Summary
T cell exhaustion is a major impediment to anti-tumor immunity. However, it remains elusive how other immune cells in the tumor microenvironment (TME) contribute to this dysfunctional state. Here we show that the biology of tumor-associated macrophages (TAM) and exhausted T cells (Tex) in the TME is extensively linked. We demonstrate that in vivo depletion of TAM reduces exhaustion programs in tumor-infiltrating CD8+ T cells and reinvigorates their effector potential. Reciprocally, transcriptional and epigenetic profiling reveals that Tex express factors that actively recruit monocytes to the TME and shape their differentiation. Using lattice light sheet microscopy, we show that TAM and CD8+ T cells engage in unique long-lasting antigen-specific synaptic interactions that fail to activate T cells, but prime them for exhaustion, which is then accelerated in hypoxic conditions. Spatially resolved sequencing supports a spatiotemporal self-enforcing positive feedback circuit that is aligned to protect rather than destroy a tumor.
Introduction
Cancer immunotherapy – harnessing the patient’s immune system to fight cancer – has revolutionized cancer treatment strategies. However, a large proportion of patients does not show clinical response, and the mechanisms underlying intrinsic and acquired resistance are still poorly understood. CD8+ T cells are critical mediators of anti-tumor immune responses, and the main target for current immunotherapy approaches. Tumor infiltration of CD8+ T cells correlates with improved prognosis and beneficial responses to immune checkpoint blockade as compared to non-infiltrated tumors (Galon et al., 2006; Tumeh et al., 2014). However, those CD8+ T cells are frequently non-functional due to establishment of a state of exhaustion, characterized by the expression of inhibitory molecules including PD1, CD38 and TOX and the loss of cytotoxic effector function (Wherry et al., 2007; Doering et al., 2012; Schietinger et al., 2016; Pauken et al., 2016; Scott et al., 2019; Khan et al., 2019). Several studies have shown that chronic antigen exposure and stimulation of the T cell receptor (TCR) are required for exhaustion programs in T cells (Utzschneider et al., 2016; Scott et al., 2019; Oliveira et al., 2021). However, how this is orchestrated in the tumor microenvironment (TME) is unclear.
The immune composition of the TME plays an important role in regulating effective anti-tumor T cell responses (Binnewies et al., 2018). Across solid tumors, the majority of immune cells in the TME is frequently comprised of antigen-presenting myeloid cells (APC), of which tumor-associated macrophages (TAM) are typically the most abundant (DeNardo et al., 2011; Ruffell et al., 2012; Broz et al., 2014). TAM abundance is correlated with poor prognosis in a variety of solid tumor types (Zhang et al., 2012; Gentles et al., 2015), and many studies report on their immunosuppressive role in cancer progression and dissemination (DeNardo and Ruffell, 2019). Conversely, some studies report immunostimulatory and anti-tumor functions of TAM through expression of TNF and iNOS, or upon treatment with CD40 agonists (Beatty et al., 2011; Klug et al., 2013). Similar to a rare population of dendritic cells (cDC1), which have been described to be potent activators of anti-tumor T cells (Broz et al., 2014; Roberts et al., 2016; Salmon et al., 2016; Spranger et al., 2017), TAMs have the potential to phagocytose large amounts of tumor-associated antigens, but fail to successfully support T cell activation (Engelhardt et al., 2012; Broz et al., 2014). Interestingly, intravital imaging studies have shown that antigen-specific CD8+ T cells preferentially localize in TAM-rich areas in the TME, and form tight interactions that persist over time (Boissonnas et al., 2013; Broz et al., 2014; Peranzoni et al., 2018).
Here we dissect the molecular mechanisms of a novel immune co-differentiation, by which TAM and exhausted CD8+ T cells (Tex) sustain each other’s maturation and presence in the TME through long-lived antigen-specific synaptic contacts. Our study reveals a novel mechanistic link through an antigen-driven positive feedback loop and offers a possible path whereby T cells and TAM might equally contribute to initial and sustained tumor immune evasion.
Results
CD8+ T cell exhaustion correlates with macrophage abundance in the TME
To study how myeloid immune cells contribute to CD8+ T cell exhaustion in the TME, we first focused on the concurrent events during the onset of exhaustion programs in tumor-infiltrating CD8+ T cells in mouse models of melanoma (B78ChOVA and B16ChOVA) and spontaneous breast cancer (MMTV-PyMTChOVA) (Fig. S1A). At different time points during tumor growth, we adoptively transferred Ovalbumin (OVA)-specific OT-I CD8+ T cells into tumor-bearing mice, focusing on: (1) early arrival T cells that were recently recruited to the TME (Tex d4) and (2) T cells that have resided in the TME for 14 days and have demonstrably upregulated PD1, CD38, TOX and CD5 as assessed by flow cytometry (Tex d14) (Fig. S1B). Both of these populations demonstrated reduced production of the cytokines IFNγ and TNFα when compared to activated CD44+ endogenous CD8+ T cells in the tumor-draining lymph node (TdLN) (Fig. S1C, D). The onset of exhaustion and dysfunction in CD8+ T cells upon tumor infiltration was antigen-specific, because irrelevant LCMV-specific P14 CD8+ T cells did not acquire phenotypic markers of exhaustion when compared to OT-I CD8+ T cells (Fig. S1E-H). However, the ability to produce effector cytokines IFNγ and TNFα was blunted equivalently in endogenous, P14 and OT-I CD8+ T cells upon tumor residence (Fig. S1I, J). This may occur if loss of those markers represented a natural decay process post-activation or if additional non-antigen-specific and more universal suppression mechanisms are at work in the TME.
Tumor-associated macrophages (TAM) comprised the majority of myeloid cells in B78ChOVA melanomas in line with previous studies (Broz et al., 2014; Gentles et al., 2015; Cheng et al., 2021), and their abundance increased during tumor progression, while the fraction of CD103+ cDC1 and CD11b+ cDC2 diminished (Fig. S1K). Recognizing the stoichiometric abundance of TAM as possible APC, we sought to study the role of TAM in the onset of CD8+ T cell exhaustion by subjecting tumor-bearing mice to antibody-mediated blockade of CSF1/CSF1R signaling (Fig. 1A and Fig. S1L, O). This treatment resulted in significant reduction of the proportion of CD11b+F4/80+ TAM in tumors (Fig. 1B,C and Fig. S1M, P). TAM-depletion also resulted in a concurrent reduction in the expression of exhaustion markers PD1, CD38 and TOX on tumor-infiltrating CD44+ OT-I CD8+ T cells (Fig. 1D and Fig. S1N, Q). Moreover, the resultant CD44+ OT-I CD8+ T cells produced higher levels of IFNγ and TNFα in anti-CSF1R-treated compared to isotype-treated mice (Fig. 1E, F). We also found that the expression of PD1, CD38 and TOX on tumor-infiltrating CD44+ OT-I CD8+ T cells positively correlates with the abundance of TAM in the TME, measured across 25 mice in three independent experiments subjected to anti-CSF1R or isotype treatment (Fig. 1G). In line with this, flow cytometric profiling of CD8+ tumor-infiltrating lymphocytes (TIL) in a cohort of 20 patients with renal cell carcinomas – which are rich in myeloid and T cells (Combes et al., 2021; Mujal et al., 2021) –, demonstrated a strong association between PD1 and CD38 (but not CTLA4) expression and the degree to which patient’s myeloid cells had differentiated toward macrophages as compared to monocytes in the TME (Fig. 1H). We used this ratio as the pure number of macrophages did not show this association (data not shown).
CD8+ Tex express monocyte/macrophage-related factors upon prolonged residence in the TME of mouse and human cancers
To test whether there might be a mechanistic link between TAM abundance and CD8+ T cell exhaustion, we isolated early (Tex d4) and late exhausted OT-I CD8+ T cells (Tex d14) from B78ChOVA tumors and compared their transcriptional profile to that of splenic naïve CD44− OT-I CD8+ T cells by RNAseq. As expected, Tex d14 showed enhanced expression of known markers associated with exhaustion (Sade-Feldman et al., 2018), including but not limited to Cd44, Pdcd1, Cd38, Tox, Irf4, Havcr2, Lag3, while expression of naïve precursor genes Sell, Tcf7 and Il7r were enriched in Tnaïve cells (Fig. 2A). Interestingly, pathway enrichment analysis of genes with a fold enrichment >5 in Tex revealed dramatic enrichment of pathways involved in “abnormal cytokine secretion” and “impaired macrophage chemotaxis” (Fig. 2B). A closer analysis of individual genes demonstrated that expression of genes associated with naïve precursor T cell states decreased and genes previously associated with exhaustion increased in Tex d14 vs Tex d4 (Fig. 2C), consistent with previous reports (Pauken et al., 2016; Schietinger et al., 2016; Sade-Feldman et al., 2018). Interestingly, a large set of myeloid-related genes was highly upregulated in exhausted CD8+ T cells and most of these increased with prolonged residence in the TME (Fig. 2C and Fig. S2A). Increased expression of Csf1, Ccl3 and Ccl5 transcripts was confirmed by qRT-PCR on an independent sample set of Tnaïve, Tex d4 and Tex d14 (Fig. S2B). Of note, the majority of these transcriptional changes was not observed in in vitro activated effector OT-I CD8+ T cells (Teff) (Fig. 2C), suggesting a unique exhaustion-related transcriptional profile in CD8+ T cells that is acquired in vivo upon prolonged residence in the TME.
Epigenetic profiling using assay for transposase-accessible chromatin with sequencing (ATACseq) confirmed a significant enhancement of overall chromatin accessibility near the transcription start site of the genes encoding these myeloid-related genes in Tex d14 versus Tnaïve cells (Fig. 2D). A more detailed analysis of signal tracks of chromatin accessibility peaks at different gene loci revealed that Tox – a major transcriptional and epigenetic regulator of T cell exhaustion (Alfei et al., 2019; Khan et al., 2019; Scott et al., 2019; Yao et al., 2019) – as well as other well-known genes associated with exhaustion programs (Pdcd1, Cd38, Havcr2, Ctla4, Lag3 and Entpd1 (Pauken et al., 2016; Philip et al., 2017)) showed increased chromatin accessibility at promoter regions in Tex when compared to Tnaïve (Fig. 2E and Fig. S2C). This enhanced accessibility was also observed for myeloid-related genes Csf1, Ccl3, Ccl4 and Ccl5 in Tex when compared to Tnaïve (Fig. 2F). In addition, utilizing transcriptional profiles of T cells isolated from human cancers, we found that increased expression of a T cell exhaustion score correlated significantly with the expression of Csf1, Ccl3 and Ccl5 in T cells in a dataset comprising hundreds of patients across a dozen cancer indications (Combes et al., 2021) (Fig. 2G).
CD8+ Tex shape the myeloid compartment in mouse melanoma
To directly study the functional significance of chemokine gene expression by Tex, we adopted a transwell experimental system using OT-I T cells with varying activation states in the bottom well, and bone marrow-derived monocytes in the upper transwell insert (Fig. 3A). After 24 hrs of culture, a significantly higher number of monocytes had migrated through the transwell membrane towards Tex when compared to Teff, Tnaïve or no T cells (Fig. 3B), demonstrating that Tex actively secrete factors that recruit monocytes while T cells in other activation states do not. Phenotypic analysis by flow cytometry revealed that monocytes co-cultured for 2 days with Tex also show increased uniformity and/or magnitude of expression of CD80, CD86, H2kB and MHCII when compared to Teff, Tnaïve or no T cells (Fig. 3C), suggesting that Tex-derived factors augment antigen-presentation potential in differentiating myeloid cells.
To assess whether Tex actively shape the myeloid compartment in tumors in vivo, we were faced with a dearth of available models that allow for specific (conditional) deletion of exhausted T cells from the TME. Therefore, we decided to use a more widely used approach to systemically deplete CD4+ or CD8+ T cells from B78ChOVA-bearing mice using depleting antibodies (Fig. 3D). Interestingly, while the ratio of monocytes to macrophages did not significantly change upon CD4- or CD8-depletion (Fig. 3E), their phenotype was drastically affected by CD8+ T cell depletion in ways consistent with the results of our in vitro co-cultures. CD11b+F4/80+ TAM showed significantly reduced expression of H2kB, MHCII and CD11c in the absence of CD8+ T cells, but not CD4+ T cells (Fig. 3F, G). In addition, CD8+ T cell depletion resulted in increased expression of ‘pro-tumorigenic M2-marker’ CD206 on TAM (Fig. 3F, G), suggesting that CD8+ T cells specifically shape myeloid cell phenotype in the TME favoring an ‘M1-like’ antigen-presenting state.
To test whether the production of CSF1 by lymphocytes had any functional relevance for myeloid composition in the TME, we generated mixed bone marrow chimeras in which Rag1−/− bone marrow was mixed 50:50 with either CSF1op/op or CSF1op/+ bone marrow and transferred into lethally irradiated Rag1−/− recipient mice (Fig. S3A). After a recovery period of 6-10 weeks, these mice were inoculated with B78ChOVA melanomas for 21 days after which the myeloid compartment in the TME was analyzed by flow cytometry. CSF1-deficiency in lymphocytes (Rag1−/−:CSF1op/op chimeras) did not affect primary tumor growth (Fig. S3B) and modestly reduced the influx of total CD45+ leukocytes compared to control animals (Rag1−/−:CSF1op/+) (Fig. S3C). In the myeloid compartment, the proportion of Ly6C+ monocytes was significantly enriched in Rag1−/−:CSF1op/op chimeras, while macrophage proportions were lower, resulting in an increased monocyte/macrophage ratio in Rag1−/−:CSF1op/op chimeras (Fig. S3D). Of note, the proportion of CD103+ cDC1 and CD11b+ cDC2 were not appreciably modulated by CSF1-deficiency in lymphocytes (Fig. S3E). In line with the results presented in Fig. 3F and G, the levels of H2kB, MHCII and CD11c on CD11b+F4/80+ TAM were lower in Rag1−/−:CSF1op/op chimeras when compared to Rag1−/−:CSF1op/+ chimeras (Fig. S3F), while expression of CD86 and CD206 were modestly increased. Despite the large biological variation among these samples, the results support the notion that exhausted CD8+ T cells, at least partially through expression of CSF1, contribute to TAM maturation and induce an antigen-presentation phenotype in the TME.
Macrophages and CD8+ T cells engage in unique long-lived interactions and synapse formation
Since our data suggests that CD8+ Tex shape myeloid cell phenotype towards an antigen-presenting state, we took a more detailed look at the interactions between TAM and CD8+ T cells in the TME. Using 2-photon microscopy, we have previously shown that newly infiltrated antigen-specific CD8+ T cells in the TME preferentially localize in TAM-rich areas, and are captured in prolonged interactions with TAM that result in the onset of exhaustion programs (Engelhardt et al., 2012; Broz et al., 2014; Boldajipour et al., 2016). Conventional wide field imaging demonstrated that, also ex vivo and outside of the context of the TME, previously activated OT-I CD8+ T cells interact significantly longer with TAM sorted from OVA-expressing tumors as compared to in vitro generated bone marrow-derived dendritic cells (BMDC) unloaded or loaded with the cognate peptide SIINFEKL (SL8) (Fig. 4A, B), suggesting a stable and persistent interaction between TAM and CD8+ T cells despite their consistent inability to stimulate T cell proliferation.
Using lattice light sheet microscopy, we found that this stable interaction between TAMs and CD8+ T cells results in small-scale clustering of TCR at the TAM-interaction site on CD8+ T cells (Fig. 4C-E and Fig. S4A), consistent with this being a signaling interaction. Calcium imaging revealed that, unlike CD103+ cDC1 – potent inducers of CD8+ T cell activation (Broz et al., 2014; Roberts et al., 2016; Salmon et al., 2016; Spranger et al., 2017) – that trigger a transient flux, B16ChOVA-derived TAM induce a weak, but long-lasting Ca2+ flux in CD8+ T cells upon recognition of cognate antigen (Fig. 4F (left)). This flux was likely antigen-dependent since TAM and CD103+ cDC1 isolated from non-OVA-expressing B16F10 melanomas did not induce a TCR trigger in OT-I CD8+ T cells (Fig. 4F (right)). While the transient Ca2+ flux triggered by CD103+ cDC1 is sufficient to induce proliferation of CD8+ T cells, the antigen-specific trigger provided by TAM fails to support proliferation (Fig. 4G). Thus, despite actively and profoundly engaging T cells in a unique long-lasting antigen-specific signal-generating synaptic interaction, TAM fail to fully support CD8+ T cell activation and proliferation.
Unique TCR engagement by TAM induces exhaustion programs in CD8+ T cells
To take a more detailed look at the TAM-induced ‘dysfunctional’ TCR trigger, we examined ex vivo co-cultures of previously activated OT-I CD8+ T cells with TAM or CD103+ cDC1 isolated from B16ChOVA and B16F10 melanomas, or in vitro generated BMDC devoid of any antigen as a negative control. As expected, after 3 days of co-culture only CD8+ T cells that had encountered CD103+ cDC1 expressing their cognate antigen displayed a CD44hi IRF4hi fully activated phenotype (Fig. 5A). Interestingly, when compared to the successful signal provided by CD103+ cDC1, TAM only modestly induced expression of activation marker CD44, while providing a similar strength in TCR trigger as suggested by the level of IRF4 expression (Fig. 5A). In chronic viral infections, IRF4 has been implicated as a main regulator of transcriptional circuits inducing and sustaining T cell exhaustion (Man et al., 2017). In line with this notion, TAM induced a significant increase in expression of PD1 and TOX in CD8+ T cells in an antigen-specific manner and similar to CD103+ cDC1 (Fig. 5A), but nevertheless fail to support proliferation (Fig. 4G).
Previous work has shown that hypoxia – in combination with chronic antigenic stimulation but not in the absence of stimulation – is required to obtain a ‘full blown’ exhausted phenotype in CD8+ T cells in vitro (Scharping et al., 2021). To study whether the signals from TAM-Tex interactions are sufficient to ‘prime’ T cells for exhaustion despite being deficient at inducing their proliferation, we performed co-culture experiments under hypoxic (1.5% O2) and normoxic (21% O2) conditions. Interestingly, the proportion of PD1+ CD8+ T cells induced by B16ChOVA-derived TAM was much more pronounced when cultured in hypoxic conditions when compared to normoxia, especially when TAM were numerically in excess (Fig. 5B, C). Together these data demonstrate that TAM can prime the onset of exhaustion programs in CD8+ T cells, a process that is exacerbated in hypoxic conditions.
Using the hypoxia tracer pimonidazole, we found that B78ChOVA melanomas show highly hypoxic areas towards the inner regions of the tumor, and away from CD31+ blood vessels (Fig. 5D). Moreover, we find that these hypoxic regions are surrounded by patches of CD11b+ macrophages (Fig. 5D). In line with this observation, expression of hypoxia-related genes (Car9, Hif1a, Hif2a, Glut1, Vegfa, Vhl) is upregulated in exhausted CD8+ T cells upon prolonged residence in the TME (Fig. 5E), suggesting that tumor-infiltrated Tex experience severe hypoxia in the TME. Interestingly, flow cytometric analysis revealed that TAM experience more severe levels of hypoxia compared to exhausted CD8+ T cells and CD45— tumor cells (Fig. 5F, G). Moreover, the degree of hypoxia was dramatically reduced in residual TAM, but not in Tex, after CSF1R blockade (Fig. 5F, G).
ZipSeq mapping reveals spatial coordination of TAM-Tex interaction dynamics in the TME
To better understand the spatial coordination of the dynamic interplay between TAM and Tex in the TME, we utilized ZipSeq, a spatial transcriptomics approach that allows us to map gene expression patterns in single cells based on their localization in the TME by printing barcodes directly onto cells in tissue (Hu et al., 2020). We used a novel CD206-LSL-Venus-DTR mouse model in which expression of a Venus fluorescent reporter is driven by the endogenous CD206 promoter (Fig. S5A). When crossed to the CSF1RCre strain, we found Venus-labeling of the majority of tumor-associated myeloid cells in B78ChOVA melanomas (Ray A. et al, in preparation). We utilized this model to clearly define distinct regions in the tumor such as the outer rim, middle and inner compartment based on the mCherry signal in cancer cells and the Venus-expression in CD206+ myeloid cells (Fig. 6A) and apply unique Zipcodes to the surface of immune cells in each of those regions. After dissociation of tumors, we sorted CD45+ immune cells and encapsulated them for our modified 10x Genomics scRNA-seq workflow (Hu et al., 2020).
uMAP analysis of the entire immune compartment revealed prototypical and predominant clusters of T cells and monocytes/macrophages, and smaller clusters of natural killer (NK) cells and dendritic cells (DC) (Fig. 6B and Fig. S6A, B). Some of these populations were enriched in specific regions in the TME (Fig. 6C). We further subsampled the CD8+ T cell subset and thereby revealed that CD8+ T cells with a more naïve phenotype are enriched at the outer regions of the tumor, while exhausted CD8+ T cells mainly localize deeper inside the TME (Fig. 6D). Subsampling the monocyte/macrophage subset revealed a distinct localization pattern for different subsets of macrophages; namely, RetnlaHI macrophages were exclusively found at the outer regions of the TME, while ApoeHI, Ms4a7HI and proliferating macrophages are skewed towards the interior of the TME (Fig. 6E).
In line with our previous data obtained using the MMTV-PyMTChOVA mouse model for spontaneous breast cancer (Hu et al., 2020), we found that CD8+ T cells show an increased exhaustion score (Wherry et al., 2007) when located in the inner regions of the TME (Fig. 6F, left). Moreover, we applied pseudotime analysis (Cao et al., 2019) on the monocyte/macrophage subset, and specified the Ly6c2HI monocyte-like cells as the root state of the trajectory which resulted in the ApoeHI, Ms4a7HI and proliferating macrophages as the terminally states (Fig. S6C, D), consistent with our previous findings (Hu et al., 2020; Mujal et al., 2021). When we overlaid Zipseq spatial localization on our pseudotime trajectory, we found a correlation with pseudotime progressing from monocyte-like early states at the outer regions to terminally differentiated TAM states in the tumor core (Fig S6E). This was also reflected in an advanced monocyte/macrophage pseudotime score when moving from outer towards the interior of the tumor (Fig. 6G, left). Interestingly, the positive correlation between the expression of exhaustion-related genes in CD8+ T cells and macrophage maturation towards the inner regions of the TME coincided with an increased glycolytic score (Argüello et al., 2020) in CD8+ T cells, but not in the monocyte/macrophage fraction (Fig. 6F, right and Fig. 6G, right), consistent with a more hypoxic microenvironment which correlates with this dysfunctional crosstalk between Tex and TAM.
To predict interaction likelihood between different cell types in distinct regions in the TME, we used CellChat analysis, which uses a curated database of receptor-ligand interactions to highlight likely cell-cell interactions (Jin et al., 2021). These analyses revealed that expression of the CSF1-CSF1R ligand-receptor pair is significantly enriched in likelihood, and especially in the inner region versus outer region of tumors (Fig. 6H). Interestingly, CSF1 is found to be exclusively expressed by Tex in those regions (Fig. 6H). CellChat analysis also predicted that the main receivers of Tex-derived CSF1 in the inner regions of the tumor are Ms4a7HI macrophages, proliferating macrophages and ApoeHI macrophages (Fig. 6H), pointing towards a co-dependency between TAM and Tex. In line with this, we found a positive correlation between an ‘exhaustion’ signature, as well as normalized Csf1 and Ccl4 expression in CD8+ T cells and macrophage maturation (monocyte/macrophage pseudotime score), when moving from the outer towards the inner regions of the tumor (Fig. 6I). Conversely, expression of genes associated with antigen presentation in monocyte/macrophages gradually decreased when moving closer towards the inner regions of the tumor (Fig. 6I). These data support a model in which monocytes, as they move inward and differentiate toward terminal TAM, downregulate antigen presentation in concert with the development of the exhausted state in T cells.
Discussion
We mechanistically dissected a cellular co-alignment by which tumor-associated macrophages (TAM) and exhausted CD8+ T cells (Tex) in the TME co-exist in a self-enforcing positive feedback loop in mouse and human cancers. This includes finding that the secretion of growth factors and chemokines by one induces the other, the key interaction biology — a weakly stimulatory, yet long-duration synapse that ‘primes’ T cells for exhaustion — and spatial transcriptomics that demonstrate the co-evolution of these differentiated cell states, across space, in tumor tissue. Together, this demonstrates a principle of co-evolution of immunosuppressive cell types in the TME that supports immune evasion rather than destruction of the tumor.
The presence of TAM in solid tumors often correlates with poor prognosis and failure of response to anti-cancer therapies (Zhang et al., 2012; De Palma and Lewis, 2013). These findings are consistent with the established role of TAM in suppressing anti-tumor T cell immunity (DeNardo and Ruffell, 2019), and recent single cell RNA sequencing studies and other immune profiling approaches have hinted towards a potential link between the presence of TAM and exhausted CD8+ T cells in several different cancer types (Bi et al., 2021; Braun et al., 2021; Combes et al., 2021; Hong et al., 2021; Hu et al., 2020; Mujal et al., 2021; O’Connell et al., 2021; Wagner et al., 2019). However, these computational predictions still require experimental investigation to establish causality. Building on previous findings from our own lab and others that CD8+ T cells preferentially localize in TAM-rich areas in the TME (Boissonnas et al., 2013; Boldajipour et al., 2016; Broz et al., 2014; Engelhardt et al., 2012; Peranzoni et al., 2018), we show here that the evolution of these long considered immunosuppressive cell types in the TME is extensively linked in a causal circuit.
Macrophages are known to display a remarkable heterogeneity and plasticity that is dependent on a variety of environmental cues, some of which are derived from T cells (DeNardo and Ruffell, 2019; Guerriero, 2019). However, prior to our study, a role for Tex in shaping macrophage phenotype and function has not been reported. We find in our models that intratumoral Tex are the main immune population producing CSF1 to actively recruit monocytes and modulate their differentiation trajectory favoring antigen-presentation. Thinking beyond these models, other cell types in the TME, including tumor cells and fibroblasts, can also modulate myeloid biology through secretion of CSF1 (Buechler et al., 2021), and so there may be settings in which additional cell types also contribute to the establishment of the TAM-Tex axis. However, in our data CSF1 is a predominant and consistent ‘exhaustion’ gene, and indeed previous reports have identified CCL3, 4 and 5 as amongst the most differentially expressed genes in chronic viral infection-associated exhaustion (Wherry et al., 2007). We consider it likely that Tex express additional myeloid-modulating factors that have yet to be identified and may now be sought, based on these studies.
Computational analysis of our spatial ZipSeq data suggests that Tex-derived CSF1 is most likely to affect specific subpopulations of terminally differentiated macrophages that are enriched in the inner regions of the TME, including Ms4a7HI, ApoeHI, and proliferating TAM. While other scRNAseq studies have also reported on the existence of multiple different subpopulations of monocytes and TAM in the TME (Hu et al., 2020; Katzenelenbogen et al., 2020; Molgora et al., 2020; Mujal et al., 2021), it remains to be determined whether these are functionally distinct from one another in their interactions with CD8+ T cells and their abilities to modulate the onset of T cell exhaustion.
Regardless of their ability to phagocytose large amounts of antigen, TAM are often considered inferior in antigen processing and presentation as compared to conventional dendritic cells. Here we show that antigen-presenting TAM capture CD8+ T cells in uniquely long-lasting synaptic interactions characterized by the formation of variegated TCR microclusters. Despite expressing similar levels of MHC class I and II, co-stimulatory molecules and genes involved in cross-presentation as do cDC1 (Broz et al., 2014), TAM trigger only a weak TCR stimulation that fails to support proliferation, but clearly primes the onset of T cell exhaustion which is not observed in the absence of TCR-ligands presented by these TAM. Notably, blockade of immune checkpoint molecules PD1/PD-L1 and CTLA4 is unable to license proliferation in T cells responding to TAM (Engelhardt et al., 2012). It is clear that more work is required to better understand the fundamental nature of disparate TCR triggers and co-stimulation over time that contribute to the hyporesponsive state in T cells during tumorigenesis, as elegantly reviewed recently (Philip and Schietinger, 2021).
Recent studies have reported that development of T cell exhaustion during chronic infection and cancer occurs in a multistep fashion, revealing distinct subtypes with unique transcriptional and epigenetic dynamics, as well as their ability to respond to immune checkpoint blockade (Im et al., 2016; Philip et al., 2017; Satpathy et al., 2019; Siddiqui et al., 2019; Jansen et al., 2019; Miller et al., 2019; Beltra et al., 2020; Pritykin et al., 2021). The decision-making during this bifurcative process seems to be tightly regulated by transcription factors like IRF4 (Utzschneider et al., 2020; Chen et al., 2021; Seo et al., 2021) which we find is strongly upregulated by TAM despite only subtle other signs of TCR engagement. Furthermore, our in vitro co-culture studies suggest that TAM at least are capable of mediating the early stages of exhaustion, which is exacerbated in hypoxic conditions. Hypoxia was recently shown to be an important co-factor in the induction of exhaustion in vitro (Scharping et al., 2021), and ZipSeq transcriptomics here places terminal exhaustion in vivo preferentially in hypoxic regions of the TME. In line with this, other recent studies have elegantly demonstrated how metabolic insufficiencies drive mitochondrial stress in T cells contributing to exhaustion phenotypes in chronic infections and cancer (Bengsch et al., 2016; Thommen et al., 2018; Vardhana et al., 2020; Scharping et al., 2021).
Taken together, our work dissects a novel spatiotemporal co-evolution between exhausted CD8+ T cells and TAM in the TME that supports tumor evasion rather than tumor destruction. We believe that co-dependency of different lineages may explain some of the resistance of the TME to targeting — removing just one cell population will still leave the other, to influence re-establishment of the targeted one. Thus, therapeutic strategies may need to break the biology underlying the TAM-Tex axis at multiple points. Doing so may work in conjunction with existing immunotherapies to enhance anti-tumor immunity, and thereby expand the proportion of cancer patients who benefit from immunotherapies.
Author contributions
K.K. designed and conducted most experiments, data analysis, and drafted the manuscript. K.H.H. and A.R. designed, performed and analyzed ZipSeq experiments. A.J.C. and B.S. analyzed human cancer patient datasets. T.H. assisted with co-culture experiments. A.A.R. analyzed RNAseq data. E.C., K.M. and J.A. provided technical support and assisted with imaging experiments. T.C. discussed data and project direction. Q.S., J.A.B. and A.T.S. performed ATACseq and assisted with data analysis. M.F.K. designed the experiments, interpreted data, and with other co-authors, developed the manuscript.
Declaration of Interest
M.F.K. is a founder, and shareholder in Pionyr Immunotherapeutics and in Foundery Innovations, that prosecute and develop novel immunotherapeutics respectively.
STAR Methods
Key Resource Table
See Excel Key Resource Table.
Contact for Reagent and Resource Sharing
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Matthew F. Krummel (matthew.krummel{at}ucsf.edu).
Experimental Model and Subject Details
Human tumor samples
Flow cytometry on RCC samples: RCC were transported from various cancer operating rooms or outpatient clinics. All patients consented by the UCSF IPI clinical coordinator group for tissue collection under a UCSF IRB approved protocol (UCSF IRB# 20-31740). Samples were obtained after surgical excision with biopsies taken by Pathology Assistants to confirm the presence of tumor cells. Patients were selected without regard to prior treatment. Freshly resected samples were placed in ice-cold DPBS or Leibovitz’s L-15 medium in a 50 mL conical tube and immediately transported to the laboratory for sample labeling and processing. The whole tissue underwent digestion and processing to generate a single-cell suspension. In the event that part of the tissue was sliced and preserved for imaging analysis, the remaining portion of the tissue sample was used for flow cytometry analysis as described in Combes et al (Combes et al., 2021).
Samples from the following indications were used for RNAseq on FACS-isolated cell fractions performed as described previously (Combes et al., 2021): Bladder cancer (BLAD), colorectal cancer (CRC), glioblastoma multiforme (GBM), endometrial and ovarian cancer (GYN), hepatocellular carcinoma (HEP), head and neck squamous cell carcinoma (HNSC), (KID), lung adenocarcinoma (LUNG), skin cutaneous melanoma (MEL), pancreatic ductal adenocarcinoma (PDAC), pancreatic neuroendocrine tumors (PNET), sarcoma (SRC).
Mice
All mice were treated in accordance with the regulatory standards of the National Institutes of Health and American Association of Laboratory Animal Care and were approved by the UCSF Institution of Animal Care and Use Committee. The following mice were purchased for acute use or maintained under specific pathogen-free conditions at the University of California, San Francisco Animal Barrier Facility: C57BL6/J, C57BL6/J CD45.1, OT-I, P14 LCMV, Rag1-/-, CSF1op/op, mTmG. With the exception of CSF1op/op, all mice used in experimentation were bred to a C57BL6/J background. Mice of either sex ranging in age from 6-12 weeks were used for experimentation. For experiments using the transgenic MMTV-PyMTChOVA strain (Engelhardt et al., 2012), only mammary tumor-bearing females were used ranging in age from 12-20 weeks. Treatments in MMTV-PyMTChOVA mice were started when mammary tumors reached ∼25mm2 in size. CSF1RCreCD206-LSL-Venus-DTR mice (Ray, A. et al. in preparation) were generated and used for ZipSeq. Food and water were provided ad libitum.
Tumor cell lines
Tumor cell lines B16F10 (CRL-6475, ATCC), B16ChOVA (Roberts et al., 2016; Binnewies et al., 2019) and B78ChOVA (Engelhardt et al., 2012; Broz et al., 2014) were cultured under standard conditions 37°C in 5% CO2 in DMEM (GIBCO), 10% FCS (Benchmark), 1% Pen/Strep/Glut (Invitrogen).
Method Details
Tumor growth experiments
For tumor studies, adherent tumor cells were grown to confluency and harvested using 0.05% Trypsin-EDTA (GIBCO) and washed 3x with PBS (GIBCO). 1.0×105 – 2.5×105 cells in PBS were resuspended in a 1:1 ratio with Growth Factor Reduced Matrigel (Corning) and a final volume of 50µl was injected subcutaneously into the flanks of anaesthetized and shaved mice. Tumors were allowed to grow for 14-21 days unless otherwise noted, before tumors and tumor-draining lymph nodes were harvested for analysis.
Adoptive T cell transfers
Inguinal, axillary, brachial and mesenteric lymph nodes (LN) or spleens were isolated from CD45.1 OT-I or P14 LCMV mice. LN and spleens were meshed through 70µm filters and treated with ACK red blood cell lysis buffer. CD8+ T cells were purified using EasySep CD8 negative selection kits (Stemcell Technologies). 1×105 (for >14 day read-out) – 2×106 T cells (for day 4 read-out) were adoptively transferred through retro-orbital injection in 100µl PBS.
For the comparison of OT-I T cells and p14 LCMV T cells, mice received a 1:1 mix of both T cells in 100µl of PBS through retro-orbital injection. The following day mice were inoculated with a bolus of CFA containing gp33-peptide (50µg/mouse; Anaspec) and SL8/SIINFEKL peptide (50µg/mouse; Anaspec) subcutaneously, to sustain both T cell populations.
In vivo antibody treatment
For macrophage depletions, mice received anti-CSF1 (clone 5A1; BioXCell), anti-CSF1R (clone AFS98; BioXCell) or corresponding isotype controls, Rat IgG1k (clone HRPN; BioXCell) and Rat IgG2a (clone 2A3; BioXCell), respectively. Antibodies were injected intraperitoneally at an initial dose of 1mg/mouse followed by 0.5mg/mouse every 7 days.
For T cell depletion studies, mice received anti-CD4 (clone GK1.5; BioXCell), anti-CD8a (clone 2.43; BioXCell) or corresponding isotype control, Rat IgG2b (clone LTF-2; BioXCell) dosed at 250 µg/mouse every 3-4 days.
Generation of mixed bone marrow chimeras
Mixed bone marrow chimeras were generated as described previously (Barry et al., 2018). Briefly, Rag1-/- mice were lethally irradiated with 1,100 rads of irradiation in two doses 3-5 hours apart. 2-5×106 bone marrow cells, consisting of 50% Rag1-/- and 50% CSF1op/op or CSF1op/+ bone marrow, were injected retro-orbitally to reconstitute irradiated mice. Chimeric mice were allowed to recover for 6-10 weeks, upon which mice were inoculated with B78ChOVA tumors subcutaneously.
Mouse tissue digestion and flow cytometry
Tumors were harvested and processed to single cell suspensions as described previously (Barry et al., 2018; Binnewies et al., 2019). Briefly, tumors were isolated and mechanically minced, followed by enzymatic digestion with 200µg/ml DNAse (Sigma-Aldrich), 100U/ml Collagenase I (Worthington Biochemical) and 500U/ml Collagenase Type IV (Worthington Biochemical) for 30 minutes at 37°C while shaking. Enzymatic activity was quenched by adding equal amounts of FACS buffer (2% FCS in PBS), and cell suspensions were filtered to obtain single cell suspensions. TdLN were isolated and meshed over 70µm filters in PBS to generate single cell suspensions. For each sample, 5-10×106 cells were used for staining for flow cytometry. Cells were washed with PBS prior to staining with Zombie NIR Fixable live/dead dye (Biolegend) for 20 min at 4°C. Cells were washed in PBS followed by surface staining for 30 min at 4°C with directly conjugated antibodies diluted in FACS buffer containing anti-CD16/32 (BioXCell) to block non-specific binding. Cells were washed again with FACS buffer. For intracellular staining, cells were fixed for 20 min at 4°C using the FOXP3 Fix/Perm kit (BD Biosciences), and washed in permeabilization buffer. Antibodies against intracellular targets were diluted in permeabilization buffer and cells were incubated for 30 min at 4°C followed by another wash prior to read-out on a BD LSR Fortessa SORP cytometer.
Fluorescence Activated Cell Sorting
Single cell suspensions from tumors were prepared as described above. For T cell isolations, single cell suspensions were enriched for mononuclear cells using Ficoll-Paque Premium 1.084 (GE Healthcare). For isolation of myeloid cells, single cell tumor suspensions were enriched for CD45+ cells using EasySep biotin positive selection kit (Stemcell Technologies). Enriched cells were stained for 30 min at 4°C with directly conjugated antibodies diluted in FACS buffer containing anti-CD16/32 (BioXCell) to block non-specific binding. Cells were washed again with FACS buffer and filtered over a 70µm mesh. Immediately prior to sorting, DAPI was added to exclude dead cells. Cells were sorted on a BD FACSAria Fusion and BD FACSAria2. Sorted T cells were collected directly in lysis buffer (Invitrogen) for RNA sequencing or in RPMI (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen) and 50µM β-mercaptoethanol (GIBCO) at 4°C for further use ex vivo. Sorted myeloid cells were collected in DMEM (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen) at 4°C for further use ex vivo.
T cell cytokine analysis
For analysis of cytokine production by endogenous and adoptively transferred T cells, 5-10×106 LN and tumor cells were re-stimulated for 3-5 hours in RPMI (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen), 50µM β-mercaptoethanol (GIBCO) containing PMA (50ng/ml; Sigma-Aldrich), ionomycin (500ng/ml; Invitrogen) and brefeldin A (3µg/ml; Sigma-Aldrich) at 37°C in 5% CO2. Cells were washed and stained for intracellular flow cytometric analysis.
RNA sequencing
mRNA from cells were isolated using DynaBead Direct and then converted into amplified cDNA using the Tecan Ovation RNA-Seq System V2 kit, following the manufacturer guidelines. The dsDNA is tagmented, amplified and undergoes clean up with AMPure XP bead, using the Illumina Nextera XT DNA Library Prep Kit. The resulting sequencing library is QC’d using an Agilent Bioanalyzer HS DNA chip to assess fragment size distribution and concentration. Libraries were pooled prior to single-end sequencing on and Illumina MiSeq/MiniSeq to ensure quantify library complexity. Libraries with less than 10 percent of the reads aligned to coding regions, or fewer than 1,000 unique reads in total were rejected. The validated libraries were re-pooled based on the percentage of reads in coding regions and submitted to the UCSF Center for Advanced Technology for 150bp paired end sequencing on an Illumina NovaSeq 6000.
Raw fastq reads were QC’d and trimmed to remove adapter contamination, and poly-G artifacts using using fastp version 0.19.6 (Chen et al., 2018). Reads with fewer than 20bp post-trimming were discarded. Trimmed reads were aligned to the GRCm38 reference sequence annotated with Gencode V25 (Frankish et al., 2019) using STAR version 2.6.1b (Dobin et al., 2013) with the following parameters (--quantMode GeneCounts–outFilterMismatchNoverLmax 0.04 --alignIntronMax 100000 --alignMatesGapMax 100000 -- alignSJDBoverhangMin 10 --alignSJstitchMismatchNmax 5 -1 5 5 --chimSegmentMin 12 -- chimJunctionOverhangMin 12 --chimSegmentReadGapMax 3 --chimMultimapScoreRange 10 -- chimMultimapNmax 10 --chimNonchimScoreDropMin 10 --peOverlapNbasesMin 12 --peOverlapMMp 0.1) STAR-generated reads counts from each library were processed using the limma/Voom pipeline (Law et al., 2014; Smyth, 2005) using the edgeR package (Robinson et al., 2010). Briefly, the read counts are loaded into a DGEList object to generate Counts Per Million (CPM), and then filtered to retain only genes with at least 10 counts in a worthwhile number of samples and at least 15 counts across all samples. The CPM matrix is normalized using TMM Trimmed mean of M-values and processed using voom to estimate the mean-variance relationship to identify edge weights that can be used to fit to a linear model with limma lmFit. Differential gene expression between two groups of empirical Bayes moderation of the standard errors towards a global value. A list of transcriptional DEGs between Tnaïve and Tex d14 with a FC equal to or >5 was generated and gene set enrichment analysis was performed using the MGI Mammalian Phenotype Level 4 database in Enrichr (Chen et al., 2013; Kuleshov et al., 2016; Xie et al., 2021).
ATAC sequencing
ATAC-seq samples were processed according to the Omni-ATAC protocol (Corces et al., 2017). 5×104 cells per replicate were lysed in 50 µL ATAC resuspension buffer supplemented with 0.1% NP40, 0.1% Tween-20, and 0.01% Digitonin. After lysis, nuclei were transposed using 2.5 µl Tn5 transposase in a 50 µl reaction for 30 min at 37°C. Finally, the transposed DNA was purified using a commercial PCR cleanup kit and libraries were prepared for sequencing. 2×75 paired end sequencing was performed on an Illumina sequencer.
ATAC-seq computational analysis was performed as previously described (Weber et al., 2021). Briefly, read trimming and filtering was performed with fastp. Reads were mapped to the hg38 reference genome using hisat2 with the --no-spliced-alignment option. Picard was used to remove duplicates from bam files. We removed any reads not mapping to chromosomes 1-22 and chrX (ie chrY reads, mitochondrial reads, and other reads were discarded). The deduplicated and filtered fragments were then formatted into a bed file. Peaks were called using MACS2. Peaks from each sample were iteratively merged into a high confidence union peak set for all samples as previously described (Corces et al., 2018). A peak by sample matrix was created by overlapping fragments in each sample with each peak, and this matrix was used to perform differential peak analysis in DESeq2. Genome coverage files were created from the fragments file by loading the fragments into R and then exporting bigwig files normalized by reads in transcription start sites using ‘rtracklayer::export’. Normalized track files were visualized using the Integrative Genomics Viewer.
qRT-PCR
RNA was extracted from FACS-sorted immune cell populations using Qiagen RNeasy Micro kit (Qiagen) and the yield was measured using Nanodrop. cDNA first-strand synthesis was performed using High-Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific) using random primers. qRT-PCR analysis was performed using Taqman probes targeting Ccl3 (Mm00441259_g1), Ccl5 (Mm01302427_m1), Csf1 (Mm00432686_m1) and Gapdh (Mm99999915_g1). All probes were obtained from Life Technologies. For amplification reactions, iTaq Universal Probes Supermix was used according to manufacturer’s instructions. qRT-PCR was performed on a QuantStudio 12K Flex lightcycler (Applied Biosystems by Life Technologies). For quantification the delta Ct method was used: delta Ct sample — delta Ct reference gene. All transcripts were normalized to Gapdh.
Monocyte recruitment transwell assays
Bone marrow was obtained from femurs and tibia of CD45.1 mice, and monocytes were isolated using EasySep Mouse Monocyte Isolation kits (Stemcell Technologies). For transwell assays, 1×105 monocytes were added to top inserts containing 5.0 µm pore polycarbonate membrane (Corning). 0.5×105 naïve, previously activated or exhausted OT-I T cells were cultured in bottom wells in RPMI (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen) and 50µM β-mercaptoethanol (GIBCO). Migration through the membrane was analyzed after 24 hours of culture. Plate was briefly centrifuged briefly at 1000rpm for 1 min to collect cells stuck to the membrane. Cells were collected for analysis by flow cytometry. Absolute counting beads (Life Technologies) were added for quantification of the number of migrated cells.
Generation of activated T cells
OT-I T cells were activated in vitro as described previously (Broz et al., 2014). Briefly, OT-I lymph node cells were stimulated with B6 splenocytes pulsed with SL8 peptide (100ng/ml; Anaspec) for 30 min at 37°C and then washed 3 times. On day 2-3, cells were expanded by adding human IL-2 (2U/ml; Peprotech) to fresh RPMI (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen), 50µM β-mercaptoethanol (GIBCO). Cells were used for co-culture assays on day 4-5. Prior to use dead cells were excluded using Ficoll-Paque PLUS (GE Healthcare).
Bone marrow-derived dendritic cells
BMDC were generated as described previously (Broz et al., 2014). Briefly, bone marrow was obtained from femurs and tibia of C57BL6/J mice and cultured in DMEM (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen) in the presence of 7.5 ng/ml GM-CSF (Peprotech) for 6-8 days, followed by the addition of 60ng/ml IL4 (Peprotech) for the last 2 days. Media was refreshed every 3-4 days. For co-cultures studies, BMDC were pulsed with SL8 peptide (100ng/ml; Anaspec) for 30 min at 37°C and then washed 3 times prior to use.
Quantification of APC-T cell interactions
APC (BMDC, BMDC+SL8 or sorted TAM) were obtained from mTmG mice, and previously activated OT-I T cells were stained for 15 minutes at 37°C with 2 µM CFSE (Invitrogen) in PBS and washed in RMPI prior to use. Cells were co-cultured in NUNC 8 well chamber slides (Thermo Scientific) that were coated with fibronectin (2µg/ml; EMD Millipore) in PBS at 37°C for 1 hour before use. APCs in phenol red-free RPMI were allowed to attach to the chamber slides for 20-30 minutes at 37°C and 5% CO2. Right before imaging, T cells (resuspended in 0.1% agarose) were added to the wells and slides were loaded for imaging. To visualize the interaction between different APC populations and T cells, a conventional widefield Zeiss Axiovert 200M was used with a Sutter Lambda XL illumination source, running on µMagellan software. Images were acquired every 2 minutes for 6 hours using a 20x objective. Samples were kept at 37°C using a heated robotic stage. Image analysis was performed in Imaris (Bitplane) and ImageJ.
Lattice light-sheet imaging
Lattice light-sheet (LLS) imaging was performed in a manner previously described (Cai et al., 2017). Briefly, 5 mm diameter round coverslips were cleaned by a plasma cleaner and coated with fibronectin (2µg/ml; EMD Millipore) in PBS at 37°C for 1 hour before use. TAM sorted from B78ChOVA-bearing mTmG mice were dropped onto the coverslip and incubated at 37°C, 5% CO2 for 20–30 min. Previously activated OT-I CD8+ T cells were labeled with CD45-AF647 (clone 30-F11; Biolegend) and TCRβ AF488 (clone H57-597; BioLegend) for 30 min and washed in FACS buffer. Right before imaging, T cells were dropped onto the coverslip containing TAM. The sample was then loaded into the previously conditioned sample bath and secured. Imaging was performed with a 488-nm, 560-nm, or 642-nm laser (MPBC, Canada) dependent upon sample labeling in single or two-color mode. Exposure time was 10 ms per frame leading to a temporal resolution of 4.5 s. Image renderings were created using Imaris software (Bitplane). Quantification of TCR clustering was performed using ImageJ. Briefly, channels were separated and entire T cell membrane versus TAM interaction site was outlined manually. Signal intensity for red (CD45) and green (TCR) channel was calculated. The following formula was used to determine TCR signal intensity for both channels at the synaptic TAM-T cell interaction site: signaling intensity = (intensity at synapse/intensity total membrane).
FURA-2AM Calcium imaging
TAM or CD103+ DC were sorted from B16ChOVA or B16F10 tumors as per description above, and stained with 2 µM CMTMR (Thermo Fisher Scientific) in PBS for 15 minutes at 37°C, followed by a wash in RPMI. Cells were co-cultured in NUNC 8 well chamber slides (Thermo Scientific) that were coated with fibronectin (2µg/ml; EMD Millipore) in PBS at 37°C for 1 hour before use. APCs in phenol red-free RPMI were allowed to attach to the chamber slides for 20-30 minutes at 37°C and 5% CO2. Right before imaging, previously activated OT-I T cells were labeled with FURA-2 AM (0.5µM; Invitrogen) for 15 minutes at RT. Cells were washed and resuspended in phenol red-free RPMI (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen), 50µM β-mercaptoethanol (GIBCO) supplemented with 0.1% agarose and were added to the wells. Imaging was performed using an Zeiss Axiovert 200M microscope with a Sutter Lambda XL illumination source equipped with a 40x oil objective. Images were acquired every 5 seconds for 18-21 minutes. Samples were kept at 37°C using a heated robotic stage. Image analysis was performed in Imaris (Bitplane). Briefly, background subtraction was performed and surfaces were created for APCs and T cells to quantify dwell time to determine whether cells were touching (cut-off equal to or >3). Calcium2+ flux was determined by calculating the average 340/380 ratiometric fluorescence per cell after contact for each time point.
Hypoxyprobe imaging
Mice were injected with pimonidazole hydrochloride in PBS (80mg/kg; Hypoxyprobe) intraperitoneally 1.5 hours prior to sacrifice. Tissues were dissected and processed as described above. For flow cytometry studies, pimonidazole was visualized using anti-pimonidazole antibodies (Pacific Blue Mab-1 clone 4.3.11.3; Hypoxyprobe) after cells were fixed for 20 min at 4°C using the FOXP3 Fix/Perm kit (BD Biosciences), and washed in permeabilization buffer. For imaging studies, dissected tumors were embedded in OCT and sectioned into 10µm cryosections. Cryosections were stored at -80°C until further use. For immunostaining, sections were fixed in 4% PFA (Electron Microscopy Sciences) for 20 minutes at RT, followed by a rinse in PBS containing 1% BSA (Sigma). Sections were blocked in 1% BSA in PBS containing anti-CD16/32 (BioXCell) for 1 hour at RT, and washed. Sections were stained with CD11b-AF594 (clone M1/70;BioLegend), CD31-AF647 (clone 390;BioLegend) and Pacific Blue Mab-1 (clone 4.3.11.3;Hypoxyprobe) for 1 hour at RT, followed by a wash and mounted using Vectashield (Vector Laboratories) and sealed with nail polish. Images were acquired on a Leica SP8 confocal microscope. Data analysis was performed using Imaris (Bitplane).
In vitro APC-T cell co-culture assays
APC populations were sorted from tumors as described above, and co-cultured with 1×105 previously activated OT-I CD8+ T cells labeled with Violet Proliferation dye (VPD; BD Biosciences) at a 1:5 ratio (unless otherwise noted) in RPMI (GIBCO), 10% FCS (Benchmark), Pen/Strep/Glut (Invitrogen), 50µM β-mercaptoethanol (GIBCO) in 96-well round bottom plates. Cells were harvested for analysis 3 days later, unless otherwise noted. In vitro generated BMDC and BMDC+SL8 were used as negative and positive controls, respectively.
For hypoxia experiments, plates containing exact same experimental groups were incubated in a Avatar hypoxic bioreactor (XcellBio) at 1.5% O2 or under ambient 21% O2 for comparison and cells were harvested after 3 days.
ZipSeq
ZipSeq spatial transcriptomics was performed as described previously (Hu et al., 2020). Briefly, B78ChOVA cells were injected subcutaneously as described above and were harvested day 16 post-injection. Tumors were sectioned while live using a compresstome (Precisionary Instruments VFZ-310-0Z) to generate ∼160 µm sections. The sectioning, imaging, spatial barcoding, tumor dissociation, sorting, 10X encapsulation and library construction were identical to the methods described in (Hu et al., 2020). The targeted number of cells for loading was 5000. With this in mind, we aimed for 30,000 reads per cell during sequencing on an Illumina S4 flowcell with a 1:10 molar ratio of Zipcode reads to gene expression reads. Resulting fastq files were processed using the CellRanger 4.0.0 pipeline, aligning to the GRCm38 Mus musculus assembly. CellRanger output thus resulted in ∼359k reads for the gene expression library and ∼40k reads for the Zipcode library.
Analysis of scRNA-Seq
The raw feature-barcode matrix generated by 10X CellRanger was loaded into Seurat (Satija et al., 2015). Cells with mitochondrial read % over 20% and those with less than 500 genes detected were excluded from analysis. Zipcode read counts from CellRanger were also loaded into Seurat as a separate ‘ADT’ assay and using CLR normalized counts, cells with either too few Zipcode reads or mixed Zipcode reads were also excluded from analysis. Following built-in Seurat methods for gene expression normalization and variance stabilization (Single Cell Transform (Hafemeister and Satija, 2019)), cells underwent one more round of clean-up, removing a small cluster of contaminating CD45– cells and another small cluster dominated by mitochondrial and ribosomal genes. This yielded 2765 cells. At this stage, the mean # of UMI’s and # of detected genes was: (25,566 and 4,199 respectively) while for the antibody derived Zipcode tags the mean UMI was 1,891 reads. At this stage, we also determined cluster identities using Seurat’s FindAllMarkers function performed on the log-normalized read counts which by default uses the Wilcoxon Rank Sum test.
For CellChat analysis, this cleaned object was fed into the CellChat workflow, using the built-in mouse ligand-receptor database, and a tri-mean thresholding for significance of interaction. The Seurat object was split into 3 sub-objects based on regional assignment and these 3 objects were separately analyzed using the CellChat workflow for multiple datasets and then merged. For signature score generation, we used Seurat’s built-in AddModuleScore function with gene lists for Glycolysis (Argüello et al., 2020), T cell exhaustion (Wherry et al., 2007), and Antigen Presentation (GO term 0048002) using 50 control features. Full gene lists can be found in the Extended Data. For pseudotime analysis, the monocyte/macrophage sub-object was passed into Monocle v3 without any changes to the UMAP dimensional reduction.
Quantification and Statistical Analysis
Statistical analysis
Unless specifically noted, all data are representative of ≥ 2 separate experiments. Experimental group assignment was determined by random designation. Statistical analyses were performed using GraphPad Prism software. Error bars represent ± standard error of the mean (S.E.M.) calculated using Prism. Specific statistical tests used were paired t-tests, unpaired t-tests, and one-way ANOVA unless otherwise noted. P-values <0.05 were considered statistically significant. Investigators were not blinded to group assignment during experimental procedures or analysis.
Data and Software Availability
Supplementary Figure Legends
Key Resource Table
Acknowledgements
We thank all members of the Krummel laboratory for discussion and support. We also thank the ImmunoX CoLabs at UCSF for technical assistance and support. Figures were created using BioRender.com. Funding: K.K. was supported by a Rubicon postdoctoral fellowship 019.163LW.006 from the Netherlands Organization for Scientific Research (NWO), and a Parker Scholar Award from the Parker Institute for Cancer Immunotherapy. K.H.H. was supported by a Parker Scholar Award from the Parker Institute for Cancer Immunotherapy, and a Jean Perkins postdoctoral fellowship from the American Cancer Society. A.R. and E.C. were supported by postdoctoral fellowships from the Cancer Research Institute. J.A.B was supported by a Stanford Graduate Fellowship and the National Science Foundation Graduate Research Fellowship under Grant No. DGE-1656518. A.T.S. was supported by NIH U01CA260852, a Cancer Research Institute Technology Impact Award, and a Career Award for Medical Scientists from the Burroughs Wellcome Fund. This work was supported by NIH/NCI U54CA163123, 5U01CA217864, R21CA191428 and R01CA197363 to M.F.K.