multiplexed immunohistochemistry imaging and quantitation a review with an assessment of tyramide signal amplification multispectral imaging and multiplex analysis

Our purpose here is to review and assess methods for multiplexed, quantitative, image analysis based approaches, using new multicolor immuno-histochemistry methods, automated multispectr

Multiplexed immunohistochemistry, imaging, and quantitation:

A review, with an assessment of Tyramide signal amplification,

multispectral imaging and multiplex analysis

PerkinElmer, Inc., Waltham, MA 02451, USA

a r t i c l e i n f o

Article history:

Received 23 April 2014

Revised 12 August 2014

Accepted 29 August 2014

Available online xxxx

Keywords:

Multiplexed

Multispectral

TSA

Quantitative

Pathology

Biomarkers

Cancer

a b s t r a c t

Tissue sections offer the opportunity to understand a patient’s condition, to make better prognostic eval-uations and to select optimum treatments, as evidenced by the place pathology holds today in clinical practice Yet, there is a wealth of information locked up in a tissue section that is only partially accessed, due mainly to the limitations of tools and methods Often tissues are assessed primarily based on visual analysis of one or two proteins, or 2–3 DNA or RNA molecules Even while analysis is still based on visual perception, image analysis is starting to address the variability of human perception This is in contrast to measuring characteristics that are substantially out of reach of human perception, such as parameters revealed through co-expression, spatial relationships, heterogeneity, and low abundance molecules What is not routinely accessed is the information revealed through simultaneous detection of multiple markers, the spatial relationships among cells and tissue in disease, and the heterogeneity now under-stood to be critical to developing effective therapeutic strategies Our purpose here is to review and assess methods for multiplexed, quantitative, image analysis based approaches, using new multicolor immuno-histochemistry methods, automated multispectral slide imaging, and advanced trainable pattern recognition software A key aspect of our approach is presenting imagery in a workflow that engages the pathologist to utilize the strengths of human perception and judgment, while significantly expanding the range of metrics collectable from tissue sections and also provide a level of consistency and precision needed to support the complexities of personalized medicine

Ó 2014 The Authors Published by Elsevier Inc This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/3.0/)

1 Introduction

In the current push to drive a tailored approach to clinical care

using clues hidden in tissue samples, there is significant effort

underway to understand genomic alterations in order to match

small molecule drugs to specific disease types Yet genetics has

not yielded the significant successes hoped for[1,2] Perhaps the

untapped contextual information, remaining in the tissue and

cap-tured through multiplexed labeling of proteins with subsequent

image analysis, can help reveal additional needed information

There have been various strategies employed in an attempt to

characterize tissues from a histological perspective Current

pathology practice utilizes chromogenic immunohistochemistry

(IHC)[3] However, far more powerful multiplexed IHC (mIHC)

approaches are now available, offering greater insights into disease heterogeneity and the characterization of systems biology mecha-nisms driving disease, as well as helping to conserve limited tis-sues An added benefit of mIHC is improved accuracy through application of image analysis, with the use of landmark markers specifically to indicate tissue architecture Landmark markers can also accelerate scan times This workflow also can increase a pathologist’s productivity by automatically measuring parameters hard to achieve reliably by eye, while needing the pathologist as an integral part of the workflow to review analysis results

Though mIHC offers greater insight into molecular cascades and preserves tissue context, in current practice multiplexed stained samples can be difficult to interpret Since mIHC often employs fluorescence, where multiple targets can blend together complicat-ing resolution, this has the potential to muddle visual assessment With formalin-fixed, paraffin-embedded (FFPE) tissues, there is also the potential for tissue autofluorescence, further complicating visual interpretation

http://dx.doi.org/10.1016/j.ymeth.2014.08.016

1046-2023/Ó 2014 The Authors Published by Elsevier Inc.

This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/3.0/ ).

⇑ Corresponding author at: PerkinElmer, 68 Elm Street, Hopkinton, MA 01748,

USA.

E-mail address: clifford.hoyt@perkinelmer.com (C.C Hoyt).

Contents lists available atScienceDirect

Methods

j o u r n a l h o m e p a g e : w w w e l s e v i e r c o m / l o c a t e / y m e t h

Current imaging metrics can effectively address multiplexing

complications, through multispectral unmixing strategies Within

oncology, this has broad potential for designing combinatorial

therapeutic approaches by revealing co-expression, pathway

con-figurations, and spatial relationships among cell types With

improved accuracy of data, automation, and faster scanning,

mul-tispectral mIHC has the capacity to induce a significant paradigm

shift in tissue analysis In this review, we will describe the varied

methodologies that support multiplexing, with a particular focus

on mIHC in the pathology workflow In particular, we will assess

a practical application of mIHC using Tyramide signal amplification

(TSA) in conjunction with multispectral image analysis, which

offers improved mIHC using similar species antibodies, while also

providing quantitatively reproducible multiplexed data, batch to

batch

2 Review

2.1 Multiplexed staining methods

2.1.1 Brightfield multiplexing

In order to perform mIHC on FFPE tissues in brightfield

micros-copy, chromogenic deposition of various chromogens/enzyme

pairs is used While this is useful when distinguishing different cell

types, it is more challenging to assess when trying to co-localize

targets within cells[4] Some of the specific chromogens available

for brightfield mIHC include: 3,30-diaminiobenzidine (DAB) and

nickel enhanced DAB (DAB-Ni), which produce an insoluble brown

or black precipitate, respectively; 3-amino-9-ethylcarbazole (AEC),

which produces a red precipitate that is susceptible to organic

sol-vents; Vector VIP which produces an insoluble purple precipitate;

and nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl-phosphate

(NBT/BCIP), which produces a deep blue precipitate (a more

com-plete listing is available inTable 1) These chromogens can be

visu-alized with either horseradish peroxidase (HRP; DAB, DAB + Ni,

AEC and VIP) or alkaline phosphatase (AP; NBT/BCIP) In addition,

several counterstaining dyes can enhance brightfield multiplexing,

such as methyl green or hematoxylin, which stain nuclei green or

blue, respectively

While brightfield multiplexing is possible, there are several

fac-tors to consider A primary concern remains the need to set up

pri-mary antibodies so that there is no cross reactivity, through

separate conjugations with HRP or AP But this limits multiplexing

capacity, which can be partially overcome with sequential staining

strategies[5–8] This results in a labor intensive protocol where

multiplexing is limited due to tissue degradation after successive

serial IHC assays[9] There is also the potential for chromogenic

overlap, and the significant risk of obscuring one color by another

In particular, there is also overlap of chromogenic spectra, which

limits the degree of chromogenic multiplexing in brightfield[4]

Hence while chromogenic IHC is a valuable tool and widely utilized

in many pathology labs, the ability to practically multiplex beyond

3 targets is limited

2.1.2 Fluorescent multiplexing Fluorescence mIHC takes advantage of light emission with dif-ferent spectral peaks against a dark background The basic princi-ple behind fluorescent IHC relies on the ability of individual fluorophores to be excited by one wavelength and emit at a longer specific wavelength (a phenomena known as a Stokes shift) For IHC utilizing fluorophores as reporters, there are two basic ways this can be accomplished: via direct or indirect labeling In direct IHC fluorescence, a fluorophore is directly conjugated to a primary antibody In indirect IHC fluorescence, the fluorophore is conju-gated to a secondary antibody, which is specific for the primary antibody Indirect IHC fluorescence can also take advantage of amplification strategies, through either multiple secondary anti-bodies binding to a single primary, or via robust amplification approaches, such as the avidin–biotin complex (ABC)

There are multiple fluorophores available for IHC applications (such as Alexa or Cy dyes, seeTable 2for Cy dyes we have success-fully multiplexed), and more recently, fluorescent quantum dot nanocrystals, which have narrower emission peaks when compared

to standard fluorophores In order to utilize fluorophores, there are certain microscopic requirements enabling proper visualization These include a very bright light source, as well as paired excita-tion/emission filter sets specific to the fluorophores employed For example, if using fluorescein, a filter set needs to provide an excita-tion wavelength (k) of 494 nm, and an emission filter needs to pass

an emission k of 517 nm Similarly, a 525 nm quantum dot nano-crystal, where emission k is a function of nanocrystal size, requires

an excitation filter that provides a k of 400 nm, and an emission filter for a k of 525 nm Choice of filter sets is an important consid-eration, as it represents a physical limitation in multiplexing capac-ity Often, the number of wavelength band passes that can be fit into the visible wavelength range will limit the number of fluorophores that can be utilized without crosstalk to 3 marker fluorophores, along with DAPI Generally, number of filter cube sets each fluores-cent microscope can accommodate is 4 or 5 Another consideration when planning any fluorescent mIHC assay is the potential for, and likelihood of, co-localization of different fluorophores In the case where co-localization occurs, complications can occur in analysis

if the co-localization causes colors to mix, as red and green might,

to provide a degree of yellow In this instance, the relative contribu-tion of red and green is extremely difficult to determine using stan-dard image analysis, and as such must be planned to achieve successful multiplexing

Indeed, fluorescent mIHC has been successfully demonstrated

in FFPE tissue in differing multiplex levels Mason et al.[10] dem-onstrated 2-plex fluorescent IHC in FFPE tonsil interrogating CD79 and PCNA, while recently, Bogusz et al.[11] interrogated active BCR signaling in diffuse large B-cell lymphoma with various 2-plex combinations of CD20 coupled with either pLYN, pSYK, or pBTK In contrast, 4-plex fluorescent IHC using quantum dots to interrogate the tumor microenvironment in gastric cancer has also been dem-onstrated[12] In all instances, indirect labeling with primary-spe-cific conjugated secondary antibodies was performed This raises

an important issue involving antibody species and the ability to

Table 1

Typical chromogens in multiplexed IHC assays.

Table 2 Sample fluorescent dyes for multiplexed IHC assays.

Fluorescent dye Excitation k (nm) Emission k (nm)

successfully multiplex, considering the constraining limits this can

impose on multiplexing capacity In general, indirect fluorescent

mIHC can only accommodate one species of primary (e.g rabbit,

mouse, or goat) for each target of interest, each accommodated

by a specific secondary antibody conjugated with a specific

fluoro-phore This represents a significant impediment should a protocol

attempt to employ all rabbit primary antibodies There are a few

strategies to overcome such limitations, but their ability to do so

without complication is limited

For example, fluorophores can be directly conjugated to

pri-mary antibodies Several studies have demonstrated protocols for

conjugating quantum dot fluorophores to primary antibodies

[13,14] These protocols are time consuming and labor intensive,

making their practicality for many labs difficult Circumventing

this approach, alternative strategies have taken advantage of serial

staining techniques, such as multi-epitope-ligand-cartography

(MELC), a fluorescent multiplex in situ strategy with the capacity

to simultaneously visualize protein expression Berndt and

col-leagues[15]examined the inflammatory status of Barrett’s

esoph-agus and esophageal adenocarcinoma in a sequential, highly

multiplexed fluorescent MELC assay To accomplish this, the MELC

assay keeps the field of view constant, and all multiplexing takes

place there As a consequence, the MELC method excludes all other

areas of the tissue sample outside the constant field of view There

is also a significant cost associated with the robotic integration

with a standard inverted microscope that must be considered

Taken together, it is apparent that such issues demonstrate hurdles

to using standard methods for direct or indirect fluorescence IHC to

achieve any power of multiplexing

There are, however, relatively novel detection strategies that

can circumvent some of the complicating factors outlined above

Tyramide signal amplification (TSA) relies on amplification of

sig-nal, which then covalently binds to the epitope in a highly specific

manner[16,17] This is ideal for fluorescent multiplexing of similar

species antibodies, as individual primary/secondary antibody

com-plexes can be removed for serial IHC, yet the covalent signal for

each remains for interrogation This method has been successfully

employed to achieve 3-plex fluorescent IHC in intratubular germ

cell neoplasias [18] using similar species primary antibodies In

an analogous study, successful 3-plex fluorescent IHC utilized

TSA in order to multiplex rabbit primary antibodies against

vaso-pressin, corticotropin releasing factor, and thyroid hormone[19]

Importantly, the TSA immunostaining protocol has been optimized

to demonstrate excellent signal to noise through reduced

back-ground[17] While TSA is quantitative, it is important to note that

it is driven by enzymatic amplification, bearing similarly with

chromogenic deposition However, unlike chromogenic deposition,

TSA is uniquely suited for serial staining, due to covalent Tyramide

binding, as noted above Importantly, this method is similar in cost

to standard chromogenic detection methods, allows for the use of

multiple similar species primary antibodies, and supports

fluores-cent mIHC assessment across an entire tissue section

2.2 Imaging approaches and systems for multiplexed detection

2.2.1 Multiplexed imaging approaches

One main reason mIHC is a sought after assay is due to the

potential it holds for informing on biology through capturing

mul-tidimensional data related to tissue architecture, spatial

distribu-tion of multiple cell phenotypes, and co-expression of signaling

and cell cycle markers But this requires imaging of multiple

bio-markers, making imaging a central component in the mIHC

work-flow Several strategies have been proposed to accomplish

imaging of mIHC, each leveraging different multiplexing and

imag-ing modalities In ‘serial sectionimag-ing’ multipleximag-ing, or feature

analy-sis on consecutive tissue sections (FACTS; [20]), multiplexing

(brightfield or fluorescent) IHC is performed on individual serial sections, which are then individually imaged Once imaged, feature extraction is performed in order to align sections for subsequent expression analysis While this approach can stain sequential sec-tions, the true multiplexing in such an arrangement is significantly limited due to the inability to capture information regarding cellu-lar co-expression and signaling cascade linkages within individual cells Perhaps more prescient is the potential to miss significant tis-sue features due to the contribution of three-dimensional shifts which can occur through sequential tissue sections

Another multiplexing approach recently described involves extraction of proteins from a single tissue section using a series

of layered membranes (L-IHC;[21]) that extract proteins by size,

on which IHC can then be performed to interrogate protein expres-sion The signal from each of the membranes is then imaged in a microarray reader, allowing for protein expression analysis [22,23] These images can then be overlaid on the original tissue section image in order to provide expression information While this approach utilizes only one tissue section, thereby taking the first step toward a comprehensive mIHC strategy, the IHC is still sequential, and as indicated such approaches are limited by the inability to capture any information regarding co-expression or signaling cascade linkages within individual cells And in instances where L-IHC may appear to offer some degree of co-expression analysis, the tight interplay of multiple tissue types (e.g epithelial, fibroblast, inflammatory/immune cells) requires proteins to be captured in situ, within specific contextual confines, in order to truly capture co-expression profiles, thus limiting the multiplexing capacity of this approach

In summary, multiplexing and imaging as described in the examples above does not adequately capture the true power of mIHC, where multiple markers are interrogated in a single section, with the tissue context completely conserved This can be achieved through brightfield IHC using multiple chromogens, or through fluorescent IHC using different fluorophores Importantly the sub-sequent image analysis is conducted to evaluate each of the vari-ous reporters Yet as outlined above, there is a limited multiplexing capacity when using chromogens, leaving fluorescent IHC uniquely positioned to offer significant improvements in mul-tiplex detection

2.2.2 Modes of multiplexing Fluorescent IHC can be accomplished as described above In this configuration, each fluorophore is separately imaged, and then images are merged together In this way fluorescent detection can provide co-localization confirmation[24], but quantitation of each signal is somewhat limited, especially in instances where autofluo-rescence is present A potential improvement in fluorescent multi-plexed IHC is possible through the use confocal laser scanning microscopy (CLSM; [25]) More specifically, analysis of tissue involving true quantitation of fluorescent mIHC could benefit from the application of spectral CLSM[26,27] This technology utilizes multiple lasers for detection, but importantly, possesses specific optics that allow for spectral detection By using spectral, or more properly, multispectral detection, the spectra of each individual marker is ‘unmixed’ allowing for isolation of the spectra of interest from each specific marker, supporting quantitative fluorescent IHC Using spectral CLSM, Pauly et al.[28]successfully analyzed autoflu-orescent macrophages in fresh lung tissue from smokers More recently, 5-plex fluorescent IHC interrogated CD20, IgD, MIB-1, CD3, and CD68 expression in tonsil via sequential IHC of multiple species primary antisera, detected with species-specific quantum dots and assessed with spectral CLSM[29]

In contrast to spectral CLSM, multispectral analysis can also be accomplished without the need for multiple laser sources, using a standard fluorescent microscope equipped with a multispectral

camera In this configuration, the spectral information from a

mul-tiplexed panel of targets is captured through the multispectral

camera In order for the spectral information to be reliably

unmixed and quantitated, correct examples of each fluorophore

emission spectra, as well as a representative autofluorescence

spectrum from an unstained sample, in the context to be used,

must be registered in a multispectral library (Fig 1) This spectral

library forms the cornerstone of target quantitation, as the

inten-sity of each fluorescent target is extracted from the multispectral

data using linear unmixing[30]

Beyond microscopy-based multispectral imaging, recent

advances have been made through mass spectrometry-based laser

microscopy (MIBI) approaches, recently referred to as ‘next-gen

immunohistochemistry’[31] In this approach, lanthanide

metal-conjugated primary antibodies are used for interrogation in a man-ner consistent with standard IHC From there, samples on slides are subjected to mass spectrometry In one instance[32]mass cytom-etry was performed, wherein a high-resolution laser ablation sys-tem was employed to image 32 proteins at a cellular resolution

of 1lm In a somewhat different mass spectrometry approach [33], a rasterizing oxygen duoplasmatron ion beam was employed

to sweep breast tissue samples, liberating ions from lanthanide adducts This system was used to analyze human FFPE breast can-cer samples via a 10-plex assay Both studies produce multicolor composite images and quantitative capacity in a dynamic range perhaps double that of quantitative immunofluorescence While the methodologies are complex, the potential for mass-spectros-copy is developing greater practical demonstration

Fig 1 Spectral library from an mIHC assay Individual examples of ER (fluorescein), PR (A594), Her2 (Cy5), ki67 (Cy3), CK (coumarin), and DAPI, as well as autofluorescence from an unstained section, were spectrally analyzed to generate a spectral library in support of multispectral unmixing in a multiplexed assay For each epicube in the system, the spectra observed are captured, so that across all epicubes, the complete spectral properties of each independent signal can be effectively utilized for spectral unmixing.

2.2.3 Staining strategies for multiplexing

For each of the multiplexing modes above, various and specific

staining strategies can be employed to influence not only the

indi-vidual degree of multiplexing capacity, but also the degree of

com-plex tissue analysis For example, it has been recently demonstrated

that iterative staining on the same defined region of interest can

produce significant multiplexing beyond 60 targets[34] In this

analysis, fluorescent IHC was performed on a tissue sample, and

the slide was then imaged After imaging, the slide was removed,

and the fluorescent tag was inactivated through incubation in an

alkaline H2O2 buffer Another round of fluorescent IHC was then

performed for the detection of another target, and again, the slide

was returned by the stage to the previous location where the first

image was captured and another image was acquired, after which

the fluorescent reporter was inactivated and the process repeated

For all images, DAPI nuclear counterstain guided a Fourier-based

alignment In this scheme, the spectra by which each of the

fluores-cent intensities were measured results from the broad spectral

pro-files captured from each of the emission cubes of the system,

subtracted from any autofluorescence contributed by an unstained

sample This is the main reason this system functions like a standard

fluorescent system, but with higher potential multiplexing capacity

However, given the complexity of this approach, and its limited field

of view on a tissue section, there will probably be many challenges

to translating it into a practical and economical pathology

work-flow Capturing a fully representative sampling of heterogeneity

across whole sections is very important, as different regions of a

tumor can contain very significant differences of phenotype[35]

A limited area of analysis, as described above, can be contrasted

with a more stringent sampling paradigm where multiplexing can

be performed across an entire tissue section containing multiple

regions of interest (e.g the entire visible tumor) For instance, it

would be relevant when interrogating a tumor that possesses

any degree of multiplexed staining to sample the entire tumor area

and capture the complete extent of tissue and cellular

heterogene-ity, as was recently demonstrated in prostate cancer, with a

multi-plex fluorescent quantum dot panel consisting of RANKL,

phospho-NF-jB-p65, VEGF, phospho-c-Met, and Neuropillin-1[36] In this

study, the authors employed an automated morphology tool

(InForm, PerkinElmer, Waltham, MA) to accurately identify all

tumor regions at 4 magnification Once the entire section was

‘segmented’ the entire tumor area was imaged at 10 nm k from

420 nm to 720 nm using the Vectra spectral imaging system

(Perk-inElmer, Waltham, MA) Each of the images captured was then

spectrally unmixed in order to effectively quantitate each of the

biomarkers, as well as autofluorescence, based on their specific

spectral properties In order to capture all elements of tumor

het-erogeneity, the authors sampled from five individual fields and

found that activated c-Met signaling was more pronounced in

cas-trate-resistant prostate cancer compared with

hormone-respon-sive prostate cancer While the degree to which multiplexing is

achieved differs from the example of Gerdes et al.[34]above, there

are several advantages to a multispectral, multiplexed approach

Perhaps foremost the perfect pixel-registration of component label

planes and high-resolution imagery enabled by having all labels in

place and imaged simultaneously, and the accurate removal of

autofluorescence signal from label signals Also, by enabling whole

slide sampling, this approach fits more seamlessly within the

stan-dard pathology workflow, capturing an important element of

pathology-based mIHC imaging

2.3 Software for analyzing multiplexed images

2.3.1 Quantitation of chromogenic and fluorescent signals

Thus far, the focus has been on the ways in which mIHC can be

performed, and how images are captured However, the ultimate

goal is reliable and specific data to support research, and ulti-mately to advance clinical care This requires image analysis tools that can reliably identify and quantitate all targets of interest In brightfield mIHC, signals are chromogenic and provide color and contrast through absorbing different regions of the visible spec-trum Unmixing requires conversion of transmitted light signals

to optical density and then linear decomposition based on charac-teristic absorption spectra In an examination of rabbit liver tis-sues, a 3-plex IHC panel consisting of cytokeratin, CD31, and Ki67 [37]was successfully interrogated using InForm for multi-spectral image (MSI) analysis For each sample, the chromogens used to visualize cytokeratin, CD31, and Ki67 were Vector Red, DAB, and Vector Blue, respectively For analysis, cells were seg-mented, and Ki67 positivity was determined using a threshold of average optical density of the unmixed ki67 chromogen Interest-ingly, the determined Ki67 positive cells nearly mimicked that number of positive Ki67 cells quantified by two independent pathologists, supporting the successful unmixing of the Ki67 sig-nal It is worth noting the choice of chromogen for Ki67 – Vector Blue One of the practical challenges of using chromogenic mIHC remains the ability to accurately quantitate each chromogen, when optical densities are at levels typically used for visual assessment The useful range of optical densities to support reliable unmixing leads to staining ‘intensity’ substantially lighter than done typi-cally[38] There is significant debate regarding the usefulness of DAB as a quantifiable chromogen within any multiplex arrange-ment[4,38,39] Indeed, as above, DAB was used to generate a land-mark of CD31 expressing sinusoid cells to help localize Ki67 positive cells within both hepatocyte tissue and non-parenchymal tissues of the sinusoidal space

In contrast, fluorescent signals are more amenable to quantita-tion given their linear and additive nature and relatively well-defined emission spectra Indeed, fluorescein intensity falls propor-tionally with a reduction in the antigen on which it was reporting [40] With such linear responsiveness, quantitation of fluorescent intensities in a fluorescent mIHC context has the potential to advance tissue interrogation when coupled with both multispec-tral instrumentation and the image analysis tools to support proper unmixing of captured spectra An example of the effective-ness of fluorescent mIHC can be seen in the relative expression pat-terns of CK18, CD34, and cleaved Caspase 3, assessed in FFPE human tonsil [41] Each marker was interrogated using specific fluorescent quantum dots, and the multispectral images captured were subsequently analyzed using InForm By quantitating the fluorescent intensity for each marker on a per-pixel basis, the anal-ysis revealed a co-localization pattern where 12% of the total CD34 was co-localized with CK18, while only 1% of the total CK18 was co-localized with CD34 This analysis is made possible as a result

of fluorescent intensity quantitation, and demonstrates the signif-icant potential of multispectral analysis in supporting multiplexed IHC studies

In addition to InForm, other image analysis software packages can support mIHC analyses For example, Definiens image analysis software (Definiens, Carlsbad, CA) has been used in conjunction with a multispectral imaging system (Nuance, PerkinElmer) to ana-lyze CK18, AMACR, and AR expression in prostate cancer[42] In this study, the imaging software was used to segment the tissue samples based on CK18 staining, and subsequent intensity analy-ses allowed determination of AR+ and/or AMACR+ epithelial cells

It was subsequently demonstrated that AR expression in AMACR negative epithelial cells is highly correlated with prostate-specific antigen (PSA) recurrence Yet another imaging software package, Precision (Leica, Butlers Grove, IL) can also report on mIHC, by pro-viding tissue segmentation and intensity measurement capabili-ties But unlike InForm or Definiens, Precision image analysis software, which is often paired with Aperio Scanscope (Leica), does

not possess multispectral analysis capacity, thus limiting its

quan-titative potential

2.3.2 Image source: monochrome, RGB, and multispectral unmixing

In order to effectively analyze captured images, the software

employed must be able to extract fluorescent or chromogenic

intensity information from the image This is, of course, completely

depended on the type of image captured For example, in

mono-chromatic imaging, utilizing a monomono-chromatic camera, there is

only one color represented, in varying intensities Thus, for the

software, there is only one color that can be extracted and analyzed

for each filter cube available This significantly limits any

fluores-cent mIHC assay, as there is no effective way for the imaging

soft-ware to separate the various reporter fluorophores that share

similar excitation properties, but differ in emission spectra

Additionally, there is no effective way to isolate the varying

contri-bution of autofluorescence[43]present in FFPE tissues

In contrast to monochromatic images, a standard three channel

RGB camera extends the color detection capacity, providing a

greater complexity of color for the imaging software In this

con-text, images captured by an RGB camera are an amalgamation of

the three available channels (red, green, and blue) This can be

ben-eficial to a mIHC assay if the fluorescent signals fall discretely

within each of the individual detection channels, and by

maximiz-ing the filter cube sets employed However, this represents certain

imaging limitations that are difficult to overcome As noted above,

nearly all FFPE tissue sections retain a varying degree of tissue

autofluorescence, and in fluorescent IHC, this can be spread

heter-ogeneously over significant portions of the fluorescent spectra (see

Fig 1) So with three channels over which autofluorescence could

distribute, it is difficult to separate and distinguish this signal from

the signal of interest with any imaging software suite that could

extrapolate from the RGB data Use of filter cubes can improve

multiplexing capacity, but the limited channels available in an

RGB context are potentially subject to spectral overlap, which the

imaging software cannot unmix effectively

Unmixing by the imaging software is central to supporting the

promise of mIHC, and perhaps the most efficient way in which to

analyze any mIHC assay is by MSI analysis Unlike an RGB camera,

a multispectral camera can be configured to capture discrete

inter-vals (P10 nm) across the entire visible spectrum from 420 nm to

720 nm, through the utilization of a liquid crystal tunable filter

[7] This ability to capture discrete spectral intervals allows for

tai-loring of the spectra to accommodate partitioning of the specific

signals expected, i.e the emission spectra of the fluorophores

mak-ing up an mIHC assay (seeFig 1), including the autofluorescent

spectra[43] Consequently, the data from the multispectral camera

can be accessed by the imaging software (e.g InForm or Definiens),

and as a result of the discrete spectral intervals, unmixing of the

combined spectra into the individual spectra associated with the

fluorophore components of the mIHC panel is possible

2.3.3 Guides to tissue analyses: visual, semi-automated, and

automated

In devising a strategy to analyze most clinical tissue samples, a

priori decisions are needed about target contexts, such as epithelial

or stromal components of tumor Once decided, the next question

leads to how such tissue segmentation can be accomplished

Perhaps most basically, this can be accomplished through visual

assessment, requiring both the time and efforts of an experienced

pathologist While this is very common in pathology practice, it has

specific limitations in the context of fluorescent mIHC One of the

basic impediments results from the lack of familiarity pathologists

have with fluorescence imagery, which does not render tissue

architecture nearly as well as H&E or chromogenic IHC In the case

of chromogenic mIHC, overlapping colors become essentially dark

brown, making it very difficult to visually interpret stain intensity where stains are co-localized More generally, the human eye is poorly suited to assessing stain intensity, having evolved to assess patterns and very weak signals rather than intensity These issues significantly limit visual assessment of mIHC assays[44,45] Alternatively, in semi-automated analysis, the imaging software presents the images captured from each tissue sample processed for mIHC to a trained pathologist to indicate, typically through annotations drawn on the image with the mouse, areas to be ana-lyzed In this context, the pathologist individually traces the areas

to be analyzed using the images of all processed tissue samples While still somewhat labor intensive, this allows for the appropri-ate regions to be segmented for image analysis The imaging soft-ware then restricts the multispectral unmixing to the segmented regions of interest, thus getting the benefit of objective, consistent quantitation of signal intensities provided by image analysis This method, while still requiring significant efforts of a pathologist, does allow for multispectral unmixing of the fluorophores of inter-est within the regions of interinter-est specified for analysis, overcoming the limitations of the human eye in the context of mIHC

But perhaps the most efficient method for tissue segmentation

is a fully automated analysis[5] In this instance, the imaging soft-ware is trained to recognize particular regions of interest (e.g tumor) at a lower power (4) ‘survey’ of the whole slide, based

on spectral elements associated with tissue morphology (e.g., auto-fluorescence spectra, DAPI spectra, etc.) Training is performed by having a trained individual manually select regions of tissue of interest and of no interest, in a series of images that represent the full range of tissue contexts the software will need to correctly segment, based on the needs of the mIHC assay (Fig 2) With the software trained to identify tissue of interest, the slide imaging system can then take high magnification imagery of areas of inter-est for detailed quantitation and characterization

The same machine learning software can then identify cells and cellular sub-compartments in the higher power magnification to automate data collection within the specific region(s) of interest For all multispectral images, the software analyzes fluorescent intensity (fluorescent units – FU) on a per-pixel basis, and then averages these intensities to create compartment intensities (e.g nuclear, cytoplasmic, or membranous) for each cell within the seg-mented region Once the higher magnification images have been segmented, a trained pathologist reviews the segmentation maps

to ensure fidelity with the intended tissue segmentation strategy Then, by examining compartment intensities, the pathologist can set threshold values to support downstream analyses The typical time spent by a pathologist reviewing results is significantly less than the time required to manually annotate and score images for analysis Once segmentation is deemed satisfactory, the imaging software data collected from regions of interest are aggregated as detailed above This strategy is a very efficient way in which to per-form MSI analysis of mIHC, and importantly, while this method still requires pathology oversight, it can decrease the pathologist work-load, resulting in a beneficial multispectral multiplexing strategy 2.3.4 Tumor and landmark markers for automation

While automated tissue segmentation can be of great value for analysis of mIHC, there are certain considerations that must be made to aid in effective tissue segmentation In general, the ability

to separate tumor form non-tumor tissue is a straightforward exer-cise for a trained pathologist using a standard H&E But there are many times when additional IHC must be performed to aid in the diagnosis[46] In the same way, certain IHC stains can be per-formed which aid in tissue segmentation by establishing biological landmarks indicating specific cellular attributes

There are certain IHC stains that can be performed to indicate cells of an epithelial origin, such as cytokeratin This stain effectively

highlights all epithelial cells within any tissue sample, and

impor-tantly, this stain can help to accurately guide tissue segmentation

of tumors with epithelial origin, such a prostate, lung, and breast

Such landmark stains can work in conjunction with a nuclear

counterstain, such as DAPI, so as to help guide subcellular staining

patterns and pinpoint each individual cell Thus, when these

land-mark stains are coupled with fluorescent mIHC, they significantly

aid in tissue segmentation and subsequent image analysis, but it

must be remembered these stains also increase the multiplexing

by adding two additional fluorophores (for both DAPI and

cytoker-atin) to the mIHC protocol Nevertheless, these stains vastly

facilitate automated image analysis

2.3.5 IHC controls guiding assay repeatability

It is important when considering assay reproducibility that

strategies are in place to ensure successful multiplexing is

achieved for every assay performed [47] One of the principle

reproducibility strategies often employed is the use of known,

pre-viously characterized positive and negative samples (positive and

negative controls) that can be consistently assessed in successive

multiplexed assays in order to confirm assay reproducibility[46]

Additionally, this can also be accomplished with known internal

controls that are concomitantly positive An example of such a

strategy is demonstrated by cytokeratin, which stains both normal

and carcinogenic epithelial cells Regardless of how such control of

the IHC process is achieved, it is a crucial element that should not

be overlooked By ensuring assay reproducibility, a mIHC assay will

be much more easily employed, and potentially integrated into a

clinical workflow

2.4 Integrating multiplex staining and multispectral imaging in the

pathology workflow

2.4.1 Supporting, facilitating, and extending the critical role of the

pathologist

As indicated in the examples of tissue segmenting above, the

pathologist plays an integral role in the mIHC assay, while imaging

systems with the capacity to inform on mIHC assays offer signifi-cant support For example, multiplexing offers deeper insight into tissue and cellular processes, aiding diagnostic potential In addi-tion, the ability to multiplex also conserves limited pathology tis-sue resources [48] And perhaps most salient, mIHC offers the potential to extend the role of the pathologist and fill a gap in cur-rent practice between genomics on one side, and classic histopa-thology on the other[3]

2.4.2 Simulated brightfield views of fluorescence samples Another important advance afforded by the imaging software, that can support the pathology workflow is the ability to take fluo-rescent IHC images, which in and of themselves often lack suffi-cient context to be useful to the pathologists eye, and create simulated brightfield images Since all images retain multispectral information, elements of the spectra can be used to simulate vari-ous brightfield views, such as H&E, or hematoxylin with DAB chro-mogen (Fig 3) With this view, data extrapolation is wholly supported by MSI analysis, yet classic pathology views can be pro-vided, enhancing understanding of complex multiplexed staining patterns, and improving the presentation of mIHC assays 2.4.3 System training, automation, and results review

In order to effectively integrate mIHC and MSI into the work-flow, it’s important that all workflow elements cooperate to achieve the desired results For this to occur, and subsequently enhance the pathology workflow, efficient and effective tissue seg-mentation training is required (seeFig 2) By introducing sufficient examples of the tissue types which would be encountered by the imaging software, tissue segmentation can be leveraged in the imaging software to effectively identify specific patterns in the tis-sues, and segment areas based on training inputs, allowing auto-matic segmentation of all remaining samples Once all samples are segmented, the pathologist is able to review the images to ensure proper tissue segmentation, and the MSI analysis can then generate intensity measurement for all samples This data is then available to support many quantitative aspects, such as specific

Fig 2 Automated tissue segmentation Fluorescent multiplexed IHC (Panel A, see Section 5 ) of breast tissue stained with ER (green), PR (purple), Her2 (red) and Ki-67 (yellow) is used to train InForm image analysis for tissue segmentation (Panel B) Once segmentation is verified (Panel C), a spectral composite of the multiplexed IHC array is created (Panel D), based on the multispectral unmixed spectra for each fluorescent probe Bar in A – 100lm.

cellular compositions, target co-localization, and positive staining

percentages This spectrally unmixed data is contextually rich,

allowing any number of analytical questions to be adequately

addressed

3 Results and discussion

The expression of estrogen (ER), progesterone (PR) and human

epidermal growth factor 2 (Her2) receptors are important

bio-markers used to classify breast cancer subtypes[49], and the IHC

classification of each of these markers in individual samples

corre-lates well with gene expression The ability to assess these markers

within the context of the tumor is becoming increasingly more

necessary, as the importance of tumor heterogeneity is more

widely recognized[50] Given the prognostic importance of these

biomarkers their assessment is critical in an era of improved

detec-tion where shifts in biopsy are emerging as new standards of care

[51], demonstrating overall improvements in breast cancer care

Yet this also suggests that tissue sample availability for biomarker

analysis may be increasingly limited[48] This led us to assess a

novel mIHC approach that could provide accurate assessment of

ER, PR, Her2 and Ki67 expression To this end, we evaluated ER,

PR, Her2, and Ki67 in human breast cancer using a novel mIHC

pro-tocol called Opal, which allows for the detection of similar species

antibodies in a single section Expression profiles of each marker

were evaluated alone in a singleplex assay, or comingled within

a multiplexed assay, in order to verify the utility of this novel mIHC

approach

3.1 Opal mIHC method

The Opal method utilizes individual TSA conjugated

fluoro-phores to detect multiple targets within a mIHC assay This

requires initial assessment of each target in a singleplex IHC assay

to generate spectral libraries required for mIHC image analysis (see

Fig 1) Once each target has been optimized as a singleplex, the

Opal multiplexed assay is performed utilizing an iterative staining

method (Fig 4) Using this method offers the advantage of

multi-plexing similar species primary antibodies without species

cross-reactivity, due to the antibody stripping protocol (microwave

treatment), which removes both the primary and the

HRP-conju-gated secondary When performed manually to achieve a 5-plex

assay (ER, PR, Her2, Ki67 and CK), the Opal method can be

performed over two days While this may require more time than

individually stained samples, when the batching sizes increase (e.g

20 samples), the Opal method takes perhaps less time than

per-forming 5 stains on 20 slides (100 total assays) Another

consider-ation worth noting is the ability to increase the numbers of targets

within a multiplexed assay Our group has successfully

multiplexed 6 biomarkers following this protocol (data not shown), demonstrating to a degree the practical scalability of this method Additionally, most of the assays our group has performed have focused on phenotypic cell markers While we have not examined biomarkers whose expression may be more variable, it

is likely that even they can be interrogated with this method For any expressed biomarker, thresholds for staining can be assigned (as positive or negative) or in instances where threshold values are not assigned, the data can be interrogated as a continuous var-iable, which would support a potentially greater understanding of

a biomarkers specific expression pattern

In practice, TSA supports the Opal method very well when com-bined with multispectral image analysis Through covalent bind-ing, the TSA conjugated fluorophores remain bound to the targeted epitope, allowing for sequential IHC to achieve the higher multiplexing levels of Opal Evaluation of staining with each TSA fluorophore revealed a very favorable signal-to-noise ratio, with

no aberrant background staining, as well as very specific compart-mental staining (e.g nuclear or membranous) consistent with accepted biological roles It is important to note that such staining requires diligent optimization Antigen retrieval, as performed using microwave technology, requires optimization to ensure both proper antigen retrieval and endogenous HRP quenching, all while ensuring complete antibody striping and tissue viability In addi-tion, properly balanced HRP concentrations are required to prevent TSA dimer formation, typically achieved through titration of pri-mary antibodies, though also can be modified through titration

of the secondary antibody

3.2 Opal sensitivity and interassay reproducibility

To assess Opal sensitivity, the expression of each target was assessed within singleplex and multiplex contexts in order to eval-uate potential alterations in a multiplexed arrangement For each target, the mean intensity was calculated for ER, PR, Her2, and Ki67 staining, and correlational analyses were performed to com-pare the levels of target expression alone versus its expression in

a multiplexed assay In this instance, mean intensity is a surrogate for positivity calls For ER, analysis of mean intensity alone and multiplexed revealed a significant correlation (Fig 5, R2= 0.8320, Pearson r = 0.9122, p < 0.0001) Similarly, comparison of the mean intensity of PR staining (Fig 6, R2= 0.6211, Pearson r = 0.7881,

p < 0.0001), Her2 staining (Fig 7, R2= 0.9884, Pearson r = 0.9942,

p < 0.0001), and Ki67 staining (Fig 8, R2= 0.7720, Pearson

r = 0.8786, p < 0.0001), alone and multiplexed, also revealed signif-icant correlations For all markers, staining profiles indicated that each marker performed similarly, whether alone or in combination within a multiplexed panel, highlighting the potential benefits of mIHC analyses

Fig 3 Simulated IHC images Fluorescent multiplexed IHC, examining the expression of ER (green), PR (purple), Her2 (red) and Ki-67 (yellow) in breast cancer (Panel A, see Section 5 ), was spectrally unmixed using InForm Subsequent simulated H&E (Panel B) and simulated hematoxylin and DAB indicating Her2 expression (Panel C) images were generated to create classic pathology views Bar in C = 50lm.

Fig 4 Opal mIHC method The Opal mIHC assay protocol is very similar to a standard IHC assay, with the notable exception being the iterative nature of the protocol enabled

by a post-visualization microwave step in a citrate buffer (similar to that used for antigen retrieval and endogenous HRP quenching) This allows for a simplified multiplexing strategy, where primary antibodies can be chosen based on performance, and not on species.

Fig 5 Analysis of ER expression in multiplex and singleplex contexts To assess expression of ER within a multiplexed context (ER, PR, Her2, and Ki67 combined spectral composite, panel A), ER was isolated for fluorescent intensity (FU) measurement (spectral composite of ER from multiplex, panel B) ER intensity was also assessed in a sister serial section as a singleplex (panel C) Analyses of average ER fluorescent intensities for all cases (N = 31) between single and multiplexed contexts demonstrated a highly significant correlation (R 2

= 0.8320, Pearson r = 0.9122, p < 0.0001, panel D) Bar in C = 100lm.

To assess interassay variability in the Opal multiplexed

stain-ing method, two serial stained breast cancer TMA sections were

evaluated, with the mean intensity calculated for ER, PR, Her2,

and Ki67 staining Again, mean intensity serves as a surrogate

for positivity calls Correlational analyses were performed to

compare the levels of target expression between each multi-plexed assay, and for all targets, the correlations were highly sig-nificant (Fig 9) For ER, analysis of mean intensity between each multiplexed section revealed a significant correlation (Fig 9A,

R2= 0.9451, Pearson r = 0.9722, p < 0.0001), while analysis of

Fig 6 Analysis of PR expression in multiplex and singleplex contexts To assess expression of PR within a multiplexed context (ER, PR, Her2, and Ki67 combined spectral composite, panel A), PR was isolated for fluorescent intensity (FU) measurement (spectral composite of PR from multiplex, panel B) PR intensity was also assessed in a sister serial section as a singleplex (panel C) Analyses of average PR fluorescent intensities for all cases (N = 33) between single and multiplexed contexts demonstrated

a significant correlation (R 2

= 0.6211, Pearson r = 0.7881, p < 0.0001, panel D) Bar in C = 100lm.

Fig 7 Analysis of Her2 expression in multiplex and singleplex contexts To assess Her2 expression within a multiplexed context (ER, PR, Her2, and Ki67 combined spectral composite, panel A), Her2 was isolated for fluorescent intensity (FU) measurement (spectral composite of Her2 from multiplex, panel B) Her2 fluorescent intensity was also assessed in a sister serial section as a singleplex (panel C) Analyses of average Her2 intensities for all cases (N = 34) between single and multiplexed contexts demonstrated a highly significant correlation (R 2

= 0.9884, Pearson r = 0.9942, p < 0.0001, panel D) Bar in C = 100lm.