (iii) BCL6+ cells clustered at the center of the follicle are stained in panel C but not in panel A; (iv) EDTA-based antigen retrieval did not fully strip anti-CD68 antibodies, which can still be detected in panel B, after incubation with AEC. Of note, biomarkers that require basic-pH antigen retrieval can be placed at the beginning of the panel, which avoids epitopes on these biomarkers to undergo unnecessary stripping cycles.

Digitalization of image data promises to accelerate our understanding of complex biological systems (9, 29, 34–36). However, each imaging technique presents unique challenges in the quantification of image data. To extract the highest amount of digital data without sacrificing data quality and accessibility, we applied advanced imaging analysis techniques using KNOSSOS (Ariadne.ai GmbH). The main steps of the analysis pipeline we developed are: alignment, segmentation, location, and intensity extraction, subset identification, and spatial analysis (Fig. 7A). As an example of the type of information that can be extracted by digitalizing histologic images, we focused on antibody producing cells known as plasmablasts. To identify potentially meaningful interactions with other immune subsets, we calculated the distance between each plasmablast and cells expressing antibody receptors (i.e., Fc receptors). These cells are mostly of myeloid origin, with the exception of NK cells (37). We observed that pDCs were often in contact with plasmablasts, as evidenced by the fact that almost 50% of them were within 10 μm (center to center) from pDCs (Fig. 7B). When we visually inspected these two immune populations, we observed that they are often found closely interacting with each other (Fig. 7C). By comparing the locations of plasmablasts with respect to pDCs at a tissue scale, we found that indeed they have contact interactions in both the subcapsular and medullary regions (Fig. 7D). This was not the case for sinusoidal macrophages or granulocytes. Importantly, the frequence of contact interactions was not dependent on cell abundance (Fig. 7D). Overall, the quantitative analysis we developed allows to study the cellular composition and location of the tissue under investigation.

In this work, we provide key information to users aiming to adopt sequential chromogenic IHC as the approach of choice for spatial analyses. We describe how to order biomarkers within a cycle, how to carefully choose secondary HRP polymers and we discuss which situations recommend doubling of the endogenous peroxidase blocking step. Overall, these considerations effectively increased biomarker detection accuracy while eliminating issues like carryover and cross reactivity. We have summarized our approach to scIHC panel development in Table 1.

Sequential chromogenic IHC has several advantages over fluorescence-based techniques: it does not require balancing fluorochrome brightness with marker abundance (38), provides higher sensitivity of detection, avoiding the need for signal amplification (39), up to 15–20 patient samples can be processed by an operator at a time, pathology departments are already equipped to perform scIHC (including image scanners), and generally lower entry costs. Despite its advantages, scIHC is not without caveats. This technique can be relatively time consuming because only one or two biomarkers can be assessed at the same time (e.g., by using a red and blue chromogen). Obtaining data from one staining round can take 4–24 hours to generate, depending on incubation times. Thus, the maximum daily output is four biomarkers (i.e., two staining rounds per day). As we consider in this article, there are constraints in arranging antibodies within a panel, so that antibodies that are sensitive to repeated stripping cycles need to be employed first. While these caveats should not discourage use of this technique, they should be taken into account when determining feasibility and design of experiments. As it is the case for other sequential histologic approaches, operators must pay special attention to the expected distribution of signal for current biomarker and compare it with previous biomarkers to detect potential issues with panel design. Expected signal distribution for the tissue under investigation can be learnt from Protein Atlas website (https://www.proteinatlas.org/).

The procedure we describe can be upgraded relatively easily by multiplexing two chromogens, which can further increase the throughput without sacrificing quality. To multiplex the scIHC with two chromogens (40), two different primary antibodies (from different species) can be employed within each round, followed by the corresponding secondary detection reagents conjugated to different enzymes (e.g., HRP and alkaline phosphatase). These two enzymes will generate precipitate of different color (red and blue, in the example above), that can be imaged at the same time. Simple computational approaches would then separate the signals from the two biomarkers, effectively doubling the throughput of the technique.

A major hurdle in any sequential histologic approach is the analysis pipeline. We developed a pipeline based on commonly used steps (11, 41) but implementing cutting-edge computational approaches, including non-rigid tissue deformation correction and Cellpose-inspired automated cell segmentation (see Materials and Methods). The result is a plug-and-play approach able to identify the localization of different cell subsets, which is a prerequisite for spatial studies. A key factor in the success of the pipeline development was frequent iteration between wet and dry lab scientists, which allowed to quickly spot issues and identify solutions to correct them. For example, because biomarkers by definition work only in the tissue where they have been defined, having a mix of tissues on the slide (e.g., lymph node, fat, and connective tissues) leads to the inclusion of spurious cells within certain biomarkers (e.g., CD141, which labels both cDC1 and endothelial cells). To fix this issue, we defined a tissue of interest mask that excludes segmented cells not pertaining to the lymph node tissue proper. By exporting the biomarker intensities and cell locations in FCS format, we were able to avoid the use of FCSExpress (Denovo software), currently the only commercial software able to handle image cytometry data. However, FCSExpress frequent crashing and software bugs render image-based inspection of the gating strategy a nightmare. We visually validated our gating strategy by importing gated events into KNOSSOS. KNOSSOS is developed by the Max Planck Institute in Germany, and it can display tissue masks, segmented cells and chromogenic signals (including hematoxylin) all together or individually, as layers. Although the neural network-based cell segmentation of a densely cellularized tissue like the lymph node worked very well in distinguishing both round and elongated nuclei (Supplementary Movie S1), a future improvement of our pipeline will be to consider chromogenic signals during segmentation, to avoid spillover of said signals into neighboring cells, which would be especially important for membrane biomarkers.

As proof of concept of what can be discovered by implementing our procedures and bioinformatics pipeline, we present novel data on the neighborhood of lymph node resident plasmablasts. To our surprise, we identified pDCs as the cell type with contact interactions with plasmablasts in HNC patient lymph nodes. The communication between these two immune subsets has been reported only in vitro (42, 43), and here we confirm that plasmablasts and pDCs display contact interactions in vivo. Given the role of Toll-like receptors (highly expressed in pDCs) in vaccine efficacy (42, 44–46), the significance of such interaction decisively warrants further studies.

M.-T. Nguyen is an employee of ariadne.ai ag at the time of data analysis. A. Wanner reports being founder and owner of ariadne.ai ag. No disclosures were reported by the other authors.

N. Claudio: Formal analysis, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. M.-T. Nguyen: Data curation, software, formal analysis, visualization, methodology, project administration. A. Wanner: Resources, software, formal analysis, supervision, visualization, methodology, writing-original draft. F. Pucci: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

We would like to thank: scientists in the Coussens lab for helpful inputs and advice, in particular Dr. Courtney Betts; Ohoud Al Malki, intern student in Pucci lab; Knight BioLibrary for providing patient specimens (IRB approved, IRB00004918); Dr. Rosemary Makar for pathology review; the V Foundation for Cancer Research (V2019-012); Cancer Research UK-OHSU project award (A29681).