Thoughts on Kite's Investor Day Presentations - TCR developments

Kite Pharma had their Investor Day this week (you can watch here). I actually thought there were a number of interesting presentations and discussions that might be worth mentioning & I'll give my thoughts on them more below.

I've skipped around in the beginning, which is mostly introduction and review of all the current data.

Pipeline: Slide 9 & 10, description at 15:00 ***Slide numbers are based on those in the webcast

They gave the following timeline for their pre-IND products:
HPV E7 entering clinical trials later this year at NCI
HPV E6 entering clinical trials 2016 as a Kite IND
1st Amgen collaboration CAR product 2nd half 2016 IND

I've previously written about Kite's neoantigen TCR program, and there were some interesting developments that were mentioned at the end of the investor day - see Steven Rosenberg section below.

KRAS TCR
Additionally,  I had not heard of their KRAS TCR product, which sounds very interesting. They stated the KRAS TCR program is entering clinical trials later this year at NCI, quite soon. People have tried to find ways to target "undruggable" but very commonly mutated genes in cancer like KRAS by trying to generate a T cell response (vaccines etc.) against the mutated portion of the protein, but with little success so far. If one could generate an effective response, it would be very exciting, as the mutations are clearly cancer driver mutations, present in a large percentage of tumors, and importantly also relatively homogeneous within the tumor, and would be difficult for the cancer to evolve around. So, while the idea is great if you had an effective KRAS mutant-specific TCR, the attempts so far have been underwhelming, so why would it be different here, and how did they develop this TCR?

A clue may have come later in the presentation by Dr. Steven Rosenberg of NCI, presenting on identification of neoantigen-reactive TCRs, which I'll get back to later, put up this slide (slide 96, 1:39:20)

While identifying neoantigen-reactive TCRs, they found one that was reactive against the common KRAS G12D mutation. G12D is extremely prevalent in a number of tumors:

From fantastic Ras review by McCormick and colleagues, Open Archive Here

So, it appears that by chance, they've discovered a TCR that is reactive against G12D, and like all TCRs, it is restricted against a specific HLA allele, HLA-A11, disclosed here (HT @SkepticalPhD). I am assuming that this naturally occurring TCR is the one that they are using in their upcoming clinical trial later this year, but that does not appear to have been disclosed yet. I expect since Dr. Rosenberg presented this currently unpublished data that we will be seeing a publication relatively soon on all of this. Also, if they continue to catalog what are the neoantigens and identify TCRs from TILs (tumor-infiltrating lymphocytes) that can react against them, they may be able to build up a bigger library of TCRs against shared driver mutations across a variety of HLA alleles. For me, this was the most exciting development coming out of Kite's presentation, and I'll be following the KRAS TCR story closely. There were a number of other interesting topics discussed and developments.

Multiple CAR Inputs - Inhibitory CARs
CARs are probably the most sensitive of antibody-targeted approaches for killing cancer cells. This has obvious advantages, but the disadvantage is that you need an exquisitely specific target that is not present at all (or certainly not at significant levels) on essential normal tissue. This severely limits the number of targets that make sense for CARs to be designed against. Recently a number of groups, including the Sadelain lab (Juno IP), have designed CARs that incorporate multiple inputs to improve their specificity. Essentially you are adding logic gates to the T cell so that it is designed to target cells that have:
Target A AND Target B: http://www.ncbi.nlm.nih.gov/pubmed/23242161
Target A NOT Target B: http://www.ncbi.nlm.nih.gov/pubmed/24337479
Target A OR Target B: http://www.ncbi.nlm.nih.gov/pubmed/23839099

While the first two examples were from the Sadelain lab, other groups are working on building similar systems, such as Martin Pule (IP I believe licensed to Autolus). Other labs seem to be working on additional approaches as well.

On Slide 43 (53:55) Kite revealed plans for an inhibitory CAR (the NOT approach above) to improve on-target off-tumor specificity & potentially open up new targets. The presenter said they could not reveal the details yet as this is still ongoing work, but it will be interesting to see what targets they go after if these make it to the clinic at Kite, Juno, or other CAR players.

Screening for TCRs

Ton Schumacher, now CSO at Kite EU, presented on technology he has developed to screen, similar to how antibodies can be screened, for TCR binding to a given target (starting at slide 48, 1:00:00).

In addition to screen for TCRs with optimal properties, it is possible that this approach could help overcome one of the drawbacks of TCR-based therapies, which is that they have to be specific to a patient's HLA allele, and any given TCR will not be able to be given to every patient. This screening technology was specifically mentioned as a way to generate TCRs against different HLA-alleles for HPV antigens to be able to cover about 90% of patients (at 2:17:00). Schumacher's group has also developed ways of screening patient TCRs for tumor reactivity, a potential approach for developing patient-specific neoantigen TCRs.

Steven Rosenberg

Rosenberg presented starting at slide 56, 1:20:40, and even though he was summarizing previous results, I thought the whole thing was worth a listen.

I'll just quickly mention some interesting slides he presented on upcoming CAR & TCR programs first.

Slide 79, 1:31:50 

He mentioned a Thyroglobulin CAR for thyroid cancers entering clinical trials at NCI. I thought this was interesting as it's a tissue-specific target, that might be the most similar to CD19 in terms of a target present on a non-essential tissue that you it is acceptable to ablate (the thyroid here, vs. B cells with CD19). I would be interested in other targets present on non-essential tissues as CAR targets, perhaps others will still need added specificity of multiple CAR inputs, as previously mentioned above.

Slide 82, 1:34:00

When discussing targeting Cancer/Testes (CT) antigens, he mentioned that while in certain tumor types NY-ESO-1 is re-expressed frequently, across all epithelial tumors, only 1.8% express NY-ESO-1 on >50% of their cells. NY-ESO-1 directed TCRs have already shown efficacy in synovial sarcoma and melanoma.
 

MAGE A3 on the other hand is present in 17.8% of epithelial tumors on >50% of cells. They are now concentrating making better TCRs here against MAGE A3, as numerous, and fatal cross-reactivity had occurred with previous TCRs against this target, describe here, here and here.
If they can develop a safe & effective TCR against MAGE-A3 it could b

Slide 95, 1:38:30 & Slide 96

Now getting to patient-specific neoantigen TCRs, Rosenberg pointed out the potential benefit of inserting TCRs into T cells as opposed to using TILs directly, specifically that you can put the TCR into the right T cell, not an exhausted, or more differentiated T cell that might be more prevalent in TIL therapies. TILs have already had some remarkable activity, specifically in melanoma, and additionally Rosenberg has recently published a case report in Science, where they purified a specific T cell clone reactive against a patient-specific neoantigen and expanded those to make a new T cell product given to the patient with impressive efficacy. While this first patient, presumably patient 1 in the table below, was treated with purified TILs, Rosenberg's group has gone on to identify the neoantigens the TILs can react against as well as identify the TCRs that are reacting against them.

I am not sure if the patients listed above were just analyzed to see if they could identify TCRs against patient-specific mutations, or if they were treated with either a specific TIL clone reactive against the antigen or with a neoantigen-specific TCR. It will certainly be interesting to see what comes next with these approaches that presumably will be coming relatively soon in publications or presentations.

Manufacturing of an autologous T cell product reactive against a patient-specific mutation seems like a daunting task, but, perhaps it's not quite as daunting as I previously thought. In the Q&A at 2:01:30 - Rosenberg stated that it only takes 48 hours to sequence the exome from a tumor, and then only an additional 48 hours to identify which peptides potentially bind the patient's MHC. So maybe with approaches Rosenberg has described in reviews here, or those previously mentioned above developed by Ton Schumacher, the development of patient-specific neoantigen-reactive TCRs will be within the realm of feasibility.

T Cell Product Composition

Another interesting point that is not mentioned very often, but probably influences the quality & persistence of all T cell products, whether CAR or TCR, is the differentiation status of the T cells in the cell product before administration. I don't think it's completely understood what's the ideal mix of the different levels of differentiated T cells. This topic was discussed on slides 41 & 42, starting at 49:35, for those interested:

Researchers at the NCI have published a methods paper for generating high central-memory phenotype T cell products here, which might be what they already use: http://www.ncbi.nlm.nih.gov/pubmed/20551831

Kite is looking at using a small molecule to be able to expand T cells while maintaining a more immature differentiation state for future T cell products (slide 42)

Here are some papers from NCI on modulating the differentiation state of T cells:

http://www.ncbi.nlm.nih.gov/pubmed/23207483

http://www.ncbi.nlm.nih.gov/pubmed/25432172

Other Q&A Notes (1:41:30)

2:02:50 - Allogeneic CAR-T cells

In general there are a number of barriers to effective allogeneic CAR-T therapy.

Immune suppress patients (normally not continued indefinitely)

Rejected even with major MHC matched or removed, because of minor MHC

GvH dangerous —> Rosenberg doesn’t see allo happening now

Universal NK cells —> 10^11 NK cells that can recognize cancer cell lines in vitro & no effect in patients (Rosenberg), maybe some evidence in hematologic cancers, but in solids not aware, but maybe NK approaches can be used to improve ADCC.

2:11:45 - Persistence of T cells

Long term T cell persistence might not be necessary, but it might be just correlated with good T cells that can recognize tumor.

Appears only a few weeks needed to get CR (After 1 month after treatment in Kite study, even if persistence waned after that the CRs could still be maintained - presented at AACR15).
This might have implications for thinking about how long allogeneic T cells might need to persist for sufficient activity before they are rejected by the patient's immune system.

2:14:10 - HPV E6 Antigen - why so excited?

TIL data good

Already have responses in HPV E6 TCR program, but this is not the place to disclose the full dataset. Goals now to increase response rates, need to make HLA-specific —> Schumacher tech allows identification of HPV TCRs against multiple HLA types —> so can cover 90% of the population.

Bluebird collaboration for next-gen: http://ir.kitepharma.com/releasedetail.cfm?ReleaseID=918816
Sounds like PD-1 knockout, or something similar.

2:19:05 - Target multiple mutations simultaneously?

Ongoing CRs from TILs —> TILs recognizing 1 or more, but normally number is pretty small, just a few 1, 2, 3 & still the can avoid escape.

Would like to target multiple or target a driver mutation

Mesothelin CAR-T —> probably not a good target because on a normal tissue (although they are still doing dose escalation trial too).

Conclusions
I feel like Kite/NCI has been building a really exciting TCR platform, but TCRs seem to be getting overshadowed by the hype (somewhat deserved) around CARs. I think Kite is going after the targets that seem to make the most sense, with the highest tumor-specificity -
Cancer/Testes antigens --> such as NY-ESO-1 & MAGE A3
Viral antigens --> such as HPV E6 & E7
Neoantigens --> which in general are patient-specific, but through this approach, they've already found a TCR targeting the ubiquitous KRAS G12D mutation.

So I will be very interested in following the ongoing developments in the TCR space, particularly against neoantigens & KRAS G12D, and I expect, based on Rosenberg's presentation, that we will be seeing additional publications or presentations in the not too distant future.




Disclosure: I own shares of KITE

Thoughts on Bluebird's EHA LentiGlobin Presentation & Sickle Cell Disease

Bluebird bio presented updated data on Saturday at the European Hematology Association conference on their first sickle cell patient treated with their LentiGlobin gene therapy product with follow up out to 6 months post-transplant. I wrote about my expectations for the 6 month data in a previous post here. I also wrote short posts before and after the abstract was released containing the 4.5 month data, where I touched upon the differences between treating Sickle Cell & B-thalassemia with LentiGlobin, as well as the characteristics of the drug product the first patient with SCD received, such as CD34+ cell dose and average vector copy number (VCN).

The major point in my most recent post was that the large amount of normal HbA hemoglobin the patient received from a donor blood transfusion might obscure the actual relative amounts of T87Q the patient-derived red blood cells contained. The patient was being weaned off blood transfusions, with the last transfusion coming on Day +88, presumably just before the 3 month data was collected. The majority of the patient's blood, and thus hemoglobin, came from the transfused blood at 3 months, and as the transfused blood cells died off, the patient's own blood cells became a larger and larger fraction of their total blood. So while the %T87Q hemoglobin increased dramatically from 9.6% to 24% from 3 to 4.5 months as a percentage of total hemoglobin, it only increased from 39.5% to 42% of patient-derived hemoglobin (this only considered T87Q and HbS hemoglobin & excluded HbF since it was not reported at 3 months). The figure below attempts to illustrate what I mean about the relative amounts of patient-derived and donor red blood cells and hemoglobin at 3 vs 4.5 months:

Here are updated charts from the previous post, now including the 6 month data:

The % changes as a fraction of total hemoglobin - patient & donor-derived
(assumes all non HbS or T87Q is HbA from transfused blood):

This is how the data is usually reported, the T87Q% is growing rapidly relative to total hemoglobin, but this is mainly due to the decrease in transfused hemoglobin as those cells are being replaced by patient hemoglobin from 3 months on.

Relative amounts of T87Q and HbS coming from patient's blood at 3, 4.5 & 6 months:

If we look at just the relative contributions of sickle hemoglobin (HbS) and T87Q, we can see that as a proportion of the patient's own hemoglobin, the percentage that is coming from T87Q is actually growing more slowly from 3 to 6 months. This excludes fetal hemoglobin, generally present at relatively low levels, and which wasn't reported at 3 months.

Adding back in HbF levels (only have data at 4.5 & 6 months):

The more complete picture, with HbF added back in at 4.5 and 6 months, shows that while the amount of patient hemoglobin coming from T87Q increased nicely from 37% to 43% at 6 months, that these gains came mainly at the expense of fetal hemoglobin, with the HbS percentage remaining at 51%. I wouldn't necessarily read too much into the HbF levels showing a relative decrease. HbF is normally elevated in sickle cell patients, and perhaps if blood function is normalizing it could go down (although 5% is still elevated), but this is pure speculation. It is also possible that T87Q will continue to slowly rise, but it seems like the majority of the increase from 9.6% to 24% to 40% was driven by loss of the transfused cells more than a relative increase in T87Q versus HbS. So, speculatively, I could see it starting to plateau here around the 50% range.

So what can we conclude about the state of the patient's blood based on these numbers? Bluebird has reported promising phenotypic data, and will perhaps report more details, on the patient having markers of improving blood function, such as reduced hemolysis markers, a stoppage in pain medication, and no hospitalizations. All of this in line with reduced sickling, although I agree with others that the data is still early here.

There are a few ways to think about what amounts of T87Q will be efficacious, and for a deeper read on all of the following topics, I would suggest taking a look at the PropThink article I co-wrote with Zack (@BioTerp), it discusses more thoroughly these topics and has a lot of useful references. To briefly summarize, the major arguments come from mixed chimerism post-allogeneic transplant in SCD patients and the levels of fetal hemoglobin in Hereditary Persistence of Fetal Hemoglobin (HPFH).

I thought I'd make some minimalist images to illustrate what the red blood cells look like in all the different scenarios, and how the different types of hemoglobin are possibly distributed. Patients with SCD have two copies of sickle-mutant hemoglobin beta (HbS), whereas patients who have sickle cell trait, who only have one mutant copy of HbS, make ~60% normal HbA and ~40% HbS. Since all the cells are genetically identical, presumably the HbA is evenly distributed throughout the red blood cells. These levels and distribution of HbS are generally asymptomatic.

Sickle cell patients can also get blood transfusions, like the patient in the trial was receiving, and generally the goal is to keep HbS levels under 30% to reduce symptoms, particularly the life-threatening ones. In this case, you are just diluting the number of cells that can potentially sickle, but the patient's cells still contain mostly sickle-prone HbS hemoglobin. In HPFH, patients can have around 30% of their total hemoglobin come from fetal hemoglobin, which can block the ability of HbS to polymerize and cause sickling. These patients generally have a very mild disease or are even asymptomatic, which is why bluebird mentions 30% as a target threshold for T87Q levels. Since the cells in HPFH are genetically identical, most likely the fetal hemoglobin is roughly evenly distributed, and the vast majority of the cells have similar amounts of protection.

The reason that I don't think you can make the direct comparison to LentiGlobin-treated red blood cells is because the distribution of T87Q hemoglobin is most likely more heterogeneous than fetal hemoglobin in HPFH, so the cells might not all have similar levels of protection. For the initial cell product, while the average VCN was around 1, that does not mean that every stem cell had 1 copy of the vector integrated. A recent paper using a similar lentiviral approach found that for an average VCN around 0.92, only 30% of progenitor cells were actually modified with between 1 and 9 copies of the vector (the majority 1-2 copies). So in this case there will be genetic heterogeneity in the amount of T87Q vectors integrated, so most likely the expression will be more heterogeneous in red blood cells, and could affect what percentage of cells are protected from sickling.

For instance, the 40% T87Q could be evenly distributed, with all the cells having 40% of their hemoglobin coming from T87Q. Alternatively, T87Q could be concentrated in a few cells, with a significant fraction of cells being unmodified and still able to sickle. Both of those are unlikely, the first due to the heterogeneity of vector integration and the second due to the survival advantage of red blood cells that don't sickle. Most likely there will be some intermediate level of heterogeneity, but that makes me more cautious in assuming that 30% total T87Q will be exactly like 30% fetal hemoglobin in HPFH patients.

We know that there is a survival advantage of normal red blood cells versus SCD red blood cells from mixed chimerism in humans after receiving allogeneic transplants. This is the situation where a patient receives an allogeneic transplant, but not all of their stem cells are replaced, and a percentage of their stem cells come from the donor, but a percentage of their original stem cells remain. It was found that even when <30% of the patient's bone marrow was from the healthy donor, that the vast majority, frequently >90%, of their red blood cells were donor derived. These patients are generally asymptomatic, even with only small amounts of normal stem cells. This suggests there is a selective advantage for non-sickling red blood cells, which should hold true for LentiGlobin-modified RBCs with sufficient T87Q:

Corrected Red Blood Cells Should Have a Selective Advantage:

At least in this paper, which documented a handful of patients with mixed chimerism, the relative advantage of normal RBCs seemed to play out quickly with HbS dropping within 3-6 months, before reaching an equilibrium:

From Walters et al. 2001

The other thing that mixed chimerism teaches us, is that you don't need to correct all the stem cells to get functional benefit approaching that seen with complete repopulation of the bone marrow by donor stem cells after transplant. The LentiGlobin-treated CD34+ cells should be above 30% corrected cells, given that their average VCN was around 1.1, and as previously mentioned, with average 0.92 VCN, there were 30% vector-modified progenitor cells by another group (although single experiment, & in vitro). However, not all LentiGlobin-modified CD34+ cells may produce red blood cells that make sufficient T87Q to completely prevent sickling in those cells. Further, the VCN appears to actually have increased over time in the patient, as at 4.5 months, the average VCN was at 2.4 in peripheral nucleated blood cells. These are not red blood cells, and should not have any selective advantage for having the vector, so this most likely reflects the current VCN in the patient's stem cells. This is also the second highest VCN seen for any patients treated with LentiGlobin, only patient 1202 had higher (with now >4 VCN). So the effective quantity of T87Q produced and number of vector-modified stem cells might be higher than would have been predicted based on the drug product given to the patient.

The higher, and increasing, VCN could be due to random variation, or a slight selective advantage of vector-modified stem cells. It is good to see that even patient 1202, who has had increasing VCN in peripheral blood, now over 4 VCN, does not show any sign of a clonal dominance.

I don't know if you could assume this increase in VCN will happen in all SCD patients treated by bluebird, it hasn't happened for all of the B-thal patients. So, it's possible this first SCD patient will have a better than average response to LentiGlobin in terms of the amounts of T87Q produced, but we'll have to see how this plays out with more treated patients.

In all, I think the data are a promising step forward for LentiGlobin in SCD.






Disclosure: I own share of bluebird

Predictions for %T87Q for Bluebird at 6 Month EHA Presentation

Bluebird bio will be presenting what is expected to be 6 months of follow up on their first sickle cell disease (SCD) patient this weekend at the European Hematology Association (EHA) conference. There is a lot of investor interest in the first data for SCD coming out using their LentiGlobin gene therapy approach. So here are some thoughts about what I might expect the data to look like at 6 months.

To catch up, here are my previous posts on bluebird's LentiGlobin in SCD before the EHA abstract release, and analysis after the abstract release.

My interpretation of bluebird's abstract including 4.5 months of follow up has changed slightly after further considering a comment by @Sharma1981N after my previous writeup. The comment was: Do you think much of the increase in %T87Q from one point to the next could be attributed to decreasing % of transfused RBCs?

I initially responded thinking in terms of the g/dl of T87Q, and that likely it had indeed gone up considerably from month 3 to month 4.5. I wasn't, however, thinking about the relative amount of T87Q coming from the patient's own blood and how that was tracking. This is probably what's actually important, because if the patient is no longer receiving transfusions, then their blood should be made up entirely of HbS, T87Q & some HbF. Here is a table from the abstract breaking down the relative amounts of T87Q, HbS, HbF, and importantly the last pRBC transfusion:

The Day +88 transfusion was presumably right before the 3 month follow up. At the 3 month follow up only 9.6% of total Hb came from T87Q, but only 14.7% came from HbS, with presumably the majority of remainder (HbA) from the transfused blood (>70%). At 4.5 month follow up, not having any transfusions since, the % T87Q and % HbS both rose, to 24% T87Q & 33% HbS (with 7.6% HbF).

When looking at it this way, the amount of T87Q hemoglobin as a proportion of the hemoglobin made by the patient only increased slightly more than HbS did. So, on the downside, this suggests the relative levels of T87Q aren't actually climbing as fast as it seemed like based on the 9.6% to 24% jump. Since we don't have HbF levels for the 3 month visit, if you take that out of the equation and assume the entirety of the blood came from HbS and T87Q, then T87Q only went up from 39.5% of patient hemoglobin at 3 months to 42% of patient hemoglobin at 4.5 months. So, on the downside, T87Q as a proportion of patient hemoglobin might not be increasing that quickly, and may not increase that much more from 4.5 months on. On the plus side, this suggests that the patient is already making greater than 35% T87Q as a percentage of their total hemoglobin. If we include the 4.5 month fetal hemoglobin HbF, then the combined T87Q+HbF% is already up to 49% (37% coming from T87Q & 12% coming from HbF), with just 51% HbS. This is considerably above the 30% T87Q+HbF threshold bluebird has touted as being potentially ameliorative of symptoms.

Here are charts illustrating the above:

The % changes as in the abstract (assumes all non HbS or T87Q is transfused blood):

 Relative amounts of T87Q and HbS coming from patient's blood at 3 & 4.5 months:

 Adding back in HbF levels (only have data at 4.5 months):

So, in what range can we expect the %T87Q of total hemoglobin (patient + donor, as reported in abstract and bluebird) to be in at 6 months. Assuming no transfusions between 3 and 6 months, a major determinant will be how much HbA from transfused blood will still be around 3 months after the last transfusion. Based on the rate of decline of transfused HbA in B-thalassemia patients (below), if the patient hasn't received a blood transfusion since day +88, it is very likely that the vast majority of transfused HbA would be gone. If this is the case, the %T87Q+HbF would seem to likely be >40% and the %T87Q by itself may even be near or greater than 40%.

Briefly, my thoughts on what %T87Q (+HbF) would be likely to achieve a functional cure - the short answer is I don't think we can know for sure yet. I've previously voiced my concerns over making that direct comparison to the 30% amount seen in Hereditary Persistence of Fetal Hemoglobin (HPFH) patients - my main argument focusing on the relatively even distribution of HbF across all red blood cells in those patients, versus most likely heterogeneous T87Q expression in LentiGlobin treated patients' red blood cells. However, even if somewhat heterogeneous, 30% would likely be clinically meaningful based on the amount of HbF increase typically seen with hydroxyurea treatment. I will probably include a slightly more detailed version of my thoughts on what I think about targets for %T87Q in a post after the data is presented.

So, in summary, when looking at the reported %T87Q changes from 3 to 4.5 months as a proportion of total hemoglobin (either patient or donor), the percentage appears to be rising dramatically. However, if we just look at the relative amounts of T87Q to HbS, the relative T87Q amount is rising much more slowly. If we even assume that this relative amount does not increase much more, if the patient is completely weaned off blood transfusions, then they would already produce in the range of 40% T87Q as a percentage of their total hemoglobin, and maybe only 50% HbS. If the patient hasn't received a transfusion since Day +88, it could very well allow the reported %T87Q (from either patient or donor) to rise to near 40% by the 6 month readout expected to be presented this weekend. So what I will be looking for is not only the %T87Q as a proportion of total hemoglobin, but also how the relative amounts of T87Q and HbS have changed from 3 to 4.5 to 6 months. Maybe I'll be surprised and the relative amount will increase substantially from 4.5 to 6 months. There are surely a number of assumptions in the above analysis, but hopefully it gives a good ballpark sense of where things seem to be tracking. Any additional phenotypic data on the patient or red blood cell function will clearly be important too.

Edit: Here is an image to demonstrate what I mean about the relative T87Q amount compared to all hemoglobin, versus the relative T87Q amount compared to patient-derived hemoglobin:

Thoughts on Alnylam's HBV Program

by Robert Kruse

Most of the discussion in HBV siRNA therapies has centered around Arrowhead and Tekmira, since their programs are further advanced and seem to be the primary focus of investors for each company. Lagging behind is Alnylam, which has its own siRNA assets and program against HBV. I wanted to quickly review its HBV targeting strategy with comparisons to Arrowhead and Tekmira, and see if there are any potential advantages.

I listed below a brief summary of the three companies for anyone unfamiliar with their HBV portfolios.

Arrowhead: Currently using IV delivery, composed of two different siRNA molecules against HBV; recently published a generation that can be delivered subcutaneous and target the liver via GalNac.

Tekmira: Lipid nanoparticle based formulation, delivered IV. composed of three different siRNA molecules against HBV

Alynylam: Has both lipid nanoparticle technology delivered IV and GalNac conjugation and subcutaneous delivery strategies, will use two different siRNA molecules; proposed combination with PDL1 siRNA


What intrigues me about Alynylam is their intention to use siRNA against PD-L1 in order to remove the break on virus specific T cells in the liver. This is an added feature that Arrowhead and Tekmira lack in their programs (although Tekmira has begun a multi-pronged approach buying small molecule assets recently). It is well known that chronic viral infections induce exhaustion in CD8 T cell effectors, including in HBV. PD-L1 can be upregulated on hepatocytes in chronic infection, in addition to PD1 upregulation on T cells, leading to inhibition of virus T cell responses. Thus, the PD1 pathways has been proposed as a target to re-awaken the fight against infections, similar to how it is being targeted in cancer immunotherapy currently. A proof of principle for this siRNA PD-L1 therapy was already modeled against murine cytomegalovirus infection of the liver in mice. Furthermore, there has been a published report using the anti-PD1 antibodies as therapies for HBV and that it can enhance activity of T cells in mice. The HBV mouse model used for this purpose has limited direct applications for human anti-HBV immune modeling, however, and direct testing in chimpanzees or phase I trials in patients is needed to truly test the exhausted HBV specific T cell phenotype. The reports of leveraging the PD1 / PD-L1 pathway against viruses looks promising in summation though.

The downside of PD-L1 siRNA could be unlocked any inflammatory effects of T cells against the liver. Currently, anti-PD1 antibodies have well studied toxicities, but the specific targeting here to the liver could uncover potential new toxicities. For example, there could be auto-reactive T cells, held in check by PD-L1 hepatocyte expression, that are then activated by the siRNA. It's worrisome in such a vital organ that a large hepatotoxicity side effect could occur. I would imagine the FDA will be monitoring this closely going forward in any clinical trials. On this point, anti-PD1 antibodies have been used in early trials for HBV+ liver cancer patients, and so perhaps there is already some safety data on inhibiting this pathway in patients with HBV. Additionally, Bristol Meyers-Squibb is planning to test anti-PD-L1 antibodies in HBV infected patients. It remains to be seen whether an siRNA against PD-L1 is similarly safe.

Those concerns aside, the angle of combining HBV and immune targeting with siRNA is very enticing. If the REPLICor reports are to be believed, then a combined HBsAg knockdown and immune stimulation (IFN alpha in their trial) is required for sustained virology response and cure. Given that interferon is challenging for patients to take, an siRNA against PD-L1 could be a more specific and potent immune activator, with less systemic side effects. It seems more and more likely that siRNA against HBV will not be enough, and that other triggers might be needed in order to induce HBV cure. In this respect, Alnylam is ahead of the game having it built into its strategy already.

Thoughts on Seattle Genetics and Unum Therapeutics Deal

by Robert Kruse

The big biotech news of yesterday was a deal struck between Seattle Genetics and Unum Therapeutics to share platform antibodies targeting tumor antigens between the two companies. I wanted to share some of my thoughts on Unum's technology and the potential upside of Seattle Genetics in leveraging it.

Unum is introducing Fc receptors (CD16) into T cells, which otherwise normally don't express these molecules. This allows T cells to bind to antibodies, and those antibodies to trigger T cell activation through 4-1BB and CD3-zeta signaling. The idea builds off of CAR T cell concepts, but adds a twist that the antigen targets could be changed in vivo in the patient, something no other technology offers currently.

Unum technology analysis:

1. This might not be obvious to investors, but why T cells versus NK cells for therapy, which are also cytotoxic but already express CD16 and interact with Fc receptors. The answer lies in manufacturing ease and to a certain extent potency of T cells versus NK cells. T cells currently can be readily expanded from a patient with anti-CD3 and anti-CD28 antibody coated beads to a scale ready for a patient in a 1-3 weeks. The best NK cell expansion protocols rely on K562 cells expressing 41BB ligand as stimulants for expansion and take longer. Using a second cell line, K562 adds cost and complexity to the manufacturing process. The Campana lab, where Unum's technology originated, is actually an expert on NK cells, meaning that if they thought they should leverage T cells, it should tell you something about current NK cell limitations.

From Unum Therapeutics website

2. One of the central advantages of Unum is the ability to re-boost modified T cells with anti-cancer antibodies once the T cells are already in the patient. However, I have concerns about the idea that one could administer new anti-cancer antibodies into the patient to re-arm the T cells at the tumor site. One obstacle is the anti-tumor antibodies reaching deep into the solid tumor, which we already know is difficult. In this regard, CAR T cells with scFv domain on board are possibly better in being able to migrate through dense extracellular barriers. Unum might be channeling the lower hanging fruit of hematopoietic cancers first, and indeed their first targets are CLL and NHL via CD20 antibody.

Assuming this re-arming is possible, the second obstacle is if pulsatile activation of T cells with antibodies will lead to their persistence in vivo. There are many facets to consider with this strategy. The positive spin would be that it is the continuous activation of CAR T cells now that lead to their exhaustion, such that a pulsatile activation would lead to continued persistence and Unum has a genius breakthrough on hand. The negative spin would be that T cells need a more chronic antigen stimulation to persist, and that lack of activation would lead to their cell death. The persistence parameter will be crucial for any future clinical trial, and might require perhaps similar selection for central memory T cell phenotypes pre-infusion in order to optimize the process.

3. I posted this article on Twitter before, but in general, CARs are the most sensitive molecule for detecting antigen on the surface of tumor cells. This has led to their great efficacy, but perhaps also to cytotoxic effects (on target, off tissue). By comparison, bispecific antibodies engaging CD3 need higher antigen thresholds on tumors in order to activate T cell cytotoxicity when tested against the same target antigen head to head.

The Unum play can be seen as a hybrid of CAR and BiTE approaches. Cross-linking CD3 on T cell surface has certain geometric constraints limiting efficacy, which the Unum strategy lacks since they mimic natural CD16 binding on T cell surface. Furthermore, Unum can tailor make a T cell receptor to recognize antibodies via CD16, but in contrast to CD16, transduce a more maximal signal (with 4-1BB costimulation) than most bispecific approaches can through CD3 epsilon engagement alone. Antibodies should be more flexible than their BiTE counterparts and can engage two different antigens leading to higher affinity binding. Thus, Unum might have significant advantages over Amgen's BiTEs for example, but with the big caveat of needing an additional cell therapy product to leverage this benefit.

The Campana paper that founded Unum suggested an approach where one transfects mRNA for the receptor inside T cells, and then infuses T cells into the patient. Antibodies would then be infused into the patient, allowing for any number of antibodies to do the trick. Incubation with antibody before infusion is possible, but the duration of efficacy should be limited since receptors will be turned over and the accompanying antibodies degraded fairly quickly. By comparison, a similar approach of transfecting mRNA into T cells will produce new CAR molecules continuously for 2-3 days. Of course, with the mRNA approach, the downside is optimizing different scFv CAR variants, whereas antibodies won't have this limitation. One could also mix and match various antibodies for the initial infusion, which is an advantage of Unum's approach. Note that Unum's cells, if potent, might degrade the antibodies more quickly than current pharmacokinetics studies with antibodies, something that should be monitored in trials.

Unum could be seen as competition for mRNA CAR players, but their later programs are all using stable viral vector expression, indicating they intended to try repeated antibody dosing to provoke CD16 - T cells inside the body.

4. I think an underrated concern about Unum's strategy is the effects of host antibodies on the T cells administered. Unum relies on a dance of two specific things finding each other at the tumor site to pair for destruction. One can imagine that additional layers could be added to this parlay to make it even more specific. The problem is that the events might just be so rare as to be inefficient.

This inefficiency could be driven by host antibodies outcompeting Rituxumab, for example, for the CD16 receptors on the outside of T cells. Unum can create high affinity CD16 receptors, but these should react to all antibodies, including endogenous ones. Upon infusion, their cells might rapidly be saturated with these decoy antibodies. In the Campana lab paper, they used an immunodeficient mouse model, which does not make any host antibodies, such that the question of competing antibodies remains.

I'm not sure if Unum is pursuing this, but this inefficiency could paradoxically turn into an advantage in other settings. For example, if the body is beginning to fight an infection, it has produced some antiviral IgG's for instance. Perhaps infusing Unum's modified T cells into the patient would allow increased potency of those hose antibodies for their effects.

The unknown dangerous side of this is to what extent people today have auto-reactive host antibodies circulating in us. It could be possible that a large bolous of Unum's T cells could enhance the activity of these molecules, and thereby lead to dangerous effects. There is some unexplored risk there in transfusing essentially un-naturally enhanced Fc-dependent effector cells. Patients out there do have the V158 mutation in CD16, but linking that to 41BB-Zeta signaling is completely novel addition and could create toxicity.

Summary: The best spin for Unum versus Juno and Kite would be if antibodies do readily penetrate the tumor, and the T cells persist and localize to the tumor well, in which case, the technology for doctors to change on the fly which antigen is being targeted would be very powerful, particularly when escape mutants are expected.


Seattle Genetics side:

From Seattle Genetics perspective, it's certainly a low risk and high reward move. They are already developing anti-tumor antibodies. They can give Unum a unique portfolio of new targets for their cell therapy. One question is if Seattle Genetics will eventually modify their antibodies to tailor them for Unum's purposes, instead of a principal purpose for their antibody drug conjugate programs. I would keep an eye on this going forward, since it would indicate an investment of their R&D resources and production facilities toward really pushing cell therapy forward and signal a departure from previous strategies.