Integrating Reproducibility into Academic Incentive Structures

by Robert Kruse

The reproducibility crisis is affecting many areas of science, threatening our progress. Biotech in particular is susceptible to these issues, since most technology is transferred into organizations, as opposed to internal research driving innovation. The NIH and venture capitalists have studied the problem as a barrier to effective drug development, and many VC's have to spend money funding the validation of research rather than advancing it forward.

As a whole, publishing irreproducible science is not a victimless crime, as millions of dollars are wasted each year by labs trying to unwittingly build upon faulty science. For the biotech industry, billion dollar drug development failures can be based around flawed science. One estimate put the total amount in wasted preclinical research at 28 billion dollars every year.  Some scientists today argue that if the original study does not have follow up citations thereby suggesting flaws, that this is already a self-policing system. Unfortunately, that logic is just an interpretation and a lack of follow up could be due to a myriad of other reasons. Given literature now is online forever, we need a formal process of ensuring that the results are accurate and robust, and that researchers can quickly verify which research has been reproduced.

From http://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.1002165

In Silicon Valley, the second or third player into the market is usually the winner, having taken a validated concept and then improved it. Facebook is the big success, not the earlier forbearers of social networks. Yet in science, we discourage the second and third comers to ideas and techniques, labeling them derivative and lacking significant improvement. Herein, we need to change this mantra. However, significant hurdles remain in changing current practices. The main hurdle is that there is no incentive in academic researchers today to actively try to reproduce other studies, whether spending the money or the valuable time to do such experiments. The idea is let someone else do that work, or continue focusing on your own research, and the inevitable tragedy of the commons occurs, where no one reproduces the data. This particularly hurts industry now, which is tasked with taking science and turning it into useful products. In order to get a drug to work, the pathway its based on must actually be important to the target disease. However, investors and companies waste millions of dollars each year chasing drugs based on junk science. Therein lies another crucial difference of motivations: academics are rewarded by journals for great stories and eye-catching headlines. The companies are rewarded by profits from products that actually work to help people.

How do we change these motivations? While many proposals out there consist of establishing new schemes and mandates, the easiest implementation route would build seamlessly into current incentive structures for academicians. Academic researchers are hired based on the impact factor of journals they publish in. There is relentless competition to continually try to get into the best journals, which leads to over-claiming scientific findings. Indeed, Nature and Science, the two best journals, also have the highest retraction rates. In order to motivate researchers to do reproducibility research, they need to be rewarded with Nature, Science, and other high impact papers on their resume. The same papers that will help them get hired into faculty positions, and provide the resume for strong grant funding for their entire careers.

Of course, Nature and Science are not in the business of publishing papers that are reproducing previous work. That is not high impact, since the convention today is for the original paper to be cited by other works. Furthermore, the reproducibility paper never generates the headline, upon which these journals thrive on. In order to solve these two contrasting motivations, it is proposed that journals adopt a new convention for a paper format for reproducibility. This paper would receive the exact same citation credentials as normal publications in that journal, thereby helping the researchers, but the paper would not be indexed by Thomson Reuters, thus not diminishing the journal’s impact factor.

As a test example, let’s consider the case of the STAP cells, published in two different articles in Nature in 2014, consisting of stress induction to produce induced pluripotent stem cells. Many labs across the world tried to reproduce the findings to no avail. If a lab tried to replicate it, whether it worked or not, they could write the work up and then publish it in Nature, to complement the original article, which could then be linked off of it. Since journals and researchers might still want to maintain the prestige of publishing in Nature, the amount of reproducibility papers off any single article could be limited at 2. If 3 labs total can independently verify the same result, than it should be a good sign that the rest of the field can trust the findings. These reproducibility papers would be peer reviewed under the same standards of the field, but without the expectation of impact, importance, and fit to the journal, since they are specifically connected to a single paper already published in that journal. In many ways, they would be similar to the open access journals that are proliferating today. Ideally, these reproducibility papers would be freely accessible and not restricted under any journal pay-walls in order facilitate the dissemination of knowledge about the original finding.

The journals might have a pushback that the proliferation of reproducibility papers might dilute their brand, but that should not be the case. If a study is truly groundbreaking, then it demands other studies to verify its findings. Furthermore, the successful reproduction of that result makes the original study even more powerful and important. While it is not envisioned that reproducibility studies would be actively advertised to the public, one could imagine the journal Nature celebrating the fact that its content has been found to be reproducible, affording it another opportunity to circulate an article in the press citing both the original publication and the new reproducibility finding.

Content is king in the developing digital world, whether from online streaming video services like Netflix, to cable companies signing live sports deal, to news sites such as the Huffington Post accumulating blog writers. Journals were born to serve a different world, where paper copies of journals were mailed out to academic scientists across the world in order to inform them of the latest discoveries. Indeed, the first Impact Factor rankings were issued in 1975, long before the advent of the Internet. Journals today should realize that it is in their interest to capitalize and grow content within their platform, in order to attract more readers on their websites. Rather than continually rejecting papers for not being important or innovative enough, journals should actively draw in more content through publishing online reproducibility studies within that same journal. Instead, journals such as Nature have created a tiered system, wherein they can keep the same studies in their journal family in order to secure payment/content, but this only hurts academic researchers. Instead, the system should be a win-win for both the journals and researchers, by allowing for the creation of a new reproducibility category.

To implement this plan, the main player that would need to acquiesce would be Thomson Reuters, agreeing that this new category of paper within a journal would not count toward impact factor. This agreement would not upset the current system at all, so should be amenable. The journals would also need to agree to this article format, but with it not being indexed, that should be the lesser step, since it won’t hurt their own impact factors. The third step would follow as researchers would flock to this new format, seeking to publish in high impact journals. Important for researchers would be that the articles would appear the same as others in the journal, such that no stigma for research articles would develop. Just as “Brief Communications” are seen the same as “Research Articles” today under the purviews of Pubmed searching, so would reproducibility articles be seen the same under Pubmed. Researchers would then benefit from these improved publication resume in facilitating getting new grants from the NIH, which are in part based on resume today. This proposal would also compliment proposed grant reforms that have called for increased levels of funding based on track record alone, as opposed to current evaluation and scoring of research proposals. The ability to do quality science needs to be rewarded in grants, and not prior reputation or speculative ideas.

The incentives of this new system would draw the highest fervor of reproducibility toward the highest impact journals, just as it should be since these articles are regarded as the most important in science. Furthermore, given the competition for limited reproducibility slots per one paper, the studies would come quickly, as opposed to protracted years between verification that studies are not reproducible. Science would begin to operate as a team across institutions, moving new discoveries together forward. Changing the norms and incentives of any societal system is challenging, but with building reproducibility into the current system, it is our best chance to reform the system through seamless integration.

Summary

• There is a lack of incentive for other academic scientists to invest money in reproducing other published papers since the negative results won't benefit their career.
• A potential solution to this problem would be to create a new category in journals that would allow validation studies to be published in prestigious journals thereby benefiting academic scientists career.
• In order for these validation studies to not hurt the impact factor of the journal since they will likely be cited less and lack the "prestige" element, it can be agreed by Thomson Reuters that these articles will not be counted in the traditional impact factor calculation.
• Journals will thus be able to keep all related content to high-profile papers, academic scientists will have a new route to publish in high impact journals for their CV, and the current system of academic/journal/NIH incentive structures can be maintained.

Thoughts on Bluebird's Striking BCMA CAR-T Data

Bluebird bio just released early data from their initial first-in-human trial of a BCMA CAR-T product for multiple myeloma that will be presented tomorrow at the #ENA2016 meeting. This is a 2nd generation 4-1BB CAR targeted to BCMA with a cyclophosphamide + fludarabine based conditioning regimen. The early results are quite impressive, with a 78% response rate in all 9 patients evaluable for response, and 100% in the 6 patients treated at the two highest cell doses. Even more impressive, and surprising, has been the lack of grade 3 or higher cytokine release syndrome or neurotoxicity that is frequently observed in CD19 CAR trials, as well as other BCMA CAR trials. While this data is still very early, with only 11 patients treated, dose escalation ongoing, and less than 1 year follow-up, the results compare very favorably to other BCMA-directed CARs, which I'll compare below.

Here is the data from Bluebird's press release:

The time to response seems a bit slow (2 & 4 months) so it's possible that some of the PRs in the highest dose cohort will improve with further follow-up. They also mention that all patients in the two highest dose cohorts that had detectable MM in bone marrow had no detectable MM in bone marrow by day 14 and beyond. They do not mention if any of the patients have relapsed, although the wording regarding the sCRs suggests that they have not relapsed at 4 and 6 months follow-up.

These data compare favorably to a 4-1BB BCMA CAR being developed by UPenn/Novartis that had some preliminary data on 6 patients in their recent ASH abstract. In this first cohort, they did not lymphodeplete with conditioning chemotherapy prior to CAR-T infusion. Even without lymphodepletion, they were able to get significant expansion of CAR-T cells in two of the six patients. These two patients also had significant responses - a VGPR and an sCR, with the sCR ongoing at 7 months, but the patient with the VGPR relapsing after 5 months. They administered a similar number of cells as Bluebird, 1.8-5 x 10^8 cells. Importantly, both patients who responded, and had significant expansion of CAR-T cells, also had grade 3 cytokine release syndrome, and one of the two had grade 4 reversible neurotoxicity. CAR-T cells were found in the CSF of this patient. This data is summarized in the table below:

These results were fairly similar, but perhaps more variable, than results published from a clinical trial on an earlier CD28 BCMA CAR at the NCI. In this study, they dose escalated up to 9x10^6 cells/kg and were given conditioning chemotherapy of cyclophosphamide and fludarabine. They treated 12 total patients, with both patients treated at the highest dose having responses of greater than 8 weeks (sCR of 17 weeks and VGPR ongoing at 26 weeks). Patients treated with lower doses of CAR-T cells had either no response, or responses of shorter duration.

From Ali et al., 2016

Importantly, similar to the UPenn study, it was the patients (10 & 11) that had the most dramatic expansion of CAR-T cells that had significant clinical responses:

From Ali et al., 2016

It was also the two patients with these dramatic expansions of CAR-T cells that had CRS and neurotoxicity (reversible).

Thus what is surprising is that the patients in Bluebird's trial achieved significant responses without these significant toxicities. This is quite different from both the UPenn trial and original NCI trials, where activity came with grade 3 and higher CRS and/or neurotoxicity. I'll discuss this difference further below.

These trials also began to characterize relapses after initial responses. In CD19 CAR trials, relapses frequently fall into two categories - loss of the target antigen (CD19) or (/and) loss of persistence of the CAR-T cells. Loss of antigen expression/availability is a concern for all CAR-T therapies. In UPenn's trial, one patient relapsed with loss of BCMA expression and concomitant loss of CAR-T persistence. The patient with an ongoing response still had detectable CAR-T levels. A patient in the NCI CD28 BCMA CAR trial who had a transient response also was found to have lost BCMA expression on their MM cells, but a number of patients who relapsed were not evaluable for BCMA expression, so this may be an underestimation of antigen loss frequencies. Importantly, in this trial, the CAR-T persistence was fairly poor, with loss of CAR-T persistence by 3 months, which could also partially explain the transient responses. The general thinking seems to be that 4-1BB CAR-T cells have better persistence, and CD28 CAR-T cells have greater expansion (and perhaps greater toxicity).

Bluebird will present their data tomorrow, so we should get some additional details.

Questions I'm thinking about going into their presentation:

What was the disease burden in their treated patients? Lower disease burden may cause less CAR-T cell expansion and toxicity. In the NCI trial, both responding patients had bone marrow disease burdens of >80%, and in the UPenn trial, both responding patients had bone marrow disease burdens of >70%.

What were the kinetics of tumor response? What was the speed of decreases in tumor biomarkers and soluble BCMA levels - these decreases were fairly rapid (on the order of weeks) in the NCI trial.

What were the kinetics and peak of CAR-T cell expansion? Did the cells expand to less dramatic levels, or less rapidly, than in previous BCMA CAR-T trials, but were still able to induce responses?

How well do their CAR-T cells persist over time?

What was the state of their CAR-T cells at peak levels? %CD8+ T cells and levels of activation markers.

How are the conditions different (if different at all) for how Bluebird manufactures their T cells compared to others? What is the composition of their final product in terms of T cell markers? At last year's ASH they had presented their finding that adding a PI3K inhibitor during manufacturing led to better efficacy preclinically.

How durable have the responses been? Bluebird started treating patients in February, and the data cutoff for the abstract was November 18th, so I am not expecting a lot of data to answer this question.

Have there been any relapses, and have these been associated with antigen loss or loss of CAR-T persistence (perhaps at the lower dose levels)?

*Update 12/1
Bluebird presented their data in a conference call this morning. Slides of their presentation are available here.

Of the above questions, Bluebird mainly presented data on the initial kinetics of CAR-T cell expansion and tumor marker decrease, as well as cytokine levels. They did not, however, want to speculate on differences between their trial and other BCMA CAR-T trials, and did not appear to provide any additional information on the initial tumor burden levels in these patients. They also did not characterize the T cell product for CD4:CD8 ratios or other T cell markers either prior to infusion or during treatment. They did not want to speculate on manufacturing differences between their product and others, but mentioned they plan to bring along their next generation BCMA-CAR program, which will use the PI3K inhibitor during manufacturing over the next year.

Below are some of the key additional data they presented. While they did not want to compare data too closely to other BCMA CAR-T trials, I will make some very preliminary, premature, any other diminishing adjective you want, comparisons just to start getting a sense of some of the early similarities and differences. Of course, this is all subject to change with further follow-up and additional patients, so CAUTION is warranted when trying to draw any conclusions.

Response timing and durability:

Above, you can see the time to first response and time to best response in their treated patients, as well as time to disease progression. In the first three patients treated at the lowest dose level, only one achieved a PR, and then subsequently relapsed after >12 weeks, but they did not disclose any data on the nature of this relapse (although persistence of CAR-T cells was shorter for this dose level, as shown below). Another thing to note is the conversion of initial responses into better responses over time in the cohort treated at the second dose level. Thus, it is still possible that the PRs with shorter follow-up in the highest dose cohort will continue to improve.

We can see some of the kinetics of response over time below:

The serum M-protein levels appear to continue to be trending down in the highest treated dose-cohort, so again, it is possible that the responses will continue to improve. The rapidity of the response appears to be roughly in line with what was seen previously in the NCI trial, but it's too few responses to really be able to have a good sense of this yet.

Bluebird also presented some initial CAR-T expansion and persistence data:

The CAR-T cells expanded for all dose cohorts, but appeared to expand to a greater extent for the two higher dose levels. The peak levels of the CAR-T cells - 10^5-10^6 vector copies per ug genomic DNA, appears to be similar to the peak expansion of 1.74 x 10^5 and 2.20 x 10^5 vc/ug genomic DNA in the two patients who had significant responses in the UPenn trial (table above). So it does not appear that the, as of yet, unobserved grade 3 or higher CRS is solely due to poor peak expansion of Bluebird's CAR-T cells.

The persistence also looks potentially improved versus the CD28 CAR in the NCI trial, which had minimal levels of CAR detectable by 3 months in any of their patients. While it is still very early in follow-up and patient numbers, Bluebird's 4-1BB CAR-T cells seem to maintain detectable levels in the majority of patients treated with the two highest doses through 8 weeks. Preliminary data from UPenn's 4-1BB BCMA CAR trial, as mentioned above, also had increased persistence, with CAR detectable for up to 7+ months in patients.

Decreases in serum BCMA levels correlated with response. For one patient with the longest follow-up in this figure (through 24 weeks), the CAR-T persistence appeared to wane by 8 weeks, and serum BCMA levels have begun to rise from 16-24 weeks, with re-emergence of CAR detection by 24 weeks. Further follow-up will be necessary to determine if this preludes a relapse.

Bluebird also presented a more detailed list of treatment emergent adverse events:

The grade 3 & higher toxicities were attributable to the lymphodepleting regimen. Again the headline has been the lack of observed grade 3 or higher CRS or neurotoxicity to date, even in patients who have had clinical responses.

They also gave additional details about the peak cytokine levels in their patients:

They show that their peak levels of cytokines in blood were all orders of magnitude below the very high levels observed in the NCI trial, in the two patients who had responses but also CRS (shown below). This is somewhat surprising that even with apparently significant CAR-T expansion they did not see very high levels of cytokines produced.

From Ali et al., 2016

Lastly, they said they plan on having a flexible trial design, potentially with different approaches for patients with higher disease burden versus low (using 50% bone marrow MM cells as a threshold). They also stated that patients enrolled so far were required to have BCMA detectable on >50% of their MM cells, but were considering loosening this criterion due to the impressive responses seen to date, hoping that their CAR will be able to detect even very low levels of antigen present on cells.

Summary

The data released by Bluebird on their still very early BCMA CAR-T program are certainly quite encouraging. They appear to compare favorably in terms of response rate to other early anti-BCMA CAR-T trials, although we are comparing small numbers here, and dose escalation in Bluebird's trial (and others) is still ongoing. More surprisingly, and impressively, unlike in previous trials, the activity of Bluebird's CAR-T cells has not been accompanied by grade 3 or higher CRS or neurotoxicity. One question that Bluebird has still not given further clarity on is what the disease burden was in these patients and we will have to see how disease burden will correlate with responses and toxicity going forward. The CAR-T cells did appear to expand significantly, but this has surprisingly not come with severe increases in cytokine levels and subsequent toxicity. These are definitely exciting results, but seeing how durable these responses are, as well as if these preliminary levels of activity and safety hold up with greater patient numbers will be of critical importance. As with CD19, loss of the target antigen, BCMA, will most likely be a difficulty going forward. If the safety and efficacy are, in fact, differentiated from other BCMA CAR-T programs, understanding why will be essential as well.



Disclosure: I own shares of Bluebird Bio

Juno Halts JCAR015 Trial Again

Juno announced its lead CD19 targeted CAR-T product, JCAR015, in its phase II ROCKET study in adult ALL, was placed on clinical hold (again) after two patients experienced severe neurotoxicity, leading to death. The fate of this program is currently undecided, whether or not it will be modified or shuttered completely. This is after an earlier clinical hold in the same trial after similarly fatal neurotoxicity events. They resolved that hold very quickly by removing fludarabine from the conditioning regimen, which they thought would prevent, or at least reduce, severe neurotoxicity. Unfortunately, this did not appear to be sufficient to prevent this toxicity. Robert wrote his thoughts on the initial hold, and a lot of those concepts are again important after this hold. Juno held a conference call to discuss the hold, which is available here. The following is a quick summary of what they said, and my thoughts on this toxicity and things to consider with CAR-T development going forward.

While Juno still believes that fludarabine was a contributing factor to neurotoxicity, they stated that they always considered it to have multifactorial causes, but hoped removal of fludarabine was a significant contributing factor. They feel that ALL (compared to NHL and CLL) is particularly prone to CAR-T toxicity, both CRS and neurotoxicity, due to the high levels of accessible antigen, which allows for rapid expansion of CAR-T cells, and may lead to toxicity. Other factors are potentially the costimulatory domain used (CD28 vs 4-1BB), dose of cells, and the conditioning regimen, which all can impact the expansion of CAR-T cells. The rapid cell expansion in patients and an early high fever (2-3 days after CAR-T administration) appear to be fairly closely correlated with later development of neurotoxicity, which typically appears approximately 1 week after CAR-T administration and progresses rapidly. Going forward, Juno highlighted the lower levels of toxicity in both ALL and other B-cell malignancies using their 4-1BB CAR-T products with defined CD4:CD8 composition (JCAR017, and non-commercial product JCAR014). The CD28 costimulatory domain used in JCAR015 may lead to more rapid expansion of T cells, and thus potentially more toxicity, although I would not say this has been definitively proven yet.

In terms of how to deal with this toxicity, Juno mentioned that both patients were given an intervention, as per protocol, either with anti-IL-6 antibodies and/or corticosteroids, at the time of their high fevers. Dealing with the subsequent neurotoxicity they speculated would be difficult because it progresses so rapidly that makes it tough to use an intervention, such as a safety switch, to intervene in time while maintaining sufficient T cell activity. They also said this would make it difficult to treat prophylactically with steroids or anti-IL-6, due to concerns it would dampen CAR-T engraftment and activity across all patients.

The question arose whether this toxicity may be due to either on-target off-tumor activity against low levels of CD19 somewhere in the CNS, or on-target on-tumor toxicity from tumor burden in the CNS. Juno did not believe it was due to targeting CD19, either tumor burden in the CNS or off-tumor activity, as they have successfully been able to treat a patient with active CNS lymphoma with JCAR017 without neurotoxicity, although importantly, this is a different CAR product. However, there is some evidence that other therapies that activate T cells against CD19 also cause neurotoxicity. In particular, the CD19-CD3 BiTE, blinatumomab, which redirects T cells to CD19+ cells, has also shown signs of neurotoxicity. In particular, the CNS toxicity has been partially attributed to activating T cells against CD19 positive cells in the CNS causing cytokine release and disruption of the blood-brain-barrier, although I have not seen the primary data indicating this is the definitive mechanism.

From Ribera et al., 2015

Some evidence that neurotoxicity is not due to CAR activity against CNS disease comes from studies of CARs against other targets that can also induce neurotoxicity as an adverse event. In particular, two early studies of BCMA-targeted CARs in multiple myeloma have found that both a CD28 CAR as well as a 4-1BB CAR could cause neurotoxicity after large expansions of CAR-T cells. Strikingly, the 4-1BB study did not use any pre-conditioning chemotherapy, demonstrating that at least for this CAR, you can have significant CAR-T cell expansion and toxicity in the absence of conditioning. Surprisingly, a new 4-1BB anti-BCMA CAR from Bluebird Bio has recently shown responses without grade 3 or higher CRS or neurotoxicity in early initial clinical data, which I write about more in a post here. While very promising, it is too early to say if this construct will have a better safety profile than others, and what are the reasons for lower toxicity. Multiple myeloma only very rarely enters the CNS, so it is unlikely that the neurotoxicity is due to BCMA+ myeloma cells in the CNS. However, soluble BCMA can exist in the CSF. Soluble antigen, in general, is not thought to significantly interfere with, or activate, CAR-T signaling, such as in a mesothelin CAR:

From Lanitis et al., 2012

This has been found to be true for other CAR-T targets as well that can exist as soluble antigen, with the thinking being that a multimer of the antigen is required for stimulation of the CAR.

My thoughts on future CAR-T development:
Other groups (such as Kite/NCI) using CD28 CARs appear to have not had as significant troubles with neurotoxicity, even though they also use cyclophosphamide and fludarabine for conditioning. As Robert had previously discussed, an additional factor to consider, although we have no sense of whether it is playing a role, is the manufacturing process of the CAR-T cells leading to additional differences between the different products.

Expansion of CAR-T cells appears to be a double edged sword: while Juno's trials may suggest that rapid expansion of CAR-T cells can potentially lead to dangerous toxicity, a lack of robust expansion if not enough antigen is available immediately is correlated with worse activity. This has been shown preclinically with a mesothelin CAR where local administration of CAR-T cells near antigen-positive cells improved expansion and activity. Additionally, a recent clinical study by Kochenderfer using CD19 CAR-T cells from donor post-allogeneic transplant showed that low B cells (containing target antigen) correlated with poor expansion and activity of the CAR-T cells. Importantly, in another CD19 CAR trial in ALL by Juno and its collaborators, the addition of fludarabine to the conditioning regimen allowed significantly better expansion of CAR-T cells and durability of responses.

From Turtle et al., 2016

Going forward, synthetic biology tools or CAR constructs that would allow for a consistent, but less rapid, increase in CAR-T expansion may be useful to get sufficient activity and persistence but reduce excessive toxicity.

Another area that may be worth exploring is to try to identify ways to better separate the activity of CAR-T cells from the causes of cytokine release syndrome and neurotoxicity. For instance, IL-6 may be dispensable for CAR-T activity, but contributing to CRS, so anti-IL-6 antibodies might be able to be used more aggressively. However, the spike in IL-6 during CAR-T expansion may not be coming from the CAR-T cells, so knockout of IL-6 in CAR-T cells may not make for a safer CAR-T product. Continuing to look for ways to separate activity from toxicity in CAR-T cells to either engineer safer T cell therapies or allow better prophylactic toxicity management will probably be important in the continued development of these therapies.

In summary, I think it is still not extremely clear why Juno has, particularly recently, been running into such severe neurotoxicity in their JCAR015 trial. I thought removing fludarabine would lead to poorer activity and durability of responses, but would, at least substantially, reduce the levels of neurotoxicity, as fludarabine appeared to allow for significantly increased CAR-T expansion. However, there are clearly other factors that are important for modulating rapid CAR-T expansion, and whether or not rapid CAR-T expansion is even the primary factor leading to neurotoxicity is still unknown. The potentially better safety profile of the JCAR017 program as well as CD19 programs developed by others suggests that neurotoxicity may be particularly problematic for JCAR015, but I would say it's too early to draw strong conclusions. Even these other programs have severe CRS and neurotoxicity, so there is still a lot of work to be done in continuing to improve the safety of CAR-T therapies while maintaining their efficacy.

CAR-T safety: Traffic laws needed

by Robert Kruse

The Juno CAR-T cell clinical trial halt brought shockwaves throughout the immuno-oncology world. While the previous toxicities for CD19 CAR-T's were well known, the death of several patients from them during later phase trials that were going to enabling for FDA approval was alarming. Is this therapy really ready for clinical approval and launch if the toxicities are not currently manageable?

The addition of fludarabine to the conditioning regiment was quickly noted by Juno to be the cause of the toxicity issue, since the patient deaths only occurred when fludarabine was added to cyclophosphamide for the conditioning regimen. Perhaps even more surprisingly, the FDA quickly agreed with them, and the trial halt only lasted 48 hours. The scare to the CAR-T sector was over, and the stocks could rise again.

The question for scientists, clinicians, and investors is if the correct culprit was really identified, or if fludarabine was a scapegoat for larger issues for the CD19 CAR-T strategy. Such issues will affect adoption and patient outcomes, even if the FDA approves the therapy. We wanted to delve into the issue more at the blog, in order to consider the opposing sides on this still controversial finding.

Why is lymphodepletion a part of CAR-T regimens?
As a starting point, those unfamiliar to the CAR-T space might ask why lymphodepletion (using chemotherapy to kill the patient's lymphocytes) is even needed. Why can't one just infuse CAR-T cells into the patient, after which they will home to the tumor and destroy it. The reason is that CAR-T cells, so far, have not been noted to remarkably persist under such conditions. Instead, helping them by depleting the host of its own T cells allows the infused CAR-T cells to expand and fill this niche. Growing into larger numbers and feeding on host cytokines like IL-7 facilitates their long-term engraftment. Furthermore, the expanded numbers help facilitate the clearance of the tumor, and the cells don't solely have to depend on target antigen driven expansion, but also utilizing homeostatic expansion mechanisms.

Drugs to lymphodeplete patients are many, having regular use in bone marrow transplants and now being adopted into CAR-T preparation regimens. Previously, cytokine release syndrome was a regular occurrence with CD19 CAR-T trial, and was in fact associated with good outcomes and complete tumor response. The lymphodepletion helped foment the cytokine release syndrome and clinical response. Cytokine release syndrome has been known to cause some neurotoxicity, thought to be in part from the systemic inflammation. Indeed, in clinical trials before the most recent death, the neurotoxicity could resolve with the administration of anti-IL-6 antibodies, suggesting the link between the inflammation and the neurological findings. It has been found that IL-6 could be transported across the blood brain barrier, forming a connection to the observed clinical findings. So before this recent adverse event then, the paradigm had been established for lymphodepletion and then toxicity management with anti-IL-6 antibodies.

Is cyclophosphamide + fludarabine combination truly the issue?
The FDA seems to be satisfied that the added fludarabine was the culprit, with the trial now already resumed. However, it is interesting that other academic investigators and companies have used the "cy/flu" combo without any CNS death attributable to the CAR-T cells. For example, Juno's competitor, Kite Pharma, is currently using it in their trials right now. What is the difference then? While I can only speculate, the differences might have to do with the potency of the CAR-T cells themselves. Differences in signaling domains between CD28 and 41BB can alter the persistence and cytokine release of the CAR-T cells, however, both Kite and Juno's CAR used here use CD28. Novartis employs the 41BB endodomain by contrast. Less reported and focused on are differences between the preparation of the cells, the cytokines they are cultured in, how long they were cultured in vitro before infusion, and donor to donor differences. All of these parameters could influence the peak CAR-T potency post infusion. Other trials using cy/flu might lack toxicity simply because the CAR-T cells are not as potent leading to a lack of profound cytokine release initially.

Is cyclophosphamide & fludarabine integral to the therapy?
Another question is if the function of cy/flu is needed for therapeutic efficacy. As discussed above, conditioning is important for establishing a niche for CAR-T cells to expand in, replacing the normal T cells. However, more than that, the conditioning might be needed to suppress the host immune system against immune responses against the CAR molecule. Most CARs in trials represent early generations that use mouse-derived single chain variable fragments to bind to CD19 antigen. Humans can develop human anti-mouse antibodies (HAMA) that would lead to elimination of CAR-T cells. This would, of course, would also prevent the elimination of tumor cells and the desired therapeutic effect. Previous studies offer clues that fludarabine might be needed for treatment efficacy. I will list the quotes in full from the paper below:

"Although a high rate of relapse was observed among adults treated at FHCRC, this may reflect inadequate lymphodepletion in patients receiving Cy monotherapy, with rejection of the murine component of the scFv; early data suggest Flu may help to overcome this limitation."

"CAR T-cell persistence was short in most of the 12 patients treated with Cy-based lymphodepleting chemotherapy, and similar to FHCRC’s experience in adults with ALL, a cytotoxic T lymphocyte–mediated response to the murine component of the CAR transgene was observed."

Given that some data indicates conditioning with both agents might be required in the current form of CAR-T cells for optimal therapeutic effects, the reduced conditioning could diminish the high level of complete remissions that clinicians, payers, and investors are looking for. This potential concern will be worth monitoring as Juno generates data with the Cy only regimen. The JCI paper reporting Juno trial data reiterates these findings that responses were much better in the cy/flu group, likely due to a lack of anti-CAR immune response, similar to the Blood paper above. In the long run, though, the immune response issue will likely be solved as almost all new generation CARs employ human scFv's. In the short run, this might prove problematic for initial product launches.

Another aspect on this point is whether cy/flu and conditioning is required for CAR-T efficacy, or if there will be future potential to do without it. Notably, the NCI had a study with allogeneic CAR-T cells post transplant where there was still anti-tumor activity with no lymphodepletion. However, since the patient was already bone marrow transplanted, forms of ablation had already occurred before the adjuvant CAR-T cell infusion period making this an uncertain comparison. We do know that early CD19 CAR-T cell trials failed without the lymphodepletion, as mentioned above. 

Other explanations for the adverse events
The other possibility that has received some attention is that the CD19 antigen might exist in the central nervous system in some form, which then leads to an on-target response in the brain that causes intense inflammation and in this case, too much brain swelling and and death.

There are two different possibilities for this. One is that the brain tissue expresses low levels of CD19 protein that have evaded detection on histological stains. Given that CAR-T cells are exquisitely sensitive to very low amounts of antigen, this possibility can not necessarily be ruled out. However, it would be unexpected given that our current understanding is that CD19 is a B cell restricted antigen. Similarly toxicities with blinatumomab might suggest that it is restricted to CD19, but it could still be driven by cytokine release syndrome.

The second is that the CD19 tumor has infiltrated the CNS, so it is an on-target on-tumor effect as intended, but that the location of the vigorous immune response leads to these adverse events. It should be possible to look for tumor by imaging, but this could also be missed on MRI or CT imaging. Perhaps more vigorous screening is needed before CAR-T infusion. On the flip side, early reports have shown that even with CNS leukemia involvement, that it wasn't necessarily predictive of toxicity development. Clearly more research is needed in order to robustly predict what will happen when patients are treated with CD19 CAR-T cells. In that respect, the autopsies of these brave patients will be crucial in educating the entire CAR-T field on the safety of these therapies, and their sacrifice should hopefully afford many more patients in the future to be successfully treated of cancer.

Importance of a safety switch
The other question from this result is the necessity of a safety switch going forward. The first question to ask is if the Juno CD19 CAR-T cells had the inducible caspase-9 safety switch from Bellicum in them, would this neurotoxicity death have been prevented. Importantly, the caspase-9 mechanism triggers an apoptotic event within the T cells that does not produce any further inflammation. This compares to the other suicide methods that leverage therapeutic antibodies to target CAR-T cells for their elimination. These approaches could cause more inflammation, and even if they didn't, it's uncertain that antibodies could fully penetrate CNS and cross the blood brain barrier to eliminate CAR-T cell that have trafficked there.

Neurotoxicity can occur concurrently or after cytokine release syndrome, and is generally less responsive to corticosteroids or anti-IL-6 antibodies as reported here. Development of severe neurotoxicity did correlate with early cytokine levels in patients, but it may be difficult to determine when to activate the safety switch to allow sufficient activity, but prevent the delayed neurotoxicity. Additionally, if CAR-T cells in the CNS are playing a role in the neurotoxicity, the dimerizer drug would also have to reach the cells there. This has not necessarily been tested in any clinical trials that I have seen for Bellicum, which focus on T cells in their periphery. In all, while Bellicum's safety switch may turn out to be the most effective way to reduce CAR-T levels safely, there may be difficulty in determining when and in which patients to titrate their CAR-T levels to prevent neurotoxicity. However, this may be able to be determined effectively with experience in the clinic. It may, in fact, be easier for the safety switch to deal with acute cytokine release syndrome, which seems more immediately responsive to interventions.

Implications for other approaches
Currently, many efforts are being made to make CAR-T cells even more powerful, in order to solve the problem of lack of efficacy in solid tumors. One push has been to use CRISPR/Cas9 gene editing to make the T cells resistant to immune checkpoints and suppressive signaling, which is commonly co-opted by tumors as a defensive mechanism. Indeed, the clinical trials to edit PD-1 in CAR-T cells for therapy will begin shortly. In light of the recent CD19 CAR data, however, the question arises if this is indeed safe, and if the FDA should reconsider. If we are getting unanticipated side effects for a well described antigen target, what about for less validated tumor antigen targets, which may occur in healthy tissue and now suddenly have permanent mutations rendering the hosts ability to combat them limited. Perhaps regulators need to take a closer look at these consequences before approving them. The same will be true for TIL and TCR trials where the antigens are more defined, but the off target toxicity could be unknown. Notably, cytokine release syndrome has been observed in a trial with virus specific T cells against EBV, so this is not just a phenomenon with CAR-T cells, and we must consider the safety for all cell therapy approaches.

Conclusion
As exciting as the triumphs of CD19 CAR-T cells have been, the recent adverse events show that the therapy still has an inherent identity crisis, trying to become a scalable standardized therapeutic moving beyond its days as a boutique therapy, where every patient was almost its own self-contained experiment. CAR-T cells in their current form represent a series of several procedures, therapies, and extended patient management, with only one step of that process being cell infusion. CAR-T cell efficacy is still dependent on these other factors, such as donor influences, cell collection, lymphodepletion, and post-infusion toxicity management. Unless the problems of toxicity can truly be standardized and contained within rigorous protocols, the therapy may never take off among clinicians. However, there will be a strong desire to achieve this, due to the impressive efficacy of these therapies. Of course, more robust safeguards like safety switches might help for managing toxicities. Furthermore, it will be crucial going forward to know if this is just a CD19 antigen issue that is less likely to occur in other CAR-T therapies. However, until these other targeted therapies start having success, we won't know about their respective efficacy/safety profiles. I'm still bullish and hopeful about this field, but events like this should be considered more going forward. Every patient and antigen is truly novel, and the tumor xenograft mouse models often utilized aren't informative for these toxic events. On the hopeful side, though, we've come a long way from the early days of CAR-T therapy where no efficacy was seen, so if the problem now is having efficacy but needing to manage toxicity, progress has been made offering hope for the future that potency and safety will be maximized together.

Evolving Therapeutics Part I: Immunotherapy

I agree with fellow science guy, Bill Nye, when he said "Evolution is the fundamental idea in all of life science - in all of biology". One can view anything in biology through the lens of evolution. Thinking about how biology is shaped by the forces of evolution is a lot like thinking about how human behavior in economics can be shaped by the forces of incentives.

With that in mind, I wanted to take a look at a number of topics, relevant to therapeutics and disease, wearing evolution-colored glasses. I think this approach gives an interesting perspective on both the challenges evolutionary forces create in treating certain diseases, like cancer, as well as the ways in which evolution can be harnessed for our own benefit through things like selection screens or directed evolution.

Considering evolutionary forces can be critical for determining the best therapeutic approaches to take in a given disease, particularly cancer, which is essentially a disease of evolution and selection. So for my first post in this series, I would like to focus on the dynamic relationship between the immune system and cancer, two systems with great diversity and proliferative capacity - the fuel for an evolutionary arms race. In particular, I will be highlighting three recent papers that add to the growing amount of research using human immunotherapy clinical trial data to further our understanding of this dynamic battle between cancer and the immune system.


A Tumor-Immune System Arms Race - T Cells Try to Keep Up with Mutating Tumor Targets

One of the most appealing aspects of using the immune system to target cancer is both its ability to evolve along with the tumor, as well as its ability to potentially recognize multiple targets simultaneously in a tumor. These factors may be key for why some patients have been able to achieve very durable responses to immunotherapies.

In particular, a recent paper by Verdegaal et al. has formally demonstrated this evolving arms race between cancer and the immune system, and shows the ability of the immune system to adapt in response to tumor evolution (and vice versa).

The authors studied two melanoma patients treated with an adoptive T cell therapy. The patients had tumors removed and used to generate T cells reactive to that tumor, and also had those tumors sequenced to identify the potential mutations the T cells were reacting against. The T cells were not generated from TILs, as has been done by Steve Rosenberg's group, instead peripheral blood cells from the patient were stimulated repeatedly for reactivity against their own tumor-derived cell lines. The paper follows the clinical course of two patients, BO, who has an ongoing complete response after receiving the adoptive cell therapy, and patient AB, who unfortunately died of the disease.

For brevity, I will just go through the example of patient AB, as this case shows both the tumor evolving to evade the T cell response, as well as the T cell response evolving to react to additional neoantigens (novel peptides presented to the immune system due to tumor-specific mutations) over time.

Patient AB had a metastasis resected in 2004 (MEL04.01) that was used to produce the T cell product, which reacted against 3 neoantigens present in that tumor. The patient was treated with the T cell product, but progressed, and had further resections of metastases in 2008 and 2012. In the relapsed metastases, it was found that for one of the targeted neoantigens there was a loss of the mutation, essentially removing the target those T cells were reacting against. Another mutated gene producing one of the neoantigens was significantly reduced in its expression in the relapse, also potentially reducing the ability of T cells to react against that target.

While the immune system appeared to have applied a selective pressure that led to the loss of specific targets, surprisingly, the authors found that T cells found in the tumors of the later metastases were now reacting against additional neoantigen targets. Specifically, they found that T cells were reacting against a neoantigen that, while the mutation was present in the initial tumor in 2004, was expressed at much higher levels upon relapse. This correlated with the presence of T cells reactive against that mutation in that metastasis, whereas no T cells reacting against this target were found in earlier tumors. This shows that the immune system is also dynamic, and has the ability to change its reactivity toward a tumor depending on the presence of targets it can recognize. While this may not be surprising, this is the first time that immunoediting, the loss of specific antigens in response to pressure from the immune system, has been demonstrated in human tumors, and further establishes the neoantigen-reactive T cell as a mediator of anti-tumor activity.

Considering that the vast majority of mutations found in a tumor are probably not contributing significantly to its growth, and are so-called passenger mutations, it is not surprising that the tumor selects for loss of these mutations. The ability of TIL therapy to generate potential cures in a percentage of melanoma patients is most likely then due to the simultaneous targeting of multiple neoantigens by diverse T cell clones. In fact, the Rosenberg lab recently charaterized a TIL product, and found the T cells were able to react against 10 distinct neoantigens.

In all, this work shows that the relationship between the tumor and the immune system is not static, with both the tumor and the immune system able to change in response to each other. This has implications for targeting cancers with immunotherapies. Specifically, that you will need to generate a response against a multitude of neoantigens to get the potential durable responses desired, whether that is through the use of broad immunomodulation with checkpoint inhibitors, or designing adoptive cell or vaccine products with reactivities against neoantigens.


Are All Neoantigens Created Equal? - Clonality Matters

Next, a recent paper from Charles Swanton's group asked the question if all of these potential neoantigen targets created equal. This paper was also discussed nicely on this blog specifically focused on the concept of cancer and evolution. There is significant evidence that the neoantigen-reactive T cell is a/the critical mediator of checkpoint antibody efficacy, and the number of predicted neoantigens correlates with efficacy. However, the authors of this paper ask the question whether its just the total number of neoantigens that matters, or if also the clonality of those neoantigens matter. This is something I have wondered about as well. Specifically, does it matter if the neoantigens are present in almost every cell in the tumor (clonal), or only present in a subfraction of the tumor cells (subclonal). This is illustrated in a theoretical tumor below: shown is a clonal "trunk" mutation A, and subclonal "branch" mutations B-F.

From Translational Implications of Tumor Heterogeneity

The authors measured the clonality of the total of a tumor's predicted neoantigens and looked to see if that influenced survival and response to checkpoint blockade. Starting in lung cancer, they found that patients with lung adenocarcinomas with high neoantigen load had longer overall survival, not taking into account treatment, and that this stratification improved further when taking into account the clonality of the the neoantigens. Specifically, the subset of patients with high neoantigen load and low neoantigen heterogeneity had the best survival. Interestingly, they found that lung squamous cell carcinomas did not have this same association, and the authors present some evidence that these tumors might have reduced processing/presentation of antigens compared to adenocarcinomas.

The authors were able to identify neoantigen-reactive T cells in patients with tumors with both high and low intratumoral heterogeneity, but they could only find neoantigen-reactive T cells to clonal neoantigens and not subclonal neoantigens. This suggests that subclonal neoantigens may not induce a robust T cell response. Unsurprisingly, neoantigen-reactive T cells tended to express more inhibitory markers, such as PD-1 and LAG-3, which has also been seen by the Rosenberg group.

Next, they went on to see if the intratumoral heterogeneity of neoantigens, in addition to the total number of neoantigens, could effect the response to checkpoint blockade. Using sequencing data from recent PD-1 studies in melanoma, they stratified patients who had a durable clinical response to PD-1 blockade versus those who did not. They found that for a given threshold of high neoantigen burden and low intratumoral heterogeneity, 12 out of 13 patients had durable clinical benefit, compared to only 2 of 18 patients who had either high intratumoral heterogeneity or low neoantigen burden (shown below).

Further, they found that applying a threshold for the amount of intratumoral heterogeneity of neoantigens helped to better predict the outcomes of patients treated with PD-1 blockade in NSCLC (Rizvi cohort) or CTLA-4 blockade in Melanoma (Snyder cohort) than neoantigen burden alone.

These data raise the question - does inducing non-clonal neoantigens actually weaken the response a patient has to checkpoint blockade? This could have implications for treating patients who are heavily pretreated with radiation/chemotherapy which may have induced non-clonal neoantigens. The authors specifically looked at data from heavily pretreated melanoma patients, and found that treatment led to high levels of non-clonal neoantigens. In this population, the predictive power of neoantigens alone to predict response is lost, but if you combine this with intratumoral heterogeneity of neoantigens, this now almost able to reach statistical significance to predict response. The authors stated there were too few patients to sufficiently power this analysis, so this is still speculation.

A potential drawback of this study is that it is correlational, and low intratumoral heterogeneity tends to correlate with things already known to be positive predictors of response to therapy, such as higher PD-L1 expression. Also, surprisingly to me, high neoantigen burden, another positive predictor, correlates with those neoantigens being more clonal: 

I would have expected a higher neoantigen burden would be indicative of greater intratumoral heterogeneity, not less. This also makes me curious what this analysis would look like for for mismatch-repair deficient cancers, since their tumors might have higher ongoing genomic instability, instead of just accumulated mutations caused by previous carcinogens, like UV or smoking. Tumors with this deficiency have been found to be sensitive to checkpoint blockade, presumably due to their high number of mutations increasing the "tumor foreignness" for the immune system to recognize. Specifically, it would be interesting to see if mismatch-repair deficient tumors have high intratumoral heterogeneity, and if patients can have positive outcomes with checkpoint blockade despite their high levels of heterogeneity.

While mouse models have drawbacks, especially in the study of tumor immunology, it will be interesting to see if any groups try to experimentally test whether increasing or decreasing the heterogeneity of tumor antigens effects the immune response to them.

In addition, this work suggests that the strategy of mutagenizing a tumor to increase the number of neoantigens may not lead to a more productive immune response, and could potentially hamper it. The etiology of these mutations would be different from the mutations initially present in a tumor, which accumulated under the stringent selective pressures of initial tumor development. A way to recreate this accumulation of mutations but with higher clonality might be to first mutagenize the tumor (with chemo/radiation or disruption of DNA repair) and then try to create a new evolutionary bottleneck with a targeted therapy with a high response rate that could apply a selective pressure to increase the clonality of the tumor. You might need a really effective second therapy to reduce the heterogeneity sufficiently, though. It would be interesting to see how different therapies effect the clonality of a tumor, and Swanton's group is performing a longitudinal study, called TRACERx, looking at how clonality changes during the course of treatment, which may shed light on therapies that increase and decrease tumor heterogeneity.

While like other studies retrospectively looking for biomarkers of response, this study is correlational and further validation is needed, intratumoral heterogeneity is now on the map, and should be considered as a potentially predictive biomarker of response to checkpoint blockade, which are nicely summarized in this review. Better biomarkers may allow doctors to better stratify patients, and determine which are most likely to benefit from these therapies. This study also emphasizes the importance of targeting the clonal "trunk" neoantigens for cell or vaccine based therapies directly targeting neoantigens. It also suggests the potential importance of developing strategies to reduce intratumoral heterogeneity or targeting the causes of heterogeneity.


Tumors Go Silent - Loss of Antigen Presentation and Interferon Responsiveness

While the ability of T cell populations to recognize multiple neoantigens in a tumor may reduce the risk of relapse due to loss of single neoantigens, there is the potential for escape mutations that globally make the tumor less sensitive to T cell recognition or attack. These mutations could potentially be the most pernicious, as they could essentially make the tumor unresponsive to any therapy dependent on T cell recognition of the tumor. Unfortunately, evidence of these types of mutations has been observed in a recent paper looking for the causes of relapse after an initial response to checkpoint blockade.

The authors examined the cases of 4 patients with melanoma who had this clinical course of a relapse after a relatively prolonged period of response to therapy. By sequencing sequential biopsies of tumors pre-therapy and post-relapse, the authors were able to look for new mutations present at relapse that could explain their loss of responsiveness. The vast majority of mutations were conserved between the tumors at both time points, but they found homozygous loss-of-function mutations in JAK1 in one patient and JAK2 in a second patient.

Looking at the biopsies of from these tumors at relapse, they found that both of these patients lost PD-L1 expression at the interface of the tumor and stromal cells, despite the continued presence of CD8+ T cells at this interface, and presence of PD-L1 expression in adjacent stromal cells (shown below).

The authors wondered if these JAK1/2 mutations could be causing a loss of responsiveness in the tumor to interferon-gamma, which normally induces PD-L1 expression. They found that T cells could still recognize tumor cells, however, tumor cells from both patients lost responsiveness to the growth-inhibitory effects of IFN-gamma, and to all interferons for the JAK1 mutation (below). They subsequently showed these mutations were sufficient to cause this loss of responsiveness.

They went on to look for relapse-specific mutations in a 3rd patient, and found B2M mutations, which have previously been shown to cause loss of MHC class I surface expression, and inability of tumor cells to be recognized by T cells. Immunohistochemistry from this patient showed this loss of surface localization for MHC class I.

IHC stain for melanoma marker (left) and MHC class I expression (right)

Interestingly, B2M mutations have been observed in about 30% of colorectal cancers with microsatellite instability (mismatch-repair deficient). Surprisingly, these mutations were not a negative prognostic marker, and if anything, trended toward being a positive prognostic biomarker. This raises the question if there are negative consequences for the tumor for mutating B2M. Another question this raises is in mismatch-repair deficient CRC, which as a class are quite responsive to checkpoint blockade, if patients that already have baseline B2M mutations are more resistant to checkpoint blockade. This could be used as a predictive biomarker if found to be the case.

The results from this paper also may have implications for combining PD-1 + JAK inhibitors such as in this trial. The results in this paper would suggest sustained inhibition of JAK signaling could reduce tumor responsiveness to T cells.

In summary, this paper demonstrates how the tumor can develop adaptive resistance to checkpoint blockade by either reducing the ability of T cells to react to the tumor or the tumor to react to T cells. Considering the ability of cancer to evolve resistance to basically any therapy we throw at it, it is almost surprising that these mutations are not essentially inevitable, and that it is possible to get these extremely durable responses with current therapies.This paper, however, highlights the need for future strategies to deal with these types of mutations. For B2M mutations perhaps by engaging other elements of the immune system, such as NK cells, that might be able to recognize this "missing self", and react to the loss of MHC surface expression. And for JAK1/2 mutations, perhaps using strategies that are not reliant on interferon for efficacy. Additionally, it is possible these B2M or JAK1/2 mutations produce compensatory vulnerabilities that can be exploited therapeutically. It will be worthwhile to further study the mechanisms of adaptive resistance to checkpoint blockade to develop the best strategies for dealing with it.


Conclusions - Designing the Next Generation of Immunotherapies

These three papers highlight the ability for both the immune system to adapt to the evolving tumor, as well as the ability for tumor evolution to produce problems for the immune system. It will be critical to understand, for each individual patient, what is the status of this tumor-immune system interaction, and how the tumor is avoiding immune surveillance. While we may not currently have therapeutics to intervene in patients for all the different mechanisms of immune evasion, further understanding the mechanisms critical for tumor survival will help identify new targets and rationally combine therapeutics.

This thinking is summarized nicely in a review called "The Cancer Immunogram", which attempts to think about treating cancer in terms of defining the different aspects of tumor-immune interactions critical for an immune response in a patient's tumor. The authors describe seven categories critical for effective recognition of cancer by the immune system, they are (shown below): tumor foreignness, general immune status, immune cell infiltration, absence of checkpoints, absence of soluble inhibitors, absence of inhibitory tumor metabolism, and tumor sensitivity to immune effectors.

In summary, here are some of the implications of the above three papers for the design and implementation of immunotherapeutic approaches in the future:

- It will probably be necessary to target multiple mutations simultaneously to reduce the likelihood of immunoediting of antigens and tumor escape.

- It is important to understand the intratumoral heterogeneity of a potential target neoantigen, and, if possible, target clonal mutations.

- Intratumoral heterogeneity may become a useful predictive biomarker for response to checkpoint therapies, and become part of how we define tumor foreignness in the cancer immunogram.

- Intratumoral heterogeneity is a problem, and it will be important to develop strategies that either reduce heterogeneity or target the mechanisms that lead to heterogeneity directly.

- Mutagenizing a tumor, by itself, may not generate productive tumor-foreignness and immune recognition. Perhaps if we can provide a new evolutionary bottleneck for the tumor, we can increase the clonality of these induced mutations.

- Unfortunately, there are mutations that make tumors less sensitive to T cell attack generally, and it will be important to develop strategies that target tumors that evolve this response.