ARTICLE
Received 7 Dec 2016 | Accepted 27 Feb 2017 | Published 27 Apr 2017
Julienne L. Carstens1,*, Pedro Correa de Sampaio1,*, Dalu Yang2,3, Souptik Barua2,3, Huamin Wang4, Arvind Rao2,3,5, James P. Allison6, Valerie S. LeBleu1 & Raghu Kalluri1,7,8
The exact nature and dynamics of pancreatic ductal adenocarcinoma (PDAC) immune composition remains largely unknown. Desmoplasia is suggested to polarize PDAC immunity. Therefore, a comprehensive evaluation of the composition and distribution of desmoplastic elements and T-cell inltration is necessary to delineate their roles. Here we develop a novel computational imaging technology for the simultaneous evaluation of eight distinct markers, allowing for spatial analysis of distinct populations within the same section. We report a heterogeneous population of inltrating T lymphocytes. Spatial distribution of cytotoxic Tcells in proximity to cancer cells correlates with increased overall patient survival. Collagen-I and aSMA broblasts do not correlate with paucity in T-cell accumulation, suggesting that PDAC desmoplasia may not be a simple physical barrier. Further exploration of this technology may improve our understanding of how specic stromal composition could impact T-cell activity, with potential impact on the optimization of immune-modulatory therapies.
1 Department of Cancer Biology, Metastasis Research Center, The University of Texas MD Anderson Cancer Center, Houston, Texas 77030, USA.
2 Department of Bioinformatics and Computational Biology, The University of Texas MD Anderson Cancer Center, Houston, Texas 77030, USA. 3 Department of Electrical and Computer Engineering, Rice University, Houston, Texas 77005, USA. 4 Department of Pathology, The University of TexasMD Anderson Cancer Center, Houston, Texas 77030, USA. 5Department of Radiation Oncology, The University of Texas MD Anderson Cancer Center, Houston, Texas 77030, USA. 6Department of Immunology, The University of Texas MD Anderson Cancer Center, Houston, Texas 77030, USA. 7Department of Bioengineering, Rice University, Houston, Texas 77005, USA. 8Department of Molecular and Cellular Biology, Baylor College of Medicine, Houston, Texas 77030, USA. * These authors contributed equally to this work. Correspondence and requests for materials should be addressed toR.K. (email: mailto:[email protected]
Web End [email protected] ).
NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications 1
DOI: 10.1038/ncomms15095 OPEN
Spatial computation of intratumoral T cells correlates with survival of patients with pancreatic cancer
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095
The immune contexture of pancreatic adenocarcinoma (PDAC) is often considered immunosuppressive in nature, with minimal antitumour T-cell inltration1.
However, PDAC presents with the inherent capacity to activate a T cell-mediated antitumour response26, and patients with PDAC possess tumour reactive memory T cells resident in their bone marrow2. A study has also shown that T cells are the dominant immune component found in the stroma of primary tumour samples obtained from PDAC patients3 and patients with higher levels of CD4 and/or CD8 T cells have signicantly prolonged survival46. Nonetheless, PDAC is considered to develop an immunosuppressive microenvironment that restricts antitumour T-cell inltration1,7,8. This may, in part, result from the proposed role of activated broblasts or myobroblasts and the extracellular matrix in PDAC. These major constituents of PDAC desmoplasia have been hypothesized to sequester T cells away from cancer cells5,9. Recent studies in mice also suggest that focal adhesion kinase activity in cancer cells mediates an inverse correlation between brosis in the desmoplastic stroma and T-cell inltration in PDAC10. While these mouse studies suggest that PDAC desmoplasia might act as a barrier for T-cell inltration5,9,10, promising early results seen with T-cell vaccines (reviewed in ref. 8) provide evidence that T cells have the capability to inltrate the PDAC microenvironment. Regulatory T-cell (Treg) inltration within the PDAC stroma is observed adjacent to cancer cells, providing additional evidence for the existence of a complex regulation of T-cell inltration as a part of the evolving PDAC desmoplasia11. The exact nature of the complex interaction between desmoplastic brotic stroma and T-cell inltration and its impact on PDAC patient prognosis and overall survival remains to be elucidated.
The function of PDAC-inltrating T cells may be attenuated by the co-inltration of immune suppressive cells, such as Treg cells, or myeloid-derived suppressor cells and M2 macrophages3. The abundance of these cells correlates with poor tumour differentiation and/or survival in preclinical and clinical studies3,1216. These observations offered support for the development of clinical efforts to target these immune cell populations using GVAX (a granulocyte-macrophage colony-stimulating factor gene-transfected tumour cell vaccine) or agonistic CD40 antibodies. The survival benets of these strategies are lacking in PDAC preclinical models without T cells and diminished in patients with low T-cell numbers17,18. Furthermore, the antitumour efcacy of these therapies is best realized in the presence of endogenous antitumour T cells, evidenced by the combination with immune-checkpoint blockade therapies (anti-PD-1, -PD-L1 and/or CTLA4 (cytotoxic T-lymphocyte-associated protein 4)) enhancing their antitumour efcacy15,16,19. These studies suggest that modulation of the immune composition in PDAC, in particular T cells, may offer clinical benet to control and suppress PDAC progression. To harness such clinical benet, a better understanding of the dynamic PDAC immune composition is essential.
The exploration of the microenvironmental composition of treatment-naive PDAC samples would offer critical insights into the complex and heterogeneous immune landscape associated with the growth and progression of this tumour. We thus set out to query the desmoplastic, mesenchymal and lymphocytic contexture of resected human PDAC tissue samples obtained from patients who did not receive neoadjuvant therapies. We probed formalin-xed, parafn-embedded (FFPE) tissue sections using a novel tyramide signal amplication (TSA) multiplexing technique to enable the simultaneous examination of eight distinct markers. The abundance and spatial organization of aSMA, Collagen-I, cytokeratin 8, CD3,
CD8, CD4 and Foxp3 immunolabelled cells (nuclei labelled with
4,6-diamidino-2-phenylindole (DAPI)) were studied along with clinical features to carefully annotate the aforementioned stromal elements and their correlation with patient survival. Our study shows that distinct T-cell subpopulations inltrate PDAC with specic spatial distributions. We also observe that stromal broblasts and type I collagen (Collagen-I) do not serve as absolute inhibitors of T-cell inltration.
ResultsHeterogeneous T lymphocyte subpopulations inltrate PDAC. We developed a novel multiplex immunolabelling protocol based on TSA, using Opal uorophores (Supplementary Fig. 1A), which allowed for the simultaneous evaluation of eight markers in a single FFPE tissue section. Multispectral imaging was applied to the eight-marker-stained samples. This entailed the capturing of an image every 10 nm through the full emission spectrum of each lter cube (DAPI, uorescein isothiocyanate (FITC), Cy3, Texas Red and Cy5, Supplementary Fig. 1B), which were then combined into one image cube per eld of view (henceforth termed raw images, Fig. 1a). A spectral signature for each uorophore was obtained using the same multispectral imaging protocol on single stained slides, as well as a no primary control slide to obtain the autouorescence signature of the tissue. These spectral signatures were then used to separate the raw image cube into its individual uorophores, in a process termed spectral unmixing (Supplementary Fig. 1 and Fig. 1bj). We used this technology to probe human and mouse PDAC tissue samples for multiple combinations of stromal markers (Fig. 1 and Fig. 2). As anticipated, a complex and heterogeneous tumour stroma was noted, which included mesenchymal and T lymphocytic components with varying abundance and distribution (Fig. 1 and Fig. 2). In order to study the inltration of different T-cell subpopulations in PDAC and their potential interactions with the mesenchymal stroma, we focussed on the following set of markers: alpha-smooth muscle actin (aSMA), Collagen-I, cytokeratin 8,
CD3, CD8, CD4, Foxp3, and DAPI (nuclear stain) (Fig. 1 and Supplementary Fig. 1). Multispectral imaging followed by spectral unmixing allowed for the simultaneous evaluation of all markers in each tissue sample (Fig. 1bj). A comparison of the unmixed images from multiplex stained tissues with tissue sections individually stained for each marker demonstrated the efcacy of the spectral unmixing algorithm (Supplementary Fig. 1C). The spectrally unmixed images were then analysed to identify different cellular phenotypes, based on the aforementioned markers as well as cellular size and morphology (phenotyping) (Fig. 1k,l). The marker CD3, in addition to the shape of the nucleus, was used to identify all T cells. Subpopulations of T cells were then dened by the presence and absence of three additional markers: CD8, CD4, and Foxp3. For the purpose of this study, cytotoxic T cells were dened as CD3 CD8 CD4 Foxp3 ,
CD4 effector T cells (CD4 Teff) as CD3CD8 CD4 Foxp3 , Treg cells as CD3 CD8 CD4 Foxp3 and any
CD3 cells negative for the three other markers were dened as other T cells. Cytokeratin 8 was used to identify epithelial cancer cells in tumour samples and benign pancreatic ductal cells in uninvolved tissue samples. Pancreatic acinar cells were observed in the uninvolved pancreatic tissue and in a small percentage of tumour tissue. These cells were negative for all markers but were distinguishable from other populations based on morphology, as detected through the autouorescence signal. We dened this population as normal, to reect their non-transformed histological nature. Finally, all other cells not dened in our phenotyping categories (that is, myobroblasts, blood vessels, nerves, pancreatic islets, macrophages and so on) were grouped into one category, labelled as other. Representative
2 NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095 ARTICLE
a b
c d e f
g h i j
k
l
Phenotype
Markers
Description
Tumour (Cytokeratin 8+)
CK8+CD3
Irregular nuclei and cell shape
Normal
Negative for all markers, CK8 epithelium and/or acini
Other
CK8CD3
CK8CD3
CD3+CD8+
Non-epithelial morphology (blood vessels, myofibroblasts, etc), includes SMA+ cells
Cytotoxic T cell
Small round shape, circular ring around nucleus
CD4+ Effector T cell
CD3+CD4+FoxP3
Small round shape, circular ring around nucleus
Regulatory T cell
CD3+CD4+FoxP3+
Small round shape, circular ring around nucleus
Other T cell
CD3+
Small round shape, circular ring around nucleus, not expressing markers CD4 or CD8
Figure 1 | Opal eight-colour multiplex analysis of human PDAC identies unique cellular subpopulations. (a,b) Representative images displaying the same TMA core after multispectral imaging (raw image (a)) and after spectral unmixing (composite image (b)). (cj) Enlarged subsection of the core highlighted in b, showing each of the individual markers in the composite image after spectral unmixing, together with the DAPI nuclear marker (pseudocoloured blue) and the autouorescence signal (pseudocoloured black); (c) all markers, (d) cytokeratin 8 (cytoplasmic, labelled with FITC, pseudocoloured green), (e) CD8 (membrane, Opal10, pseudocoloured orange), (f) CD3 (membrane, Opal9, pseudocoloured cyan), (g) Foxp3 (nuclear, Cy3, pseudocoloured white), (h) CD4 (membrane, Cy5, pseudocoloured magenta), (i) aSMA (cytoplasmic, Cy5.5, pseudocoloured red) and (j) Collagen-I (extracellular, Coumarin, pseudocoloured teal blue). (k) Cell phenotype map identifying the cell populations dened by the individual markers in the multiplex stain, overlaid on the raw image. (l) Summary of each dened cell phenotype, colour code and associated markers. All scale bars equal 100 mm.
images of all analysed cellular phenotypes are shown in Supplementary Fig. 2.
Tissue microarrays (TMAs) comprised of tumour and uninvolved tissue samples obtained from 132 PDAC patients were used for phenotyping (Supplementary Fig. 3). Whenever
available, two TMA tumour tissue cores collected from different FFPE blocks were analysed and the combined percentage of cell counts were calculated per patient. Cores were only excluded if no analysable tissue were present and not on the account of heterogeneity between the cores, as is common, in order to
NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications 3
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095
Mouse
HumanCK8 Collagen-I SMA CD31 CD4 CD8 Foxp3 DAPI
CK8 Ki-67 CD31 FSP1 SMA Collagen-I DAPI
Figure 2 | Applications of multiplex staining protocols in PDAC tissue sections. Spectrally unmixed images obtained after multispectral imaging of a mouse PDAC tumour section from a mutant Kras and mutant p53-driven PDAC mouse model and a tumour section from a PDAC patient after multiplex staining with different markeruorophore combinations. Mouse: cytokeratin 8 (CK8, cytoplasmic, labelled with Cy3.5, pseudocoloured green), Ki-67 (nuclear, Cy5, pseudocoloured white), CD31 (membrane, Cy3, pseudocoloured cyan), FSP1 (membrane, 680, pseudocoloured magenta), aSMA (cytoplasmic, Coumarin, pseudocoloured red), Collagen-I (extracellular, FITC, pseudocoloured yellow), autouorescence (pseudocoloured black) and the DAPI nuclear marker (pseudocoloured blue). Human: cytokeratin 8 (CK8, cytoplasmic, Cy3.5, pseudocoloured green), Collagen-I (extracellular, Coumarin, pseudocoloured teal blue), aSMA (cytoplasmic, FITC, pseudocoloured red), CD31 (membrane, Cy3, pseudocoloured cyan), CD4 (membrane, 680, pseudocoloured magenta),
CD8 (membrane, Cy5, pseudocoloured orange), Foxp3 (nuclear, Biotin-Streptavidin 594, pseudocoloured white), autouorescence (pseudocoloured black) and the DAPI nuclear marker (pseudocoloured blue). All scale bars equal 100 mm.
capture the heterogeneity of the tumour. The clinical characteristics of the patients represented in the TMAs are detailed in Table 1. Of note, uninvolved pancreatic tissue specimens of good quality and absent of tumour tissue were only available for 50 patients. Direct comparisons between the uninvolved pancreatic tissue and tumour specimens for these 50 patients are presented in Supplementary Fig. 4AI. When all available cancer and uninvolved cores were taken into account, we observed the tumour samples presented with signicantly less normal cells than the uninvolved tissue (Fig. 3a,b). Other components, as dened above, which include aSMA myobroblasts and associated desmoplasia, were also more abundant in the tumour tissue as compared to the uninvolved pancreatic tissue (Fig. 3a,c). Interestingly, the uninvolved pancreatic tissue
samples contained signicant amounts of ductal and acinar cells that expressed cytokeratin 8 (Fig. 3a,d, Supplementary Fig. 4J,K). This resulted in no signicant differences in the percentage of cytokeratin 8 cells between tumour and uninvolved pancreatic tissues (Fig. 3d). These adjacent uninvolved pancreatic tissue cores, as opposed to fully normal samples, may have exhibited chronic pancreatitis and early stages of transformation (that is, acinar to ductal metaplasia) that could be reected by their expression of cytokeratin 8. Finally, the total number of T cells, as well as the numbers of all T-cell subpopulations analysed, were higher in the tumour tissue when directly compared to the uninvolved tissue (Fig. 3a,ei).
When we divided all CD3 cells into the different T-cell subpopulations, we observed that 480% of the CD3 T cells were either CD4 or CD8 in both tumour and uninvolved pancreatic tissues (Fig. 3j, middle panel). Signicantly fewer
CD3 T cells were noted in the uninvolved pancreatic tissue compared to the tumour tissue (Fig. 3e), yet the relative proportion of CD8 T cells within total CD3 cells was greater in uninvolved (67%) compared to tumour (47%, P value o0.001) and the relative proportion of CD4 within total CD3 cells was lower in uninvolved (17%) compared to tumour tissue (36%, P value o0.001) (Fig. 3j, middle panel). The higher proportion of cytotoxic T cells within total CD3 T cells in the uninvolved tissue may reect an active immune reaction against the abundant cytokeratin 8 cells, which could represent initial stages of cancer cell transformation. Alternatively, the high number of cytotoxic T cells in the uninvolved tissue might represent a reaction to the presence of the tumour. Among CD4 T cells, both CD4 Teffs (CD4 FoxP3 ) and Tregs (CD4 FoxP3 ) inltrated the tumour and uninvolved pancreatic tissue in equal relative proportions in both tumours and uninvolved tissue (Fig. 3j, lower panel). Collectively, these results indicated that a signicantly more abundant CD3 T-cell inltration is found in tumours when compared to uninvolved pancreatic tissue. These tumour-inltrating T-cell populations are mostly comprised of CD8 cytotoxic T cells and CD4 helper
T cells (CD4 Teff and Treg).
T-cell inltration correlates with PDAC patient survival. We next sought to dene whether specic T-cell inltration in PDAC independently correlated with patient survival. We computed the percentage of T-cell subpopulations out of the total number of cells for each patient, calculating the combined cell percentages in both cores when multiple cores where available per patient. Patients were stratied into low or high tumour-inltration groups based upon the median percentages for each T-cell subpopulation. We observed that high levels of total T-cell inltration were associated with prolonged survival (Fig. 4a). This highlights the clinical relevance and possible functional importance of T-cell inltration for PDAC progression. High inltration of cytotoxic T cells and CD4 Teff cells also correlated with a signicant increase in patient survival (Fig. 4b,c), whereas inltration of Treg or other T cells did not signicantly associate with survival (Fig. 4d,e). In light of the opposing functions of CD4 Teffs and Tregs, we examined the ratio of
CD4 Teff/Treg cells but observed no association with outcome in this cohort of patients (Fig. 4f). We also observed a positive correlation between the three T-cell subpopulations (Fig. 4gi), inferring that Tregs are increased when there is an increase in CD4 Teff or cytotoxic T cells, possibly accounting for the lack of differential inltration pattern with predictive survival value for Tregs. Altogether these results suggested that CD4 Teff and cytotoxic T cells are dening determinants of patient survival in PDAC.
4 NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095 ARTICLE
Table 1 | Patient information.
N 132 Median age at surgery (range), years 64.4 (2585) Median time to recurrence (range), months 9.7 (0.4153) Median survival (range), months 21 (0.4153) Median primary tumour diameter (range), cm 3 (0.712) Reached cancer survival end point 96
SexM 75 F 57
EthnicityWhite 112 Black 7 Hispanic 8 Asian 3 Other 2
Adjuvant radiation therapyYes 57 No 75
Adjuvant chemotherapyYes 95 No 37
Surgical margins0 108
1 24
AJCC stageIB 1 IIA 27 IIB 101 IV 3
Positive lymph nodeYes 104 No 28
DifferentiationWell/moderate 91 Poor 41
Surgery typePancreaticoduodenectomy 109 Distal pancreatectomy 21 Total pancreatectomy 2
AJCC, American Joint Committee on Cancer; F, female; M, male.
The clinical and pathological data for all 132 PDAC patients analysed.
Beyond their association with survival, inltration levels of total T cells did not signicantly associate with any other clinical parameters (Supplementary Table 1). This was also true for the inltration levels of all T-cell subpopulations (Supplementary Tables 26). In the cohort studied here, a subset of measured clinical parameters also had a signicant impact on survival, namely, surgical margins, lymph node status and adjuvant chemotherapy (KaplanMeier/univariate analysis, Table 2). We therefore investigated whether the inltration levels of the T-cell subpopulations shown to associate with survival did so independently of these factors. A stepwise multivariate Cox regression analysis was performed comparing each of the T-cell populations with the clinical features listed above. We observed that total T-cell inltration (P value 0.014), cytotoxic T-cell
inltration (P value 0.002) and CD4 Teff inltration
(P value 0.032) all maintained an independent association with
survival even in the presence of varying surgical margins, lymph node positivity or adjuvant chemotherapy (Table 2).
Cancer cell-adjacent cytotoxic T cells correlate with survival. To explore whether spatial distributions of intratumoral T cells with respect to cancer cells correlate with patient outcome, we determined the spatial coordinates of each T-cell subpopulation. These coordinates were used to characterize the spatial point patterns of T cells relative to cytokeratin 8 cancer cells, using Ripleys L-function20,21. This methodology was previously used to demonstrate a prognostic value of heterogeneous cellular spatial patterns in breast cancer using hematoxylin and eosin-stained tissue specimens22,23. An L-function is dened such that, in this case, the number of T cells within a specied radius (r) distributed from a given point (nuclear centres of cytokeratin 8 cancer cells) is p L r0
2. If the
T-cell population is randomly distributed relative to the cancer cells, the underlying theoretical L-function will have the form L(r) r,
represented by a linear slope (Fig. 5a and Supplementary Fig. 5, dashed line). The area under the L-function curve (AUC) can therefore be used to measure the inltration of T cells within a specied radius around the cancer cells (Fig. 5a and Supplementary Fig. 5). Low T-cell inltration into the tumour will correspond to a low AUC value (Fig. 5a, red), whereas high T-cell inltration will correspond to a high AUC value (Fig. 5a, blue). This is further exemplied in Supplementary Fig. 5, where patients AC have increasing levels of cytotoxic T-cell inltration in relation to cancer cells, corresponding to increasing AUC levels. We focussed on a 20 mm radius around the cancer cells, which represents an enhanced probability for cellcell contact (Fig. 5b). The AUC values were calculated for each T-cell subpopulation (all T cells, cytotoxic T cells, CD4 Teff, Treg and other T cells) within a 20 mm radius of cytokeratin 8 cancer cells. Our results showed that the inltration levels of most T-cell subpopulations within this radius did not signicantly associate with survival (Fig. 5cg). However, high inltration of cytotoxic T cells, within a 20 mm radius of cytokeratin 8 cancer cells, signicantly correlated with prolonged patient survival (Fig. 5e). This suggests that cytotoxic T cells within the direct vicinity of cancer cells may perform an important biological function. This is in accordance with the required cell-cell contact necessary for cytotoxic T cells antitumour activity24.
Desmoplastic elements do not limit lymphocytic inltration. The dense desmoplasia surrounding PDAC has been proposed to impede lymphocyte inltration5,9. Therefore, we investigated whether inltration of cytotoxic T cells adjacent to cancer cells was associated with reduced levels of brotic stroma. The uorescent pixel intensity within a 20 mm radius of cytokeratin 8 cancer cells was determined for both aSMA and Collagen-I (Fig. 6a). Unspecic background uorescence levels were determined using unstained tissue within the image and were subtracted from aSMA or Collagen-I component images. The remaining intensity level (in grey counts), representing aSMA or Collagen-I uorescent pixel intensity, was determined for each pixel within the 20 mm radius of each cancer cell (Fig. 6a). In patients stratied by the AUC levels of cytotoxic T cells within the same radius (high versus low levels of cancer cell-adjacent cytotoxic T cells), the aSMA or Collagen-I intensities were similar (Fig. 6b,c), suggesting that varying levels of aSMA or Collagen-I did not affect cytotoxic T-cell inltration.
These results challenge the postulated role of aSMA cells or Collagen-I content as negative regulators of T-cell inltration5,9.
We next compared aSMA or Collagen-I content adjacent to cancer cells with and without cytotoxic T cells inltration (Fig. 6d). A similar result was obtained with respect to aSMA intensity levels: cancer cells in close contact with cytotoxic T cells
NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications 5
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095
showed insignicant difference (P value 0.0539) in pericellular
aSMA expression compared to cancer cells with no cytotoxic T-cell inltration (Fig. 6e). In contrast, pericancer cell areas
containing cytotoxic T cells had signicantly higher Collagen-I content (Fig. 6f). We additionally compared the aSMA or
Collagen-I content adjacent to cancer cells with or without each
a
Tumour (N=132)
Cytokeratin 8+
Other
Normal
Treg
CD4+ Teff
Cytotoxic T cells
Other T cells
Uninvolved (N=50)
0 2 4 6 8 1020 40 60 80 100 120
Percent distribution
b c
Other
100
Percent normal
****
100
Percent other
****
j
Normal
80
80
Cytokeratin 8+
cells per patient
cells per patient
60
60
Percent of total cells per patient
40
40
140
120
20
20
100
0 Tumour Uninvolved
0 Tumour Uninvolved
80
60
40
d e
NS
****
20
100
60
4
Percent cytokeratin 8+
cells per patient
30
2
cells per patient
80
0
15
Percent all T
60
Tumour (N=132)
Uninvolved (N=50)
CD3+
10
40
20
5
140
Percent of
CD3+ cells per patient
0 Tumour Uninvolved
0
120
100
Other
CD8+
CD4+
Tumour Uninvolved
80
f g
***
****
60
20
10
40
68
Percent cytotoxic
cells per patient
Percent CD4+ Teff
cells per patient
15
10
20
8
4
0
6
2
Tumour (N=132)
Uninvolved (N=50)
4
2
0
0
140
Percent of
CD3+ CD4+ cells per patient
FoxP3+
Tumour Uninvolved
Tumour Uninvolved
120
FoxP3
100
h i
****
**
80
8
40
60
4
20
40
Percent Treg
cells per patient
cells per patient
3
10
20
2
Percent other T
0
1
5
Tumour (N=132)
Uninvolved (N=50)
0
0
1 Tumour Uninvolved
Tumour Uninvolved
Figure 3 | PDAC tissue samples display increased inltration of heterogeneous T-cell subpopulations. (a) Relative distribution of all analysed cell phenotypes in PDAC and uninvolved pancreatic tissue samples. (bi) Pairwise comparisons of the percentage of cells per patient for normal cells (b), other cells (c), cytokeratin 8 cells (d), all T cells (e), cytotoxic T cells (f), CD4 Teff cells (g), Treg cells (h) and other T cells (i) between tumour and uninvolved pancreatic tissue samples. Signicance determined by unpaired t-test. (j) Relative distribution analysis of different T-cell phenotypes within dened groups, by separating the total cell number initially into other, normal, cytokeratin 8 cells or total CD3 Tcells (includes all T-cell subpopulations;
upper panel); then focussing only on CD3 Tcells and dividing them into CD4 (helper Tcells), CD8 (cytotoxic Tcells) and CD4 CD8 (other Tcells) subpopulations (middle panel); and nally focussing on CD3 CD4 T cells and dividing them into FoxP3 (Tregs) and FoxP3 (CD4 Teff)
subpopulations (lower panel). Data presented as the means.d. **Po0.01, ***Po0.001, ****Po0.0001, NS, not signicant.
6 NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095 ARTICLE
a
b
c
100
100
100
**
Low T cells (N=52) High T cells (N=44)
P=0.0094
0 0 50 100 150 200
**
Low cytotoxic T cells (N=53) High cytotoxic T cells (N=43) P=0.0040
*
Low CD4+ Teff (N=53) High CD4+ Teff (N=43) P=0.0338
Percent survival
50
Percent survival
50
Percent survival
50
0 0 50 100 150 200
0
0 50 100 150 200
Months
Months
Months
d e
f
NS Low Treg (N=52)
High Treg (N=44) P=0.0515
100
100 NS Low other T cells (N=50)
High other T cells (N=46) P=0.6854
100 NS Low CD4+ Teff/Treg ratio (N=45)
High CD4+ Teff/Treg ratio (N=48) P=0.7422
Percent survival
50
Percent survival
50
Percent survival
50
0
0
0 50 100 150 200
0 0 50 100 150 200
0 50 100 150 200
Months
Months
Months
g h i
r = 0.527P value < 0.001
r = 0.372P value < 0.001
Number of CD4+ Teffs per core
1,000
10,000
1,000
r = 0.561P value < 0.001
Number of Tregs per core
Number of Tregs per core
100
1,000
100
100
10
10
10
1
1
1
0.1
0.1
0.1 1 10 100 1,000 10,000
0.1 0.1 1 10 100 1,000
0.1 1 10 100 1,000
Number of CD4+ Teffs per core
Number of cytotoxic T cells per core
Number of cytotoxic T cells per core
Figure 4 | T-cell inltration signicantly straties patient survival. (af) Survival analysis of all 132 patients based on percentage of cell numbers per patient of all CD3 T cells (a), CD3CD8 cytotoxic T cells (b), CD3CD4Foxp3 Teffs (c), CD3CD4 FoxP3 Tregs (d), CD3 CD4 CD8
other T cells (e) and the ratio of CD4 Teff/Treg per patient (f). N values correspond to uncensored patients (who reached cancer survival end point). In f, patients with no Tregs were excluded (N 3 patients excluded) as no ratio could be calculated. High and low inltration values were divided based on
the median percentage of positive cells or ratio. Signicance was determined using the Log-rank MantelCox test. (gi) Correlation analysis between Treg and CD4 Teff cell counts (g), cytotoxic T cells and CD4 Teff cell counts (h) and cytotoxic T cells and Treg cell counts (i) per core. Pearson correlation coefcient (r) and signicance levels (P value) are presented for each correlation. Axis values are shown in log scale for clarity. *Po0.05, **Po0.01, NS, not signicant.
Table 2 | Univariate and multivariate survival analysis.
Univariate Multivariate Negative/low Positive/high P value B P value Surgical margins 27.5 17.7 0.009 0.75 0.004
Lymph nodes/AJCC stage 44.1 21.8 0.004 1.026 0.001 Adjuvant chemotherapy 15.9 29.4 0.003 0.944 o0.001
All T cells 18.9 33.1 0.009 0.51 0.014
Cytotoxic T cells 18.9 27.5 0.004 0.687 0.002
CD4 T effector cells 21.2 29.4 0.034 0.447 0.032
AJCC, American Joint Committee on Cancer.
Univariate (MantelCox) and multivariate (Cox regression) survival analyses for each clinical parameter that signicantly impacted survival. Surgical margins, lymph nodes and adjuvant chemotherapy were divided into negative and positive groups, while T-cell inltrations are divided into low and high groups. The median survival times (months) were reported for each of the negative/low and positive/ high groups. Of note, as the majority of the patients fall into AJCC stages IIA and IIB (difference in lymph node status, Table 1) the lymph node and AJCC stage parameters were treated as equivalent and combined in the multivariate analysis. The multivariate B coefcient and signicance are reported, demonstrating all parameters maintain independent effects on survival.
of the remaining pericellular T-cell subpopulations and noted that all T cells combined, CD4 Teffs, Tregs and other T-cell subpopulations were found adjacent to cancer cells associated with higher levels of either aSMA or Collagen-I (Fig. 6gn). We nally looked at the correlation between the expression of
markers of desmoplasia and T-cell inltration in the full context of the TMA cores. While aSMA levels did not signicantly correlate with T-cell inltration (Supplementary Fig. 6), we observed a positive correlation between Collagen-I deposition and the percentage of tissue-inltrating cytotoxic T cells and CD4
NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications 7
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095
a
b
High AUC
Low AUC
High AUC
c d
100
100
Low T cells AUC (N=47)
High T cells AUC (N=49)
P = 0.8278
Percent survival
NS
Low other T cells AUC (N=45)
High other T cells AUC (N=51)
P = 0.4150
NS
NS
Percent survival
50
50
0 0 50 100 150 200
0
0 50 100 150 200
Months
Months
e f
100
Low CD4+ Teff AUC (N=52)
High CD4+ Teff AUC (N=44)
P = 0.1513
100
Percent survival
Low cytotoxic T cells AUC (N=52)
High cytotoxic T cells AUC (N=44)
P = 0.0383
*
50
Percent survival
50
0
0
0 50 100 150 200
0 50 100 150 200
Months
Months
g
100
Low Treg AUC (N=53)
High Treg AUC (N=43)
P = 0.1303
NS
Percent survival
50
0
0 50 100 150 200
Months
Figure 5 | Cancer cell-adjacent cytotoxic T cells signicantly correlate with survival. (a) Schematic representation of the AUC levels based on an L-function calculated for T-cell inltration. Dashed linerandom distribution, Blue areahigh AUC levels/High inltration, Red arealow AUC levels/low inltration. (b) Schematic representing the calculation of the L-function based on the distribution of Tcells within a radius of 20 mm from the nuclear centre of a cytokeratin 8 cancer cell. (cg) Survival analysis of all 132 patients based on inltration as determined by the AUC levels of all CD3 T cells (c), CD3 CD4 CD8 other T cells (d), CD3CD8 cytotoxic T cells (e), CD3CD4 Foxp3 Teffs (f) and CD3CD4 FoxP3 Tregs (g) within 20 mm of cytokeratin 8 cancer cells. N values correspond to uncensored patients (who reached cancer survival end point). High and low inltration values were calculated using the L-function AUC values and divided based on the median inltration values. Signicance was determined using the Log-rank
MantelCox test. *Po0.05, NS, not signicant.
Teff (Fig. 6o,p). These results demonstrated the relationships between T lymphocytes and cellular (myobroblasts) and non-cellular (Collagen-I) desmoplasia were more heterogeneous than had been previously appreciated. Our results indicated that a brotic reaction in the PDAC microenvironment may not impair the inltration of T cells and that increased levels of Collagen-I deposition rather appear to correlate with the presence of different T-cell subpopulations. Future detailed mechanistic exploration is, however, required.
DiscussionThe precise nature of PDAC immunity requires a comprehensive understanding of its microenvironmental complexities. Spatial relationships of individual cellular and acellular components in PDAC may offer novel insights into the dynamic and complex functions of PDAC desmoplasia. Here we have established a novel computational methodology to probe the spatial features of both mesenchymal and T lymphocytic components in the PDAC stroma. This technology benets from the combined bioinformatics power of ow cytometry, with its use of several
8 NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095 ARTICLE
a
b
c
Cancer cell
SMA or Collagen-I pixels
SMA grey values within
20 m of cancer cells
Collagen-I grey values within
20 m of cancer cells
8,000
NS
NS
2,500
NS
6,000
2,000
1,500
4,000
1,000
2,000
500
0
0
Low cytotoxic T-cell infiltration
High cytotoxic T-cell infiltration
Low cytotoxic T-cell infiltration
High cytotoxic T-cell infiltration
Low cytotoxic T-cell infiltration
High cytotoxic T-cell infiltration
d e f
2,000 *
SMA grey values within
20 m of cancer cells
Collagen-I grey values within
20 m of cancer cells
8,000
6,000
1,500
4,000
1,000
2,000
500
0
0
Without T cells With T cells
Without cytotoxic T cells
With cytotoxic T cells
Without cytotoxic T cells
With cytotoxic T cells
g
h
i
j
****
SMA grey values within
20 m of cancer cells
Collagen-I grey values within
20 m of cancer cells
SMA grey values within
20 m of cancer cells
Collagen-I grey values within
20 m of cancer cells
6,000
1,500
****
15,000
****
6,000
****
4,000
1,000
10,000
4,000
2,000
500
5,000
2,000
0
0
0
0
Without T cells
With T cells
Without T cells
With T cells
Without CD4+ Teff
With CD4+ Teff
Without CD4+ Teff
With CD4+ Teff
k
l
m
n
SMA grey values within
20 m of cancer cells
Collagen-I grey values within
20 m of cancer cells
SMA grey values within
20 m of cancer cells
Collagen-I grey values within
20 m of cancer cells
8,000
*
2,500
*
8,000
*
5,000
****
6,000
2,000
6,000
4,000
1,500
3,000
4,000
4,000
1,000
2,000
2,000
500
2,000
1,000
0 Without Treg With Treg
0
0
0
Without Treg With Treg
Without other T cells
With other T cells
Without other T cells
With other T cells
o
p
20
r=0.2666P value=0.002
Percent cytotoxic T cells
per patient
10 r=0.2066P value=0.0175
8
Percent CD4+ Teffs
per patient
15
6
10
4
5
2
0
0
0 20 40 60
0 20 40 60
Percent Collagen-I+ area per patient
Percent Collagen-I+ area per patient
Figure 6 | Desmoplastic elements associate with T-cell inltration. (a) Schematic representing the parameters analysed in b,c. (b,c) The mean intensity of aSMA (b) and Collagen-I (c) (grey values) for pixels within 20 mm of cancer cells for each patient separated by low or high cytotoxic T-cell inltration as determined by the L-function AUC; signicance determined by an unpaired t-test. (d) Schematic representing the parameters analysed in en.
(en) The mean intensity of aSMA (e,g,i,k,m) and Collagen-I (f,h,j,l,n) (grey values) for pixels within 20 mm of cancer cells for each patient separated by cancer cells with or without adjacent cytotoxic Tcells (e,f), all Tcells (g,h), CD4 Teffs (i,j), Tregs (k,l) and other Tcells (m,n); signicance determined by an unpaired t-test. (o,p) Correlation analysis between area of Collagen-I deposition and the percentage of cytotoxic T cells (o) and area of Collagen-I deposition and the percentage of CD4 Teffs (p) per patient. Pearson correlation coefcient (r) and signicance levels (P value) are presented for each correlation. Data presented as the means.e.m. *Po0.05, ****Po0.0001, NS, not signicant.
NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications 9
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095
combined markers, and the spatial information obtained from immunohistochemistry. The use of multiple markers is important as it allows for the identication of distinct subpopulations in one tissue section. For instance, Foxp3, commonly used to identify Tregs5,25, is also expressed in cancer cells26, therefore, the use of just this one marker could overestimate the abundance of tumour-inltrating Tregs. Additionally, the spatial relationships between T-cell subpopulations and cancer cells are retained in our experimental strategy, allowing for a more precise appreciation of their biological interactions.
We applied this technology to a TMA comprised of tissue obtained upon pancreatectomy of 132 patients with PDAC without neoadjuvant therapy. Two tumour cores as well as a core with non-involved tissue were available for each patient. In order to elucidate the complex nature of PDAC immunity, we utilized this technology to demonstrate that human PDAC contains a heterogeneous T-cell population. Among these, cytotoxic T cells were the predominant T-cell subpopulation (approximately 47% of all T cells) and high inltration of cytotoxic T cells was associated with prolonged survival. Additionally, tumour samples showed a signicant increase in the proportion of CD4 T helper cells when compared to normal tissue, suggesting the activation of a systemic immune response to disease progression. Our data suggested that the distribution and relative abundance of tumour-promoting Tregs may not inuence patient outcome to the extent that has been suggested1,6,27, as Treg-inltration levels did not signicantly associate with patient survival in our analysis. Interestingly, the number of PDAC-inltrating Tregs in our study (average of 25 cells mm 2) was equivalent to the numbers of Tregs reported in melanoma (B20 cells mm 2)
assayed using similar techniques28. Considering melanoma is widely regarded as an immunogenic cancer, our results suggest the number of inltrating Tregs may not be a signicant determinant of PDAC low immunogenicity. In contrast, CD4
Teff and cytotoxic T-cell inltration emerged as independent indicators of survival for PDAC patients. Additionally, the distribution of cytotoxic T cells was particularly signicant when found in the direct vicinity of cancer cells. These observations were performed on all evaluable PDAC tissue areas, thereby providing an overall snapshot of the PDAC T lymphocytic landscape. This also removed the bias of previous observations, which focussed only on high areas of immune inltration for T-cell counts using single marker immunohistochemistry4,6. When both tumour cores were available for a patient, both were included for evaluation. Our analysis further supports the published data that PDAC with high levels of cytotoxic T cells have prolonged survival. It should be noted, however, that while the overall inltration of these cells had a signicant impact on survival, their functional status was not measured and remains to be explored. Combining our panel with additional multiplex panels for markers of T-cell activation, as well as markers of additional components of the PDAC microenvironment, such as myeloid cells, may elucidate the differences between PDAC and melanoma and their immunotherapy responsiveness.
The desmoplastic stroma has been hypothesized to sequester T cells away from cancer cells5,9. The multiplex technology provided us an opportunity to address this hypothesis. Here we show tumour cells with high or low pericellular cytotoxic T-cell inltration did not exhibit differences in the associated levels of aSMA and Collagen-I. This suggested that the desmoplastic reaction as determined by aSMA and Collagen-I deposition may have insignicant impact on the inltration of cytotoxic T cells. Recent observations have suggested that focal adhesion kinase signalling in cancer cells can mediate a brotic reaction with immunosuppressive consequences10. This was shown only in
mouse models of PDAC using generic measurements of desmoplasia without spatial analysis. Using patient samples, we show that the desmoplastic reaction in the direct vicinity of cancer cells does not negatively correlate with T-cell inltration. We further demonstrate that enhanced Collagen-I levels correlated with increased levels of cytotoxic T-cell and CD4
Teff inltration. Cancer cells devoid of pericellular cytotoxic T cells exhibited diminished Collagen-I in their vicinity, suggesting that Collagen-I deposition and remodelling may favour, rather than hinder, cytotoxic T-cell inltration adjacent to cancer cells.
The efcacy of single-agent immune checkpoint blockade in the treatment of PDAC patients has so far been underwhelming29, contributing to the perception that PDAC is a poorly immunogenic tumour. Notably, in the treatment of melanoma using immunotherapy, patients with programmed death ligand 1 (PD-L1) expression in their tumours had the most benecial response to anti-PD-L1 therapy3032. These studies are prompting ongoing efforts to stratify patients according to dened biomarkers (that is, PD-1, PD-L1 expression or cytotoxic T cells)30. It is conceivable that stratication of PDAC patients could also enable the realization of signicant benet with immunotherapy. Our collective analyses suggest that such stratication in PDAC may require a detailed audit of the tumour microenvironmental components due to the complex and dense composition of mesenchymal cells and immune cells. We propose that tissue-typing the microenvironmental composition of PDAC (that is, the number of cytotoxic or helper T cells and desmoplastic stroma) may aid in dening patient populations that would most benet from immune therapies, as proposed previously14. This is in accordance with growing consensus that T lymphocytic inltration should be included in standard tumour pathological scoring33. In addition to measures of relative abundance, the spatial distribution analysis we report could also aid in retrospective evaluation of responses to therapy. In summary, our current observations suggest that T cells inltrate PDAC tumours and may not be impeded by aSMA or Collagen-I stromal deposition. This study offers new insights into the nature of PDAC immunity and how this information can be harnessed towards effective immunotherapy strategies.
Methods
Patient cohorts. This study was approved by the institutional review board at the University of Texas MD Anderson Cancer Center (MDACC). Our study population consisted of 132 patients with PDAC who underwent pancreatectomy with curative intent (Table 1) at MDACC. Informed consent was obtained from all patients. None of the patients received neoadjuvant therapy. The PDAC TMAs were constructed from FFPE blocks of archived PDAC specimens using the method described previously34. Briey, representative areas of tumour and matched uninvolved pancreatic tissue were selected based on the review of hematoxylin and eosin-stained slides. The corresponding FFPE tissue blocks were retrieved. For each patient, two 1.0 mm cores from representative areas of the tumour and one 1.0 mm core of matched uninvolved pancreas were used for TMA construction35. Clinical information was obtained from the electronic medical records.
Samples for trial multiplex stains. Sections from full FFPE blocks that were used to generate the TMA were utilized in the initial optimizations of the multiplex tissues, as presented in Fig. 2. Sections of mouse PDAC samples (end point Pdx1- cre;LSL-KrasG12D;P53R172H/ tumours) that were obtained from archived blocks from previously published work14 were also used in initial multiplex optimizations as presented in Fig. 2.
Eight-colour immunohistochemical multiplex. In all, 5 mm sections obtained from the TMA blocks were deparafnized and tissues were xed with formaldehyde:methanol (1:10) prior to antigen retrieval in heated Citric Acid Buffer (pH 6.0) for 15 min (EZ Retriever microwave, BioGenex). Each section was put through seven sequential rounds of staining, each including a protein block with 1% BSA followed by primary antibody and corresponding secondary horseradish peroxidase-conjugated polymer (Table 3 and Supplementary Fig. 1A).
10 NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095 ARTICLE
Table 3 | Sequential Opal multiplex staining protocol.
Antigen Primary antibody Catalogue number Secondary polymer TSA uorophore Concentration Provider Polymer ProviderFoxP3 1:50 Abcam ab20034 Super Picture Invitrogen Cy3
CD4 1:20 Thermo MS-1528 Super Picture Invitrogen Cy5 Collagen-I 1:500 AbDSerotec 1310-01 Goat Po-link1 GBI Coumarin CD8 1:100 Dako M7103 Super Picture Invitrogen Opal10 Cytokeratin 8 1:50 DSHB Troma-1 Rat Po-link1 GBI FITC aSMA 1:2,000 Dako M0851 Super Picture Invitrogen Cy5.5
CD3 1:500 Dako A0452 Super Picture Invitrogen Opal9
Mouse (Fig. 2)
FSP1 1:6,000 Dako A5114 Super Picture Invitrogen 680 Cytokeratin 8 1:50 DSHB Troma-1 Rat-on-Mouse BioCare Cy3.5 CD31 1:4,000 SantaCruz sc-1506 Goat Po-link1 GBI Cy3 Collagen-I 1:3,000 AbDSerotec 1310-01 Goat Po-link1 GBI FITC Ki67 1:200 Thermo RM-9106-S Super Picture Invitrogen Cy5 aSMA 1:2,000 Dako M0851 Super Picture Invitrogen Coumarin
Human (Fig. 2)
CD4 1:25 BioCare CM153CK Super Picture Invitrogen 680 Collagen-I 1:500 AbDSerotec 1310-01 Goat Po-link1 GBI Coumarin CD31 1:3,000 SantaCruz sc-1506 Goat Po-link1 GBI Cy3 CD8 1:1,000 Dako M7103 Super Picture Invitrogen Cy5 Cytokeratin 8 1:100 DSHB Troma-1 Rat-on-Mouse BioCare Cy3.5 aSMA 1:2,000 Dako M0851 Super Picture Invitrogen FITC
FoxP3 1:25 Abcam ab20034 Super Picture Invitrogen Biotin-Strep 594
aSMA, alpha-smooth muscle actin; FITC, uorescein isothiocyanate; TSA, tyramide signal amplication.
Each row represents one step in the sequential staining protocol with each individual primary antibody and corresponding secondary HRP polymer and TSA uorophore. The table is divided into three sections, corresponding to three staining protocols reported. The rst section describes the multiplex analysed in the majority of the paper and the second and third sections correspond to the multiplex protocols presented in Fig. 2.
Each horseradish peroxidase-conjugated polymer mediated the covalent binding of a different uorophore using tyramide signal amplication (Table 3 and Supplementary Fig. 1A). This covalent reaction was followed by additional antigen retrieval in heated Citric Acid Buffer (pH 6.0) for 15 min to remove bound antibodies before the next step in the sequence. After all seven sequential reactions, sections were counterstained with DAPI (Life Tech) and mounted with Vectashield uorescence mounting medium (Vector Labs, Burlingame, CA).
Multispectral imaging. Multiplex stained TMA slides were imaged using the Vectra Multispectral Imaging System version 2 (Perkin Elmer), where one raw image comprising four stitched 200 multispectral image cubes was obtained for
each TMA core. Each 200 multispectral image cube was created by combining
images obtained every 10 nm of the emission light spectrum across the range of each emission lter cube. Filter cubes used for multispectral imaging were DAPI (440680 nm), FITC (520 nm-680 nm), Cy3 (570690 nm), Texas Red(580700 nm) and Cy5 (670720 nm) (Supplementary Fig. 1B).
Spectral unmixing and phenotyping. A spectral library containing the emitting spectral peaks of all uorophores was created with the Nuance Image Analysis software (Perkin Elmer) (Supplementary Fig. 1B), using multispectral images obtained from single stained slides for each marker and associated uorophore. This spectral library was then used to separate each multispectral image cube into its individual components (spectral unmixing) allowing for the colour-based identication of all eight markers of interest in a single image using the inForm 2.1 image analysis software. All spectrally unmixed and segmented images were subsequently subjected to a proprietary inForm active learning phenotyping algorithm. This allows for the individual identication of each DAPI-stainedcell according to their pattern of uorophore expression and nuclear/cell morphological features, associating their phenotype with specic x,y spatial coordinates. Cells were phenotyped into one of seven different classes according to our markers of interest (Fig. 1l) as follows: cancer/cytokeratin 8 cells (CK8 ), cytotoxic T cells (CD3 CD8 ), CD4 Teff cells (CD3 CD4 Foxp3 ), Treg cells (CD3 CD4 FoxP3 ), other T cells (CD3 CD4 CD8 ), normal (CK8 CD3 and epithelial or acinar cell morphology as determined through the autouorescence signal), and other (CK8 CD3 ). All phenotyping and subsequent quantications were performed blinded to the sample identity and clinical outcomes.
Quantication of T-cell spatial distribution. In each spectrally unmixed and phenotyped core, the relative spatial distribution of cancer cells and each individual
subpopulation of T cells were considered as a bivariate point pattern. This bivariate point pattern can be characterized by bivariate K- and L-functions, generalized from Ripleys K- and L-functions21. The bivariate K-function in our application is dened as follows:
Kxy r
ly E number of cells of type y within a distance r of a randomly selected cell of type x
1
where ly is the number of type y cells per unit area in the region of interest and E[.] evaluates the expected value of the quantity in the bracket. Theoretically, if the spatial distribution of the two types of cells x and y are completely independent (Poisson hypothesis), the value of K-function is Kxy r pr2 (ref. 36). The
bivariate L-function is a transformation of K-function, dened as:
Lxy r
Kxyr=p
q
Under the Poisson hypothesis, we have Lxy r r, which is a more convenient
representation than the K-function. Also, the estimation of Lxy r has a more
stabilized variance36. We used the toolbox spatstat in R for the estimation of the L-function37. For each of the TMA images, the L-function was estimated fora range of r from 0 to 20 mm. The level of inltration of different T cells is represented by the AUC of their L-function and dichotomized at the median to discriminate patient survival.
Quantication of uorescent pixel intensity. Monochromatic tiff imagesof the spectrally unmixed cores containing only the Cy5.5 (aSMA) or Coumarin (Collagen-I) signal components (processed separately) were analysed by ImageJ to determine the background pixel intensity levels (grey values), as determined by the unstained tissue areas. Absolute intensity levels of all pixels within 20 mm of each CK8 cancer cell were determined and background intensity levels were subtracted using Matlab. Percentage of positive area for the whole core was determined by setting the positivity threshold on ImageJ using the Huang algorithm (aSMA 20/255, or Collagen-I 35/255). This positivity is represented as the percentage of all signal areas calculated from the multiplexed composite image displaying all channels, including autouoresence (region of interest selected from all pixels over the brightness threshold of 30/255).
Statistics. All statistical analyses were performed using the GraphPad Prism software unless stated otherwise. Statistical analyses of immunohistochemical quantications were performed using a Students t-test or analysis of variance as appropriate. For survival analyses, KaplanMeier plots were drawn and statistical differences were evaluated using the log-rank MantelCox test (a.k.a univariate).
NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications 11
ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095
Multivariate analyses of the survival data were performed for each new T-cell inltration parameter using a Cox regression analysis in SPSS. For the correlation analyses between different T-cell distributions, the Pearsons correlation coefcient (r) was calculated using the SPSS statistical software. Pairwise comparisons between clinical variables and all survival stratications were determined by a chi-square analysis using SPSS. A P value o0.05 was considered statistically signicant.
Data availability. The authors declare that the data supporting the ndingsof this study are available within the paper and its Supplementary Information les (Supplementary Data 18). All computer codes used for spatial distribution analyses as well as any additional clarications are available from the corresponding author upon request.
References
1. Vonderheide, R. H. & Bayne, L. J. Inammatory networks and immune surveillance of pancreatic carcinoma. Curr. Opin. Immunol. 25, 200205 (2013).
2. Schmitz-Winnenthal, F. H. et al. High frequencies of functional tumour-reactive T cells in bone marrow and blood of pancreatic cancer patients. Cancer Res. 65, 1007910087 (2005).
3. Shibuya, K. C. et al. Pancreatic ductal adenocarcinoma contains an effector and regulatory immune cell inltrate that is altered by multimodal neoadjuvant treatment. PLoS ONE 9, e96565 (2014).
4. Fukunaga, A. et al. CD8 tumour-inltrating lymphocytes together with
CD4 tumour-inltrating lymphocytes and dendritic cells improve the
prognosis of patients with pancreatic adenocarcinoma. Pancreas 28, e26-31 (2004).5. Ene-Obong, A. et al. Activated pancreatic stellate cells sequester CD8 T cells
to reduce their inltration of the juxtatumoral compartment of pancreatic ductal adenocarcinoma. Gastroenterology 145, 11211132 (2013).6. Ino, Y. et al. Immune cell inltration as an indicator of the immune microenvironment of pancreatic cancer. Br. J. Cancer 108, 914923 (2013).
7. Clark, C. E. et al. Dynamics of the immune reaction to pancreatic cancer from inception to invasion. Cancer Res. 67, 95189527 (2007).
8. Alrifai, D., Sarker, D. & Maher, J. Prospects for adoptive immunotherapy of pancreatic cancer using chimeric antigen receptor-engineered T-cells. Immunopharmacol. Immunotoxicol. 38, 5060 (2016).
9. Feig, C. et al. Targeting CXCL12 from FAP-expressing carcinoma-associated broblasts synergizes with anti-PD-L1 immunotherapy in pancreatic cancer. Proc. Natl Acad. Sci. USA 110, 2021220217 (2013).
10. Jiang, H. et al. Targeting focal adhesion kinase renders pancreatic cancers responsive to checkpoint immunotherapy. Nat. Med. 22, 851860 (2016).
11. Hiraoka, N., Onozato, K., Kosuge, T. & Hirohashi, S. Prevalence of FOXP3
regulatory T cells increases during the progression of pancreatic ductal adenocarcinoma and its premalignant lesions. Clin. Cancer Res. 12, 54235434 (2006).12. Tassi, E. et al. Carcinoembryonic antigen-specic but not antiviral CD4
T cell immunity is impaired in pancreatic carcinoma patients. J. Immunol. 181, 65956603 (2008).13. De Monte, L. et al. Intratumor T helper type 2 cell inltrate correlates with cancer-associated broblast thymic stromal lymphopoietin production and reduced survival in pancreatic cancer. J. Exp. Med. 208, 469478 (2011).
14. Ozdemir, B. C. et al. Depletion of carcinoma-associated broblasts and brosis induces immunosuppression and accelerates pancreas cancer with reduced survival. Cancer Cell 25, 719734 (2014).
15. Soares, K. C. et al. TGF-beta blockade depletes T regulatory cells from metastatic pancreatic tumours in a vaccine dependent manner. Oncotarget 6, 4300543015 (2015).
16. Soares, K. C. et al. PD-1/PD-L1 blockade together with vaccine therapy facilitates effector T-cell inltration into pancreatic tumours. J. Immunother. 38, 111 (2015).
17. Beatty, G. L. et al. Exclusion of T cells from pancreatic carcinomas in mice is regulated by Ly6Clow F4/80 extratumoral macrophages. Gastroenterology 149, 201210 (2015).
18. Schueneman, A. J. et al. Low total lymphocyte count is associated with poor survival in patients with resected pancreatic adenocarcinoma receiving a GM-CSF secreting pancreatic tumour vaccine. Ann. Surg. Oncol. 20, S725S730 (2013).
19. Winograd, R. et al. Induction of T-cell immunity overcomes complete resistance to PD-1 and CTLA-4 blockade and improves survival in pancreatic carcinoma. Cancer Immunol. Res. 3, 399411 (2015).
20. Besag, J. Comments on Ripleys paper. J. R. Stat. Soc. B 39, 193195 (1977).21. Ripley, B. The second-order analysis of stationary point processes. J. Appl. Prob. 13, 255266 (1976).
22. Heindl, A., Nawaz, S. & Yuan, Y. Mapping spatial heterogeneity in the tumour microenvironment: a new era for digital pathology. Lab. Invest. 95, 377384 (2015).
23. Yuan, Y. et al. Quantitative image analysis of cellular heterogeneity inbreast tumours complements genomic proling. Sci. Transl. Med. 4, 157ra143 (2012).
24. Janeway, C., Travers, P. & Walport, M. Immunobiology: the Immune System in Health and Disease 5th edn (Garland Science, 2001).
25. Wartenberg, M. et al. Accumulation of FOXP3 T-cells in the tumour
microenvironment is associated with an epithelial-mesenchymal-transition-type tumour budding phenotype and is an independent prognostic factor in surgically resected pancreatic ductal adenocarcinoma. Oncotarget 6, 41904201 (2015).26. Hinz, S. et al. Foxp3 expression in pancreatic carcinoma cells as a novel mechanism of immune evasion in cancer. Cancer Res. 67, 83448350 (2007).
27. Lutz, E. R. et al. Immunotherapy converts nonimmunogenic pancreatic tumours into immunogenic foci of immune regulation. Cancer Immunol. Res. 2, 616631 (2014).
28. Feng, Z. et al. Multispectral imaging of formalin-xed tissue predicts ability to generate tumour-inltrating lymphocytes from melanoma. J. Immunother. Cancer 3, 111 (2015).
29. Royal, R. E. et al. Phase 2 trial of single agent Ipilimumab (anti-CTLA-4) for locally advanced or metastatic pancreatic adenocarcinoma. J. Immunother. 33, 828833 (2010).
30. Redman, J. M., Gibney, G. T. & Atkins, M. B. Advances in immunotherapy for melanoma. BMC Med. 14, 20 (2016).
31. Tumeh, P. C. et al. PD-1 blockade induces responses by inhibiting adaptive immune resistance. Nature 515, 568571 (2014).
32. Haanen, J. B. Immunotherapy of melanoma. EJC Suppl. 11, 97105 (2013).
33. Galon, J. et al. Cancer classication using the Immunoscore: a worldwide task force. J. Transl. Med. 10, 205 (2012).
34. Wang, H., Wang, H., Zhang, W. & Fuller, G. N. Tissue microarrays: applications in neuropathology research, diagnosis, and education. Brain Pathol. 12, 95107 (2002).
35. Wang, O. H. et al. Prognostic and functional signicance of MAP4K5 in pancreatic cancer. PLoS ONE 11, e0152300 (2016).
36. Harkness, R. & Isham, V. A bivariate spatial point pattern of ants nests. Appl. Stat. 32, 293303 (1983).
37. Baddeley, A., Rubak, E. & Turner, R. Spatial Point Patterns: Methodology and Applications with R (CRC Press, 2015).
Acknowledgements
We thank Dr Jared Burks, Director at the MDACC Flow Cytometry and Cellular Imaging Facility, for originally introducing J.L.C. and P.C.S. to the Opal Multiplexing technology and for his assistance with the initial optimization of the eight-colour multiplexed panel. We also thank Peter Miller, Cliff Hoyt, Aaron Risinger, Damara Gebauer, Edward Stack, Mike Campisano and Kent Johnson of Perkin Elmer for optimization support concerning multiplex uorophore physics, multispectral imaging, spectral unmixing and phenotyping algorithm optimizations, as well as the generous gift of the Opal 9 and 10 TSA uorophores. This study was primarily supported by funding from the Lustgarten Foundation to R.K. and J.P.A. R.K. and J.P.A. are also supported by NCI P01 CA117969 grant. Research in the R.K. laboratory is in part supported by the Cancer Prevention and Research Institute of Texas. Research in the V.S.L. laboratory is supported by the UT MDACC Khalifa Bin Zayed Al Nahya Foundation. Research in the A.R. laboratory is supported by the NCI Cancer Center Support Grant NCI P30 CA016672, CPRIT RP110532, American Cancer Society (RSG-16-005-01), Center for Radiation Oncology Research Pilot Grant, NIH NCI U01 (CA196403), Institutional research grant from The University of Texas MD Anderson Cancer Center, Career Development Award from the Brain Tumor SPORE P50CA127001-07 and Equipment grant from NVidia. Some of this research was performed at the MDACC Flow Cytometry and Cellular Imaging Facility, which is supported in part by the NIH through MDACC Support Grant CA016672.
Author contributions
J.L.C. and P.C.S. developed the technology, designed and performed the experiments, trained and validated the phenotyping algorithms, analysed the data, generated the gures and wrote the manuscript. D.Y. and S.B. generated the spatial analysis algorithms and data, developed computer code for analyses, analysed data and participated in discussions related to the spatial and survival analyses. H.W. generated the tissue array and provided intellectual input in the interpretation and presentation of the clinical data. A.R. conceived and designed the analytic strategy for spatial inltration analysis, supervised S.B. and D.Y. toward implementation of the analysis algorithms, provided intellectual input and participated in discussions related to the spatial and survival analysis. J.P.A. provided intellectual input in the design and interpretation of immunological signicance. V.S.L. helped design experimental strategies, provided intellectual input and helped write the manuscript. R.K. conceptually designed the strategy for this study, participated in discussions, provided intellectual input, supervised experimental discussion and helped write the manuscript.
12 NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15095 ARTICLE
Additional information
Supplementary Information accompanies this paper at http://www.nature.com/naturecommunications
Web End =http://www.nature.com/ http://www.nature.com/naturecommunications
Web End =naturecommunications
Competing interests: J.P.A. is an inventor of intellectual property owned by the University of California, Berkeley and licensed to Bristol Myers-Squibb. The remaining authors declare no competing nancial interests.
Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/
Web End =http://npg.nature.com/ http://npg.nature.com/reprintsandpermissions/
Web End =reprintsandpermissions/
How to cite this article: Carstens, J. L. et al. Spatial computation of intratumoral T cells correlates with survival of patients with pancreatic cancer. Nat. Commun. 8, 15095 doi: 10.1038/ncomms15095 (2017).
Publishers note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afliations.
This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the articles Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/
Web End =http://creativecommons.org/licenses/by/4.0/
r The Author(s) 2017
NATURE COMMUNICATIONS | 8:15095 | DOI: 10.1038/ncomms15095 | http://www.nature.com/naturecommunications
Web End =www.nature.com/naturecommunications 13
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Copyright Nature Publishing Group Apr 2017
Abstract
The exact nature and dynamics of pancreatic ductal adenocarcinoma (PDAC) immune composition remains largely unknown. Desmoplasia is suggested to polarize PDAC immunity. Therefore, a comprehensive evaluation of the composition and distribution of desmoplastic elements and T-cell infiltration is necessary to delineate their roles. Here we develop a novel computational imaging technology for the simultaneous evaluation of eight distinct markers, allowing for spatial analysis of distinct populations within the same section. We report a heterogeneous population of infiltrating T lymphocytes. Spatial distribution of cytotoxic T cells in proximity to cancer cells correlates with increased overall patient survival. Collagen-I and αSMA+ fibroblasts do not correlate with paucity in T-cell accumulation, suggesting that PDAC desmoplasia may not be a simple physical barrier. Further exploration of this technology may improve our understanding of how specific stromal composition could impact T-cell activity, with potential impact on the optimization of immune-modulatory therapies.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer