Research ArticleCancer

CD3 bispecific antibody–induced cytokine release is dispensable for cytotoxic T cell activity

See allHide authors and affiliations

Science Translational Medicine  04 Sep 2019:
Vol. 11, Issue 508, eaax8861
DOI: 10.1126/scitranslmed.aax8861

Taming the cytokine beast

Bispecific antibodies, which are engineered to engage a cancer cell antigen and activate T cells to kill the cancer cell, are showing clinical promise. Unfortunately, they can also cause major side effects as a result of uncontrolled immune activation and cytokine release. Li et al. found a way to separate the beneficial effects from the harmful ones by showing that activation of tumor necrosis factor–α signaling is necessary for the toxic systemic cytokine release but not for successful cancer treatment. The authors identified several ways to inhibit the dangerous signaling pathway and demonstrated them in mouse models, with no loss of anticancer efficacy.


T cell–retargeting therapies have transformed the therapeutic landscape of oncology. Regardless of the modality, T cell activating therapies are commonly accompanied by systemic cytokine release, which can progress to deadly cytokine release syndrome (CRS). Because of incomplete mechanistic understanding of the relationship between T cell activation and systemic cytokine release, optimal toxicity management that retains full therapeutic potential remains unclear. Here, we report the cell type–specific cellular mechanisms that link CD3 bispecific antibody–mediated killing to toxic cytokine release. The immunologic cascade is initiated by T cell triggering, whereas monocytes and macrophages are the primary source of systemic toxic cytokine release. We demonstrate that T cell–generated tumor necrosis factor–α (TNF-α) is the primary mechanism mediating monocyte activation and systemic cytokine release after CD3 bispecific treatment. Prevention of TNF-α release is sufficient to impair systemic release of monocyte cytokines without affecting antitumor efficacy. Systemic cytokine release is only observed upon initial exposure to CD3 bispecific antibody not subsequent doses, indicating a biological distinction between doses. Despite impaired cytokine release after second exposure, T cell cytotoxicity remained unaffected, demonstrating that cytolytic activity of T cells can be achieved in the absence of cytokine release. The mechanistic uncoupling of toxic cytokines and T cell cytolytic activity in the context of CD3 bispecifics provides a biological rationale to clinically explore preventative treatment approaches to mitigate toxicity.


T cell–retargeting therapies [chimeric antigen receptor (CAR) T and CD3 bispecific antibodies] have transformed the therapeutic landscape of oncology. However, uncontrolled systemic cytokine release is the primary challenge that hinders the broad use of practically all T cell–retargeting therapies regardless of the target or treatment modality (16). Mechanistic understanding of cytokine release in response to these reagents is currently poor and is an urgent medical need to facilitate the design and implementation of rational mitigation strategies. Undoubtedly, mechanistic insights into the determinants of cytokine release in response to T cell–retargeting therapies are critical for maximizing their therapeutic index and clinical success.

Whereas checkpoint blockade therapies rely on the concept of exhausted preexisting tumor immunity (7), CD3 bispecific antibodies can be used to override the endogenous T cell receptor (TCR)–mediated specificity of T cells. CD3 bispecific antibodies function as conditional agonists that trigger a polyclonal T cell response, which allows on-demand, synchronous T cell triggering and chemokine-mediated recruitment of additional T cells (8). Our previous studies (9) have concluded that despite the fact that these antibodies bypass the initial canonical cell-cell interaction mediated via TCR and peptide major histocompatibility complex (pMHC) contact, the subsequent molecular events in synapse formation closely resemble T cell triggering initiated by TCR-pMHC interaction.

Natural TCR-pMHC interaction-induced T cell activation has been suggested to have different signaling thresholds for triggering cytotoxicity and cytokine production by T cells (1012). In this study, we found that granzyme and perforin-mediated T cell killing induced by anti-HER2 (human epidermal growth factor receptor 2)/CD3 (13) is completely disconnected from cytokine release. Although our findings may mechanistically differ from previously presented results, they support the general concept of uncoupling killing and cytokine release and challenge the interpretation of cytokine production as a marker for direct “T cell activity.”

We further found the cell type–specific cellular mechanism and sequence of events that links CD3 bispecific antibody–mediated killing to toxic cytokine release. Specifically, we demonstrate that T cell–produced tumor necrosis factor–α (TNF-α) is the signal that initiates from the T cell triggering event and causes monocyte activation, which results in systemic production of toxic cytokines.

Last, we describe multiple nodes for potential therapeutic intervention in the context of CD3 bispecific antibody–induced cytokine release, including blockade of interleukin-6 (IL-6), IL-1β, and the upstream trigger for monocyte activation, TNF-α. We show that all of the above cytokines can be effectively and specifically blocked without affecting the therapeutic activity of T cells mediated by CD3 bispecific treatment. Our studies suggest that cytokine blockade should be applied as early as possible, ideally prophylactically, and not after clinical manifestation of severe cytokine release syndrome (CRS) symptoms. Our findings could be quickly translated to clinical use because approved therapeutics are available for all of the above cytokines.


Anti-HER2/CD3–induced cytokine release is not required for T cell cytotoxic activity

To investigate anti-HER2/CD3 T cell–dependent bispecific (TDB)–induced cytokine release in mice, we used immunocompetent transgenic MMTV.huHER2 mice (14). These mice have high expression of HER2 in the mammary epithelial cells, causing spontaneous development of mammary tumors. Anti-HER2/CD3 induces robust chemokine-mediated recruitment of T cells (8) into these tumors, resulting in tumor regression (13). TDB treatment also substantially increases systemic cytokines, such as IL-6, TNF-α, and IL-2, which are detectable 2 hours after treatment (Fig. 1A and fig. S1A). Systemic serum IL-6 and TNF-α correlated with respective intratumoral cytokines (fig. S1B). The serum cytokine profile of TDB-treated mice was similar to that reported in clinical CD19 CAR T studies (fig. S1C) (1, 3, 5). TDB treatment induced weight loss in mice but did not result in body temperature changes (fig. S1D), even when administered at supratherapeutic doses. Administration of a second TDB dose did not result in detectable systemic cytokine release when the interval between doses was 1 or 7 days (Fig. 1A and fig. S1A). Recovery of some TDB-induced cytokine release was observed when a 2-week interval was used, namely, increase in TNF-α (Fig. 1A). Recovery of other cytokines, such as IL-6 and IL-2, was substantially slower, and for IL-2, recovery was not observable even when using a 4-week interval (Fig. 1A and fig. S1A). The mechanism for the lack of second-dose cytokine release is not understood. Transient lymphopenia (margination) is a pronounced pharmacological effect of CD3-targeting antibodies (9, 15, 16). Initial TDB treatment resulted in reduction of peripheral T cell numbers (fig. S2A), a slight down-regulation of free CD3 on the T cell membrane (fig. S2B), and an increase in the fraction of CD8+ central memory and effector memory T cells (fig. S2C). TDB-induced chemokine receptor up-regulation in CD8+ cells (8) persisted for more than 21 days (fig. S2D). In summary, the analysis of T cell status in TDB-treated mice did not provide a satisfactory explanation for the lack of second-dose cytokine release.

Fig. 1 T cell–mediated cytotoxicity induced by anti-HER2/CD3 TDB is decoupled from cytokine release.

(A) MMTV.huHER2 transgenic (TG) mice were treated with anti-HER2/CD3 TDB (0.5 mg/kg) (4D5:2C11) on day 0. Second dose was administered 1 to 28 days later. Systemic cytokines were detected 2 hours after each dose using Luminex. (B) Human PBMC-based in vitro model developed to monitor TDB responses. HER2-amplified breast cancer cells (KPL-4) were used as targets and cytokines were detected 1 day after initial T cell activation (first dose) or on day 8 (second dose). (C) Effect of anti-HER2/CD3 TDB treatment on human CD8+ cell phenotype. (D) Human PBMC-mediated in vitro killing of target cells upon first or second dose. (E) Mice with MMTV.huHER2 Fo5 tumors were treated with a single dose of anti-HER2/CD3 TDB (0.5 mg/kg) for 7 days. Spleen- or tumor-derived T cells from TDB-treated mice were used as effector cells in an ex vivo apoptosis assay that monitors caspase-3/7 activity in HER2-amplified KPL-4 cells. Data are shown as means ± SEM (A) or means ± SD (B, D, and E). Data color (A, B, and D): red, first dose; blue, second dose; and black, nontreated (NT).

Next, we tested whether the lack of second-dose cytokine release can be recapitulated using human T cells (Fig. 1B). TDB treatment of healthy donor peripheral blood mononuclear cells (PBMCs) ex vivo induced a robust increase in cytokines, which was substantially lower when a second dose was administered 7 days later (Fig. 1B). Because PBMCs were supplemented with new HER2+ target cells before the second dose, depletion of target cells could not explain the reduced cytokine release. As expected, TDB treatment induced expression of activation marker programmed cell death-1 in human T cells and conversion of naïve T cells to the effector phenotype (Fig. 1C). The ability of T cells to secrete cytokines is commonly used as a marker for T cell activity and to reflect the potential to kill target cells; however, human peripheral T cells induced a robust dose-dependent killing of target cells on the second dose (Fig. 1D). Therefore, T cells with prior TDB exposure retained killing potency despite the lack of cytokine induction, demonstrating that TDB-induced cytokine production is disconnected from the cytotoxic potential of T cells. To confirm that tumor-infiltrating T cells retain the capacity to kill in the absence of detectable systemic cytokines, we harvested tumor-infiltrating T cells from orthotopically transplanted HER2 transgenic tumors previously treated with HER2-TDB for 7 days and performed an ex vivo apoptosis assay. T cells extracted from these tumors robustly induced apoptosis of target cells (Fig. 1E), confirming that tumor-infiltrating T cells retain functionality after initial TDB activation despite the lack of cytokine release upon second dose.

Monocytes and macrophages are the major source of anti-HER2/CD3 TDB–induced IL-6 and IL-1β

To better deconvolute the cellular source of cytokine secretion and signaling dynamics within immune cell subsets triggered by TDB-induced killing, we performed single-cell transcriptional analysis using the in vitro human PBMC cytokine release model 4 hours after the first or second TDB treatment (fig. S3A). On the basis of canonical transcriptional signatures and gene expression, we were able to detect all major immune cell subtypes, including B cells, CD4+, CD8+, dendritic cells (DCs), monocytes/macrophages, and natural killer cells (Fig. 2A and fig. S3B).

Fig. 2 Monocytes and macrophages are the primary source of TDB-induced toxic cytokine secretion.

(A) Single-cell RNA-seq of PBMC-based in vitro model: (Left) Visualization of single cells using t-distributed stochastic neighbor embedding (tSNE). Individual cells are colored on the basis of their immune cell type assignment. (Right) Heat map depicting the classification of cells (columns) into distinct immune cell types using established genes and gene sets (rows; data file S1). (B) Violin plot of single-cell RNA-seq expression for perforin (PRF1) and granzyme B (GZMB) in CD8 cells stratified by treatment. (C) Granzyme B protein in cell supernatant 24 hours after first or second dose of treatment, quantified by Luminex analysis (n = 3 human healthy donors of PBMCs). (D) Violin plot of single-cell RNA-seq expression for IL-6 and IL-1β across immune cell types and stratified by treatment. Data color (A, B, and D): B cells (red), CD4 (blue), CD8 (green), DCs (purple), macrophage/monocytes (orange), and natural killer (NK) cells (yellow). Data color (B, C, and D): Nontreated (gray), 4 hours after first TDB dose (red), 7 days after first TDB dose (pink), and 4 hours after second TDB dose (blue). (E) MMTV.huHER2 nontumor-bearing mice were treated with 1 mg of liposomal chlodronate 3 and 1 day before anti-HER2/CD3 TDB (0.5 mg/kg). Serum was harvested 2 hours after TDB dose for Luminex analysis. Vehicle (black), TDB (red), and TDB and clodronate (magenta). Significance of differences in single-cell RNA-seq expression was assessed using a likelihood-ratio test (B and D). *P < 1 × 10−10, **P < 1 × 10−20, ***P < 1 × 10−50, and ****P < 1 × 10−100. Bars represent means ± SEM (C and E). Paired t test was used for statistical analysis (C). Unpaired t test with Welch’s correction was used for statistical analysis in (E). NS, not significant.

Cell type representation was consistent across time points and treatments (fig. S4A), allowing us to interrogate cell type–specific gene expression. As expected, induction of early T cell activation markers, such as CD69 (fig. S4, A and B), was primarily seen after the first TDB dose, whereas late activation markers, such as IL2RA (CD25; fig. S4C), were strongly induced after the second dose. TDB treatment induced transcriptional expression of perforin and granzyme B in CD8+ T cells (Fig. 2B), resulting in high amounts of secreted granzyme B (Fig. 2C). In accordance with our in vitro cell killing assay results (Fig. 1D), these key cytotoxic molecules were higher after the second dose, consistent with their retained ability to kill upon second TDB exposure. IL-6 and IL-1 have been proposed as potential mediators of CRS toxicities (17, 18). TDB treatment induced expression of IL-6 and IL-1 signaling components primarily in monocytes (Fig. 2D and fig. S4D). In contrast, marginal expression was detected in T cells, suggesting that T cells are not a major source of IL-6 or IL-1. Induction of monocyte IL-6 and IL-1 was substantially higher on first dose, consistent with our inability to detect these cytokines in culture medium after the second dose treatment. Depletion of macrophages and monocytes from the mouse model using liposomal clodronate blocked systemic IL-6 induction, suggesting that these cells are necessary for TDB-induced IL-6 release in vivo (Fig. 2E). Macrophage and monocyte depletion had no effect on interferon-γ (IFN-γ), IL-2, or TNF-α (Fig. 2E; IL-1 was undetected). As a summary, our studies indicate that monocytes and macrophages serve as major sources of both IL-6 and IL-1 secretion after TDB treatment.

Prophylactic blockade of IL-6 does not impair TDB-mediated antitumor efficacy

IL-6 has been postulated as a central mediator of CRS toxicity, and tocilizumab (IL-6R antagonist) is effectively used to manage high-grade CRS in patients treated with CAR T cells and CD3 bispecific antibodies (3, 5, 19). However, it is not understood whether blocking IL-6 signaling affects efficacy of T cell–redirecting therapies and whether earlier IL-6–targeted intervention is feasible. We therefore tested the impact of prophylactic intervention against IL-6 signaling on TDB-induced cytokine release, pharmacology, and antitumor activity (fig. S5A). Anti–IL-6 pretreatment before anti-HER2/CD3 TDB dosing selectively blocked TDB-induced IL-6 from healthy donor human PBMCs in vitro (Fig. 3A) and in tumor-bearing huHER2 transgenic mice (Fig. 3B) but had no impact on the other cytokines tested, including IFN-γ, TNF-α, IL-2, and IL-1β (Fig. 3, A and B, and fig. S5, B and C). Blockade of IL-6R was tested in nontumor-bearing transgenic mice and resulted in the expected increase of systemic IL-6 after TDB treatment (fig. S5D). Similar to anti–IL-6 treatment, receptor antagonism had no effect on other systemic cytokines after TDB treatment (fig. S5D). An increase in serum IL-6 and no impact on other cytokines were detected when TDB responses in IL-6R knockout (KO) mice were compared to wild-type (WT) animals (fig. S5E).

Fig. 3 Prophylactic blockade of IL-6 does not impair anti-HER2/CD3 TDB–mediated antitumor efficacy.

(A) Impact of anti-human IL-6 (1 μg/ml) (clone 6708) and 1 μM dexamethasone cotreatment on TDB-induced in vitro cytokine release using human PBMCs as effector cells. (B) In vivo effect of anti–IL-6 treatment on anti-HER2/CD3 TDB–induced systemic cytokine release in nontumor-bearing MMTV.huHER2 transgenic mice. Anti–IL-6 (MP5-20F3, 200 μg per animal) was administered twice, 1 day and 1 hour before anti-HER2/CD3 TDB (0.5 mg/kg). Serum was harvested 2 hours after TDB dose for Luminex analysis. (C) Influence of anti–IL-6 and dexamethasone cotreatment on TDB-induced in vitro human T cell activation (24 hours) and target cell killing after 24 hours (first dose) or 8-day exposure (second dose). Schematic of experimental design is presented in Fig. 1B. (D to F) Effect of anti–IL-6 pretreatment on lymphocyte margination (D; at 24 hours after TDB), tumor-infiltrating lymphocytes (E; 6 days after TDB), and antitumor efficacy (F; maximum percent change from baseline over 21 days) in treated MMTV.huHER2 transgenic mice with palpable mammary tumors. Data color: Red, TDB; magenta, dexamethasone (A and C); cyan, anti–IL-6 (D and F); blue, TDB and anti–IL-6; and black, vehicle (B, D, E, and F). Unpaired t test with Welch’s correction was used for statistical analysis. Data are shown as means ± SEM (B, D, E, and F) or means ± SD (A and C).

Anti–IL-6 pretreatment had no detectable effect on TDB-induced CD8+ T cell activation or killing activity (Fig. 3C) in vitro. In contrast, dexamethasone, which is commonly used as a clinical mitigation strategy for CRS, affected in vitro killing activity (Fig. 3C). TDB-induced lymphocyte margination was not affected by either pharmacological (Fig. 3D) or genetic (fig. S5F) blocking of IL-6 signaling. Anti-HER2/CD3 TDB induces robust chemokine-mediated T cell recruitment to murine mammary tumors (8). Anti–IL-6 pretreatment did not impair the TDB-induced T cell recruitment to the tumors (Fig. 3E). Consistent with unaffected in vitro activity and in vivo pharmacology, IL-6 pretreatment did not affect the antitumor activity of anti-HER2/CD3 TDB in the treatment of HER2-positive spontaneous mammary tumors (Fig. 3F and fig. S5G).

Preclinical studies have indicated IL-1 as a potential mediator of CAR T–induced CRS and central nervous system events in preclinical mouse models (17, 18). Similar to IL-6, prophylactic cotreatment with a therapeutic recombinant IL-1 receptor antagonist had no impact on cytokine release (fig. S6, A and B), human T cell activation and cytotoxicity (fig. S6C), lymphocyte margination (fig. S6D), T cell recruitment and activation status in tumors (fig. S6E), or antitumor activity (fig. S6F). In summary, the data demonstrate that pharmacological or genetic obstruction of IL-6 or IL-1 signaling has no evident impact on TDB-induced pharmacology or antitumor activity and suggest that early intervention is feasible without compromising the CD3 bispecific’s antitumor activity.

TDB-induced TNF-α triggers production of monocyte IL-6 and IL-1

To elucidate the upstream mechanism triggered by TDB treatment in T cells that results in cytokine release from monocytes and macrophages, we performed RNA sequencing (RNA-seq) analysis of purified CD8+ T cells induced by anti-HER2/CD3-TDB (fig. S7A). The treatment caused immediate and broad changes in gene expression in CD8+ T cells (4800 genes affected; fig. S7, B to D). The most strongly affected pathways included TNF-α, IL-6/JAK/STAT-3, IFN-γ, MYC targets, IL-2/STAT-5, and inflammation (fig. S7, C to E).

Next, we analyzed the expression pattern of TNF-α and its receptors. Single-cell RNA-seq analysis confirmed robust TNF-α transcription in CD8+ T cells and, to a lesser extent, in monocytes and CD4+ cells (Fig. 4A). TNF-α expression was substantially reduced upon second-dose exposure (Fig. 4A). TNFR1 was selectively expressed in monocytes, and expression was increased after TDB treatment (fig. S8A). TNFR2 was constitutively expressed in several cell types including monocytes, CD4+, and CD8+ T cells (fig. S8B). TDB treatment induced a robust and durable elevation of TNFR2 in T cells (fig. S8B), whereas monocytes responded with initial up-regulation followed by substantial down-regulation at later time points. In summary, these results demonstrate that activation of T cells by TDB induces TNF-α production in T cells, whereas monocytes are positive for the receptors for TNF-α, thus suggesting that TDB-induced TNF-α may be a factor that links T cell activation and monocyte-derived cytokine production.

Fig. 4 Systemic monocyte-derived cytokine release is initiated by T cell–derived TNF-α and can be prevented by anti–TNF-α without impact on antitumor activity.

(A) tSNE plot indicating single-cell expression of TNF-α (left). Single-cell expression of TNF-α by cell type and treatment (right). Nontreated (gray), 4 hours after first TDB dose (red), 7 days after first TDB dose (pink), and 4 hours after second TDB dose (blue). (B) Effect of CRISPR-Cas9–mediated gene KO of TNF-α in primary human CD3+ cells on PMA/ionomycin/BFA stimulation-induced cytokine production. (C) TDB-induced cytokine production and (D) TDB-induced T cell activation (top) and target cell killing (bottom). N/A, not applicable. (E) Influence of anti-human TNF-α (5 μg/ml) (adalimumab) cotreatment on TDB-induced in vitro human T cell activation (24 hours, left) and target cell killing after 24 hours (first dose, middle) or 8-day exposure (second dose, right). Schematic of experimental design is presented in Fig. 1B. (F) In vivo effect of anti–TNF-α cotreatment on anti-HER2/CD3 TDB–induced systemic cytokine release in nontumor-bearing MMTV.huHER2 TG mice. Anti-mouse TNF-α (XT3.11, 500 μg per dose) was administered twice, 1 day and 1 hour before anti-HER2/CD3 TDB (0.5 mg/kg). Serum was harvested 2 hours after TDB dose for Luminex analysis. (G) Anti–TNF-α cotreatment effect on antitumor efficacy of anti-HER2/CD3 in treatment of MMTV.huHER2 TG mice with palpable mammary tumors. Data color: (D) Red, WT T cells and blue, TNF-α KO T cells. (E, F, and G) Red, TDB; cyan, anti–TNF-α; blue, TDB and anti–TNF-α; and black, vehicle. Significance of differences in single-cell RNA-seq expression was assessed using a likelihood-ratio test (A). *P < 1 × 10−10 and ***P < 1 × 10−50. Unpaired t test with Welch’s correction was used for statistical analysis (F and G). NS, not significant. Data are shown as means ± SEM (F and G) or mean ± SD (C, D, and E).

To directly test whether TNF-α initiates monocyte-derived cytokine release, we genetically removed TNF-α via CRISPR-Cas9 selectively from human peripheral T cells and assessed cytokine release in the context of our in vitro killing assay. The cell type–specific CRISPR-Cas9 KO of TNF-α reduced the number of TNF-α–positive T cells by 95 to 98% (Fig. 4B) upon phorbol 12-myristate 13-acetate(PMA)/ionomycin/brefeldin A (BFA) stimulation without an effect on the number of IFN-γ–positive cells. No TNF-α release was detected when TNF-α KO CD3+ T cells were incubated with HER2+ target cells and anti-HER2/CD3 TDB (Fig. 4C). Robust secretion of TNF-α was induced by the TDB when WT CD3+ T cells were mixed with an autologous CD3-depleted fraction of PBMC. TNF-α was reduced by 69% using CD3+-specific TNF-α KO cells, demonstrating that T cells are the main, but not the only, source of TDB-induced TNF-α (Fig. 4C). No cytokine release was detected in the absence of CD3+ cells, demonstrating that T cell triggering is required to initiate the cytokine cascade (Fig. 4C). As expected, no IL-6 was detectable when purified CD3+ cells were used as effector cells in the killing assays in the absence of monocytes (Fig. 4C). Addition of the CD3 PBMC fraction together with CD3+ T cells induced IL-6 secretion. Induction of IL-6 was correlated with TNF-α and was 69% lower when TNF-α KO CD3+ cells were used in the assay compared to WT T cells (Fig. 4C). Genetic deletion of TNF-α had only a minor effect on T cell activation and did not impair the ability of T cells to kill target cells when exposed to TDB treatment (Fig. 4D). In summary, these results demonstrate that T cell triggering is required for TDB-induced cytokine release. TDB-activated T cells are the initial and major source of TNF-α, which is upstream of monocyte-derived IL-6 secretion.

Anti–TNF-α prevents TDB-induced systemic monocyte-derived cytokine release without impact on antitumor activity

Our data suggested that upstream T cell–derived TNF-α may be the key link that mediates monocyte activation and could therefore provide a breakpoint to disconnect treatment efficacy from monocyte-derived toxic cytokine production and clinical CRS. Cotreatment of our in vitro model with anti–TNF-α and anti-HER2/CD3 TDB not only resulted in reduction of monocyte-derived IL-6 and IL-1β (fig. S9A) but also reduced T cell–derived cytokines IFN-γ and IL-2 (fig. S9A) and the T cell activation markers CD69 and CD25 (Fig. 4E). However, anti–TNF-α pretreatment had no impact on the ability of T cells to kill tumor cells (Fig. 4E). In vivo, anti–TNF-α pretreatment suppressed TDB-induced IL-6 and IL-1β (Fig. 4F) without an effect on IFN-γ (Fig. 4F), confirming that TNF-α is the key upstream factor mediating monocyte activation in vivo. Correlation between serum TNF-α and IL-6 (Fig. 4F) further supports the relationship between these cytokines. Anti–TNF-α pretreatment had no impact on TDB-induced T cell recruitment or lymphocyte margination response (fig. S9, B and C). Consistent with our in vitro experiments, anti–TNF-α had no impact on antitumor activity of the TDB (Fig. 4G and fig. S9D). In summary, our data demonstrate that TDB treatment induces TNF-α production in T cells, which is required for activation of macrophages and systemic monocyte cytokine release. TNF-α blockade prevents macrophage activation without affecting the T cell killing and effectively mitigates TDB-induced cytokine release.


T cell–retargeting therapies are a relatively recent medical advancement with many open questions regarding their optimal clinical use, such as management of CRS. The in vivo immunologic cascade resulting from CD3 bispecific antibody–induced T cell triggering is generally poorly understood.

The current understanding is mostly extrapolated from CAR T cell studies. Although the clinical presentation of CRS in patients treated with CD3 bispecific molecules resembles CRS observed in CAR T cell–treated patients, these treatment modalities have major differences, which may limit the direct applicability of the findings associated with each approach. For example, the kinetics of CRS onset and its severity for T cell–retargeting therapies are dependent on the magnitude of initial T cell activation, which is substantially less defined for cell-based therapies compared to CD3 bispecific antibodies. Also, in contrast to bispecific antibodies, cell-based treatments are typically administered as a single dose with high variability stemming from differences in the expansion and persistence of the engineered cells.

The ability of T cells to secrete cytokines is commonly used as a marker for T cell activity and to reflect their cytolytic potential. This is clearly reflected in the focus on cytokines as a surrogate biomarker for activity in T cell–redirecting clinical trials. Cytokine secretion has also been proposed as a means to characterize the exhausted state of T cells (20, 21). Although these efforts to characterize exhausted T cell phenotypes have certainly expanded our understanding of the capacity for cytokine expression, the cytolytic functionality requires independent assessment. For example, our data demonstrate that in response to CD3 bispecific antibody therapeutics, T cells clearly retain their cytolytic activity despite their inability to release cytokines. The exhausted T cell state may therefore need to be more directly assessed with functional readouts and more precisely defined with respect to treatment context. Together, these data introduce insights that could change treatment practices and expand the use of T cell–redirecting therapies.

Increased IL-6 is frequently detected in CRS, and IL-6 has been suggested to play a central role in the clinical progression of CRS (4). Genetic experiments using IL-6–deficient mice demonstrated that IL-6 is critical for fever induced by lipopolysaccharide, TNF-α, and IL-1β (22, 23). In addition, clinical CAR T cell (2428) and CD19 BiTE (19) studies report that anti–IL-6R can revert CRS (2428). However, the limited uncontrolled clinical data also indicate that IL-6 blockade is not universally effective in reversing symptoms and preventing cytokine release–related death (3, 5). In addition, conclusions regarding the benefit and extent of IL-6 signaling blockade and the effect on T cell activity are difficult to make from the existing clinical data. Our results indicate that although T cells are required to trigger the immunologic cascade, T cells are not the major source of IL-6 and IL-1 after TDB treatment. Despite the differences between therapies, CAR T and CD3 bispecific antibody–induced cytokine release share some commonalities. In vitro experiments with CD19 CAR T cells suggest that monocyte-lineage cells, rather than CAR T cells, are the major source of IL-6 (29). Furthermore, our work is consistent with previous in vivo studies, suggesting that modulation of macrophage function or depletion of monocytes protected mice from CAR T–induced CRS (17, 18). A subset of patients treated with blinatumomab or CD19 CAR T cells was reported to develop clinical and laboratory findings similar to those of patients with macrophage activation syndrome (MAS)/hemophagocytic lymphohistiocytosis (HLH) (2, 19), supporting the role of macrophages in the clinical presentation of CRS. In summary, our work demonstrates that therapeutic intervention blocking monocyte/macrophage activation downstream of T cell triggering may be an optimal mitigation strategy that would retain the T cell–mediated antitumor activity of TDBs and minimize release of toxic cytokines.

We acknowledge limitations in the studies presented that should be considered when interpreting the data. The hypothesis was tested using one solid tumor-targeting CD3 bispecific antibody, and therefore the applicability to reagents targeting other tumor antigens, including solid tumor and hematological oncology indications, requires further studies. In addition, our in vivo effects were tested using limited mouse models, which may not broadly recapitulate the immune contexture of human tumors. Moreover, although the systemic cytokine release in our model systems demonstrates fidelity with cytokine release in human CRS, our murine models do not manifest the clinical symptoms associated with the disease.

Although IL-6 is often associated with CRS, increases in multiple other cytokines have been noted (24, 28, 30). Moreover, preclinical data indicate that there are differences between individual cytokines and potential related toxicities (17, 18, 22), suggesting that blocking a single downstream cytokine may not be sufficient to eliminate CRS. Stimulation of monocytes/macrophages and neutrophils by exogenous pyrogens is a key step causing stimulation of endogenous pyrogens that initiate the febrile response. Along with IL-1 and IL-6, TNF-α is a major endogenous pyrogen and can induce several clinical symptoms associated with CRS, as well as the synthesis of acute phase proteins (31). Previous studies suggest that IL-6 is downstream of IL-1 and TNF-α signaling, although functional hierarchy of endogenous pyrogens is complicated due to cross-induction of these molecules (23, 32).

In addition to the elevation of TNF-α, TDB treatment induced dynamic expression changes of TNF family receptors, further implicating regulation of TNF signaling. Our data show that monocytes express both TNFR1 and TNFR2, indicating that monocytes remain poised to respond to either soluble or transmembrane forms of TNF-α (33). The induced and sustained expression of TNFR2 in T cells observed after TDB treatment is intriguing and perhaps partly accounted for by the known role of IFN signaling in inducing TNFR expression (34). TNFR2 signaling mediates cytotoxic activity and TNF-α production in T cells in the context of antigen stimulation (35), which may implicate a potential role in mediating second-dose TDB T cell phenotypes and feedback effects on T cells; however, the requirements for TNFR2 in the context of TDB-induced killing need to be explored further.

Although clinical studies report increased TNF-α (24, 27, 36) in response to T cell–retargeting therapies, it has not been the focal point of clinical biomarker studies. For instance, a marked TNF-α increase was detected in all patients treated with CD28 agonist (TGN1412) and preceded other cytokine increases in the immunologic progression of severe CRS (37), supporting an upstream, triggering role for TNF-α. Anecdotal reports of TNF-α inhibitor use to treat CRS exist, but none have demonstrated benefit thus far (2, 38). These studies focused on patients with refractory CRS, which may have biased against success. Our data suggest that TNF-α release upon initial T cell activation is the proximal trigger, indicating that preventative treatment with anti–TNF-α would be crucial for successful intervention. Furthermore, previous reports were focused on anti–TNF-α treatment in the context of CAR T–induced CRS, which may not directly translate to CD3 bispecific–induced CRS. In the context of MAS as a complication of arthritis, high TNF-α concentrations have been reported, and case studies demonstrate that anti–TNF-α can be used to effectively treat MAS (39, 40). Together with our findings, these results indicate that TNF-α is a critical upstream node that mediates inflammatory processes and has the potential to be targeted. Accurate timing of cytokine interventional treatment is likely critical for the success of T cell–retargeting therapies. If the key role for TNF-α is to trigger macrophage activation, prophylactic use may be optimal as compared to therapeutic administration upon clinical manifestation of CRS. Given that we show that TNF-α is not required for CD3 bispecific T cell cytolytic activity, prophylactic use of anti–TNF-α is likely to be a viable therapeutic combination treatment because it should not compromise antitumor efficacy.


Study design

The main objective of this study was to characterize details of anti-HER2/CD3 TDB–induced cytokine release and find optimal nodes for mitigating CRS to aid clinical development of T cell therapeutics. Cytokine release was investigated using cultured human blood cells and mouse models. In vitro studies were based on one to three biological repeats. For in vivo studies, 4 to 10 mice were randomized to each treatment group. The number of mice used was determined using minimum necessary animals to ensure no waste of animal life. Study size determination was influenced by the robustness of the biological endpoint. For tumor response studies, larger cohort sizes were accommodated to allow for equal representation of tumor sizes among groups. Data analysis was performed unblinded.

Antibodies and reagents

Bispecific immunoglobulin G (IgG) TDB assembly from half-antibodies was performed as described previously (13, 41). Human IgG1 anti-HER2(hu4D5)/CD3 TDB was used in all assays with human T cells and included N297G substitution to attenuate the Fc-mediated effector functions. Experiments with murine T cells were done using mIgG2-based TDB with L234A, L235A, and P329G substitutions to attenuate Fc. For muIgG2a, the “knob” arm was murine anti-HER2 4D5 (42) and the “hole” was chimeric anti-murine CD3 2C11 (43). mu4D5-2C11 was used in all in vivo mouse studies, with the exception of Fig. 3B and figs. S1C and S5D, where TDB had antihuman CD3 arm (Genentech). TDBs were diluted in 20 mM histidine acetate, 240 mM sucrose, 0.02% Tween 20 (pH 5.5 buffer) (vehicle). Antibodies for flow cytometry were from BD Biosciences or BioLegend unless indicated otherwise. IL-6 signaling was blocked using antibodies against human (clone 6708, R&D Systems) or mouse IL-6 (clone MP5-20F3, BioXcell) or IL-6 receptor (clone 15A7, BioXCell). Recombinant, nonglycosylated human IL-1 receptor antagonist (IL-1Ra; anakinra, Swedish Orphan Biovitrum AB) was used to block IL-1 signaling. Adalimumab (AbbVie) was used to block human TNF-α, and anti-mouse TNF-α (clone XT3.11) was from BioXCell. Dexamethasone was from STEMCELL Technologies.

Human blood cell fractionation

Human blood samples were obtained from healthy volunteers through the Genentech employee donation program, under an Institutional Review Board–approved protocol using an informed consent process. PBMCs were separated from the blood of healthy volunteers using Lymphoprep medium (STEMCELL Technologies). Miltenyi CD8 T cell isolation kit (catalog no. 130-096-495) and Miltenyi XS column (catalog no. 130-041-202) were used for isolation of CD8+ cells.

In vitro cell killing, T cell activation, and cytokine analysis

For the killing assays, human PBMCs and HER2-amplified target cells (KPL-4; at a density of 20,000 cells per well) were incubated in 10:1 ratio in the presence of anti-HER2/CD3 TDB for 24 hours in black, clear-bottomed 96-well plates. Viability was measured using CellTiter-Glo luminescent cell viability reagent (Promega, catalog no. G7570). For T cell activation and cytokine analysis, human PBMCs and HER2-amplified target cells KPL-4 in 10:1 ratio were incubated in the presence of anti-HER2/CD3 TDB for 24 hours in flat-bottom 96-well plates (BD). CD69 and CD25 surface markers were used to detect T cell activation by flow cytometry. Cytokines and chemokines from culture supernatants were analyzed using human cytokine MILLIPLEX Luminex (EMD Millipore). Briefly, 25 μl of sample and 25 μl of analyte-specific color-coded magnetic beads coated with capture antibodies were mixed. Biotinylated detection antibodies were added followed by incubating with streptavidin-phycoerythrin. The median fluorescent intensity (MFI) data were analyzed using a five-parameter logistic method for calculating cytokine/chemokine concentrations in samples.

Two-dose in vitro model

Human PBMCs and KPL-4 cells (in 10:1 ratio) were incubated with anti-HER2/CD3 TDB (50 ng/ml) in six-well plates for 7 days. All cells were harvested by repeated pipetting and digesting the cells by nonenzyme digestion buffer (Sigma) at room temperature for 5 min to achieve single-cell suspension. Live PBMCs were enriched using Lymphoprep medium gradient centrifugation and cocultured with fresh KPL-4 and a second dose of anti-HER2/CD3 TDB at the indicated concentration.

CRISPR KO of TNF-α in human T cells

A mixture of three guide RNAs (GCTGAGGAACAAGCACCGCC, CTGATTAGAGAGAGGTCCCT, and ATCTCTCAGCTCCACGCCAT) was used to target TNF-α (44). Nonspecific control guide RNA sequences were used as CRISPR controls. Each guide RNA was combined separately with tracrRNA (Alt-R CRISPR-Cas9 tracrRNA, Integrated DNA Technologies) and denatured for 5 min at 95°C followed by slow cool down to room temperature. The Cas9 and RNA complex was generated by incubating guide RNA with Cas9.V3 nuclease (Integrated DNA Technologies) at room temperature for 10 min. TNF-α–targeting ribonucleoprotein complexes were pooled and electroporated into purified human CD3+ T cells (Miltenyi) using 4D-Nucleofactor system in P2 buffer (Lonza). The purity of T cell fraction was confirmed by flow cytometry (>91% CD3+). Electroporated cells were cultured for 72 hours before assays. Human non–T cells were enriched by depletion of CD3+ T cells from human PBMCs using human CD3+ microbeads (Miltenyi) followed by fluorescence-activated cell sorting (FACS) to eliminate the residual CD3+ T cells. The purity of non–T cells was >99%.

Generation and analysis of bulk RNA-seq data

SK-BR-3 cells were labeled with carboxyfluorescein diacetate succinimidyl ester (CFSE) (Thermo Fisher Scientific; CellTrace, catalog no. C34554), plated on six-well dishes, and incubated at 37°C overnight. CD8+ cells were added in an effector:target (E:T) ratio of 3:1. Anti-HER2/CD3 TDB (1 μg/ml) was added to the plates, and CD8+ T cells were harvested after 4 hours and FACS-sorted [propidium iodide (PI), CFSE] to reach ~100% purity. T cells were flash-frozen on dry ice. Total RNA from isolated T cells was extracted using the RNeasy Micro Kit (Qiagen). Three replicate samples were collected for each treatment condition. The concentration of RNA was determined using NanoDrop 8000 (Thermo Fisher Scientific), and the integrity of RNA was assessed by Fragment Analyzer (Advanced Analytical Technologies). About 500 ng of total RNA was used as an input for library preparation using the TruSeq RNA Sample Preparation Kit v2 (Illumina). Size of the libraries was confirmed using High Sensitivity D1K screen tape (Agilent Technologies), and their concentration was determined as recommended by the manufacturer by quantitative polymerase chain reaction–based method using the Library Quantification Kit (product no. KK4873, KAPA). The libraries were multiplexed and sequenced on Illumina HiSeq4000 (Illumina). Average of 84 million single-end 50–base pair (bp) reads was obtained per sample.

Reads were first aligned to ribosomal RNA sequences to remove ribosomal reads. The remaining reads were aligned to the human reference genome [National Center for Biotechnology Information (NCBI) Build 38] using Genomic Short-read Nucleotide Alignment Program (GSNAP) (45) version “2013-10-10,” allowing a maximum of two mismatches per 50 bp sequence (parameters: “-M 2 -n 10 -B 2 -i 1 -N 1 -w 200000 -E 1 --pairmax-rna=200000 --clip-overlap”). Transcript annotation was based on the RefSeq database (NCBI annotation release 106). To quantify gene expression, the number of reads mapped to the exons of each RefSeq gene was calculated using the HTSeqGenie R package. Read counts were scaled by library size and quantile-normalized, and precision weights were calculated using the “voom” R package (46). Subsequently, differential expression analysis on the normalized count data was performed using the “limma” R package (47) by contrasting HER2-TDB–treated samples with control samples. Gene expression was considered significantly different across groups if we observed a |log2-fold change| ≥ 1 (estimated from the model coefficients) associated with a false discovery rate (FDR)–adjusted P value of ≤0.05. In addition, gene expression was obtained in the form of normalized reads per kilobase gene model per million total reads, as described previously (48).

Generation, processing, and analysis of single-cell RNA-seq data

Human PBMCs from one healthy donor and target cells KPL-4 were cocultured in 10:1 ratio. For first TDB treatment, the cocultures were treated with HER2-TDB (50 ng/ml) or fresh medium as control for 4 hours. For second TDB treatment, PBMCs were treated with HER2-TDB (50 ng/ml) for 7 days, and nonadherent cells were recovered and purified by gradient centrifugation. PBMCs were washed twice with fresh medium, counted, and cocultured with fresh target cells (KPL-4) in 10:1 ratio with HER2-TDB (50 ng/ml) or in fresh medium as control for 4 hours. At the end of the incubation, nonadherent cells were collected by gently pipetting up and down. The adherent cells were treated with nonenzyme dissociation buffer (Sigma) for 5 min and gently pipetted up and down to harvest the PBMCs without disturbing the KPL-4 cells. The PBMCs were purified by gradient centrifugation to get rid of the dead cells.

Samples were processed for single-cell RNA-seq, as described previously (49), using the Chromium Single Cell 3′ Library and Gel Bead Kit v2, following the manufacturer’s manual (CG00052 Chromium Single Cell 3′ Reagent Kits v2 User Guide RevA; 10X Genomics). Cell density and viability of the single-cell suspensions were determined by Vi-CELL XR cell counter (Beckman Coulter). All of the processed samples had more than 95% viable cells. Cell density was used to impute the volume of single-cell suspension needed in the room temperature master mix, aiming to achieve ~6000 cells per sample. Complementary DNAs and libraries were prepared following the manufacturer’s manual (10X Genomics). Libraries were profiled by a Bioanalyzer High Sensitivity DNA kit (Agilent Technologies) and quantified using the Kapa Library Quantification Kit (Kapa Biosystems). Each library was sequenced in one lane of HiSeq4000 (Illumina) following the manufacturer’s sequencing specification (10X Genomics).

Sequencing data were processed as described previously (49). In short, reads were tallied and demultiplexed on the basis of their association with cell-specific barcodes. Only barcodes with >10K associated reads were considered for further processing. Reads were mapped to the human reference genome using the same GSNAP settings as for bulk RNA. The number of transcripts per gene was quantified on the basis of unique molecular identifiers (UMIs) using reads that overlapped exonic regions in the sense direction. Only cells with a mitochondrial UMI fraction of <0.25 were considered for analysis. Per-gene UMI counts were normalized to the total number of transcripts per cell and scaled by the median number of transcripts across all cells. Dataset alignment (between first and second time point), cell clustering, visualization, calculation of gene set scores, and differential expression analysis were performed according to best practices using Seurat (v3.0.0.9000) (50). In an initial round of clustering, a group of cells that had high amounts of the ERBB2 transcript (and was thus considered to be the KPL-4 cell line) was identified and removed from all further analysis. Differential expression between treatment conditions was based on the likelihood-ratio test for single-cell gene expression (51). Gene sets that were used for calculation of cell type–specific scores and cell type identification are provided in data file S1 and were derived from the top 50 most strongly up-regulated genes for relevant immune cell types in a recent study (52) or were manually assembled.

Gene set analysis

Quantitative set analysis for gene expression (53) was performed to identify relevant biological processes associated with T cell activation. For that purpose, HER2-TDB–treated samples were contrasted with control samples. Subsequently, the gene set activity (the mean difference in log2 expression of the individual genes that compose the set) was calculated for all sets present in Molecular Signatures Database (MSigDB) (54). FDR-adjusted P values were obtained by comparing the probability distribution function of log2-fold changes in a given gene set to a baseline value of zero using a one-sided test.

Genetically engineered mouse models

MMTV.huHER2.FVB/n transgenic female mice have been previously described (14) and were maintained on FVB/n background strain. HuHER2 mice were used as a double transgenic together with human CD3e (55) in Figs. 3 (B and E)and 4 (F and G), and figs. S1 (B and C), S5 (C and D), S6 (B and D to F), and S9 (B to D). Mice were maintained on an FVB/n and Balb/c mixed background. The animals were dosed and monitored according to the guidelines from the Institutional Animal Care and Use Committee at Genentech Inc. IL-6RA KO mice were generated using the conditional allele (56) interbred with Rocs26.Cre (Taconic). IL-6RA loss was validated in offspring and thereafter maintained as IL-6R.KO on C57Bl6 strain.

Mouse tumor tissue and blood preparation for flow cytometry

Tumor tissue preparation has been previously described (8). Mouse peripheral blood was collected in tubes coated with EDTA (BD). Red blood cells were lysed using ACK lysis buffer (Thermo Fisher Scientific) twice. Remaining cells were resuspended in FACS buffer. Small aliquots were taken for cell counting. Briefly, the cell suspension and Precision Count Beads (BioLegend) were mixed in 1:1 ratio, and flow cytometry was used to calculate the cell number following the manufacturer’s instruction. The rest of the samples were stained with the indicated markers for flow cytometry analysis.

Mouse cytokine and chemokine analysis

Mouse blood was transferred to a microtube (1.1 ml of Z-gel; SarStedt) and centrifuged at 10,000 rpm for 10 min. The sera were transferred into a clean 96 V-bottom plate and stored in −80°C. Cytokines and chemokines from mouse serum or tumor tissue lysates were analyzed using the mouse MILLIPLEX Luminex assay (EMD Millipore) following the manufacturer’s recommendations. Briefly, 25 μl of mouse serum samples or tumor lysate samples and 25 μl of analyte-specific color-coded magnetic beads coated with capture antibodies were mixed. Biotinylated detection antibodies were added followed by incubating with streptavidin-phycoerythrin. The MFI data were analyzed using a five-parameter logistic method for calculating cytokine/chemokine concentrations in samples.

Granzyme B protein analysis

Granzyme B was assessed using a Luminex-based platform [Magnetic Luminex Assay, Human Premixed Multi-Analyte Kit (LXSAHM); R&D Systems, USA]. Supernatants were diluted fivefold and assayed in triplicate. MFI measurements were collected using the Luminex FLEXMAP 3D System and then fitted onto a standard curve to determine picogram per milliliter concentrations.

Ex vivo target cell killing

Propagation of Fo5 tumor allograft model has been previously described (8). Mice with an average tumor size of 570 mm3 were treated with anti-HER2/CD3 TDB for 7 days, and a single-cell suspension was made from tumors as described (8). Harvested cells were stained with anti-HER2, and tumor-infiltrating lymphocytes (TILs) were enriched by FACS gating on forward scatter (FSC)/side scatter (SSC)/PI/HER2. Splenocytes were harvested from the same mice as TILs. Splenic T cells were isolated by a mouse Pan T cell isolation kit (Miltenyi). KPL-4 was seeded in 96-well clear-bottom black plates as target cells. Effector cells (either TILs from tumors or splenic T cells) were added in E:T ratio of 20:1 in the presence or absence of anti-HER2/CD3 TDB (1000 ng/ml). Apoptosis of KPL-4 cells was monitored using IncuCyte caspase-3/7 Green Apoptosis Assay Reagent (Essen BioScience) IncuCyte ZOOM system.


Welch’s t test was used to test for differences between groups. Results were determined to be significant at P < 0.05. Statistical analyses were performed using GraphPad Prism version 7. Bulk and single-cell RNA-seq analyses were performed in R (version 3.5.1). Original data are shown in data file S2.


Fig. S1. Anti-HER2/CD3 treatment-induced cytokine release in mice.

Fig. S2. Effect of anti-HER2/CD3 treatment on peripheral T cell status in mice.

Fig. S3. Characterization of single-cell RNA-seq data.

Fig. S4. Single-cell expression analysis of treatment-based expression patterns within human PBMCs.

Fig. S5. Effect of pharmacologic or genetic IL-6 signaling intervention on TDB pharmacology.

Fig. S6. Effect of prophylactic IL-1 blockade on anti-HER2/CD3 TDB pharmacology.

Fig. S7. CD8-specific expression analysis after TDB treatment.

Fig. S8. Expression of TNFR1 and TNFR2 in human T cells and monocytes.

Fig. S9. Effect of prophylactic TNF-α blockade on anti-HER2/CD3 TDB pharmacology.

Fig. S10. Schematic representation of anti-HER2/CD3 TDB–induced cytokine release.

Data file S1. Marker gene sets used for cell type identification in single-cell RNA-seq data.

Data file S2. Original data.


Acknowledgments: We thank A. Arrazate for tumor harvest; S. Rutz, A. Seki, and J. Johnston for CRISPR design and assistance; C. Kemball and D. Rigas for granzyme assay; D. Nickles for valuable bioinformatic suggestions and discussion; and A. Bruce for illustrations. Funding: All funding for the study was provided by Genentech Inc., a member of the Roche group. Author contributions: All authors contributed to the study design, data analysis, and interpretation. Ji Li, R.P., R.Y., Y.-J.J.C., Jason Li, D.S., M.H., and R.C. performed the experiments. R.P., Z.M., K.T., M.R.J., and T.T.J. provided study oversight. Ji Li, R.P., M.R.J., and T.T.J. wrote the manuscript. Competing interests: All authors are employees of Genentech Inc., a member of the Roche Group, and hold stock and options. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. Bulk and single-cell RNA-seq data have been deposited to the European Genome-Phenome Archive (EGA) under accession no. EGAS00001003734. Investigators may request materials from Genentech by submitting a request form at

Stay Connected to Science Translational Medicine

Navigate This Article