|Home | About | Journals | Submit | Contact Us | Français|
Chemical genetics represents an expanding collection of techniques applied to a variety of signaling processes. These techniques use a combination of chemical reporters and protein engineering to identify targets of a signaling enzyme in a global and non-directed manner without resorting to hypothesis-driven candidate approaches. In the last year, chemical genetics has been applied to a variety of kinases, revealing a much broader spectrum of substrates than had been appreciated. Here, we discuss recent developments in chemical genetics, including insights from our own proteomic screen for substrates of the kinase ERK2. These studies have revealed that many kinases have overlapping substrate specificity, and they often target several proteins in any particular downstream pathway. It remains to be determined whether this configuration exists to provide redundant control, or whether each target contributes a fraction of the total regulatory effect. From a general perspective, chemical genetics is applicable in principle to a broad range of posttranslational modifications (PTMs), most notably methylation and acetylation, although many challenges remain in implementing this approach. Recent developments in chemical reporters and protein engineering suggest that chemical genetics will soon be a powerful tool for mapping signal transduction through these and other PTMs.
Identification of direct substrates and their respective modification sites for any given enzyme is a challenging process, but this information is necessary to understand cellular signal transduction in normal and abnormal physiological processes. To date, much of the effort to resolve enzyme-substrate relationships has been focused on protein kinases and their substrates, but even in this area, there remains a massive knowledge gap, as over 100,000 unique phosphorylated residues have been observed in mammalian cells, but only a few thousand of those sites have been associated with a particular kinase or phosphatase. Several innovative techniques enabling large-scale, higher-throughput identification of direct substrates for given enzymes have recently emerged. Here, we will provide our perspective on the state-of-the-art in chemical genetics, a collection of techniques that have enabled discovery of many novel kinase-substrate relationships.
Historically, most kinase substrates have been identified through directed studies (e.g., in vitro kinase assays), in which a given protein of interest is incubated with a selected kinase to test the ability of the kinase to phosphorylate the protein; many of these substrates have also been confirmed in cells through altered phosphorylation in response to inhibition or knockdown of the kinase. In addition to being low-throughput, these directed studies all rely on the selection of a given kinase and putative substrate at the start of the experiment and are therefore limited in their ability to discover truly novel substrates. To increase the throughput and enable unbiased discovery of novel substrates, several labs have reformatted the in vitro kinase assay into protein microarrays, leading to the putative identification of thousands of novel putative substrates for selected kinases.1,2 Unfortunately these assays may include many false-positive or false-negative interactions, as they are typically performed on immobilized proteins, in the absence of protein-protein complexes and associated co-factors found in the cellular environment.
As an alternative strategy for substrate identification that preserves more of the natural environment and interactions occurring in the cell, the field of chemical genetics represents a rapidly growing collection of methods being applied to a wide variety of signaling processes. In general, these methods use a combination of chemical reporters and protein engineering to identify kinase targets in a global and non-directed manner. In these approaches, an enzyme is typically engineered to engender specificity toward a synthetic small molecule, allowing selective inhibition or labeling of direct substrates with chemical tags. Selective inhibition makes it possible to parse the contribution of an individual kinase to phenotypes in cell-based or mouse models, and analysis of quantitative changes in protein phosphorylation can be used to infer substrates of the given kinase (although due to the complexity of the signaling network, many of the proteins demonstrating altered phosphorylation may not be direct substrates of the kinase).
In contrast to selective inhibition, selective labeling by chemical genetics allows kinase substrates to be directly identified using proteomics and is emerging as a powerful tool to uncover novel biological roles for kinases and substrates. Unlike hypothesis-driven candidate approaches, the combination of selective chemical labeling and proteomics is able to connect a kinase to totally unexpected downstream pathways. We have recently used a chemical genetics and proteomic approach to identify targets of the ERK2 mitogen-activated protein kinase (MAPK).3 This chemical genetics strategy paired a mutant form of the kinase with a bioorthogonal adenosine triphosphate (ATP) analog to selectively label ERK2 substrates with thiophosphate.4,5 Combining this approach with mass spectrometric analysis of phosphorylated (previously thiophosphorylated, see below) substrates revealed targets of ERK2 across a broad range of pathways and biological processes, many of which were previously not connected to the ERK MAP kinases. Similar work by other groups has had the same impact for other kinases, including cyclin/CDKs, AMPKα2 and Aurora B, in that each of these studies has revealed a much wider range of target proteins than previously appreciated, suggesting a surprising amount of diversity and complexity in signal transduction networks.6-9 The highly connected nature of these signaling networks complicates the idea of discrete signaling pathways and underscores the importance of systems-level analysis in understanding signaling processes.
Ongoing developments in chemical genetics will allow similar strategies to be applied to other signaling processes, including methylation of lysine and arginine and acetylation of lysine. Recent work has suggested that both of these modifications are widespread, and that they are involved in regulation of processes including transcriptional regulation, response to DNA damage and crosstalk with signaling networks regulated by other PTMs. Our understanding of signal transduction networks regulated by lysine methylation or acetylation and arginine methylation is still nascent, but it is clear that chemical genetic approaches will have a significant positive impact on our ability to decipher the connectivity in these networks. Most importantly, the information revealed by these techniques will uncover how these different signaling processes interact to coordinate cellular behaviors.
The first implementation of chemical genetics occurred in 1997, when Shokat and coworkers screened mutations to the ATP-binding pocket of v-Src in an effort to identify isoforms that would be selectively inhibited by an ATP-competitive inhibitor (these mutant kinases are denoted analog-sensitive or AS). These engineered, analog-sensitive kinases can be transfected into cells to enable specific inhibition of a kinase of interest without affecting homologous kinases or closely related family members. For example, a recent application of chemical genetics demonstrated targeted inhibition of the mitotic kinase Aurora B without perturbing the kinase activity of the closely related kinases Aurora A and C.9 To facilitate substrate identification, in their initial efforts, the Shokat group also developed N6-modified ATP analogs with radiolabeled gamma phosphates that are not reactive toward normal biological processes (they are “bioorthogonal”), leading to the tagging of direct kinase substrates with radiolabeled phosphate.10 Even though substrates of the AS-kinase were radiolabeled and could be detected as distinct spots on 2D gels, substrate identification was hindered by the difficulty of separating these proteins from the large background of phosphorylated proteins in the cell lysate.
To address this problem, the gamma phosphate on the bioorthogonal ATP analog (Fig. 1) has recently been replaced with thiophosphate, resulting in the tagging of AS-kinase substrates with a distinctive chemical functionality.11 Since thiophosphate does not occur in natural biological systems and is not reactive toward normal biological processes, it serves as a handle for enrichment and identification of labeled protein or peptides. Both chemical and immunoaffinity procedures have been described for detection and enrichment of thiophosphorylated proteins or peptides, facilitating proteome-wide screens for kinase substrates.3,6-9,12 The first proteome-wide screen using this approach was implemented by Blethrow et al. in 2008 to map CDK1/cyclin B substrates. In this study, recombinant AS-CDK1/cyclin B and N6-(benzyl)ATP-γ-S were added to HeLa cell extracts, resulting in thiophosphorylation of AS-CDK1/cyclin B substrates.6 After proteolytic digestion, thiol-containing peptides were captured on agarose beads functionalized with immobilized iodoacetyl. Thiophosphorylated substrates were then recovered by treatment with an oxidizing agent that converted thiophosphate to phosphate, leading to the selective elution of thiophosphorylated peptides, while leaving cysteine-containing peptides permanently anchored to the beads. This analysis led to the identification of dozens of substrates of the CDK1/cyclin B complex in HeLa cells, including many involved in mRNA transcription and processing and DNA repair. Interestingly, a separate study of CDK2/cyclin A substrates using different enrichment chemistry revealed many of the same substrates, but also an unexpected set of potential substrates involved in protein translation.7 Given the overlapping set of potential substrates for these kinases, additional work is required to determine if there is functional relevance to the phosphorylation of these sites by either kinase, and to determine the cellular conditions that may lead to phosphorylation of these sites by either or both kinases. The potential for multiple kinases to phosphorylate a given site on a substrate may increase the robustness of the network but also significantly complicates traditional pathway diagrams.
In our chemical genetics screen, we have focused on identifying substrates of ERK2 due to its integral role in development and differentiation, as well as its role in cancers carrying activating mutations in B-Raf, K-Ras, EGFR and other tyrosine kinases. Through a combination of chemical genetics, metabolic labeling (see below) and analysis of multiple biological replicates by quantitative mass spectrometry, we were able to identify 80 ERK2 substrate proteins, including 67 novel substrates of this well-studied kinase. Perhaps the most striking aspect of the 67 novel substrates identified in our screen was the diversity of biological processes that they represent, which includes motor proteins, metabolic regulators, cell cycle and apoptosis proteins and proteins associated with nuclear structure, DNA organization, ubiquitination and cytoskeletal organization, among many others. Interestingly, even within this diverse group, many substrates cluster around a few biological processes; these clusters may provide a new perspective on how signaling networks exert control over complicated behaviors.
As one instance of multiple novel ERK2 substrates clustering around a particular biological function, we identified four novel substrates involved in microtubule regulation and mitosis. ERK1/2 activity has been shown to be necessary for the G2/M transition in Xenopus oocytes,13 and it has been associated with progression through early G2 and with G2/M transition in mammalian cells. A precise understanding of the timing and role of ERK activity in G2/M remains elusive, with the overall effect of ERK-pathway inhibition being dependent on cell type, timing (chronic vs. acute) and the mechanism of inhibition.14 In contrast, ERK1/2 activity has a clearly established role in mediating genomic instability as a result of oncogenic mutations in upstream Ras or Raf signaling proteins.14-16 Several mechanisms by which ERK1/2 could influence mitosis have been suggested, including activation of mitotic phosphatases Cdc25B and Cdc25C in both Xenopus and mammalian cells17,18 and regulation of microtubule and spindle assembly.19 Several high-throughput screens have also shown that activation of K-Ras in colon carcinoma cells sensitizes them to loss of mitotic protein including Polo-like kinase 1 (PLK1) and the microtubule-associated proteins survivin and targeting protein for Xenopus kinesin-like protein 2 (TPX2).20-22
In our study, new targets identified using AS-ERK2 include TPX2, DLG7, CEP170 and GTSE1, all of which are involved in microtubule assembly during mitosis.23-26 These, along with previously reported microtubule-associated substrates, such as components of the dynein complex, suggest a more integral role for ERK2 in coordinating mitosis through regulation of microtubule assembly and bundling. Molecular functions for these phosphorylation events remain to be determined, but it is intriguing that one signaling pathway seems to impinge on the mitotic apparatus at so many points. The collection of mitotic ERK2 substrates may represent potential targets in carcinoma cells carrying activated MAPK pathway that could be exploited for therapeutic effect, perhaps in combination with upstream (e.g., Raf or MEK) inhibitors.
AS-ERK2 substrates also cluster around the regulators and effector proteins of the Rho family of small GTPases, which are broadly involved in cytoskeleton dynamics and cell motility.27 ERK1/2 has been shown to localize to focal adhesions, and its activity is involved in signaling from focal adhesions to Rho and its immediate target ROCK as well as interacting with focal adhesion kinase (FAK) to control disassembly of focal adhesions (reviewed in ref. 28). However, the mechanisms by which ERK signaling regulates Rho-family guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs) are not well-understood. Novel AS-ERK2 substrates include the Rho-family GEFs DOCK1 and ARHGEF17, the Rho GAP myosin IxB (MYO9B) and the Rho-family effector proteins CDC42EP1 and CDC42EP2. It is worth noting that ERK1/2-dependent phosphorylation has also been previously observed on GEF-H1 and ARHGAP26, both of which regulate RhoA.29,30 These multiple related ERK 1/2 substrates suggest that crosstalk between ERK signaling and the Rho GTPases could occur at several points, both up- and downstream from the Rho GTPases themselves. It will be interesting to determine the functional significance of these phosphorylation sites on related proteins in this network: does phosphorylation on any site lead to a small change, with the ensemble phosphorylation then driving a much larger overall effect, or is phosphorylation of one of these sites sufficient to induce a significant phenotypic effect, with the multiple different sites then representing a fail-safe mechanism? Mutation of these phosphorylation sites to phospho-mimetic or non-phosphorylatable residues under different stimulation conditions will be needed to parse out their individual or ensemble effects.
In sum, the set of substrates identified using AS-ERK2 provides an abundance of interconnections among otherwise distinct signaling processes. Novel substrates connecting ERK2 to apparently unrelated pathways include GLI2, which may represent a mechanism for crosstalk with the Hedgehog signaling pathway, GRB10-interacting GYF protein 2 (TNRC15), which is involved in activity of the insulin receptor, the spicing factor RNA binding protein 9 (FOX2) and the microRNA-regulating protein 5'-3' exoribonuclease 2 (XRN2). Parsing how these pathways work together to coordinate cellular behaviors will require detailed characterization of molecular functions and interactions as well as careful analysis of phenotypes in a wide range of biological systems.
The chemical genetic approach used to study kinases is applicable in principle to any signaling enzyme with an active site that can be engineered to accept a bioorthogonal cofactor analog. Posttranslational modifications that may be amenable to chemical genetics include glycosylation, methylation of lysine and arginine as well as acetylation and acylation of lysine.
Methylation of lysine and arginine on histones is well-characterized and has defined roles in transcriptional regulation, but methylation of non-histone proteins has also been shown to regulate a variety of biological processes (reviewed in ref. 31). For example, methylation by the lysine methyltransferase (KMT) Set9 stabilizes the tumor suppressor p53 and regulates the transcription of genes targeted by p53.32 Methylation also regulates inflammation signaling through the RelA subunit of NFκB,33,34 and inhibits the ability of C-RAF to activate the ERK1/2 cascade.35 Although multiple chemical approaches have been developed to analyze protein methylation (reviewed in ref. 36), the full extent of proteome-wide methylation has not been characterized, as proteomic techniques to identify lysine and arginine methylation are still nascent.34,37
In a similar vein to the AS kinases, protein methyltransferases (PMTs) have been engineered to accept bioorthogonal S-adenosyl-L-methionine (SAM) analogs as methyl donors. In the first example of this approach, Gray and coworkers engineered the yeast arginine methyltransferase RMT1 to accept SAM analogs with a benzyl group at the adenosine N6 position.38 As with the initial AS-kinase studies, identification of PMT substrates is still difficult with this approach, because the N6-substituted SAM analogs carry a normal methyl group and therefore do not label target proteins with a bioorthogonal tag. More recent efforts have focused on SAM analogs carrying substitutions at the 2' and 3' positions39 and at the sulfonium methyl group.40-45 Several of these studies investigated SAM analogs with alkyne groups in place of the sulfonium methyl, leading to the modification of PMT substrates with a bioorthogonal alkyne functional group. Alkyne-tagged substrates can be coupled to functional groups, including fluorophores, epitope tags or biotin through copper-catalyzed alkyne-azide cycloaddition (CuAAC, Click chemistry), thereby enabling a wide variety of detection and affinity capture approaches.41,43,44 However, since multiple PMTs can use alkyne-functionalized SAM analogs, this strategy suffers from non-specificity and does not identify direct substrates of a specific PMT. Additionally, only a restricted set of PMTs has been shown to use any particular SAM analog, thereby limiting the broad application of this method.
To overcome these limitations, the Luo group has recently engineered mutant PMTs able to utilize bioorthogonal SAM analogs containing longer carbon chains in place of the sulfonium methyl, leading to specific labeling of PMT substrates. Alignment of catalytic SET domains of several lysine methyltransferases led to the identification of a conserved tyrosine in the enzyme G9a. Conversion of this tyrosine to alanine allowed for the utilization of (E)-hex-2-en-5-ynyl SAM (Hey-SAM) as a modified alkyl donor (Fig. 1).44 Specificity appears to be conferred by the size of Hey-SAM, as smaller analogs carrying propargyl or pent-2-en-4-ynyl can be utilized by some PMTs, but Hey-SAM may be too large to be a substrate for native PMTs. Since many SET domains share well-conserved sequence and structure, it is likely that this approach will be extensible to many PMTs. In a similar effort, the Luo group has also identified a pair of mutations that allow the protein arginine methyltransferase PRMT1 to utilize 4-propargyloxy-but-2-enyl SAM.45 Combined with high-throughput approaches to identify pairings of SAM analog with mutant PMTs, these studies raise the possibility that the combination of bulky alkyne-bearing SAM analogs and protein engineering could be used for global identification of lysine residues targeted by specific PMTs.
A variety of other posttranslational modifications, such as fatty acylation, glycosylation and acetylation, may also be amenable to chemical genetics.46-51 Lysine acetylation of histones has a well-established role regulating chromatin structure and gene transcription. Acetylation of many non-histone proteins has now been implicated in regulating many diverse biological processes, including cell metabolism, DNA damage response, cell cycle progression and age-associated hearing loss. Intriguingly, bromodomains (domains that bind to acetylated lysines) have recently emerged as potential therapeutic targets in multiple human cancers, further expanding the potential impact of lysine acetylation. Despite the emerging importance of this protein posttranslational modification, the connectivity within lysine acetylation signaling networks is poorly characterized. Proteome-wide strategies to identify lysine acetylation have revealed that the PTM occurs at thousands of positions across the proteome,52-54 yet the specific enzymes responsible for acetylation or deacetylation of any given site are typically not known. Since altered acetylation has been implicated in multiple human diseases, understanding the enzymes responsible for regulating the level of acetylation on these sites may provide therapeutic targets to intervene in these disease states.
Lysine acetyltransferases (KATs) use acetyl-coenzyme A (acetyl-CoA) as an acetyl donor. The Hang group has shown that p300, one of the prevalent KATs in mammalian cells, is able to utilize 4-pentynoyl-CoA to label substrates with a five-carbon analog of the acetyl group containing a terminal alkyne (Fig. 1). They have extended this approach to in vivo labeling by treating cells with the metabolic precursor sodium 4-pentynoate, which was then converted to 4-pentynoyl-CoA and utilized by the endogenous KATs to label substrates with the bioorthogonal tag.51 Intriguingly, utilization seemed to be somewhat specific to 4-pentynoyl-CoA, as the slightly smaller 3-butynoyl-CoA was not a substrate for p300 and 5-hexynoyl-CoA was utilized only very inefficiently. After labeling substrates with the terminal alkyne, click chemistry with CuAAC was used to couple the alkyne tag to either a fluorescent reporter or to biotin for affinity purification on Streptavidin. Through this labeling and enrichment procedure, a variety of novel putative p300 substrates were identified by conducting an in vitro labeling reaction with p300 added to HeLa nuclear extract with 4-pentynoyl-CoA.55 Unfortunately, in the current format, this approach does not specifically identify p300 substrates, as significant non-specific labeling was detected even in the absence of recombinant p300 (Fig. 1 of ref. 55), indicating that endogenous KATs were also able to utilize the acetyl-CoA analog. Following on the above demonstrations of protein engineering for specificity in chemical genetics through a bumped-enzyme approach, it should be possible to engineer lysine acetyltransferases to accept an N6-substituted bioorthogonal acetyl CoA analog, potentially with a modified acetyl group that could be used for downstream enrichment. This approach would then specifically label substrates of the given engineered KAT with a bioorthogonal affinity tag, enabling their identification by mass spectrometry. Although challenging, this overall strategy would lead to direct identification of KAT substrates, providing the connectivity that is currently missing in the acetyllysine signaling network.
AS-kinases and similar chemical genetic strategies have several inherent concerns that should be taken into account when pursuing this line of investigation. The first concern is that the chemical genetic mutation may affect enzyme activity or enzyme specificity, potentially leading to labeling of non-endogenous substrates by an overactive enzyme. Additionally, since many of the AS-enzyme substrates have been identified through in vitro reactions occurring in cell lysates, there is a significant concern that conditions of the labeling reaction may affect enzyme specificity or may allow non-physiologically relevant protein-protein interactions. Accumulated experience so far indicates that AS-kinases have the same substrate specificity as their wild-type counterparts.3,4,56 We have validated seven novel AS-ERK2 substrates biochemically and found that all seven are phosphorylated by wild-type ERK2. Despite these results, changes in enzymes specificity are still a possibility, and all substrates should be validated before beginning in-depth biological investigation into the function of a given PTM site. It has also been observed that not all kinases retain their activity following mutation of the conserved “gatekeeper” residue. In these cases it has sometimes been possible to identify a secondary mutation that restores activity.9
Another concern is that endogenous wild-type enzymes present in the reaction may utilize the ATP analog and create off-target labeling, or that the bioorthogonal chemical handle on the ATP analog (e.g., the terminal thiophosphate) may be removed from analog ATP and placed onto an endogenous ATP molecule, resulting in bioorthogonal labeling by wild-type kinases. Either case leads to false-positive detection of substrates in the analysis, as proteins that appear to be direct substrates of the AS enzyme are actually due to other enzymes in the cell. This issue of non-specific labeling is serious and warrants detailed consideration. Although we are not aware of any individual kinase with strong affinity for the bioorthogonal ATP analog, we did observe substantial labeling in control reactions from lysates of cells transfected with WT-ERK2 but containing no AS-kinase. This result was borne out in our mass spectrometry analysis, in which we routinely observed a variety of phosphorylated peptides that lacked any part of the consensus sequence for phosphorylation by ERK2. Although lack of a consensus sequence does not rule out a potential substrate, per se, it does suggest that a robust approach is necessary to eliminate false positives.
Most investigators have addressed background labeling by simply excluding any peptide observed in a negative control reaction. Unfortunately, failure to detect a peptide in a proteomic analysis does not indicate absence of the peptide. In fact, proteomic analysis of technical replicates routinely identifies only about 60% of the same peptides. Therefore, false positives may only be seen in the AS-kinase sample just due to the vagaries of data-dependent mass spectrometry. We addressed this issue by incorporating quantitative analysis into our proteomic experiment. Stable Isotope Labeling in Cell Culture (SILAC) was used to compare the AS-ERK2 labeling reaction with a matched negative reaction containing the same amount of wild-type ERK2.57 This strategy allowed us to combine positive and negative samples, so that they are subject to identical handling and analyzed as a single experiment on the mass spectrometer. Having a quantitative negative control and multiple biological replicates allowed us to define statistical thresholds to control the false-discovery rate and unambiguously separate real substrates from non-specifically labeled peptides. Examination of previously known ERK2 substrate sites also revealed that there was considerable non-specific labeling occurring on some residues targeted by ERK2. Without a truly quantitative approach, these substrates may have been detected in a negative control reaction and incorrectly excluded.
A quantitative strategy also raises the possibility of comparing kinase substrates between biological conditions (treatment, cell type, etc.) or between closely related kinases. Biological context for chemical genetic labeling is another area that needs to be considered carefully in experiments to identify direct substrates of AS enzymes. Many enzymes act as part of larger signaling complexes and may lose activity or specificity when purified as individual recombinant proteins. Even though we conducted the labeling reaction with AS-ERK2 in cell lysate, we found that substrate labeling was enhanced when AS-ERK2 was stably overexpressed in the cells of interest instead of being added exogenously. We hypothesize that pre-formed complexes improved the ability of the enzyme to interact with its substrates. Permeabilizing cells expressing an AS-kinase so that ATP analog can pass the cell membrane provides an alternate approach, enabling labeling in cells rather than in cell lysate, thereby potentially preserving more physiologically relevant localization and protein-protein interactions. This approach was recently implemented by Banko et al. to achieve in-cell labeling of AS-Aurora B substrates following permeabilization with the detergent digitonin.8 Our experience has been that in-cell labeling is less efficient compared with labeling in cell lysate due to issues including competition with normal ATP and rapid loss of cell viability, but it does reduce concerns about the biological context of novel substrates.
Chemical genetics extends the tools of chemical biology to allow exquisitely targeted experiments with selective inhibition or bioorthogonal substrate labeling. Although a variety of caveats remain, AS-kinases have shown their value in high-throughput identification of novel substrates. This work has accelerated in the last year, with reports of proteome-wide screens for substrates of ERK2, CHK1, AMPKα2 and Aurora B.3,8,9,12 We expect that research using AS-kinases will continue to accelerate, with applications to a wider set of kinases and investigation of a broad range of biological systems. Chemical genetics has also been recently applied to protein methyltransferases targeting both lysine and arginine. We believe that these experiments are likely to identify far more methylated substrates than are currently known. Since many posttranslational modifications have been investigated using bioorthogonal labeling, it is likely that the approach of chemical genetics will be applicable to an extraordinarily wide range of signaling enzymes and biological systems. We hope that these techniques will provide a new perspective on the complexity of biological signaling pathways, and that they will inform experimental and conceptual tools that will allow a deeper understanding of the signaling network.
The authors declare no conflicts of interest. This work is supported by NIH grants R01DK42816, R01CA118705, U54CA112967 and 5P30CA014051. S.M.C. has been supported by a Graduate Research Fellowship from the NSF, the David H. Koch Institute for Integrative Cancer Research Graduate Fellowship, the Whitaker Health Science Fellowship and the Siebel Foundation.
Previously published online: www.landesbioscience.com/journals/cc/article/19956