Resident Kupffer cells and neutrophils drive liver toxicity in cancer immunotherapy

Abstract

Immunotherapy is revolutionizing cancer treatment but is often restricted by toxicities. What distinguishes adverse events from concomitant antitumor reactions is poorly understood. Here, using anti-CD40-treatment in mice as a model of Th1-promoting immunotherapy, we showed that liver macrophages promoted local immune-related adverse events. Mechanistically, tissue-resident Kupffer cells mediated liver toxicity by sensing lymphocyte-derived IFN-𝛾 and subsequently producing IL-12. Conversely, dendritic cells were dispensable for toxicity but drove tumor control. IL-12 and IFN-𝛾 were not toxic themselves, but prompted a neutrophil response that determined the severity of tissue damage. We observed activation of similar inflammatory pathways following anti-PD-1 and anti-CTLA4 immunotherapies in mice and humans. These findings implicated macrophages and neutrophils as mediators and effectors of aberrant inflammation in Th1-promoting immunotherapy, suggesting distinct mechanisms of toxicity and antitumor immunity.

One sentence summary

Resident macrophages and recruited neutrophils cause toxicities in tumor-free tissues following immunotherapy

Introduction

Patients receiving immunotherapy for cancer can experience immune-related adverse events (irAEs) in normal, non-cancerous tissue, which frequently leads to discontinuation or disruption of therapy (1, 2). Toxicity appears to correlate with antitumor efficacy (3, 4); yet, whether similar or different mechanisms drive antitumor immunity and irAEs is largely unknown.

Interferon-gamma (IFN-𝛾) and interleukin-12 (IL-12) induction accompany effective antitumor immune responses, both in mice (5–8) and humans (9–11). These cytokines are characteristic of cell-mediated, T-helper 1 (Th1)-polarized immunity, and are appreciated as important in the body’s response to cancer. Myeloid cell-targeted and lymphocyte-targeted therapies can similarly promote both cytokines (7), and their induction is crucial in the rational design of cancer immunotherapeutics. However, robust activation of IFN-𝛾 and IL-12 can be toxic in both humans and mice (12–15), suggesting that these cytokines may be involved with both productive antitumor immunotherapy responses and irAEs.

To understand how the canonical antitumoral cytokines IL-12 and IFN-𝛾 might detrimentally affect tumor-free tissues in the setting of immunotherapy, we used the Th1-activating myeloid cell agonist anti-CD40 (aCD40). Similar to checkpoint inhibitors, aCD40 causes a Th1-polarized antitumor response; however, while mice tolerated checkpoint inhibitors in tumor-free tissues, aCD40 caused systemic immune activation and multi-organ toxicities. Therefore, we utilized aCD40 as a tool to robustly induce IL-12- and IFN-𝛾-dependent responses in tumor-bearing mice and to interrogate whether some features of the Th1 response could distinguish antitumor immunity from undesired therapy-driven pathology. Understanding mechanisms of toxicities associated with aCD40 treatment is of clinical interest in its own right, as CD40 agonists are potentially powerful to treat cancer, but can trigger adverse events in patients (16). While we found that aCD40 triggered an inflammatory response in various tissue sites, we predominantly focused on the liver to interrogate mechanisms governing toxicity, since this is a clinically important site of irAEs. We further tested whether immune checkpoint blockers (anti-PD-1, anti-CTLA4) could trigger similar inflammatory pathways in mice and humans.

Results

aCD40 triggered pro-inflammatory Th1 cytokines throughout a tumor-bearing host

To study Th1 cytokine responses in tumor-free tissues, we initially analyzed multiple organs throughout IL-12p40 and IFN-𝛾 reporter mice bearing MC38 flank tumors, comparing untreated mice to those receiving agonist aCD40 (aCD40) (Fig. 1A). We chose MC38 tumors because they can be controlled by systemically-delivered Th1-inducing immunotherapies (7, 17) and we have previously shown that aCD40 stimulates a robust Th1 immune response associated with MC38 tumor control (7). aCD40 elevated both IL-12p40 and IFN-𝛾 in nearly all tissues analyzed (Fig. 1B–E, Fig. S1A–B). As an exception, we could not detect an IFN-𝛾 response at the tumor site by flow cytometry, although this response has been documented when using a less invasive approach such as intravital microscopy (7).

Fig. 1. aCD40 triggered canonical anti-tumor Th1 cytokines throughout a tumor-bearing host.

(A) Rationale and workflow to study of irAEs using IL-12 and IFN-𝛾 reporter mice.

(B) Flow cytometry plots exemplifying IL-12p40-EYFP induction in liver on day two following aCD40. Y-axis = viability dye (Zombie Aqua).

(C) IL-12p40-EYFP induction across tissues on day two following aCD40. Values calculated based on % of CD45⁺ events that are EYFP⁺ (n = 3–16 mice/group).

(D) Flow cytometry plots exemplifying IFN-𝛾-EYFP induction in liver on day two following aCD40 treatment. Y-axis as in (B).

(E) IFN-𝛾-EYFP induction across tissues on day two following aCD40. Values calculated as in (C) (n = 5–14 mice/group).

(F) MC38 tumor volumes and changes in body weight for mice treated or not with aCD40 with or without IL-12 or IFN-𝛾 neutralization (n = 5–7 mice/group).

(G) TC-1 tumor volumes and changes in body weight for mice treated or not with aCD40 with or without IL-12 neutralization (n = 7–9 mice/group).

Data are represented as mean ± SEM. For comparisons between two groups, Student’s two-tailed t test was used. For comparisons between multiple groups, one-way ANOVA was used. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Mice experience weight loss in the days following aCD40 therapy (14, 18), which we initially used as a proxy for systemic toxicity. Cytokine neutralization showed that MC38 tumor control and weight loss both depended on IL-12p40 and IFN-𝛾 (Fig. 1F). Neutralizing IL-12p40 similarly diminished tumor control and weight loss following aCD40 in TC-1 tumor-bearing mice (Fig. 1G). Toxicity did not depend on IL-23, which uses IL-12p40 in its heterodimeric structure (Fig. S2). These findings supported that systemically-delivered Th1-promoting immunotherapy induced systemic effects, acting in tumor-free tissues to trigger similar inflammatory pathways to those induced in the tumor.

IL-12 and IFN-𝛾 were interdependent and causative of inflammatory pathology

Anti-CD40 treatment causes liver toxicity in mice, and this irAE is also observed in patients on Th1-promoting immunotherapies (1, 16, 19–21). IL-12 and IFN-𝛾 induction in the liver was consistently robust following aCD40. Hence, we dissected mechanisms of toxicity in this site. Histological analysis showed that liver damage featured portal and prominent lobular hepatitis and broad, confluent areas of necrosis (Fig. 2A). Pathology was linked to IL-12 and IFN-𝛾 induction, as neutralizing either cytokine eliminated liver necroinflammatory lesions (Fig. 2A–B). In another site of irAE, the gastrointestinal tract, aCD40 led to colon crypt hyperplasia in the days following treatment, which was also abrogated by neutralizing IL-12 (Fig. S3A–B). Thus, both IL-12 and IFN-𝛾 drove irAEs in the mouse liver and other tissues.

Fig. 2. IL-12 and IFN-𝛾 crosstalk after immunotherapy was causative of inflammatory pathology.

(A) Hematoxylin and eosin (H&E) staining of fixed liver tissue from mice treated with aCD40, with or without IL-12 or IFN-𝛾 neutralization, two days after aCD40. Necrotic lesions (dashed yellow lines).

(B) Quantification of necrotic lesion area as a percent of total liver area in H&E section (n = 3–7 mice/group).

(C) Diagram depicting generation of bone marrow chimeras to study the requirement for IFNgR1 signaling on hematopoietic vs. radio-resistant cells.

(D) Changes in body weight from mice as depicted in 2C, two days following aCD40 (n = 4–6 mice/group).

(E) H&E staining of fixed liver tissue from mice sufficient for IFNgR1 only in hematopoietic cells (left) or only in radio-resistant cells (right). Necrotic lesions (dashed yellow lines).

(F) Flow cytometry quantification of IL12-EYFP⁺ cells in livers of mice treated or not with aCD40, with or without IFN-𝛾 neutralization (n = 3–4 mice/group).

(G) Flow cytometry data as in (F), but from IFN-𝛾-EYFP mice with or without IL-12 neutralization (n = 5–6 mice/group).

Data are represented as mean ± SEM. For comparisons between two groups, Student’s two-tailed t test was used. For comparisons between multiple groups, one-way ANOVA was used. ***p < 0.001, ****p < 0.0001.

Since IFN-𝛾 may act on immune cells or directly upon parenchymal cells (22), we used bone marrow chimeras to interrogate the importance of each (Fig. 2C). We found that non-hematopoietic (radiation-resistant) cells did not need to sense IFN-𝛾 for liver damage or weight loss to progress; however, IFN-𝛾 sensing by hematopoietic cells was necessary and sufficient for toxicity (Fig. 2D–E). Although a proportion of recipient-derived F4/80^hi CD11b^lo/- macrophages remained as radio-resistant cells following immune reconstitution (Fig. S4A), this represented a minority of macrophages, and in mice reconstituted with IFNgR1-deficient bone marrow, residual wild type cells were not sufficient to reproduce a “wild type” toxicity phenotype. Chimerism of myeloid cells recruited to the liver was near-complete (Fig. S4B). We therefore established that immune cell, rather than parenchymal cell, sensing of IFN-𝛾 was critical for liver toxicity.

Cytokine neutralization indicated that IL-12 and IFN-𝛾 operated in a positive-feedback manner in the liver following aCD40 treatment, similar to these cytokines’ interactions in the tumor (7) following immunotherapy (Fig. 2F–G). We investigated two additional tissue sites—the bone marrow, which harbored robust induction of both cytokines, and the lungs, where cytokine induction was less extreme—and observed the same interdependence between these cytokines (Fig. S5A–B). Thus, IL-12 and IFN-𝛾 cooperatively drove tissue-damaging immune responses in both the tumor and in sites of toxicity, with IFN-𝛾 sensing by immune cells representing a key event driving toxic inflammation.

DC-independent sources of IL-12 were sufficient to drive toxicity

Tumor control in many models of immunotherapy requires Batf3-dependent cross-presenting cDC1s and antigen-specific CD8⁺ T cells (23–25). It is thought that Batf3-dependent cDC1s can give rise to IL-12-producing DC3s (26, 27), also called mregDC (8), and LAMP3⁺ DC (28); these account for most tumor-infiltrating IL-12-producing cells and are associated with antitumor immune activation (7, 26, 27). We therefore asked whether Batf3 deficiency impacted IL-12 production and toxicity in the liver after aCD40.

Batf3-deficiency abrogated tumor control following aCD40 (Fig. S6A) and limited the proportion of IL-12⁺ immune cells in the tumor compared to Batf3-sufficient mice (Fig. 3A, Fig. S6C). Conversely, we observed no deficiency in IL-12-producing cells in the livers of knockout mice compared to WT controls (Fig. 3A–C). In Batf3-sufficient mice, IL-12p40-EYFP⁺ cells were detected proximally to areas of structural aberration in the inflamed liver, along with an accumulation of labeled aCD40 two days after treatment (Fig. 3B, Fig. S7A–B). In Batf3^–/– Il12p40-Eyfp mice, we found no major difference in prevalence or distribution of EYFP⁺ cells (Fig. 3B–C). Histological examination showed that Batf3-deficiency did not prevent liver necrosis following aCD40, and IL-12 neutralization showed that necrosis was caused by Batf3-independent sources of IL-12 (Fig. 3D). Batf3 deficiency likewise did not eliminate weight loss after aCD40, although it did somewhat diminish this effect (Fig. S6B). We further interrogated the role of classical DCs (cDCs) in liver toxicity using Zbtb46-Dtr bone marrow chimeras, enabling specific depletion of cDCs (29). These mice and their cDC-sufficient counterparts similarly developed liver toxicity, indicating that Zbtb46-dependent cells were not required for driving the irAE (Fig. 3E).

Fig. 3. DC-independent sources of IL-12 were sufficient to drive toxicity.

(A) Intracellular IL-12p40 staining from tumors (left) or livers (right) of Batf3^+/+ or Batf3^–/– mice two days after aCD40 (n = 5–8 mice/group).

(B) Whole mount imaging of livers from Batf3^+/+ Il12-EYFP or Batf3^–/– Il12-EYFP mice, given or not aCD40, two days after treatment. Lectin-rhodamine (blue); IL12-EYFP (green). Tissue lesions (dashed yellow lines).

(C) Quantification of IL12-EYFP⁺ cells from livers of mice as in (B).

(D) H&E staining of livers from Batf3^+/+ or Batf3^–/– mice treated with aCD40, with or without IL-12 neutralization, two days after aCD40, quantified as in Fig. 1B (n = 3–10 mice/group). Necrotic lesions (dashed yellow lines).

(E) H&E staining of liver tissue taken from WT mice or Zbtb46-Dtr bone marrow chimeras treated with aCD40. Necrotic lesions (dashed yellow lines).

Data are represented as mean ± SEM. For comparisons between two groups, Student’s two-tailed t test was used. For comparisons between multiple groups, one-way ANOVA was used. **p < 0.01, ***p < 0.001, ****p < 0.0001.

Rag2-deficiency and antibody-mediated lymphocyte targeting indicated that toxicities were CD8- and Rag2-independent, in contrast to tumor control, which required both (Fig. S8A–D). Toxicity was independent of B, CD4⁺, and NK cells (Fig. S8A, C). The latter two, alongside Rag2 and CD8⁺ cell data, suggested that lymphoid sources of IFN-𝛾 in the liver were likely diverse and redundant following aCD40. Tumor control was preserved with B cell deficiency as well as with CD4⁺ and NK cell targeting (Fig. S8B, D).

Resident Kupffer cells were a source of IL-12

We next interrogated the identities of IL-12-producing cells in an unbiased manner using scRNAseq. By comparing IL-12p40-EYFP⁺ cells in the liver and tumor following aCD40, we probed whether toxicity-associated IL-12 producers were distinct from antitumoral IL-12⁺ cells in the tumor microenvironment (Fig. 4A). Visualizing IL-12⁺ cells from the liver and tumor together showed that these sites contained similar and distinct IL-12⁺ states (Fig. 4A–B, Fig. S10A–D). Both tissues included cells expressing transcripts associated with the DC lineage (Batf3, Zbtb46, Flt3), and more precisely the DC3 state as defined by Zilionis et al., 2019 (Fscn1, Ccr7) (Fig. 4B–D, Fig. S9A–D, Table S1–2). Nearly all IL-12⁺ cells in tumors resembled DC3s, in line with our previous findings (7). However, those in the liver showed additional heterogeneity, harboring transcripts associated with macrophage identity (Cd68, C1qa, Apoe) (Fig. 4B–D, Fig. S9A–D), as well as transcripts canonically associated with Kupffer cell (KC) identity (30) (Nr1h3, Spic, Hmox1) (Fig. 4B–D, Fig. S9A–B).

Fig. 4. Resident Kupffer cells were a source of IL-12.

(A) scRNAseq pipeline (left) and UMAP representation (right) comparing tumor and liver IL-12-EYFP⁺ cells from aCD40-treated reporter mice (light blue, tumor, n = 3 mice; dark blue, liver, n = 2 mice).

(B) UMAP of EYFP⁺ cells in tumor (n = 2295 cells) and liver (n = 5157 cells) colored by cell state annotation.

(C) Il12b expression and dendritic cell, macrophage, and Kupffer cell markers in EYFP⁺ cells from (A). Colorbar saturated at the 99.5^th expression percentile measured across all EYFP⁺ cells in tumor or liver.

(D) Quantification of transcripts depicted in (C) across Il12b⁺ cell states. Mean of each biological replicate is shown (dots) with standard error of each replicate-specific mean; count per 10,000, CP10K.

(E) Flow cytometry of liver IL12-EYFP⁺ cells from aCD40-treated reporter (representative example of n = 4 mice).

(F) Contribution of IL12-EYFP⁺ cells with different myeloid cell phenotypes as defined in (E) (average from n = 4 mice).

(G) Schematic for parabiosis study to analyze chimerism of liver EYFP⁺ cells after aCD40.

(H) Flow cytometry data gating IL-12-EYFP⁺ CD11b^–/lo F4/80⁺ KCs in IL-12 reporter (top) and non-reporter (bottom) livers.

(I) Proportions of EYFP⁺ cells with a KC (CD11b^–/lo F4/80⁺) or migratory macrophage (CD11b⁺ F4/80⁺) phenotype in each parabiont (n = 3 mice/group).

Data are represented as mean ± SEM. For comparisons between two groups, Student’s two-tailed t test was used. *p < 0.05.

In sorting IL-12p40-EYFP⁺ cells for scRNAseq, we used a restrictive gating strategy to optimize capture of viable cells and minimize debris (Fig. S10A). However, in our flow cytometry studies, which used additional phenotypic markers and a different gating scheme, we observed an abundance of liver IL-12⁺ cells with a CD11b^–/low F4/80⁺ phenotype (Fig. S10B). These cells, thereafter referred to as IL-12⁺ KCs, were likely not efficiently recovered during sorting for scRNAseq, effectively enriching for IL-12⁺ DC-like cells rather than IL-12⁺ KCs. In our subsequent flow cytometry analyses, wherein we used for example a broader initial FSC/SSC gate and incorporated these cells, we observed that KCs constituted the major EYFP⁺ subset in the liver (Fig. 4E–F, Fig. S11A). We confirmed these findings by intracellular staining with an anti-IL-12p40 mAb (Fig. S11B–C).

We next assessed whether the cells appearing as IL-12⁺ KCs by flow cytometry were tissue resident, which would support their identity as true KCs, rather than infiltrating cells adopting a KC-like phenotype upon arrival in the liver. We tested for IL-12⁺ KC residency using parabiosis. In the setting where Il12p40-Eyfp mice and WT mice are surgically joined, all EYFP⁺ cells recovered from the WT parabiont must be derived from shared circulation with the EYFP⁺ reporter mouse. Our analysis of the WT parabiont revealed a marked paucity of chimeric IL-12p40-EYFP⁺ CD11b^–/low F4/80⁺ KC-like cells in the liver following aCD40, compared to F4/80⁺ CD11b^hi macrophages, which exhibited efficient chimerism (Fig. 4G–I) and are thought to be derived from circulating Ly6C^high monocytes (31). Consequently, the majority of IL-12⁺ CD11b^–/low F4/80⁺ cells did not derive from the circulation, supporting that they were tissue-resident and therefore bona fide KCs (Fig. 4H–I). Together, these findings supported that resident Kupffer cells were a major source of IL-12 in the liver.

IFN-𝛾-responsive, IL-12-producing Kupffer cells, but not DCs, dictated toxicity

We next addressed which IL-12⁺ populations contributed to liver toxicity. Considering the importance of cDCs in IL-12-dependent immune responses (32), we designed a condition wherein Zbtb46-dependent cDCs were specifically unable to produce IL-12, while other IL-12⁺ cells, including KCs, were preserved (Fig. 5A–B). With this approach we validated that IL-12⁺ KCs were ontogenically distinct from cDCs, and also identified that mice developed similar liver toxicity in the presence or absence of IL-12⁺ DCs (Fig. 5C). These data indicated that IL-12 production from cDCs was not a requirement for Th1-driven liver toxicity.

Fig. 5. IFN-𝛾-sensing Kupffer Cells drove liver toxicity.

(A) Schematic for bone marrow chimeras sufficient or deficient for IL-12-producing cDCs.

(B) Quantification of IL-12-producing DCs (left) or KCs (right) from livers of mice shown in (A) two days after aCD40.

(C) H&E of livers from mice as depicted in (A). Necrotic lesions (dashed yellow lines).

(D) Flow cytometry quantification of IL-12-producing DCs (F4/80^– CD11c⁺ MHCII⁺) or KCs (CD11b^–/lo F4/80⁺), two days after aCD40, given Control or Clodronate Liposomes (n = 4–5 mice per group).

(E) H&E of livers from aCD40-treated mice given Control or Clodronate Liposomes. Necrotic lesions (dashed yellow lines).

(F) Quantification of lesions as shown in (E) (n = 5 mice/group).

(G) H&E with quantification from livers of control (left) or Clec4f-Dtr (right) mice 2 days following aCD40 (n = 3–4 mice/group). Necrotic lesions (dashed yellow lines).

(H) Flow cytometry quantification of liver IFN-𝛾-eEYFP⁺ cells from mice given Control or Clodronate Liposomes, two days following aCD40 (n = 4 mice/group).

(I) Flow cytometry quantification of liver IL-12-producing CD11b^–/lo F4/80⁺ from mice treated or not with aCD40, with or without IL-12 or IFN-𝛾 neutralization (n = 3–5 mice/group).

(J) Diagram of mice containing both WT and Ifngr1^–/– hematopoietic cells (left). Flow cytometry data comparing IL-12 production in Ifngr1^+/+ vs. Ifngr1^–/– CD11b^–/lo F4/80⁺ KCs from livers of bone marrow chimeras two days after aCD40 (right).

(K) Flow cytometry data comparing IL-12 production in CD11b^–/lo F4/80⁺ KCs from livers of Clec4f-cre^o/o Ifngr1^fl/fl vs. Clec4f-cre^+/o Ifngr1^fl/fl mice two days after aCD40.

(L) H&E of livers from Clec4f-cre^o/o Ifngr1^fl/fl (left) or Clec4f-cre^+/o Ifngr1^fl/fl (right) mice 2 days following aCD40. Necrotic lesions (dashed yellow lines).

Data are represented as mean ± SEM. For comparisons between two groups, Student’s two-tailed t test was used. For comparisons between multiple groups, one-way ANOVA was used. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

To test the importance of IL-12 produced by macrophages, we sought to manipulate these cells while sparing DCs. First, we used clodronate liposomes (Clo. Lip.) to deplete the macrophages selectively. This treatment profoundly decreased the number of KCs (Fig. S12A), and by extension IL-12⁺ KCs (Fig. 5D), while preserving IL-12⁺ DCs (Fig. 5D). Macrophage depletion eliminated liver necrosis following aCD40 (Fig. 5E). While we considered that suppression of toxicity with Clo. Lip. could be linked to a deficiency in Fc receptors, which enable the action of aCD40 in vivo (33), we did not see that aCD40 was rendered inert in the setting of macrophage depletion; IL-12 was still induced in other subsets of myeloid cells (i.e. DCs), indicating that the pharmacological requirements for cellular activation and IL-12 induction were not missing.

Since Clo. Lip. treatment could deplete both resident and circulation-derived macrophages, we tested a role for bona fide KCs using Clec4f-driven expression of the diphtheria toxin receptor and DT administration (30). Clec4f-cre^+/o lsl-Dtr^+/o mice (also referred to as Clec4f-Dtr), when compared to Clec4f-cre^o/o lsl-Dtr^+/o mice, showed depletion of KCs following DT injection (Fig. S12B–D), as well as suppression of liver necrosis after aCD40 (Fig. 5G). These data supported a causal role for liver resident macrophages in aCD40-induced tissue damage.

To interrogate the position of liver macrophages in the IL-12/IFN-𝛾 feedback interaction, we used IFN-𝛾 reporter mice and found that Clo. Lip. treatment dramatically diminished IFN-𝛾 induction in the liver after aCD40 treatment, supporting that macrophages stimulated IFN-𝛾 production (Fig. 5H). Second, we asked whether IFN-𝛾 sensing by KCs was critical for their activation. IFN-𝛾 neutralization markedly reduced the proportion of IL-12-producing KCs (Fig. 5I); IL-12 neutralization induced similar results, presumably due to IL-12’s effects on IFN-𝛾-producing cells (Fig. 5I). Third, we asked whether KCs lacking a functional IFN-𝛾 receptor would fail to become activated and thus produce IL-12. Using 50:50 WT:IFNgR1^–/– bone marrow chimeras to directly compare KCs sufficient or deficient for IFNgR1, we found that IFNgR1-deficient KCs had diminished expression of IL-12 (Fig. 5J) and other markers of classical activation (MHCII, CD80, CD86) (Fig. S13). Finally, we specifically interrogated the importance of IFN-𝛾 sensing in KCs for toxicity using Clec4f-cre^+/o IFNgR1^fl/fl mice. In this setting, where KCs specifically lacked IFNgR1, IL-12 production by these cells was markedly diminished (Fig. 5K) and necrosis was nearly abrogated (Fig. 5L), supporting that liver resident KCs were activated by elevated levels of IFN-𝛾 and that this was a critical event in driving forward liver damage following immunotherapy. Together, these findings support that tissue-resident liver macrophages can play a major role in propagating undesired immune-mediated toxicity, and that they are active local participants in the IL-12-IFN-𝛾 positive feedback reaction.

IFN-𝛾, IL-12 and macrophages, but not DCs, induced a toxic neutrophil response

We next used scRNAseq to unbiasedly assess inflammation in the aCD40-treated liver. We looked beyond IL-12-producing cells at sources of IFN-𝛾 and at broader patterns of inflammation, which we thought might indicate avenues for limiting Th1-driven toxicity without affecting antitumor immunity. Therefore, we sorted liver CD45⁺ cells from untreated and aCD40-treated mice, and readily identified major immune cell subsets (Fig. 6A, Fig. S14A). Dramatic changes in immune cluster prevalence and structure accompanied treatment (Fig. S14B). As suspected, multiple lymphocyte populations indicated expression of IFN-𝛾 (Fig. S15A–D), including CD4⁺ T cells, CD8⁺ T cells, NK cells, and both T and NK cells with transcripts indicating cell cycling (Mki67, Top2a) (Fig. S15A–D); all five clusters contributed to increased Ifng expression following aCD40 (Fig. S15C–E). Heterogeneity in IFN-𝛾⁺ cells was confirmed by flow cytometry (Fig. S16).

Fig. 6. IFN-𝛾, IL-12 and macrophages, but not DCs, induced a pathogenic neutrophil response.

(A) UMAP of CD45⁺ cells from livers of untreated (n = 5,879 cells) or aCD40-treated (n = 12,892 cells) mice, colored by major cell-type (n = 2 mice per condition).

(B) Fold change in the relative abundance of major cell types in sequenced CD45⁺ cells from livers of aCD40-treated vs. untreated mice.

(C) Flow cytometry data of neutrophils (CD11b⁺ Ly-6G⁺) in livers of untreated or aCD40-treated mice.

(D) Quantification of flow cytometry data as shown in (C) (n = 5 mice/group).

(E) Flow cytometry quantification of liver neutrophils from mice treated or not with aCD40, with or without IL-12 or IFN-𝛾 neutralization (n = 3–5 mice/group).

(F) Flow cytometry quantification of liver neutrophils from WT or Zbtb46-Dtr mice following aCD40 (n = 4–6 mice/group).

(G) Flow cytometry quantification of liver neutrophils from mice treated with aCD40 and control or clodronate liposomes (n = 5 mice/group).

(H) MPO staining of livers from untreated and aCD40-treated mice. ROI, region of interest (n = 5 mice/group). Necrotic lesions (dashed red lines).

(I) H&E staining of livers from aCD40-treated mice, given (or not) anti-Gr-1 or anti-Ly-6G mAbs with or without a CXCR2 inhibitor. Necrotic lesions (dashed yellow lines).

(J) Quantification of liver lesions from aCD40-treated mice, given (or not) anti-Gr-1 mAbs (n = 6–7 mice/group).

(K) Quantification of liver lesions from aCD40-treated mice, given (or not) anti-Ly-6G mAbs with or without a CXCR2 inhibitor (n = 3–5 mice/group).

(L) Quantification of liver lesions from aCD40-treated mice sufficient or deficient for CSF3R (n = 4–5 mice/group).

(M) MC38 tumor volumes for mice treated (or not) with aCD40, with or without neutrophil targeting. (n = 6–8 mice/group)

(N) Flow cytometry quantification of IL-12-EYFP⁺ cells in livers of mice treated (or not) with aCD40, given or not anti-Ly6G mAbs (n = 4–5 mice/group).

(O) Flow cytometry quantification of IFN-𝛾-EYFP⁺ cells in livers of mice treated (or not) with aCD40, given or not anti-Ly6G mAbs (n = 4–5 mice/group).

Data are represented as mean ± SEM. For comparisons between two groups, Student’s two-tailed t test was used. For comparisons between multiple groups, one-way ANOVA was used. *p < 0.05, **p < 0.01, ****p < 0.0001.

Neutrophils had the greatest fold increase as a percentage of sequenced cells when comparing aCD40-treated to untreated conditions (Fig. 6B). We confirmed this increase by flow cytometry (Fig. 6C–D). The increase in liver-associated neutrophils depended on both IL-12 and IFN-𝛾 (Fig. 6E), was preserved in the setting of cDC depletion (Fig. 6F), but was diminished with macrophage depletion (Fig. 6G). Histological analysis further showed spatial association of MPO⁺ cells (a proxy for neutrophil identity) with lesioned areas in the liver (Fig. 6H), indicating that these cells could be directly involved in liver pathology.

There is ample evidence that neutrophils can exert protumoral functions in the TME (34), yet they can also cause tissue damage (35). Therefore, we wondered whether these cells could be promoting toxicity; if this were true, neutrophil-based interventions could conceivably reduce cells with protumoral and toxic functions. To test whether neutrophils were favoring liver toxicity, we took multiple approaches to inhibit neutrophils in vivo and assessed the impact on liver damage following aCD40. First, we used an anti-Gr1 mAb (36), which can also deplete Ly6C⁺ monocytes. CD11b⁺ CXCR2⁺ neutrophils were significantly reduced by this method, while we had less of an effect on CD11b⁺ CD115⁺ monocytes in circulation (Fig. S17A). Anti-Gr1 reduced necrotic lesions in the liver (Fig. 6I–J). Second, we used an anti-Ly6G mAb, which may not efficiently deplete neutrophils (36), but can limit their trafficking to tissue sites (37) including the liver (38). Anti-Ly6G, with or without a CXCR2 inhibitor, similarly reduced neutrophil-like cells in the liver (Fig. S17B) and suppressed necrotic lesioning (Fig. 6I, K). Third, we took a genetic approach, using Csf3r^–/– mice, which have a deficiency in neutrophils (39). Csf3r deficiency significantly reduced liver-associated neutrophils following aCD40 but did not affect liver DCs, KCs, or other macrophages (Fig. S18). Absence of Csf3r reduced necrotic lesioning in the liver (Fig. 6L). These various approaches supported the notion that neutrophils exacerbated liver necrosis following aCD40.

None of the pharmacological approaches to neutrophil inhibition impeded tumor control following aCD40 (Fig. 6M). Furthermore, IL-12 and IFN-𝛾 expression in the liver remained unchanged with anti-Ly6G (Fig. 6N–O), showing that targeting neutrophils could limit liver damage even in the setting of robust IL-12 and IFN-𝛾 production. This raised the possibility that toxicities associated with activation of IL-12 and IFN-𝛾 signaling could depend on the reactivity of neutrophils to this response, rather than the cytokines themselves. Also, targeting of neutrophils revealed a potentially powerful avenue for separating antitumor effects from toxicity.

Tnf-expressing, IFN-𝛾-responsive neutrophils determined liver toxicity

Neither genetic nor pharmacological approaches achieved complete inhibition of neutrophils. This could explain the incomplete rescue of liver inflammation with these interventions (Fig. 6I). We therefore thought to target effector functions of neutrophils rather than the cells themselves in an attempt to stem their pathogenicity.

When we compared differentially expressed genes between aCD40-treated vs. untreated neutrophils (Fig. S19A, Table S3) and assessed enriched gene ontology terms, we found positive regulation of TNF production to be significantly enriched in neutrophils from the treated condition (Fig. 7A). Tnf was among the top enriched transcripts in aCD40-treated compared to untreated neutrophils (Fig. 7B, Fig. S19A). We found that neutrophils were the cell type producing this transcript at the highest level on a per cell basis (Fig. S19B). When factoring in the abundance of diverse immune populations in the inflamed liver, we found that neutrophils contributed ~92% of all Tnf detected (Fig. 7C). Consequently, neutrophils appeared to be the primary source of Tnf in the liver following aCD40.

Fig. 7. TNF-a-expressing, IFN-𝛾-responsive neutrophils determined toxicity but not tumor control.

(A) Gene Ontology (GO) results based on scRNAseq transcripts significantly enriched in liver neutrophils from aCD40-treated compared to untreated mice.

(B) Single-cell expression of Tnf in liver neutrophils from mice treated or not with aCD40. Colorbar saturated at 99.5^th expression percentile measured across all CD45⁺ immune cells.

(C) Relative contributions by different cell types to Tnf transcription based on transcript counts and relative representation of each cell type.

(D) H&E staining, with quantification, of livers from aCD40-treated mice, with or without TNF-a neutralization (n = 5 mice/group). Necrotic lesions (dashed yellow lines).

(E) Diagram of mice containing both WT and Ifngr1^–/– neutrophils. Both populations were sorted from livers two days after aCD40 and taken for RNAseq (left). Quantification of Tnf transcripts (gene of interest) in these cells (right).

(F) Single-cell expression of Cd274 in liver neutrophils from mice treated (or not) with aCD40. Colorbar as in (B).

(G) Quantification of Cd274 transcripts (gene of interest) from WT and Ifngr1^–/– neutrophils from livers of mice in (E).

(H) H&E staining, with quantification, of livers from aCD40-treated mice, given (or not) anti-PD-L1 followed by anti-Rat IgG2b depleting mAbs (n = 5 mice/group). Necrotic lesions (dashed yellow lines).

(I) Flow cytometry quantification of liver neutrophils from mice treated (or not) with aCD40, given (or not) anti-PD-L1 followed by anti-Rat IgG2b depleting mAbs (n = 6 mice/group).

(J) Flow cytometry quantification of liver IL-12⁺ cells from mice treated as in (I) (n = 6 mice/group).

(K) Flow cytometry quantification of tumor IL-12⁺ cells from mice treated as in (I) (n = 6 mice/group).

(L) MC38 tumor volumes for mice treated as in (I) (n = 7 mice/group).

Data are represented as mean ± SEM. For comparisons between two groups, Student’s two-tailed t test was used. For comparisons between multiple groups, one-way ANOVA was used. *p < 0.05, ***p < 0.001, ****p < 0.0001.

We therefore hypothesized that neutralizing TNF-α might eliminate a key pathogenic effector function of neutrophils. Indeed, anti-TNF-α prevented body weight loss triggered by aCD40 (Fig S20A), in accordance with previous observations (14), and abrogated liver necrosis (Fig. 7D). Liver neutrophils in aCD40-treated mice showed high expression of both Tnfrsf1a and Tnfrsf1b (Fig. S21), suggesting a possible feed-forward loop of TNF-α onto neutrophils themselves. In favor of this hypothesis, we found a marked reduction in neutrophils in the context of anti-TNF-α treatment (Fig. S20B). Limiting TNF-α availability also led to a decrease in IFN-𝛾⁺ and IL-12⁺ cells (Fig. S20B). The identification of neutrophils, rather than macrophages or lymphocytes, as the main Tnf expressers shed light on the cellular source of this toxic mediator and highlighted a mechanism of neutrophil pathogenicity in the context of irAEs. Further, the dramatic effects of TNF neutralization in our model illustrated the pleiotropic effects this cytokine can have on driving forward toxic inflammation.

We further asked whether pathogenic neutrophils responded to key elements of Th1 inflammation. In our gene ontology analysis of aCD40-treated vs. untreated liver neutrophils, we identified multiple pathways indicating responsiveness to interferons, notably IFN-𝛾 (Fig. 7A). When we directly compared neutrophils sufficient or deficient for the IFN-𝛾 receptor that co-existed in the same hosts (Fig. 7E), we observed similar numbers of WT and Ifngr1^–/– neutrophils in livers on day 2 after aCD40 (Fig. S22A), indicating that IFNgR1 did not control the neutrophil response quantitatively; however, Ifngr1^–/– neutrophils showed significantly less Tnf expression compared to their WT counterparts (Fig. 7E, Table S4). These data indicated that IFN-𝛾 sensing by neutrophils contributed to their Tnf production.

In addition to Tnf, we found that transcripts associated with oxidative burst (Cybb, Nos2, Acod1, Sod2), pro-inflammatory cytokines and chemokines (Cxcl9, Cxcl10, Il27), and IFN responsive factors (Ifi47, Tap1), were also expressed at lower levels in Ifngr1^–/– neutrophils compared to WT neutrophils (Fig. S22B, Table S4), overall indicating that IFN-𝛾 sensing favored a particular inflammatory transcriptional program in liver neutrophils.

Neutrophils in aCD40-treated livers also showed high expression of the canonical IFN-𝛾-responsive transcript Cd274 (encoding PD-L1; Fig. 7F), which depended on IFN-𝛾 sensing (Fig. 7G). Furthermore, both scRNAseq (Fig. 7F) and flow cytometry (Fig. S23A, B) readouts indicated that only a subset of neutrophils strongly up-regulated Cd274/PD-L1 following aCD40 treatment, suggesting that PD-L1 may mark IFN-𝛾-responsive, pro-inflammatory neutrophils. We thus sought to experimentally remove these cells by using an anti-PD-L1-mediated cell depletion approach. To this end, aCD40-treated mice received anti-PD-L1, followed by a depleting isotype of anti-Rat IgG2b. In this setting, liver necrosis was reduced (Fig. 7H), consistent with a reduction in liver-associated neutrophils (Fig. 7I). The treatment also reduced some CD11b⁺ macrophages (Fig. S23C), which should be considered when interpreting our results; however, and importantly, IL-12-producing populations remained unchanged in the liver (Fig. 7I and Fig. S23D), supporting that we did not substantially impact upstream myeloid cell players mediating Th1 toxicity. Furthermore, tumor-associated IL-12-producing cells were retained (Fig. 7K), and the antitumor response was not significantly impacted (Fig. 7L). Together, these data situate IFN-𝛾-sensing neutrophils as important players in Th1-dependent liver inflammation following aCD40 immunotherapy.

Immune checkpoint blockade (ICB)-induced Th1 responses in mice and humans

We next aimed to determine whether other Th1-promoting immunotherapies could activate the same inflammatory pathways in tumor-free tissue. First, we assessed the livers of IL-12p40-EYFP reporter mice treated with ICB (anti-PD-1 + anti-CTLA-4). We indeed found a trend towards elevated IL-12 expression in this tumor-free site (Fig. S24A), although it was of a lower magnitude than observed with aCD40, as expected, since ICB is typically well tolerated in mice. We also identified increased KC activation following ICB, indicated by higher MHCII expression (Fig. S24B). Furthermore, the same ICB treatment led to an increase in liver-associated neutrophils (Fig. S24C).

We next sought to investigate the clinical relevance of our findings. Because tissue samples from aCD40-treated patients are rare, we focused instead on patients who received ICB treatment and developed irAEs. We specifically investigated the hypothesis that, in humans, Th1 activation caused by immunotherapy agents like ICBs might achieve a sufficient magnitude in tumor-free sites to cause toxicities similar to those in aCD40-treated mice. To this end, we initially made use of published scRNAseq data from melanoma patients experiencing ICB-induced colitis (15) and compared this dataset with our scRNAseq data from mouse irAEs (Table S5). We found that expression of IFN-𝛾- and IL-12-related transcripts correlated in both direction and degree of change between mouse and human irAEs (Fig. 8A). These parallels suggested that aspects of the lymphoid and myeloid Th1 responses observed in the context of mouse irAE (triggered by aCD40) could occur in the clinical setting (triggered by ICBs). We additionally found that gene expression changes in liver T cells, monocytes/macrophages, and DCs from aCD40-treated mice correlated with changes seen in the corresponding cell populations from irAE colons (Fig. 8B, Fig S25A–B), indicating that our model of Th1 irAEs recapitulated features of clinical irAEs. Considering these parallels, we propose that IL-12 and IFN-𝛾 signaling may generally feature in, and even be determinants of, toxicities caused by Th1-promoting agents, independent of therapeutic modality.

Fig. 8. IFN-𝛾, IL-12 and neutrophil responses in human irAEs.

(A) Comparison of fold changes in single-cell gene expression for key cytokines and receptors in immune cells from human colon (immunotherapy-induced colitis vs patients not receiving immunotherapy) and mouse liver (aCD40-treated vs untreated). Red, yellow quadrants show conserved responses to therapy. See also Table S5.

(B) Comparison of gene expression changes in selected immune cell types from mouse livers and human colons from immunotherapy conditions as in (A). Pearson correlation (R) for genes changing in mice and human homologs, calculated based on direction of change. The 100 genes with greatest fold change in mice were used for the analysis. See also Table S5.

(C) Scatterplot comparing changes in gene expression for monocytes/macrophages in mouse livers and human colons from immunotherapy conditions as in (A). Up to 100 genes were selected based on: (1) FDR<0.05 and magnitude of change >2-fold in mouse; (2) existence of a 1:1 human homolog. Red, yellow quadrants show conserved responses to therapy; genes with conserved responses listed. See also Table S5.

(D) Cancer diagnosis, treatment, toxicity score, and granulocyte inflammation scores from livers of 24 cancer patients who developed irAEs. Additional information is available in Table S6. n.a. = not assessed.

(E) MPO staining of liver tissue from four patients diagnosed with cancer, treated with ICBs, and who developed hepatitis. Dashed red lines indicate lobular hepatitis. Additional information available in Table S6.

(F) Example of CD15 staining in liver tissue from patient as in (E). Additional information is available in Table S6.

(G) Quantification of neutrophil score (left) and CD15⁺/MPO⁺ granulocyte score (right) in irAE livers from cancer patients, treated with ICB who developed mild, moderate, or severe liver toxicities as assessed by histological analysis of liver biopsies.

We further tested whether ICB-induced adverse events were associated with a neutrophil response in cancer patients. For this, we could not use available ICB-induced colitis scRNAseq data, since neutrophils were removed prior to sequencing (15). However, we collected liver tissue from 24 cancer patients who developed hepatitis in the context of ICB treatment (Fig. 8D, Table S6). Histopathological analysis revealed neutrophils associated with areas of liver inflammation, as shown by H&E staining (Fig. 8D, Fig. S26). Granulocytes were also visualized by MPO staining, which showed MPO⁺ cells in inflamed areas of the livers (Fig. 8E). We confirmed granulocyte presence in irAE livers using an additional marker, CD15 (Fig. 8F, Fig. S26). When we compared the histological score, graded as mild, moderate or severe (Fig. 8D), with the neutrophil and MPO/CD15 scores, graded from 0–2, for each case of hepatitis, we found consistent granulocytic infiltration in both moderate and severe cases of irAE hepatitis (Fig. 8G). This suggested that granulocytic inflammation was associated with severity of inflammation._Overall, these human data indicate that, in the setting of ICB, both Th1 cytokine activation and neutrophil accumulation remarkably correspond with diagnosis of irAEs, implicating broad clinical importance of the mechanisms of Th1 toxicity that we revealed in aCD40-treated mice.

Discussion

This study dissected the cellular and molecular mechanisms of irAEs, which are a major clinical problem. In mice treated with the Th1-promoting immunotherapy aCD40, canonical antitumoral cytokines IL-12 and IFN-𝛾 were induced in tumor-free tissue sites, driving a pathological inflammatory response that depended on macrophages and neutrophils. Conversely, dendritic cells and CD8⁺ T cells drove tumor control but were not required for toxicity. Our findings in mice mirrored phenotypes from human irAEs caused by immune checkpoint blockade, showing similar immune reactions across species; the findings herein reported can thus serve as an important baseline when considering mechanisms of clinical relevance.

We suggest that IFN-𝛾-dependent corruption of tissue-associated macrophages may be broadly featured in irAEs triggered by Th1-promoting immunotherapies; if a therapy boosts IFN-𝛾, IFN-𝛾-vulnerable macrophages may be activated by the local cytokine milieu, lose their tolerogenic phenotype, and propagate undesired immune activation. Our findings build upon important discoveries implicating macrophages in aCD40 liver toxicity (40, 41), and models of Hepatitis B infection have likewise implicated damaging consequences of Kupffer cell activation and IL-12 production (42, 43), highlighting the importance of these cells for balancing liver homeostasis versus pathology.

We showed that the toxic effects of IL-12 and IFN-𝛾 could be tied to activation and effector functions of neutrophils, which were reactive to Th1 inflammatory signals in a manner that substantially influenced their phenotype, ultimately impacting pathology progression. Neutrophils have been proposed as a biomarker of irAEs by multiple previous studies (44, 45). Here, we found these cells in both our mouse model and clinical samples, further indicating their reproducible association with this clinical problem, and supporting translational relevance of the neutrophil-driven mechanisms of toxicity we report.

Since our manipulations of neutrophils suppressed tissue damage without affecting Th1 cytokines or tumor control, we propose that neutrophil targeting may limit irAEs without hampering antitumor immunity, especially since neutrophils are often considered to be protumoral (34). Multiple strategies for neutrophil targeting are being investigated in diverse clinical settings (46), and therapeutics such as CXCR2 inhibitors are being used in some cancer patients (47). Alternatively, chemotherapy preceding immunotherapy may benefit patients in some settings, considering that this cytotoxic treatment induces neutropenia which could limit irAEs. This idea is supported by findings from preclinical combination therapy models, published by Byrne et al. (40). However, because chemotherapy itself may cause hepatotoxicity (48) and could disrupt KCs, additional studies will be needed to define whether our findings apply in the combination setting. Finally, targeting TNF could also limit neutrophil-dependent toxicities, considering these cells were the main Tnf expressers in our system. Our findings highlight a biological rationale for TNF targeting in the clinic, and favor prioritizing anti-TNF for the management of irAEs including hepatitis.

Ultimately, there is a need to improve efficacy of cancer immunotherapy beyond current response rates, but avoiding toxicity is crucial. It is critical to understand which elements of therapy-induced immune reactions cause irAEs, so that they can be targeted without hampering antitumor Th1 responses. With the goal to untangle the pathologic processes that so often limit clinical utility of immunostimulatory agents, here we revealed potential pathways that may help to dissociate desired antitumor and undesired toxicity effects of immunotherapy.

Materials and Methods

(For full methods, see supplemental materials)

Study Design:

This study was designed to interrogate mechanisms of Th1 cytokine-driven pathologies that occur in non-malignant tissues following anti-cancer immunotherapies. Mice were treated with the agonist aCD40, and cellular and molecular mechanisms of toxicity were delineated primarily using the time point when greatest toxicity was observed in the mice (indicated by body weight loss and gross pathology of vital organs including the liver, 2–3 days after immunotherapy treatment). Liver toxicity was quantified, based on H&E stained tissue sections, by calculating the percentage of tissue appearing as confluent areas of necrosis. All toxicity studies were completed in tumor-bearing mice, and mice were randomized between treatment groups based on tumor volume on day 6–7 of tumor growth, before treatments began. End point analyses included flow cytometry, scRNAseq, RNAseq, imaging, and histological analysis. Raw data from all mouse experiments are available in Table S9.

Mice and tumor models:

Animals were bred and housed under specific pathogen free conditions at the Massachusetts General Hospital. Experiments were approved by the MGH Institutional Animal Care and Use Committee (IACUC) and were performed in accordance with MGH IACUC regulations. MC38 or TC-1 cells were implanted at 2 × 10⁶ cells in the flank. After one week, mice were treated with ~5 mg / kg of aCD40 Clone FGK4.5 (BioXCell Cat #BE0016–2) intraperitoneally and analyzed two days following treatment, unless otherwise noted. For ICB, anti-PD-1 (Clone 29F.1A12) was generously provided by Gordon J. Freeman. Mice received ~10 mg / kg anti-PD-1 and ~5 mg / kg anti-CTLA-4 (Clone 9D9, BioXCell Cat #BE0164) intraperitoneally on days 6, 7, and 8 of tumor growth, and tissues were analyzed on day 9.

Flow cytometry studies:

Briefly, all solid tissues were excised, minced, and digested with enzymes at 37 degrees before being processed through a cell strainer and resuspended for staining in PBS with 0.5% BSA.

Cytokine and lymphocyte targeting:

Cytokine neutralizing antibodies were given at 500 μg daily starting on day 7 of tumor growth and continued for 1–2 days. For the TC-1 tumor study, IL-12p40 was neutralized for 6 consecutive days. Anti-IFN-𝛾 was administered at 1 mg of antibody on day 7 then 500 μg for subsequent doses.

Anti-CD4 was dosed at 100 μg / injection (BioXCell Cat #BE0003), anti-CD8a at 200 μg / injection (BioXCell Cat #BE0004), and anti-NK1.1 at 200 μg / injection (BioXCell Cat #BE0036), and injected every other day from day 6–10 of tumor growth. All injections were intraperitoneal (i.p.).

Neutrophils targeting:

For neutrophil targeting, antibodies/inhibitors were used as follows. Anti-Gr1 (BioXCell Cat #BE0075) at 100 μg per dose on day 6 and 8 of tumor growth before sacrifice on day 10. Anti-Ly-6G (BioXCell Cat #BP0075) was administered at 500 μg / mouse on day 7 (2h before aCD40) then 250 μg / mouse on day 8 before sacrifice on day 9. For tumor growth studies, all mAbs were dosed one additional time on day 10. Cxcr2 inhibitor SB 225002 (Tocris Cat #2725/10) was injected at 200 μg / dose and scheduled as anti-Ly-6G. For IL-12 and IFN-𝛾 reporter mice, anti-Ly-6G (Absolute Antibody Cat #Ab00295–2.0) was dosed at 200 μg / mouse on days 6–8. For anti-PD-L1-based targeting of neutrophils, anti-PD-L1 (BioXCell Cat #BE0101) was injected at 12.5 mg / kg ~10 hours after aCD40 treatment and again 1 day after treatment, each time followed by 12.5 mg / kg anti-Rat IgG2b (BioXCell Cat #BE0252) ~1 hour later. All injections were i.p.

Histology:

Mouse livers were fixed in 10% formalin overnight, washed twice with PBS, and placed in 70% EtOH or PBS until processing. For colons, 3 days after aCD40, tissue was isolated, flushed with cold 5% FBS in PBS, opened longitudinally, rolled and tied loosely with a nylon suture (Ethicon). For H&E, tissues were paraffin embedded, sectioned, and stained with Hematoxylin and Eosin at the MGH Histopathology Research Core.

Whole mount liver imaging:

On day 7 of tumor growth, mice were injected i.p. with unlabeled or fluorescently-labeled aCD40 mAb (mAb: BioXCell Cat #BE0016–2; Antibody labeling kit: Thermo Fisher Cat #S30044). Two days later, mice were injected retro-orbitally with fluorescent lectin to label vasculature then sacrificed. IL-12p40-IRES-EYFP or IL-12p40-IRES-EYFP Batf3^–/– mouse livers were excised, placed in PBS between an inverted petri dish and a microscope cover slip, and imaged using an Olympus FluoView FV1000MPE confocal imaging system (Olympus America). Images were processed using Fiji from ImageJ.

Single cell RNAseq:

MC38 flank tumors in IL-12p40-IRES-EYFP mice grew for 7 days, then mice were treated or not with aCD40. Two days later, tumors and livers were processed to generate single cell suspensions. Cells were stained for CD45 (Table S8), labeled with 7AAD (Sigma-Aldrich), and CD45⁺ cells or IL-12-EYFP⁺ cells were sorted using a BD FACSAria sorter. InDrops single cell RNA sequencing was performed as described before (49) with changes to DNA primers and read lengths listed in Table S7.

Parabiosis:

CD45.1-STEM and B6.129-Il12^btm1LKy/J (IL-12-EYFP) mice were surgically joined as previously described (7). Circulatory equilibrium was confirmed 3–5 weeks post-surgery; both mice were then injected with MC38 tumors on the outer flank and 7 days later treated with aCD40; tissues were harvested 2 days later to analyze IL-12⁺ populations.

Bone marrow transfer experiments:

CD45.1 (Jackson Labs Cat #002014 or CD45.1-STEM) recipient mice were irradiated with a single dose of 1000 cGy using a cesium-137 irradiator. The next day, bone marrow from donor mice was processed for injection: For IFNgR1 deficiency experiments, either WT CD45.1 mice or CD45.2 Ifngr1^–/– mice served as donors. For Zbtb46-DTR experiments, either WT (CD45.2), Il12p40^–/–, or Zbtb46-Dtr mice served as donors. Donor mouse cells were counted manually. For 50:50 chimeras, cells were mixed at a 1:1 ratio before injection. Cells were injected retro-orbitally at 10–14 × 10⁶ total cells / mouse in 200–400 uL volume, and mice were allowed to reconstitute for 5.5–7.5 weeks. Chimerism was confirmed by cheek bleed before tumor challenge and immunotherapy treatment.

Diphtheria toxin injection:

Mice receiving diphtheria toxin (DT) (Sigma-Aldrich) were dosed at 10 ng / gram of body weight initially 0.5–1 day before aCD40 injection, then at 4 ng / gram of body weight one day after immunotherapy.

Clodronate Liposomes:

Mice receiving clodronate or control liposomes were dosed with 200 μl of liposomes (Liposoma Cat# P-010–010) retro-orbitally one day before immunotherapy and again one day following treatment.

Bulk RNAseq:

For RNAseq, 50:50 WT:Ifngr1^–/– bone marrow chimeras were injected with tumors and aCD40 as usual. On day 2 after treatment, livers were processed for FACS and CD45.1 (WT) or CD45.2 (Ifngr1^–/–) neutrophils were sorted directly into Trizol reagent and placed on ice. RNA was extracted using TRIzol™ Plus RNA Purification Kit (Thermo Fisher Cat #12183555). Libraries were prepared in collaboration with the Harvard Biopolymer Core Facility. Libraries were normalized in equimolar ratios for one final pool and sequenced using an Illumina NextSeq 500 instrument; samples were demultiplexed, and the resulting fastq files were analyzed using an RNA-seq pipeline implemented in the bcbio-nextgen project (https://bcbio-nextgen.readthedocs.org/en/latest/).

Gene Ontology:

Differentially expressed genes from aCD40-treated vs. untreated liver neutrophils were entered into the GO enrichment analysis tool for Mus musculus, and enriched biological processes were determined with PANTHER Overrepresentation Test (Released 20200728) using Fisher’s exact test; Annotation Version and Release Date: GO Ontology database DOI: 10.5281/zenodo.4081749 Released 2020–10-09 (50–52).

Human liver samples:

Patient selection and inclusion criteria:

Patients on immune checkpoint therapy who developed hepatic irAEs were identified through The Oncology Department of the Lausanne University Hospital and the Hospital of the University of Geneva. All included patients had given their written informed consent for re-use of their medical and histopathological data, with the exception of deceased persons.

Patient characterization:

Demographic, clinical, laboratory and histopathological data were retrieved from electronic medical records and archives.

Histopathology and immunochemistry:

Liver biopsy or autopsy samples were fixed in 10% formalin and paraffin embedded, and 3–5 um thick sections were stained as follows: for cases 1–20, H&E and CD15 (clone: BD Pharmingen MMA, dilution 1/1500); for patients 21–24, MPO (clone: DAKO A0398, Dilution 1:1000) and CD15 (Ventana). Cases were scored for severity, neutrophilic infiltration, and CD15/MPO positivity.

Statistical analysis of flow cytometry, histology or tumor burden data.

All statistical analyses were performed using GraphPad Prism software. Results were expressed as mean±SEM. Student’s two-tailed t-test were done to compare two groups. One-way ANOVA was used to compare multiple groups. p values > 0.05 were considered not significant (n.s.); p values < 0.05 were considered significant. * p value < 0.05, ** p value < 0.01, *** p value < 0.001, **** p value < 0.0001.