PDAC Backgrounder
Pancreatic ductal adenocarcinoma, or PDAC, is a brutal disease. Often diagnosed after widespread metastases, the disease has a median survival of well under a year - and that number has been essentially unchanged for decades. An important note is that PDAC is very different from neuroendocrine tumors of the pancreas - it is of this other type that Steve Jobs died of. On a personal note, I've had two people close to me whom I knew before and after their diagnosis, and sadly both followed the typical course of the disease. Largely due to improvements in the treatment of other major cancer types, pancreatic ductal adenocarcinomas are on track to become the deadliest cancer in the US.One near hallmark of pancreatic ductal adenocarcinomas is activating mutations in KRAS - over 90% of tumors bear this mark. RAS proteins - KRAS, HRAS and NRAS are all oncogenes - are an important signaling nexus and can be roughly imagined as a molecular hourglass. When RAS has GTP bound, then it is in the ON state and with GDP bound it is in the off state. RAS proteins are very poor GTPases, so the hydrolysis is slow - providing for an ON-to-OFF timing mechanism. They're also slow to release the product GDP, providing for an OFF-to-ON timing mechanism. Both of these timings can be altered by interacting factors - GTP exchange factors (GEFs) speed up the OFF-to-ON side and GTP activating proteins (GAPs) speed up the hydrolysis-based ON-to-OFF. RAS proteins interact with many other proteins to enable the ON/OFF state to be transmitted to downstream effector pathways.
KRAS for decades was in the "undruggable" category, but the past decade has seen the introduction of KRAS inhibitors - though these typical act against the OFF state and have not moved the needle of PDAC. There are now ON state inhibitors in clinical trials, and two of those are from Revolution Medicines, which bought Warp Drive Bio which I helped found. Indeed, while I had nothing to do with that side of the company other than cheer it on, the Revolution Medicines compounds are based on Warp's molecular glue (we called it "presentation") chemistry and the clinical compounds are elaborations of Warp molecules. These drugs act by causing abundant intracellular chaperones to stick to KRAS, blocking interactions with effector molecules. If they should really move the needle on this dreadful disease, then Revolution Medicines will do very well financially - and I feel obligated to say that I am not a bystander on that point due to Warp shares converted in the acquisition. Rumors around a $30B takeover possibility of Revolution Medicines, discussions which is now rumored to be no longer active, drove their stock to dizzying heights - and had any Warper wondering about an alternative history where we raised funding to keep going alone.
Single Cell PDAC Sequencing
But back to the genomics (though we will come back to KRAS inhibitors). The paper applied both bulk sequencing and Mission Bio to fresh/frozen samples from 24 patients. All but one of these had PDAC; one turned out to be another pancreatic tumor acinar cell carcinoma (ACC). Five of the PDAC patients plus the ACC patient had germline BRCA2 mutations, a known risk factor for PDAC. Thirty primary and 42 metastases were subjected to sequencing.
The Mission Bio protocol, detailed in a prior publication, uses droplet encapsulation of nuclei followed by PCR amplification of specific targets within the droplets. So nuclei extraction using the Singulator 100, sucrose density gradient centrifugation to clean up the nuclei prep, then into the Mission Bio Tapestri to generate the emulsions to enable amplifying 596 amplicons targeting known PDAC-relevant genes.
Since this is clinical data, it is deposited in protected fashion that I really can't get permitted access to in any short time (it's even worse for industry, let alone "I'm a blogger who doesn't currently do drug discovery"). But I tried to back-calculate the total number of reads from the supplementary table and got about 1/2 an Illumina NovaSeq X 25B flowcell. I know that number isn't quite right because the read counts for each sample aren't integers, which they should be if the calculation was right. But it's probably within a factor of 2-3, so that does point to how this sort of study is still not inexpensive on the sequencing side. One issue is that the read depth of the amplicons is hard to control - some samples have averages over 2000 whereas others are below 50. And the text discusses - though its unclear whether these were included in the supplementary table counts - that some nuclei are "incomplete" and believed to be from apoptotic cells and therefore not informative - so there's that drag on the data.
The data is good enough that copy numbers can be computed, so amplified or dropped out amplicons can be detected. So deletions and amplifications can be found. But, an inherent limitation of amplicon-based methods is that they can only find what matches both primers and was a targeted gene - so gene fusions are generally off the table. And that will pop up later. Also, some cells will be non-neoplastic, but since this is looking purely at DNA sequence the type of such cells - epithelium, immune cells, etc - cannot be determined.
From all this data it was possible to construct trees for each sample that can be used to determine the number of distinct clones represented in the data. It was also possible to identify mutually exclusive events - a key advantage of single cell data over bulk data. Intriguingly, in one patient a mutation was found that was unique to non-neoplastic cells (i.e. lacking activating mutations) but has been associated with clonal hematopoiesis. Another sample showed a different mutation which was exclusive to non-neoplastic cells. The single cell data also showed that most of the variation between neoplastic clones within a sample are from copy number variants, not new point mutations.
Several interesting patterns were observed for KRAS. For example, a wide range of allelic ratios between mutant and wild-type KRAS were observed, indicating cases of either wild-type or mutant amplification. KRAS copy variation was seen between clones within the same sample. Most strikingly - and very relevant to the possible clinical utility of KRAS-ON inhibitors and particularly of allele-specific KRAS inhibitors - is that some tumors showed clones which had lost the mutant allele. In some of these clones, plausible new driver mutations could be observed. Apparently there is already published mouse data concordant with this. So it can be hypothesized that these tumors were initially driven by KRAS mutations but then switched to alternative oncogenic signaling which might render them immune to KRAS inhibitors. Only in one patient was it observed the presence of two different KRAS mutations, and the one was interpreted to be an independent lesion not derived from the dominant clone.
Yet another pattern of heterogeneity was seen a KRAS mutant tumors sampled from multiple sites in the same patient - a subset of clones had mutations in PIK3CA but these were two different mutations which were mutually exclusive.
In all of the BRCA2 germline mutation patients it was possible to identify the mechanism of loss of the wildtype copy - so none were driven by hypermethylation which the Mission Bio approach cannot detect. In four patients loss-of-heterozygosity (LOH) for BRCA2 was observed via loss of the wild-type allele; in the other two inactivating somatic mutations are observed in what was the wildtype copy.
In one BRCA2 germline mutation patient, the single cell sequencing identified mutually-exclusive clones with different candidate oncogenic lesions. Because these clones were spatially separated, it was possible to go back into the bulk sequencing data and identify a fusion in a known oncogene which might provide KRAS-independent signaling. Another BRCA2 patient had another known fusion observed in bulk data. In yet another patient, an interesting pattern was observed in which some clones had BRCA2 homozygous deletion but intact CDKN2A, whereas others had BRCA2 LOH and homozygous deletion in CDKN2A. One conclusion from this sample made by the authors is that the timing of BRCA2 function loss is highly variable.
In multiple samples it was observed that there are minor subclones which have mutations which would be expected to inactivate inhibitory TGF-beta signaling. Mutations in SMAD4 in this pathway are common in PDAC; the observation that tumors seem to be converging on attenuating this pathway is novel.
Ruminations
Back to technical stuff. As noted before, the power and the limitations of Mission Bio's PCR-based approach can be seen. Much was learned in this study, but only the pairing with bulk sequencing was able to detect fusions - and the particular configuration of fusions with other mutations could only be resolved because it happened to be that fusions were in distinct metastases which were derived from a single clone of the primary tumor.
In an ideal world, one would have whole genome, whole transcriptome, chromatin architecture, and methylation state all resolved at single cell level. That's perhaps a pipe dream for now, but I see that Atrandi is presenting at talk at SLAS on single cell DNA+RNA. One could also imagine ways to tag DNA in the single cell droplets, perhaps with barcoded tagmentation, and then using hybrid capture to cast a net that can catch fusions as well as point mutations. And as the paper noted, the Mission Bio approach requires fresh or frozen tissue and is not compatible with FFPE.
Mission Bio has been one of few companies focused on single cell DNA. You can also use various devices to generate plates with at most one cell per well and then amplify with kits from companies such as QIAGEN, Bioskryb or Takara. BIO-RAD markets a single cell ATAC-Seq approach as does 10X Genomics and Parse Biosciences (now acquired by QIAGEN). But the field has seemed sleepy - in need of new entrants. Might Atrandi and Factorial Biosciences be the ones to provide the needed energy? Atrandi teasing both DNA and RNA from the same cells - did I read that right? And could any approach tag native long read libraries which could read out methylation? Could you even have single cell DNA chromatin state atop that via fiber-seq aka chromosome stenciling?
There is also the economics. In a paper such as this, there is a great cost to acquiring the samples, but sequencing is still not cheap. If my estimate of half a 25B is accurate, then scaling this up to hundreds or thousands of patients would get pricey. And this was just in pancreatic ductal adenocarcinoma - there are sure many other tumors which would benefit from the approach. Plus clonal hematopoiesis is an area of increasing interest. So that's one reason my eyes might roll every time I see someone saying "sequencing is cheap enough, why would anyone want it cheaper?" - because there are always bigger, deeper and broader scope projects that can't happen until the current price of sequencing undergoes another step reduction.
A paper from a Spanish lab has acquired much press interest in that it claims to have erased PDAC in multiple mouse models. The paper is paywalled, but I found a preprint I think is the same work and it appears these models are either genetically engineered mice or xenografts using tumor lines in such mice in additional immunocompetent mice. So how heterogeneous are these tumors? Do they model anything like clinical reality? It would be great to explore with single cell DNA sequencing great numbers of tumors from genetically engineered mouse models and patient derived xenografts and all the other xenograft models that have been used to gate (for better or worse) oncology drug development - but who will be willing to pay for it?
Finally, I'll throw out a question to which I certainly have no answer: will this have clinical utility? If, for example, it was seen that a PDAC patient already had clones that had lost KRAS-dependence, would that mean skipping KRAS inhibitors, adding another agent to target the subclones, or simply do nothing different and wait to see what happens?
No comments:
Post a Comment