Three years ago in a high-profile paper, researchers led by Michal Schwartz at the Weizmann Institute of Science in Rehovot, Israel, deployed an immune-boosting approach pioneered for cancer in a mouse model of Alzheimer’s disease. The results appeared encouraging. The therapeutic antibodies—checkpoint inhibitors—reportedly improved cognition and reduced amyloid plaque in the animals. Now, a follow-up study claims that checkpoint blockade is similarly effective in models of dementia driven by accumulation of pathologic tau. Their results raise the tantalizing prospect that checkpoint inhibitors, several of which are FDA-approved, could be used to treat dementia regardless of its underlying pathology. The study appeared January 28 in Nature Communications.

  • In mouse models, immune checkpoint inhibitors clear amyloid, tau pathology, improve cognition.
  • Researchers are planning to test clinically.
  • Other groups tried to replicate, unsuccessfully.

Groundbreaking, if true. Unfortunately, the earlier work (Jan 2016 news) has not been replicated independently, despite efforts from multiple labs in academia and industry. In one effort, three pharma companies tried, and failed, to recreate the amyloid-clearing effects of antibodies targeting the programmed cell death receptor PD-1, as was first reported by Schwartz in 2016 (Latta-Mahieu et al., 2018). The replication paper’s senior author, Laurent Pradier of the French pharma company Sanofi, told Alzforum that Schwartz’s new results are interesting, but that independent research teams need to confirm and generalize them to other models and antibodies.

Others agree. "This is provocative data, though it comes on the heels of a highly controversial study from this group that has been difficult for others to reproduce," said Todd Golde, University of Florida, Gainesville.

Schwartz is confident in her data. She is working with a startup, ImmunoBrain Checkpoint, Inc., of Ness-Ziona, Israel, to develop checkpoint inhibitors to treat AD. In June 2017, the Danish pharmaceutical company Lundbeck announced it would partner on the project.

Checkpoint inhibitors block the immunosuppressive activity of the programmed cell death (PD-1) receptor and its ligands, PD-L1 and PD-L2. Neutralizing antibodies to PD-1 and PD-L1 stimulate anti-tumor responses; a handful have been approved by the FDA to treat various cancers. In their previous work, Schwartz and colleagues reported that in 5XFAD transgenic mice, injected PD-1 antibodies stimulated the peripheral immune system. As part of this response, interferon-γ produced at the choroid plexus ushered phagocytic monocytes from the blood into the brain. Antibody treatment reportedly reduced amyloid plaque load and improved cognition, even in mice with advanced pathology (Baruch et al., 2016). 

In the new study, first author Neta Rosenzweig tested the same strategy in mice that express the human tau gene with two mutations associated with frontotemporal dementia, K257T and P301S. The animals develop neurofibrillary tangle-like tau pathology and neuroinflammation that worsens with age (Rosenmann et al., 2008). By eight months, their short-term spatial memory has declined, signaled by their apparent inability to distinguish old territory from new in a T maze. One month after eight-month-old mice were injected with either an anti-PD-1 or anti-PD-L1 antibody, but not with isotype-control antibodies, the mice performed almost normally, suggesting an improvement in cognition. Normal mice treated with antibody had no such change in behavior. Treatment reportedly lowered levels of the inflammatory cytokine IL1β and reduced tau hyperphosphorylation and aggregation in the hippocampus and cortex. The treated mice also had more surviving cortical neurons than their untreated counterparts.

Similar to the treated 5XFAD mice, the antibody-treated tau mice had more monocytes cross from their blood to the brain. Single-cell RNA sequencing of these interlopers revealed expression of a collection of scavenger receptors absent from brain-resident microglia. One such receptor, Msr1, also known as Scara1, was previously implicated in clearance of Aβ (Frenkel et al, 2013). Msr1 was required for cognitive improvement: In mice irradiated to destroy bone marrow and deplete blood macrophages, and then reconstituted with bone marrow from wild-type mice, the PD-L1 antibody still improved cognition. But if the bone marrow came from Msr1 knockout mice, the antibody was ineffective.  

In their previous work with AD models, the scientists tested a PD-1 antibody. Here, they report the PD-L1 antibody is just as effective in those animals.

Together, the results of both the amyloid and the tau transgenic models suggest that a dementia treatment need not directly target the pathology in the brain, Schwartz told Alzforum. “This tells us we can target the peripheral immune system, thereby initiating a cascade of immunological events, which, in synergy with the inflammatory signaling from the brain, can lead to restoration of homeostasis and modification of brain pathology, and thus, would be applicable to diverse etiologies in patients,” she said.

As with a car, taking the brakes off the immune system can be dangerous. Checkpoint blockade comes with serious, and in rare cases even fatal, autoimmune side effects (Wang et al. 2018). Schwartz said she plans to use intermittent dosing. With short exposures sufficient to mobilize monocytes to the brain, her group reduced adverse immunological side effects in animal studies, she said.

At least five laboratories have tried to corroborate Schwartz’s 2016 paper. In the only published attempt, researchers from Sanofi, Janssen, and Lilly assessed the effect of checkpoint blockade on amyloid plaque load in their respective labs, using three different mouse models and three PD-1 antibodies, including the rat monoclonal employed by the Schwarz lab. Last fall, the FDA approved Sanofi/Regeneron's cemiplimab (Migden et al, 2018). Janssen and Lilly have investigational checkpoint inhibitors targeting PD-1 or PD-L1. The companies could substantially expand their drugs’ market if they were safe and effective in Alzheimer’s disease.

The pharma researchers confirmed checkpoint blockade in the mice’s periphery, and an increase in interferon-γ in one model. Alas they found no infiltration of the brain by myeloid cells, and in no model did they find a reduction in amyloid plaque load. They did not test cognition.

Pradier, who prefers the word “generalize” to “replicate” or “repeat,” told Alzforum he knows of at least one more company lab that also tried and failed to show amyloid clearance with checkpoint blockade, and assumes there are others. An academic investigator, who asked to remain anonymous, told Alzforum his lab, too, has been unable to confirm the Schwartz lab's findings.

Why the discrepancy? Different animals, antibodies, labs, disease stages? Reproducibility disputes frequently pit the original lab, which insists that replication conditions must be exactly identical, against labs trying to replicate, which argue that for a published effect to hold up in clinical development, it must be generalizable to similar models and other variations in experimental conditions. In other words, corroborating results should be obtained across slightly different experimental paradigms. The consensus is that if a tantalizing finding is finicky, it may survive peer review at a high-profile journal, but not the rigors of a drug development program.

Some commentators suggested that even variations in microbiota between one vivarium and another can lead to incongruent results in mouse studies of immune-related treatments. If so, Pradier and other investigators questioned if the PD-1 inhibitor approach would be robust enough to succeed in people, where such variability is the rule.

Schwartz attributed the disagreement in results to a lack of optimization of the treatment in the other labs, including dose calibration. The antibody’s affinity and half-life critically affect the regimen, she said. Her new study expands on the original single-dose work with data on additional models at multiple doses, she said. Schwartz also suggested that cognition be used as the primary readout, not amyloid clearance. “Since the antibodies are not directed at amyloid plaque, the primary dose calibration should be based on disease modification, as measured by cognitive performance,” she wrote to Alzforum (see full comment below).

Pradier said that for purposes of preclinical work, his and many other groups nowadays consider cognition a “soft endpoint” that needs to be backed up with some measurable molecular pathology response to therapy. That’s why the Sanofi, Janssen, and Lilly groups looked at plaques. Use of cognitive behavior outcomes in transgenic AD mice has diminished across pharma and academic labs in the aftermath of years of poor predictive validity from mice to human trials.

“We all wished it had worked,” Pradier told Alzforum. “Many companies have these antibodies. Hopefully, others will be able to generalize the results, and we’ll figure out conditions that could make it work,” he said.

He noted that the Schwartz lab injected rat or human antibodies into mice at high doses. These foreign proteins, in addition to checkpoint blockade, could contribute to peripheral immune responses and mobilization of innate immunity in brain. “It would be important that such potential ‘sensitizing/potentiating’ factors be elucidated in the future,” he and Andreas Ebneth of Janssen wrote to Alzforum (see full comment below).

Eti Yoles of ImmunoBrain Checkpoint, Schwartz’s startup, told Alzforum that the company has replicated the result with three different anti-PDL-1 antibodies. Schwartz said that Lundbeck reproduced the results in an independent study, where anti-PD-L1 reduced cognitive deficits and brain pathology in a different tau mouse. The companies are developing new PD-L1 targeting antibodies that Schwartz said are tailored to AD treatment. She estimated they will be ready to begin testing in people in 2020. Lundbeck declined to comment.

Last October, researchers at the University of Southampton, U.K., reported that genetically deleting PD-1 in a prion model of neurodegeneration did not recruit monocytes to the diseased brain and slightly worsened the behavioral phenotype (Obst et al., 2018). 

Richard Ransohoff, Third Rock Ventures, Boston, expressed strong reservations about immune checkpoint inhibitors for Alzheimer’s or other dementias. More broadly, he questions the value of animal models as predictors of clinical efficacy in neurodegeneration (Ransohoff, 2018). While genetic studies implicate immune responses in Alzheimer’s disease, he sees no evidence for involvement of the PD1/PDL1 pathway. “Success in one animal experiment, with failure to replicate, should mandate caution in moving forward to the clinic. This concern is all the more relevant because of the known risk of autoimmune complications caused by checkpoint inhibitors. With uncertain benefit and known risk, this approach doesn't seem actionable,” Ransohoff said.

Guillaume Dorothee, INSERM, Paris, also worries about the consequences of moving to the clinic too quickly with a strategy that boosts peripheral immune responses indiscriminately. “I think we shouldn’t take for granted that it would be beneficial. Immune responses are complex, and we need to further understand and validate what different immune cells are doing in the context of amyloid and tau pathology, before going ahead with broad-spectrum boosting approaches,” he said.

Dorothee and David Blum, INSERM, Lille, recently reported that depleting peripheral T cells—the opposite of what checkpoint inhibitors do—reduced neuroinflammation and memory loss in tau mice (Laurent et al., 2017). Dorothee and others also reported that selectively amplifying regulatory T cells, which are key controllers of immune responses, was beneficial in several mouse amyloid models (Dansokho et al., 2016; Baek et al., 2016Alves et al., 2017). 

“The Schwartz data are intriguing, but there are contradictory findings that remain to be clarified. We also need supporting data from human studies, to avoid the risk that disappointing clinical results would discourage further exploration of this promising avenue of treatment by immunomodulation,” Dorothee said.—Pat McCaffrey


  1. We read with great interest the very extensive publication from Dr. Schwartz’ group, which significantly expands on their previous work on checkpoint blockade for treating neurodegenerative diseases. Using both anti-PD-1 and now anti-PD-L1 antibodies, they demonstrate in a large series of animal studies prevention and reversion of cognitive deficits in the well-characterized 5xFAD amyloid transgenic model, and significant protection against cerebral pathology, together with the new finding of increased anti-inflammatory IL10 cytokine. Very interestingly, they extend these findings to tau transgenic animals and further analyze the possible underlying mechanisms.

    We have previously tried to generalize the checkpoint blockade concept to other amyloid transgenic mouse models, with different genetic backgrounds, and in three different animal facilities in pharmaceutical companies. While the checkpoint blockade was confirmed in the periphery in one case, collectively we could not observe a significant impact on brain amyloid pathology in any model (Latta-Mahieu et al., 2018). Experimental conditions cannot be identical (discussed in detail in Latta-Mahieu et al.), and according to Rosenzweig et al. it is indeed of outmost importance to “optimize treatments with the tested antibody in each animal model at each facility.”

    We note that in the two publications from Dr. Schwartz, xenogenic rat anti-PD1 or anti PD-L1 and human anti-PD-L1 antibodies are used and most recently at significant doses, 0.5 and 1.5 mg per mouse. Those, together with the concomitant checkpoint blockade, could participate in enhancing peripheral immune responses and the mobilization of innate immunity in brain. It would be important that such potential “sensitizing/potentiating” factors be elucidated in the future.

    We welcome and congratulate Dr. Schwartz’s group on this new publication and are looking forward to more teams in the world generalizing the checkpoint blockade concept for treating neurodegenerative diseases.


    . Systemic immune-checkpoint blockade with anti-PD1 antibodies does not alter cerebral amyloid-β burden in several amyloid transgenic mouse models. Glia. 2018 Mar;66(3):492-504. Epub 2017 Nov 14 PubMed.

  2. With regard to Latta-Mahieu et al., 2018, I am confident that the three groups that published this paper would have been able to repeat the positive effect of anti-PD-1 antibody on plaque burden, if they had calibrated the antibody and the treatment protocol in the mouse model that they used, and taken into account the different stages of the disease. The manuscript describes three single experiments, employing animal models that are different from those used in either of our papers. Below, I elaborate on the factors that could explain the failure to detect a decrease in plaque burden, the only read-out that was used to monitor the treatment effect.

    The target of the treatment with anti-PD-1 or anti-PD-L1 is not directed against β-amyloid plaques, but rather, this therapy evokes an immunological pathway that starts in the immune system and facilitates, in synergy with inflammatory cues coming from the brain, the recruitment of immune cells to the brain, together leading to multiple effects that ultimately result in cognitive improvement and neuroprotection. Accordingly, the beneficial effect of the treatment on disease modification is an outcome of the modulation of many processes that contribute to disease escalation; these include, but are not limited to, the clearance of amyloid plaques. In contrast to a treatment that is specifically directed against amyloid pathology, with the primary goal of reducing plaques, and which may or may not affect cognitive performance, in the case of immune checkpoint blockade, the beneficial effect is an outcome of several events that are affected by the therapy, the primary one being immune modulation.  

    Our first paper (Baruch et al., 2016) introduced the concept of adopting immune checkpoint blockade for treating AD, by reporting results of a single dose of anti-PD-1 antibody in 5xFAD mice. In order to generalize the phenomenon, further experiments were needed, including those addressing the mechanism, optimization of the dose/regimen, and testing in additional mouse models, which are all presented in our most recent publication (Rosenzweig et al., 2019). 

    We found that the effect of the antibody is dependent on the dose, and its affinity. Moreover, we found that elevation of IFN-γ, indicating peripheral engagement, is an essential response to the treatment, but is not sufficient. Thus, dosages that were sufficient to induce a peripheral response were insufficient to evoke a central one. In line with that observation, we found that the peripheral immune response can be evoked in wild-type animals as well, though there is no effect on monocyte infiltration and immune activity in their brains, suggesting that IFN-γ is important, but is not a valid marker of efficacy on disease progression of this treatment. In addition, in animal models of dementia, which show no plaque burden, treatment using the correct dose and timing resulted in improved cognition and reduced Tau pathology.

    In Latta-Mathieu et al., 2018, results from three different facilities were presented, which summarize three single experiments. Two of the three experiments reported used Fc chimeras (mIgG1) derived from the RPM1-14 anti-PD-1 antibody; importantly, it was shown that the chimeric Fc region significantly affects the pharmacokinetic profile of such antibodies (Ober et al., 2001), and obviously requires adjustment of the therapeutic dose/regimen. The only experiment which used the same antibody that our team used (Baruch et al., 2016) was carried out by Janssen using the commercially available rat IgG2a anti-PD-1 from BioXcell (clone RMP1-14; Catalog number, BE0146), but was tested on the ThyAPP/PS1 mouse model.

    In light of the above, we suspect that the lack of amyloid-β plaque clearance in all the experiments described in Latta-Mahieu et al., 2018, indicates that the authors, using a single dose, were working under sub-optimal experimental conditions. Based on our understanding of the mechanism, we are currently establishing the nature of the immunological marker of the response induced by the antibody that most closely correlates with treatment efficacy in Alzheimer’s disease modification; such a marker could be used to predict a response leading to disease modification. Until these markers are disseminated, and since the antibodies are not directed at amyloid plaque, the primary dose calibration in any mouse model or facility should be based on disease modification as measured by cognitive performance.


    . Systemic immune-checkpoint blockade with anti-PD1 antibodies does not alter cerebral amyloid-β burden in several amyloid transgenic mouse models. Glia. 2018 Mar;66(3):492-504. Epub 2017 Nov 14 PubMed.

    . PD-1 immune checkpoint blockade reduces pathology and improves memory in mouse models of Alzheimer's disease. Nat Med. 2016 Feb;22(2):135-7. Epub 2016 Jan 18 PubMed.

    . PD-1/PD-L1 checkpoint blockade harnesses monocyte-derived macrophages to combat cognitive impairment in a tauopathy mouse model. Nat Commun. 2019 Jan 28;10(1):465. PubMed.

    . Differences in promiscuity for antibody-FcRn interactions across species: implications for therapeutic antibodies. Int Immunol. 2001 Dec;13(12):1551-9. PubMed.

  3. This paper is interesting and extends the group's previous findings. The authors show PD-1 or PD-L1 blockade improved murine AD and tauopathy models. Interestingly, they found that macrophages expressing scavenger receptors are involved in this process. However, the relationship between the PD-1/PD-L1 system, which works in T cells, and macrophage/microglial scavenger receptors has not been well-elucidated. To extend this effect observed in mice to the clinic, it is necessary to clarify the detailed mechanism of how PD-1/PD-L1 blockade reduces plaque burden. It must be cautioned that this treatment may promote neuroinflammation, like multiple sclerosis.

Make a Comment

To make a comment you must login or register.


News Citations

  1. Cancer Drug Clears Plaque, Improves Mouse Memory

Research Models Citations

  1. 5xFAD (B6SJL)

Mutations Citations

  1. MAPT K257T

Paper Citations

  1. . Systemic immune-checkpoint blockade with anti-PD1 antibodies does not alter cerebral amyloid-β burden in several amyloid transgenic mouse models. Glia. 2018 Mar;66(3):492-504. Epub 2017 Nov 14 PubMed.
  2. . PD-1 immune checkpoint blockade reduces pathology and improves memory in mouse models of Alzheimer's disease. Nat Med. 2016 Feb;22(2):135-7. Epub 2016 Jan 18 PubMed.
  3. . A novel transgenic mouse expressing double mutant tau driven by its natural promoter exhibits tauopathy characteristics. Exp Neurol. 2008 Jul;212(1):71-84. PubMed.
  4. . Scara1 deficiency impairs clearance of soluble amyloid-β by mononuclear phagocytes and accelerates Alzheimer's-like disease progression. Nat Commun. 2013;4:2030. PubMed.
  5. . Fatal Toxic Effects Associated With Immune Checkpoint Inhibitors: A Systematic Review and Meta-analysis. JAMA Oncol. 2018 Dec 1;4(12):1721-1728. PubMed.
  6. . PD-1 Blockade with Cemiplimab in Advanced Cutaneous Squamous-Cell Carcinoma. N Engl J Med. 2018 Jul 26;379(4):341-351. Epub 2018 Jun 4 PubMed.
  7. . PD-1 deficiency is not sufficient to induce myeloid mobilization to the brain or alter the inflammatory profile during chronic neurodegeneration. Brain Behav Immun. 2018 Oct;73:708-716. Epub 2018 Aug 4 PubMed.
  8. . All (animal) models (of neurodegeneration) are wrong. Are they also useful?. J Exp Med. 2018 Dec 3;215(12):2955-2958. Epub 2018 Nov 20 PubMed.
  9. . Hippocampal T cell infiltration promotes neuroinflammation and cognitive decline in a mouse model of tauopathy. Brain. 2017 Jan;140(Pt 1):184-200. Epub 2016 Nov 5 PubMed.
  10. . Regulatory T cells delay disease progression in Alzheimer-like pathology. Brain. 2016 Apr;139(Pt 4):1237-51. Epub 2016 Feb 1 PubMed.
  11. . Neuroprotective effects of CD4+CD25+Foxp3+ regulatory T cells in a 3xTg-AD Alzheimer's disease model. Oncotarget. 2016 Oct 25;7(43):69347-69357. PubMed.
  12. . Interleukin-2 improves amyloid pathology, synaptic failure and memory in Alzheimer's disease mice. Brain. 2017 Mar 1;140(3):826-842. PubMed.

External Citations

  1. FDA approved

Further Reading

No Available Further Reading

Primary Papers

  1. . PD-1/PD-L1 checkpoint blockade harnesses monocyte-derived macrophages to combat cognitive impairment in a tauopathy mouse model. Nat Commun. 2019 Jan 28;10(1):465. PubMed.