We’ve got some large sample sizes in a handful of these cancers which makes the findings very meaningful. It’s also all real life human cancer tissue samples with all their real world complexity—not factory replicated strains of cancer in mice. That’s why you see much more variety in these numbers. They say they are getting more samples for the ones they didn’t have enough of, and for more subtypes within each cancer category. So this data will build over time and help with knowing which type of cancer, sub-cancer, and stage a Sort1 targeting drug should theoretically have some level of efficacy in. The most meaningful conclusions are around the ones where they have more than 100 evaluable cases, so it focuses on those 7 cancers. Seems more will come in time as samples build.
Differential and No Sort1 Expression: In a targeting therapy, tumor cell vs. healthy cell target expression in the cancerous environment is paramount in the safety/efficacy tradeoff. The larger the differential between your target in the tumor cells versus the healthy cells, the safer it should be; and maybe the more effective it will be as more of the drug is on-target at the tumor site. We’ve all discussed how important this is on this board. These differentials are all very good in this large study [“….SORT1…was either negative (null) or low for most healthy tissues…”] It’s also very important to note where they found negligible expression of sortilin —in the blood and skin, two pretty large and important parts of the body. May contribute to why the neutropenia issue has been under control at 300mg.
Just a side-note on the H-scores they list and I talk about: The H-scores include a component of what % of the cells show intensity in staining, and also the differential versus other cells in that sample. So it’s a combo of both intensity and differential because both are important for a target to work.
They list some places where the healthy tissue had moderate (more than 10% of cells) and it includes a few parts of the colon, parts of the pancreas, breast lobules, testicles, kidney, and brain and nervous system. I think you’d expect that maybe you’d have more side effects for colon/pancreatic/some breast/testicular cancers. But at <10%, it seems still pretty small. In fact, it may have contributed to why they actually saw TH1902 work on colon cancer where taxol wasn’t known to work. That would go toward proof that the peptide is honing in on SORT1, being internalized, and releasing the toxin. Key parts of the POC. These are also probably cancers where in the future (if successful and commercial) you’d go back and trial at play with the dosage/schedule to see what might work better.
Panel B: Number of samples and average H-Scores: Lot’s of interesting data here, but sample-size limited so can’t make too many conclusions. Obviously the initial one would be the further left on the chart, the higher chances you’d see success with a drug latching on to SORT1. THTX is saying that any cancer where average H-scores are about 100 or higher is “high”, so it’s a big list. Interestingly though, we see Prostate (only 7 samples so take it with a grain of salt) has an H-score of 58, on the low side. But like we’ve surmised, it could be that the Stage 4 Prostate patients are three times that, or that certain sub-types are higher, etc… We know it worked pretty good on two patients, so something must be working on them. These are all they unique issues they’ll learn about but that are mysteries now to all —us and them.
Panel E - Sub Analysis by pathological type of cancer: This still needs to be a bit better fleshed out with more samples but there are some interesting conclusions. They really only have a significant number of sub-types in Ovarian, but that may just reflect the rough percentage of sub-types for these kinds of cancers anyway. I’m guessing that’s the case —so most Endometrial is adenocarcinoma and not the others, etc… Maybe Ovarian is more equally split between the two large sub-groups.
First conclusion is that you can see in every single case of what they are defining as a High Avg Sort1 expression cancer, you still have a pretty wide spread of low to high. The standard deviation around those averages they show is wide. So it shouldn’t surprise us that it won’t work that effectively on every case of these high SORT1 cancers. You could easily see one Ovarian patient respond strongly and another see next to nothing. This is part and parcel for all oncology drugs, but these graphs really point out that it’s not just SORT1 expression alone. There’s other factors that they will try to discover —sub-types, stages, genetic expressions, baseline factors, previous treatments, etc.. No one really knows and there’s plenty of treatments being used today where they don’t know 100% of the reasons it worked on one of their patients and not the other that “looked” like the responder. The mysteries of the human body. But that doesn’t negate it’s effectiveness for some.
Second conclusion: Let’s make an assumption —let’s assume that if you have an H-level (relative % of cells showing SORT1) that is scored as 150 or higher, you should respond to it fairly well and see tumor shrinkage and other cancer markers decline so that you will be a “responder” in RECIST terms. If we assume that, then (on average) we should see a response rate of around 60% for Endometrial Adenocarcinma, 50% for Skin Melanoma, 65% for Breast Non-TNBC Invasive Ductal, 45% for Breast TNBC Invasive Ductal, 50% for Ovarian Adenocarcinoma and High-Grade Serious, 30% for Colorectal Adenocarcinoma, and 10% for Pancreatic Adeno. In fact, Pancreatic, as the table shows, is mostly mid-levels of SORT1 and not high, so it’s not really on the high over expression list except for specific types (more on that later). I’m just eyeballing it on the charts and not counting the dots, so I could be off some. Of course, if they find that you really need to be in the 200 or higher H-Score level for solid efficacy, lower all these numbers. If you need to be 100 or higher, increase them all. These are the things they will find out in time. The other sub-types don’t have enough data to really conclude anything yet. But if they conclude an H-score of 100 or higher is “high” and “high” SORT1 is where it should work, then we should see some decent response rates over a larger sample size. But since we don’t know fully the MOA, maybe SORT1 expression is one part of the “magic” and it will work on those mid to high levels but some other factor pushes your efficacy up? We don’t know at this point, but this is a good theoretical construct to start the investigation and learn.
Panel F - Cancer Stage: One thing all the studies showed was that SORT1 is correlated with the stage of cancer you are in. What they are showing in these studies is that this is the case with Breast, Ovarian and Pancreatic, but not necessarily with Edometrial, Melanoma or Colorectal. The caveat is that they don’t really have enough data for some cancer stages to really be conclusive, but let’s just go with the data so far and learn a bit. In Endo, their data shows it’s actually negatively correlated —the highest levels of SORT1 are seen in Stage 1 and not the later stages. The numbers are small here though. But you will recall that Patient 2 in 1a was the endometrial that stayed on the trial the longest, saw single digit tumor regression and stable disease for a very long time period. This kind of data may suggest to them that, in fact, don’t look at this as just a Stage 4 and beyond, last gasp treatment. Use it throughout the stages! You know what this means — a massively larger market opportunity. In fact, their overall conclusion with these 7 cancers is “…high expression (of SORT1) is maintained throughout the progression from stages 1 to 4.” They are defining high as H-Score 100 or higher. This really is a novel finding and, while data limited, contradicts some of the previous academic articles. Where they have the human data, it’s hard not to agree with this new interpretation as the others were not huge samples either (I think just the breast cancer one had a very large sample size of tissues).
But Panel E&F Together: But along the lines we’ve all been discussing, you can see that if you put together Panel E and Panel F, you can start to understand where, in the clinic, you might reach for TH1902 should it become a viable therapy. It certainly should inform Phase 2 and beyond. For instance, you might slowly prove up that the best efficacy in TNBC is only for invasive ductal and lobular, not medullary, and for Stage 2/3/4, but not Stage 1. Or for Colorectal it’s best for Adeno and Interstitialoma, but not squamous, and should start at Stage 3 onward. For Endo and Melanoma, it works for all stages and in certain sub-types. You can see these are the kind of clinical findings that will build a case for what we may see in a further Phase trial and how it might be designed for best efficacy in order to get to market fastest. As of now, we know they are shooting for highest expressers within very late stage candidates.
So the very broad conclusions I take are:
- The tumor/healthy cell differential seems workably high for a target-based approach. SORT1 appears to have what you’d want in a genetic target for a small molecule PDC.
- For the cancers they had enough samples of, there’s a massive market potential for cancers with a large amount of “high” to “very high”. We should increase our confidence a bit that there’s a good chance we will hit that 30% or higher response rate in some of these cancers in 1b.
- What will be the “high” that works —an H-score of >100 as they seem to suggest, or >200? We will eventually learn that but likely not until P3 or later. But we may be able to infer things given this tissue sample study and that’s why it’s helpful. If we see 30% response rates in some of these, then it’s likely the H-score needs to be above 200 or higher. If 10% response rates, then 250/300 is needed. If we see 60% levels, then H-scores >100 would be roughly enough to see efficacy. This assumes all else being equal, but we know there are other factors also affecting response rates. But we may glean some parts of a POC and de-risk certain aspects of the program nonetheless.
- We have enough data to show Pancreatic may be more of a specialty approach —it will work under very specific conditions only. That’s still huge, given where pancreatic is, but that will likely be a small market opportunity (but maybe high $$ still) and not something we see in P1b or P2. There’s others that are pointing to the fact they should “work” better. Who knows, as science once applied to humans can be so very different.
- They have designed the 1b for maximum chance of success if you believe a lot of these numbers. It means they’ll likely also design P2 to have the maximum chance of success. As you learn more about SORT1 in tumors and cancers —sub-types, stages, baseline data, etc… —you play to those strengths in your design. I think they’ve done the best we could hope for in giving us many shots on goal here and looking for a real score on one….or more.
- The Panel F Stages study is really important commercially and for the clinicians. I think part of the issue we’ve had on a dragged out schedule is both competition for trials, lack of knowledge of THTX in oncology (and their lack on past contacts with oncologists), and a complete lack of understanding around SORT1, what it is, how it can be used, and if this works. All this top-notch, clinical back-up around the “theory” is very helpful to doctors who get this kind of data and information. Of course, they need the human studies, but you really need a very compelling theoretical framework around it to create some confidence and buzz. This is just one more building block and the human data will be the most important. But if you get that human data and it correlates strongly with your studies, it’s a real “provability” enhancer and makes more of what you’ve theorized about seem highly probably. If it’s more or less reflecting the theorized science, then the science becomes believable to the doctors. They’ll understand the stem cell and VM science, the ability to evade resistance — all that will build a very, very strong and compelling case to an oncologist if the human data gets it across the line.