Uniqure and the Celladon Legacy

Virus-based gene therapies are demonstrating strong proof-of-concept primarily for hematological diseases. To date, however, the data targeting solid organs has been weak. This was best exemplified by the recent results for Celladon’s Mydicar therapy providing AAV-mediated SERCA protein expression for heart failure (HF) patients. As readers might recall, prior posts outlined that Celladon’s preclinical data were exceptionally weak and failed to show increased target protein expression. Without this, there was little reason to expect success in their randomized CUPID-2 trial regardless of the results in the dose-finding CUPID-1 trial. CUPID-2 ended with a hazard ratio of 0.93 in a modified intent-to-treat population, confirming that the “therapy” was effectively a placebo. This outcome strongly encourages future investors in HF-targeted gene therapies to ensure that the preclinical data achieves one fundamental observation: verifiable change in expression of the targeted protein.

In addition to their hematological programs, European outfit Uniqure is also directing their AAV efforts to address heart failure. Based on their preclinical publications, Uniqure have seemingly come out ahead where Celladon clearly failed. Uniqure’s proof-of-concept studies in pigs receiving their AAV9-based construct to increase S100A1 protein expression seems both reliable and conclusive. At multiple points throughout this publication, the authors are able to show convincing evidence of S100A1 protein at levels above and beyond baseline. Why have Uniqure succeeded in providing convincing proof-of-concept where Celladon have failed? Although Uniqure is using an AAV2/9 construct whereas Celladon opted for AAV2/1, I do not think this makes a material impact. Preclinical data testing the tropism of these vectors have shown that both can target the heart efficiently in rodents, suggesting that the choice of capsid type is not the discriminator. Rather, Uniqure’s retroinfusion with accompanying left anterior descending (LAD) coronary occlusion is likely the key factor. Celladon’s preclinical and clinical data have conclusively shown that intracoronary infusion of AAV is not sufficient to transduce the myocardium, and Uniqure’s modification of working in a retrograde fashion is a novel and effective means of addressing delivery. Although not as simple and convenient as a daily pill, the infusion / temporary LAD occlusion method used by Uniqure is achievable and, given the state of the technology, necessary to achieve transduction of the myocardium and target protein expression.

 

Although Uniqure and their team are to be commended for their technical achievement and thorough preclinical work, I remain skeptical of the S100A1 program and, by extension, their collaboration with Bristol-Myers Squibb. Although this blog attempts to be as quantitative and data-driven as possible, the reservations surrounding these programs are, admittedly, qualitative and a “gut” feeling.

To add context to the gut feeling, we can set the stage for what these programs are attempting to achieve. Simply put, both programs from Uniqure and Celladon were attempting to restore what was once abundant. In the case of Celladon, they were attempting to capitalize on literature observations that the expression of SERCA2a declines in the context of the failing heart. Similarly, Uniqure is pointing to a decrease in S100A1 in the failing heart as a point of intervention. Simply put, this approach relies on an underlying belief that restoring something to the levels found in the “normal” heart will allow the failing heart to recover some or all of its lost function. In effect, the strategy treats heart failure as if it was analogous to the pathophysiological settings wherein enzyme replacement therapies succeed. Missing glucocerebrosidase because you have Gaucher’s disease? We can provide the enzyme to attempt to restore it to physiological levels. Is your failing heart low on S100A1? We have an AAV for you. Unfortunately, I think this strategy is very flawed when applied to heart failure.

 

It is very well established that heart failure impacts the heart in a myriad of ways. At the molecular level, we can observe changes in protein signaling and protein abundance. But at a more macro level, there are clear structural changes that occur. The figure from Wikipedia does a good job of showing representative structural changes and remodeling that are evident in various pathophysiological states. 

Simply put, this figure is the backbone for my skepticism. If we take the example of S100A1, the preclinical data suggest that ischemia-induced HF (which often takes a first step through hypertrophy) induces a state of the heart wherein S100A1 expression level is less than it was under normal conditions. Therefore, the presumption is that increasing S100A1 in the hypertrophic or failing / remodeled heart will allow the heart to return to normal. This, in my view, is a logical flaw. In effect, I believe this reflects a cognitive dissonance that treats the remodeled heart as simply a normal heart with reduced S100A1 (or reduced SERCA2a, or reduced protein X) expression. However, this simply isn’t true. The remodeled heart has altered structure including myocyte disarray, aberrant signaling, and changes in the expression level of a myriad number of proteins. In effect, the context has changed. As a simple analogy, consider the case of the smoker who develops emphysema. At the point of emphysema diagnosis, quitting smoking will not simply reverse the course of disease. The damage is done, alveoli and capillary beds have expired. In effect, cessation of smoking by an emphysematic patient does not have the same effect as cessation of smoking for someone who has not yet developed emphysema. The contexts have changed, and therefore the treatment and care strategies are not simply interchangeable. Similarly, my contention is attempting to “restore” a singular protein level to that found in the normal heart provides no guarantee of recovery for a remodeled heart. In support of this argument for S100A1 specifically, human heart tissue from nonfailing, failing and left ventricular assist device (LVAD) supported hearts show that although there is a decrease in inotropic response and S100A1 protein in failing hearts, LVAD supported hearts demonstrated restored inotropic response yet no restoration of S100A1. In that respect, these human tissue data suggest a peripheral rather than pivotal role for S100A1.

 

The counterpoint to this is, obviously, the hundreds of mouse and porcine studies showing that restoration of intracellular protein X to the failing heart restores function and rescues the model. But the level of success of each and every one of these studies is reason to be highly skeptical and, to a certain extent, dismissive of the “restoration of protein X rescues the heart” narrative. Additionally, consider the drugs used in the clinic for patients with / developing heart failure. By and large, they impact whole intracellular signaling cascades through modulation of surface receptors rather than focus on restoration of a single intracellular protein. Even in that context, the HR benefit is around 20%. Therefore, if manipulation of a whole signaling pathway in the heart yields a 20% benefit, what can reasonably be expected from focusing on an individual intracellular protein?

ITT Treasure Hunt

Celladon management engaged in a brief, ~23 minute conference call to review the company’s 4th quarter achievements. It appears only 2 or 3 investment analysts are interested in the story, as the questions posed were few. Admittedly, it can be hard to attract attention in a crowded biotechnology sector, and we wish them luck in drumming up more interest.

 

As mentioned previously here and here, Celladon is investigating the use of Mydicar, an AAV1 encoding SERCA2a, as a therapeutic for heart failure patients. The company is currently running the CUPID 2 trial, and it is due to provide top line data by the end of April / early May. Their aim is to demonstrate a 45% risk reduction (HR = 0.55) in time to recurrent heart failure rehospitalizations. As noted previously, I’m skeptical that this barrier can be met simply because the company has not provided convincing evidence of SERCA2a protein upregulation in large animal models receiving Mydicar.

During the conference call, the company articulated further their view towards the CUPID 2 data release. Disappointingly, they have steered investors away from expecting data to be presented on an intent-to-treat (ITT) basis. As a teaser, investors were told that the company approached the FDA with a plan to undertake a modified ITT (mITT) analysis wherein they could omit data from patients who did not receive placebo / Mydicar or who had events prior to dosing. As previously discussed, ITT analyses are the gold standard, and companies that attempt to mine trial results for subgroups are often putting on a brave face. In that regard, the red flag raised by this pre-emptive mITT declaration is deep crimson in color.

To be fair, it was noted that if a significant difference exists between the ITT and the reported mITT results (a >5% range was mentioned), investors may be notified of the ITT results. However, if mITT and ITT were close, then investors should not expect the ITT disclosure. But given the a priori declaration of an mITT data presentation, and the flexibility that the company may have in deciding what this modified group is, it brings up two linked considerations for the data release:

 

- The exclusion of patients who did not receive placebo / Mydicar or who had events prior to dosing may not necessarily be the modification that they use. The company noted that this modification would include less than 5% of the patients in the trial. The corollary here is that the company appears to have access to the blinded data. Otherwise, how else would they be aware of the percentage of patients that fell into this group? Therefore, interested observers have to be accepting of the fact that Celladon may have a lot of leeway in creating the specifics behind this mITT subgroup. Popular subgroups in these types of HF trials often include patients with ejection fraction (EF) less than X% or patients with NYHA HF Class X and Y. Time will tell if Celladon gets creative. In Biotechland, subgroup creativity is inversely proportional to efficacy of the drug.

- Will the mITT description completely obscure the ITT dataset? The company noted that if the ITT and mITT results are close, investors won’t be informed of the ITT at this time. But what will they consider ITT? For example, if they find a baseline EF subgroup that showed p = 0.07 for Mydicar versus placebo, but can then fall to p < 0.05 if they remove patients who did not receive Mydicar / placebo as well as removing events prior to dosing, then the EF/dosing-adjusted mITT “reaches” statistical significance. In this case, if your mITT patient base is EF/dosing-adjusted, what is your “ITT” comparator? Only EF-adjusted? That, by itself, is not true ITT either. Again, in Biotechland, your difficulty in finding the ITT outcome in a data release is inversely proportional to the efficacy of the drug.

 

I suspect the above two points already feature prominently in the canon of skeptical biotech observers. Time will tell how Celladon negotiates this, but all indications are that the data release may well be more “complicated” than first imagined.

Celladon at BioCEO 2015: A Tough Proposition

Sometimes reiteration makes tasks look harder.

Celladon recently reiterated the case for Mydicar, their AAV1-SERCA2A therapy for patients with heart failure with reduced ejection fraction (HFrEF), at the BioCEO gathering. Celladon has come to prominence based on a subgroup analysis of their CUPID1 trial, wherein the Mydicar high dose arm (n=9) showed statistically significant improvement over the placebo arm (n=14). Although some are convinced by this study result, I’ve expressed my skepticism in an earlier post. Unfortunately the recent presentation makes the Mydicar proposition look less likely to succeed.

In the earlier piece, I’d highlighted strong reasons to question the degree of cardiomyocyte transduction and SERCA2A expression induced by Mydicar therapy. Celladon’s CEO was much more upbeat at BioCEO, describing the administration as one that results in “homogenous uptake” resulting in SERCA2A being expressed “for the life of the cell.” Perhaps that’s a form of confidence in the product. Unfortunately, their data to date provide little support for such unequivocal claims. The discrepancy between the presented scientific data and the company’s claim can be a red flag in and of itself: it is one thing to be confident in your program, but it is more effective if that confidence isn’t accompanied by claims that seem to be undermined by your own published data.

Rather than getting bogged down in the details of the AAV1 mechanism and SERCA2A protein expression, another way to assess this program is to examine the upcoming clinical trial expectations and compare them to historical observations. Celladon is currently running CUPID2, their sequential follow-up to CUPID1. CUPID2 is placebo controlled and enrolling 250 patients (~86% NYHA Class III/IV) split evenly into two arms. Participants will receive the high dose of Mydicar that resulted in statistically significant improvement for 9 patients in CUPID1 compared to their 14 placebo / best available care comparators. In this subgroup analysis, CUPID1 demonstrated an 88% risk reduction. That’s a very large number, especially when compared to our current and upcoming standards of care. To note:

  1. Beta-blockers provided a 19% risk reduction in a HFrEF population that was ~60% NYHA class III/IV
  2. LCZ696, the new offering from Novartis, provided a 20% risk reduction in a HFrEF population that was ~30% NYHA class III/IV

In these multi-thousand patient trials, two drugs, one a mainstay and the other a hopeful newcomer to heart failure treatment, have shown ~20% risk reductions. The trials did not enroll identical patient populations, but nonetheless the risk reduction percentages are eerily similar. In that light, CUPID1’s 88% risk reduction can either appear stellar, or more convincingly a fluke event. It is hard to imagine that this treatment will come out with a risk reduction that is four times those observed in these large trials. This is especially difficult to reconcile since we know that the SERCA2A mechanism of action overlaps with the beta-blocker mechanism of action. In fact, preclinical models have demonstrated that beta-blocker treatment can restore low SERCA2A expression, which is Mydicar’s raison d’etre. If beta-blockers, within their broader mechanism of action, increase SERCA2A expression the way Mydicar is proposed to, is it reasonable to expect that Mydicar demonstrate a risk reduction that is twice as large as that for beta-blockers?

It is not controversial to state that the risk reductions in larger trials tend to be smaller than those in earlier, smaller, proof-of-concept trials. Celladon’s design of the currently running CUPID2 trial suggests that they’re aware of this contributing factor, as the trial was designed to have 83% power to detect a 45 percent risk reduction. This is a more modest target, and the results are due in April, 2015. The company concedes that they need a p value significantly less than 0.05 to make this trial the primary basis of approval. At the same time, the company appears to concede that CUPID2, in and of itself, is unlikely to support registration. They note that a follow-on trial, CUPID3, is under planning and will be used as a phase 3 or phase 4 trial, depending on CUPID2 results and their conversations with US and European regulators.

In effect, for CUPID2 to form the primary basis for registration of Mydicar, the 250 patient trial needs to have an outcome that would come close to replicating the 80%+ risk reduction seen in a 23 patient subgroup analysis. That is a tall order for any drug.

 

Gene Therapy For The Heart - The SERCA Example

The push behind gene therapy for the heart is gaining strength alongside efforts to develop cell-mediated therapies to induce cardiomyocyte repair and regeneration (commented previously). Gene therapy strategies, particularly those based on viral vectors such as the adenoassociated virus (AAV) are attractive due to their relative ease of preparation and administration in preclinical models. One of the furthest along is AAV-mediated delivery of the sarco/endoplasmic reticulum Ca2+ ATPase pump (SERCA) for the treatment of heart failure.

 

The premise behind this approach lies in the observation that preclinical models of heart disease demonstrate a reduction in SERCA expression. Consequently, decreased SERCA activity in the cardiomyocytes may reduce the uptake of Ca2+ into the sarcoplasmic reticulum, and may therefore allow abnormal Ca2+ levels to persist during the cardiomyocyte’s Ca2+ oscillations. This may have implications for contractile activation / relaxation of the heart, and may even help to provoke a pro-hypertrophic response. The administration of an AAV encoding SERCA aims to counter these deleterious events by raising SERCA levels back to normal physiological levels.

 

To reasonably restore physiological SERCA function through AAV-mediated expression, it is instructive to keep in mind the basic and necessary steps of the AAV-mediated gene therapy paradigm. For the AAV-mediated therapy to have a reasonable chance at success, three general steps are requisites before we can reasonably hope to see clinical efficacy:

 

1. Inject virus at a dose sufficient to transduce the target organ

2. Assure that the virus meaningfully transduces the target organ

3. Check for *protein* expression in the target organ of interest

 

In regards to step 1: It is important to note that these are replication defective viral therapies. In that regard, there will be a proportionality between the target number of cells to be transfected and the viral dose. In simpler terms, the dose of virus required to meaningfully and reliably transduce a mouse heart will not be sufficient for equal transduction of a larger rat heart. Therefore, as we increase size of the target organ, the viral dose has to increase, thereby necessitating a scaling-up of doses as we move from mice to humans. Below is a graph showing the maximum viral genomes injected per gram of typical heart weight for the AAV-SERCA construct used by Celladon in clinical trials as well as the antecedent preclinical studies:

 

Arguably, there isn’t significant difference in dosing at the various stages. These data suggest that, given equal tropism of the virus for the heart across species, the dosing may be sufficient. Arguably, dosing in the preclinical sheep model was the highest, and would therefore give the best corollary for assessing the sufficiency of dosing in the clinic.

 

This allows us to move onto Step 2 and attempt to address the ability of the virus to transduce the target organ of interest. The most complete set of data provided in the available preclinical literature is from the sheep model described by Byrne et al. For an intracoronary dose of 2.5x10E13 viral genomes, they note the detection of an average of 1651 copies per ug of DNA from the heart, but that only 3 of the 6 injected animals had any detectable copies. In other words, 3 of 6 animals had no detectable persistence of the AAV, and of the 3 that did, they averaged 1651 copies per ug of DNA. For the liver and lung from these animals, they indicated the detection of 148773 copies and 7781 copies per ug of DNA, respectively. In essence, the data suggest that a minority of the injected AAV persists in the heart, if at all. Therefore, with respect to the targeting of the injected AAV to the intended organ, the preclinical sheep data imply that a significant challenge remains. If we assume that only the heart, lung and liver had detectable AAV and that they’re identically sized organs (Footnote 1), it would suggest an upper boundary of ~1% of the injected intracoronary dose targeted to the heart.

 

Nonetheless, the low level of AAV presence in the heart may be immaterial if a significant increase in SERCA protein can be demonstrated. To best address Step 3, we can examine the publication by Byrne et al to see if there are data provided for protein expression. Unfortunately, no Western blot data are provided. There is a vague indication about mRNA expression in the intracoronary versus intracoronorary + recirculation group, but it only indicates the degree of increase with recirculation versus without. Plus, mRNA expression does not guarantee protein expression, and without protein expression no therapeutic benefit can be expected or rationalized. However, given that only 3 of 6 animals receiving intracoronary injection of 2.5x10E13 viral genomes showed any persistence of the AAV whatsoever, we can safely assume that 50% of these animals had no increase in SERCA protein expression.

 

But what about the remaining 3 animals from the high dose intracoronary group? Absent explicit data from the authors, we can try to determine whether the resident viral genomes are present in sufficient number to elicit sufficient transcription / translation to increase SERCA protein levels. For these 3 animals, it was noted that they averaged 1651 copies of the AAV per ug of DNA. Therefore:

 

  • Convert the mass of DNA to mol basepairs: 1 ug DNA / (660 g/mol basepairs) = 1.52x10E-9 mol basepairs of DNA
  • Convert the mol of DNA bp into number of bp: 1.52 x 10-9 mol DNA bp * 6.02x10E23 basepairs/mol = 9.12x10E14 basepairs
  • Convert the number of basepairs into copies of the genome: For this step, we ride the assumption that the human genome is ~3x10^9 bp, meaning that 9.12x10^14 bp / (3x10^9 bp per genome) = 3.04x10E5 genomes.

 

Put simply, their test surveyed an equivalent of 304000 copies of the genome. Within those 304000 copies, 1651 copies of the exogenous AAV DNA were found. In other words, there were ~302350 copies of the endogenous SERCA gene and 1651 copies of the exogenously introduced SERCA. Therefore, the exogenous AAV-SERCA was responsible for increasing SERCA copies by ~0.5% over the amount endogenously present. Unfortunately, the clinical data does not appear to improve on the data from the preclinical sheep model. Szebo et al note that the highest level was 561 copies per ug of DNA assay from Patient ID 091007 (Table 3). Given similar calculations, that suggests an approximately 0.2% overexpression above endogenous levels. Suffice to say, that is not a large level of expression above and beyond what is already endogenously present.

 

The above calculations give us some insight into the consideration in Step 3: is there an increase in SERCA protein expression in the target organ of interest? The calculations regarding persistence of the AAV-SERCA would suggest that noticeable protein overexpression is unlikely. Therefore, we can comb through the preclinical work in rats, swine and sheep to determine if there is a consistent and reliable demonstration of increased SERCA protein in the heart of AAV transduced animals.

 

Sakata et al suggest increased AAV-mediated SERCA expression in a rat model of aortic banding (Fig. 4). However, there are no numerical data provided to document the extent or significance of overexpression, which is visually not striking. Further, a contemporaneous control using an AAV encoding for parvalbumin shows striking overexpression of parvalbumin protein, making the claimed SERCA overexpression appear modest, if at all significant. A thematically overlapping paper by the same group suggests a doubling of SERCA protein expression in the rat.

 

Moving to larger animals, a subsequent paper using a swine model suggests an increase in protein expression following AAV-SERCA administration in the heart failure group when compared to the saline treated heart failure group. Comparing the SERCA protein level in the AAV treated heart failure group to the normal, nonfailing group demonstrated no SERCA protein overexpression. Interestingly, these data are provided in bar graph form (Fig. 4) but are the only bar graphs in the manuscript that do not have error bars. In the sheep study reported by Byrne et al, no protein data are offered.

 

To be fair, Byrne et al note:

 

“Nevertheless, the positive functional effect of AAV2/1SERCA is entirely consistent with previous work that shows SERCA2a is a low-abundance gene and that only one to two functional copies are required per cell to maintain normal levels of SERCA mRNA and protein.”

 

This suggests that very little overexpression can have a significant functional impact. In that regard, it is important to note that the sheep receiving 2.5x10E13 viral genomes of AAV-SERCA by intracoronary infusion did not show improvement over control in any of the remodeling or functional parameters presented including left ventricular internal diameter (Fig. 2), change in fractional shortening or ejection fraction (Fig. 3), or positive or negative dP/dt (Fig. 4). Therefore, the low level of persistence for 2.5x10E13 viral genomes of AAV-SERCA administered through the intracoronary route in sheep did not translate into a functional benefit as presented, suggesting that the cited low threshold of expression required to restore normal SERCA protein was not exceeded.

 

 

All told, the high dose of 1x10E13 viral genomes administered in the trials to date by Celladon provide limited corroborating data to support the hypothesis that the functional outcomes are a result of AAV-mediated increases in myocardial SERCA expression. Although the CUPID clinical trial demonstrated a benefit for the 1x10E13 viral genome high dose group versus placebo, it is reasonable to ask, given the low level of persistence of the AAV-SERCA construct and scant data for increased SERCA protein expression, exactly what is the mechanism driving this benefit? The calculations and observations based on the preclinical models do not provide a consistent line of evidence for AAV-SERCA driven benefit. The ongoing CUPID2b study, administering 1x10E13 viral genomes via intracoronary route, will determine whether the functional outcomes in CUPID1 were reflective of the technology or a fortuitous coincidence within this patient group.

 

 

 

 

 

 

Footnote 1: The assumption that the heart, lung and liver are the same size benefits the calculations in favour of the heart. In fact, the liver is 4-5x larger by mass (~1.5 kg) and the lungs 2-3x larger by mass (~800 g) as per data here and here.