Celyad — Update 7 July 2016

Celayd has been focussed on cardiac development with the important NKR-T Chimeric Antigen Receptor programme for cancer, based on the proprietary NKG2R technology, running in the background. This is because the current dosing ranging NKR-T Phase I/II study has taken time due to the safety framework. The CHART-1 Phase III cardiac data, discussed in this note, now enables Celyad to metamorphose into a focussed cancer company with the cardiac programme partnered. Once NKR-T data becomes available (the trial is at the 30m cell dose level) this programme can be re-evaluated. This transition will take time, partners need to be signed, data acquired and complied, new trial designs agreed with the FDA and clinicians, but it is underway.

C-Cure is an autologous stem cell therapy to treat chronic, ischaemic heart failure in seriously ill patients. The two C-Cure Phase III trial structures and positions are summarised in Exhibit 1.

The disclosed outcome is of overall non-significance (data not disclosed), but significance, p=0.015, was seen in a subgroup of patients with a “mid- range” EDV, numerically undisclosed. Selecting subgroups retrospectively is subject to bias, but Celyad states it used a statistical methodology. EDV was measured at baseline as it is used to calculate ejection fraction, one of the endpoints. In this case, the subgroup p value indicates a clear separation between treated and placebo sets. There are statistical methods to adjust the p value threshold in these situations as the primary endpoint threshold of p=0.05 is too high for subgroup analysis; no adjusted threshold has been discussed but management indicate that the result is robust enough for an EMA submission. The EMA has provided guidelines for subgroup analysis. These suggest that scientific credibility of the finding, which can be qualitative, is important in any regulatory decision.

The EMA process anticipated will probably run during 2017 with marketing possible from 2018. Review takes 210 days excluding clock stops for questions; Celyad has already submitted preliminary background data. If a positive opinion is issued, the EU commission will issue an approval some months later after pack inserts and other details have been approved.

Without more information, the regulatory outcome is very hard to assess. The EMA can be much more pragmatic than the FDA. If a treatment is novel, safe, as this appears to be, and meets an unmet medical need (these patients were on the best current therapy but not improving), then the EMA may recommend approval, possibly on a conditional basis involving a patient registry (database) and further studies. Celyad management noted that the data was “intensive and powerful”. As this is uncertain, no EU sales are forecast until CHART-2 data perhaps in 2021.

In European markets, each is different, cost effectiveness is a major buying decision criteria. Agencies such as the UK’s NICE will take at least a year to do a rigorous economic assessment; other countries have similar processes. Even if “cost effective”, there is no guarantee that the cash budget will be available. Germany is probably the most important market. France typically requires extensive pricing discussions. The UK is very cash constrained, maybe more so in future.

The CHART-1 study did not have EDV as a selection criteria or as a pre-specified endpoint. Edison notes that this is the sort of finding that trials, which are scientific experiments, can reveal especially when it is the first large-scale, properly conducted study of its type to report. This is cutting edge science and clinical development. The EDV finding scientifically needs confirmation by other studies, but is regarded by Celyad scientific and clinical advisers as a good indicator.

EDV is the volume of blood in the left ventricle in the instant before it contracts. Echocardiography is the usual measurement method of choice and was used by Celyad. EDV is needed to calculate the ejection fraction by subtracting end systolic volume, the fully contracted left ventricle volume, from EDV. Hence EDV was measured. Normal levels seem to be about 100mL for women and around 150mL for men. However, measurement techniques vary so this is not a standardised methodology (Rigolli 2016). There is also an EDV index, relating EDV to body size.

On contraction of the heart, blood is pushed into the aorta and round the arterial system. Some blood remains in the ventricle. The volume of blood ejected depends on the pressure in the aorta – if high, less blood is pushed out – and the strength of the left ventricle muscle. To be in CHART-1, patients only expelled 35% or less of the EDV; this is the ejection fraction (EF). A healthy person will eject 50-70%, with 60-70% being normal. We do not know the ejection fraction range in CHART-1 or how EDV related to EF.

However, it is complicated as always. The Frank-Starling effect shows that in a healthy person, the EF increases as the EDV rises until a limit where cardiac output (volume) plateaus. This appears to be related to the elasticity of the muscle fibres; if stretched more they contract better. Patients with lower blood volumes, for example undergoing major surgery or after trauma, pump less blood. Volume, and pressure, are the key to positive clinical outcomes.

In heart failure, the Frank-Starling effect is weakened or lost. The left ventricle becomes distorted and distended. The muscle is damaged in places or dead if there is no blood supply due to a heart attack (infarction) so is weaker overall. Hence, EDV is much higher, perhaps doubled, relative to healthy people. A high EDV and weaker muscle leads to a low ejection fraction. However clinically, the key issue is the volume of blood ejected. If EF falls but EDV rises, the two may cancel out, although pressure is an issue.

The reason why Celyad found an apparent effect in “mid-range” EDV patients is not certain. Patients with very weak cardiac muscles may have high EDV but have little viable tissue and be too damaged for the C-Cure cells to have an effect. In “healthier” patients, it may be that there is sufficient functional muscle so the cells make little difference particularly as the Frank-Starling effect shows that the cardiac system is adaptable. More analysis may give some indication.

In the clinical literature, Kramer (2010) reviewed 25 cardiac intervention trials covering 69,766 patients and a further 80 studies on cardiac remodelling with 19,921 patients. The analysis focused on mortality, which is the primary endpoint but not expected by Celyad to be a key indicator in CHART-1. The analysis by Kramer et al is in Exhibit 2. There is a clear correlation with mortality but at best it is 50%, so these variables are not the only factors: These factors are interlinked and they are not independent variables. Hence, as CHART-1 has a clear EDV signal, one would expect to see similar signals in EF and ESV in the same grouping. If these are not seen, then the EMA may be highly sceptical.

In a device trial Goldenberg (2011) found that patients in the top quartile of EDV had the best response. This is in contrast to the CHART-1 where high EDV (undefined) had weak response. Goldenberg also noted a very similar ESV response pattern.

A paper on the CHART-1 design was published in 2015: Bartunek et al, 2015. The study uses a Finkelstein–Schoenfeld hierarchical composite endpoint. The endpoints are, in order of testing, listed in Exhibit 2. This process appears complex but allows multiple endpoints to be included.

For example, if Patient 1 is compared to Patient 2 and they both survived, had no cardiac events and rated equal on MLHFQ, all those score zero. If Patient 1 walks more than 40m further on 6MWT but Patient 2 walks less than 40m further (compared to their baselines), Patient I scores 1. No other endpoint is then scored by Patient 1 against Patient 2. When Patient 2 is compared against Patient 1, they will score -1.

Next Patient 1 is compared with Patient 3. For example, if Patient 3 died, Patient 1 scores 1 (as a survivor).

Patient 1 is then compared with Patient 4, Patient 5 and so on. The score for each patient is then summed to give a single value. For example, when Patient 3 is compared to all other patients, if there were 19 other deaths, all those score zero. As all the other 251 patients survived, Patient 3 scores -251 in total. All 271 scores for each patient are put into order and a rank order assigned to each. The groups are then separated into placebo and treated so each group has a distribution of rankings. A two-sided 5% significance level using a modification of the generalized Wilcoxon test is used to get statistical significance. This tests if the distribution is the same between the groups.

The authors of Bartunek et al, 2015 note that the Minnesota Living with Heart Failure Questionnaire (MLHFQ) and six-minute walk (6MW) tests “may be influenced by knowledge of the treatment variable”. These tests are likely to be crucial to understanding the trial result. They are rated above cardiac quantitative measures as they are more clinically relevant. Exhibit 2 shows the design.

To take an example of scoring for a single patient, to score a mortality endpoint, a patient scores one if they survive but the comparison patient dies. If they both survive they are equal, so score zero. This is comparison is done with every other patient. However, if this patient died and the comparison survived, the patient scores minus one. If both survive the outcome is zero (the likely outcome on this measure). The second endpoint is then tested then the third and so on. Some patients will not have full data sets, for example, because they died before the designed 39-week endpoint or did not complete some tests, but the scores they gained can still be included in the data. This allows hard endpoints like mortality where few events are expected to be included (so most comparisons score zero) with endpoints that are expected to have a more variable outcome like quality-of-life scores and the six-minute walk test.

The CHART-2 study was agreed by the FDA in 2015 but the start was delayed to see the CHART-1 data. The trial has been redesigned and now comprises 400 patients with a 52 week primary endpoint. The intention now is to recruit the EDV mid-range subgroup claimed by management as significant in CHART-1; Celyad does not foresee any problems in this. Having a successful Phase III subgroup in CHART-1 may encourage recruitment. The limiting factor is the time to sign a partner. It is planned to start in December 2016 with primary data in March 2018.

Celyad is now focusing on the CAR T-cell cancer area with its NKR-T programme. Although cardiac has been a major part of the value to date, and could be if an EU approval is received with US partnering, CAR is still likely to comprise the majority of the value eventually as NKR-T moves into specific cancers and, possibly, solid tumours. The heavy focus of this set of indications means that the indicative value is very sensitive to the dollar/euro exchange rate.

If C-Cure gains EMA conditional approval, initial EU sales may have a slow start due to the disparate national markets and the patient population could, in theory, be 60% of what was anticipated, although there is a high margin of error in forecasting new markets. European marketing will be through a partner, but Celyad will run the authorisation process.

CHART-2 is a true Phase III as it will incorporate the valuable insights from CHART-1. CHART-2 assessment depends on the data due in Rome. CHART-2 has been redesigned and refocussed but can only progress with a partner, which will take time and adds risk. If successful, this could allow an EMA reappraisal, either for approval (if rejected) or for a broader indication (if approved).

The previous valuation was at €846m ($914m) or €88 ($95) per share. The indicative value will be revised again after 28 August data, but as an interim position has been modified as follows:

This gives a revised indicative value of €32 per share after management warrants and options are included. Celyad had €94.7m in cash in Q116. If C-Cure milestones are lower and there are no immediate royalties, this mean that additional capital will be needed by 2018 to develop a broad NKR-T clinical trial portfolio and progress the important allogeneic technology platform. At the moment, there is a value imbalance as the additional 2017 investment into NKR-T is becoming apparent while the clinical position is still not known, hence the value of that programme cannot yet be reassessed.

CAR T-cell cancer therapy has become one of the most active areas of cancer research and investment activity. Juno is now valued at $4.3bn. Another prominent CAR specialist is Kite Pharma at $2.5bn. Compared to the value of Juno and Kite, Celayd illustrates a massive valuation dichotomy. There seems to be little CAR value in the Celayd share price.

The NKR-T programme has a potentially huge market if it extends into solid tumour therapy. We reviewed the NKR-T programme in the note “Far from the overvalued crowd” published 8 April 2015. This gave a scenario value, at that time, of €519m for the NKR-T project if it moved into solid tumour indications. The crucial assumption underlying this estimate is that powerful tumour responses are seen like those seen initially with CD19 in lymphoma. There is, apart from published preclinical work on an ovarian cancer model, no data to support this premise. Consequently, solid tumour values are not yet included in the core valuation estimate in Exhibit 4 (above).

Allogeneic versions of NKR-T (not specifically valued as too early in development) could have longer biological protection if classed as new products. They could also extend the profitability by cutting the price, making the therapies easier to use and by greatly expanding sales.

As an independent June 2016 report from EP Vantage shows (Shifting CAR-Ts into a Higher Gear), the optimistic basis for such valuations might be undermined as the complexity of developing standard CAR approaches for solid tumours become apparent. In this area, Celyad is potentially well placed with its different, proprietary NKG2R approach as this might be applicable to multiple cancer types. Edison will revaluate this area once data form the current studies becomes available.

Sales and fees are only used in the valuation model and are not included in the financial P&L and cash-flow model. Upfront fees are included in financial projections only once deals are announced.

Cash on 31 December 2015 was €107m and €94.7m on 31 March 2016; the Q1 cash outflow included a US$5m payment relating to the Celdara acquisition. The CHART-2 costs previously expected have been removed as a partner is being sought. This may result in higher than expected cash by December 2016, which we forecast at about €77m, vs €69m previously expected. However, in 2017 we now expect investment to increase into NKR-T projects, following management guidance, so cash at year end 2017 might around €15-20m. This is lower than the previous expectation of €35-40m. This makes more funding by 2018 inevitable unless a major deal is achieved.

[Type a quote from the document or the summary of an interesting point. You can position the text box anywhere in the document. Use the Drawing Tools tab to change the formatting of the pull quote text box.]