Identifying general reaction conditions by bandit optimization.

Reaction conditions that are generally applicable to a wide variety of substrates are highly desired, especially in the pharmaceutical and chemical industries1-6. Although many approaches are available to evaluate the general applicability of developed conditions, a universal approach to efficiently discover these conditions during optimizations is rare. Here we report the design, implementation and application of reinforcement learning bandit optimization models7-10 to identify generally applicable conditions by efficient condition sampling and evaluation of experimental feedback. Performance benchmarking on existing datasets statistically showed high accuracies for identifying general conditions, with up to 31% improvement over baselines that mimic state-of-the-art optimization approaches. A palladium-catalysed imidazole C-H arylation reaction, an aniline amide coupling reaction and a phenol alkylation reaction were investigated experimentally to evaluate use cases and functionalities of the bandit optimization model in practice. In all three cases, the reaction conditions that were most generally applicable yet not well studied for the respective reaction were identified after surveying less than 15% of the expert-designed reaction space.

Predicting Reaction Yields via Supervised Learning.

Numerous disciplines, such as image recognition and language translation, have been revolutionized by using machine learning (ML) to leverage big data. In organic synthesis, providing accurate chemical reactivity predictions with supervised ML could assist chemists with reaction prediction, optimization, and mechanistic interrogation.To apply supervised ML to chemical reactions, one needs to define the object of prediction (e.g., yield, enantioselectivity, solubility, or a recommendation) and represent reactions with descriptive data. Our group's effort has focused on representing chemical reactions using DFT-derived physical features of the reacting molecules and conditions, which serve as features for building supervised ML models.In this Account, we present a review and perspective on three studies conducted by our group where ML models have been employed to predict reaction yield. First, we focus on a small reaction data set where 16 phosphine ligands were evaluated in a single Ni-catalyzed Suzuki-Miyaura cross-coupling reaction, and the reaction yield was modeled with linear regression. In this setting, where the regression complexity is strongly limited by the amount of available data, we emphasize the importance of identifying single features that are directly relevant to reactivity. Next, we focus on models trained on two larger data sets obtained with high-throughput experimentation (HTE). With hundreds to thousands of reactions available, more complex models can be explored, for example, models that algorithmically perform feature selection from a broad set of candidate features. We examine how a variety of ML algorithms model these data sets and how well these models generalize to out-of-sample substrates. Specifically, we compare the ML models that use DFT-based featurization to a baseline model that is obtained with features that carry no physical information, that is, random features, and to a naive non-ML model that averages yields of reactions that share the same conditions and substrate combinations. We find that for only one of the two data sets, DFT-based featurization leads to a significant, although moderate, out-of-sample prediction improvement. The source of this improvement was further isolated to specific features which allowed us to formulate a testable mechanistic hypothesis that was validated experimentally. Finally, we offer remarks on supervised ML model building on HTE data sets focusing on algorithmic improvements in model training.Statistical methods in chemistry have a rich history, but only recently has ML gained widespread attention in reaction development. As the untapped potential of ML is explored, novel tools are likely to arise from future research. Our studies suggest that supervised ML can lead to improved predictions of reaction yield over simpler modeling methods and facilitate mechanistic understanding of reaction dynamics. However, further research and development is required to establish ML as an indispensable tool in reactivity modeling.

Bayesian reaction optimization as a tool for chemical synthesis.

Reaction optimization is fundamental to synthetic chemistry, from optimizing the yield of industrial processes to selecting conditions for the preparation of medicinal candidates1. Likewise, parameter optimization is omnipresent in artificial intelligence, from tuning virtual personal assistants to training social media and product recommendation systems2. Owing to the high cost associated with carrying out experiments, scientists in both areas set numerous (hyper)parameter values by evaluating only a small subset of the possible configurations. Bayesian optimization, an iterative response surface-based global optimization algorithm, has demonstrated exceptional performance in the tuning of machine learning models3. Bayesian optimization has also been recently applied in chemistry4-9; however, its application and assessment for reaction optimization in synthetic chemistry has not been investigated. Here we report the development of a framework for Bayesian reaction optimization and an open-source software tool that allows chemists to easily integrate state-of-the-art optimization algorithms into their everyday laboratory practices. We collect a large benchmark dataset for a palladium-catalysed direct arylation reaction, perform a systematic study of Bayesian optimization compared to human decision-making in reaction optimization, and apply Bayesian optimization to two real-world optimization efforts (Mitsunobu and deoxyfluorination reactions). Benchmarking is accomplished via an online game that links the decisions made by expert chemists and engineers to real experiments run in the laboratory. Our findings demonstrate that Bayesian optimization outperforms human decisionmaking in both average optimization efficiency (number of experiments) and consistency (variance of outcome against initially available data). Overall, our studies suggest that adopting Bayesian optimization methods into everyday laboratory practices could facilitate more efficient synthesis of functional chemicals by enabling better-informed, data-driven decisions about which experiments to run.

T-ALL can evolve to oncogene independence.

The majority of cases of T-cell acute lymphoblastic leukemia (T-ALL) contain chromosomal abnormalities that drive overexpression of oncogenic transcription factors. However, whether these initiating oncogenes are required for leukemia maintenance is poorly understood. To address this, we developed a tetracycline-regulated mouse model of T-ALL driven by the oncogenic transcription factor Lmo2. This revealed that whilst thymus-resident pre-Leukemic Stem Cells (pre-LSCs) required continuous Lmo2 expression, the majority of leukemias relapsed despite Lmo2 withdrawal. Relapse was associated with a mature phenotype and frequent mutation or loss of tumor suppressor genes including Ikzf1 (Ikaros), with targeted deletion Ikzf1 being sufficient to transform Lmo2-dependent leukemias to Lmo2-independence. Moreover, we found that the related transcription factor TAL1 was dispensable in several human T-ALL cell lines that contain SIL-TAL1 chromosomal deletions driving its overexpression, indicating that evolution to oncogene independence can also occur in human T-ALL. Together these results indicate an evolution of oncogene addiction in murine and human T-ALL and show that loss of Ikaros is a mechanism that can promote self-renewal of T-ALL lymphoblasts in the absence of an initiating oncogenic transcription factor.

Regioselective Cross-Electrophile Coupling of Epoxides and (Hetero)aryl Iodides via Ni/Ti/Photoredox Catalysis.

A cross-electrophile coupling reaction of epoxides and (hetero)aryl iodides that operates via the merger of three catalytic cycles involving a Ni-, Ti-, and organic photoredox catalyst has been developed. Three distinct classes of epoxides, styrene oxides, cyclic epoxides, and terminal aliphatic epoxides, all undergo coupling in moderate to good yield and high regioselectivity with the use of three different nitrogen-based ligands for Ni under otherwise identical reaction conditions. The mild reaction conditions accommodate a broad scope of abundant and complex coupling partners. Mechanistic studies suggest that when styrene oxides are employed radical intermediates are involved via Ti-radical ring-opening of the epoxide. Conversely, for terminal aliphatic epoxides, involvement of an iodohydrin intermediate enables the formation of the unexpected linear product.

The EMT modulator SNAI1 contributes to AML pathogenesis via its interaction with LSD1.

Modulators of epithelial-to-mesenchymal transition (EMT) have recently emerged as novel players in the field of leukemia biology. The mechanisms by which EMT modulators contribute to leukemia pathogenesis, however, remain to be elucidated. Here we show that overexpression of SNAI1, a key modulator of EMT, is a pathologically relevant event in human acute myeloid leukemia (AML) that contributes to impaired differentiation, enhanced self-renewal, and proliferation of immature myeloid cells. We demonstrate that ectopic expression of Snai1 in hematopoietic cells predisposes mice to AML development. This effect is mediated by interaction with the histone demethylase KDM1A/LSD1. Our data shed new light on the role of SNAI1 in leukemia development and identify a novel mechanism of LSD1 corruption in cancer. This is particularly pertinent given the current interest surrounding the use of LSD1 inhibitors in the treatment of multiple different malignancies, including AML.

Nickel/Photoredox-Catalyzed Methylation of (Hetero)aryl Chlorides Using Trimethyl Orthoformate as a Methyl Radical Source.

Methylation of organohalides represents a valuable transformation, but typically requires harsh reaction conditions or reagents. We report a radical approach for the methylation of (hetero)aryl chlorides using nickel/photoredox catalysis wherein trimethyl orthoformate, a common laboratory solvent, serves as a methyl source. This method permits methylation of (hetero)aryl chlorides and acyl chlorides at an early and late stage with broad functional group compatibility. Mechanistic investigations indicate that trimethyl orthoformate serves as a source of methyl radical via ß-scission from a tertiary radical generated upon chlorine-mediated hydrogen atom transfer.

3d-d Excited States of Ni(II) Complexes Relevant to Photoredox Catalysis: Spectroscopic Identification and Mechanistic Implications.

Synthetic organic chemistry has seen major advances due to the merger of nickel and photoredox catalysis. A growing number of Ni-photoredox reactions are proposed to involve generation of excited nickel species, sometimes even in the absence of a photoredox catalyst. To gain insights about these excited states, two of our groups previously studied the photophysics of Ni(t-Bubpy)(o-Tol)Cl, which is representative of proposed intermediates in many Ni-photoredox reactions. This complex was found to have a long-lived excited state (τ = 4 ns), which was computationally assigned as a metal-to-ligand charge transfer (MLCT) state with an energy of 1.6 eV (38 kcal/mol). This work evaluates the computational assignment experimentally using a series of related complexes. Ultrafast UV-Vis and mid-IR transient absorption data suggest that a MLCT state is generated initially upon excitation but decays to a long-lived state that is 3d-d rather than 3MLCT in character. Dynamic cis,trans-isomerization of the square planar complexes was observed in the dark using 1H NMR techniques, supporting that this 3d-d state is tetrahedral and accessible at ambient temperature. Through a combination of transient absorption and NMR studies, the 3d-d state was determined to lie ∼0.5 eV (12 kcal/mol) above the ground state. Because the 3d-d state features a weak Ni-aryl bond, the excited Ni(II) complexes can undergo Ni homolysis to generate aryl radicals and Ni(I), both of which are supported experimentally. Thus, photoinduced Ni-aryl homolysis offers a novel mechanism of initiating catalysis by Ni(I).

The NUP98-HOXD13 fusion oncogene induces thymocyte self-renewal via Lmo2/Lyl1.

T cell acute lymphoblastic leukaemia (T-ALL) cases include subfamilies that overexpress the TAL1/LMO, TLX1/3 and HOXA transcription factor oncogenes. While it has been shown that TAL1/LMO transcription factors induce self-renewal of thymocytes, whether this is true for other transcription factor oncogenes is unknown. To address this, we have studied NUP98-HOXD13-transgenic (NHD13-Tg) mice, which overexpress HOXA transcription factors throughout haematopoiesis and develop both myelodysplastic syndrome (MDS) progressing to acute myeloid leukaemia (AML) as well as T-ALL. We find that thymocytes from preleukaemic NHD13-Tg mice can serially transplant, demonstrating that they have self-renewal capacity. Transcriptome analysis shows that NHD13-Tg thymocytes exhibit a stem cell-like transcriptional programme closely resembling that induced by Lmo2, including Lmo2 itself and its critical cofactor Lyl1. To determine whether Lmo2/Lyl1 are required for NHD13-induced thymocyte self-renewal, NHD13-Tg mice were crossed with Lyl1 knockout mice. This showed that Lyl1 is essential for expression of the stem cell-like gene expression programme in thymocytes and self-renewal. Surprisingly however, NHD13 transgenic mice lacking Lyl1 showed accelerated T-ALL and absence of transformation to AML, associated with a loss of multipotent progenitors in the bone marrow. Thus multiple T cell oncogenes induce thymocyte self-renewal via Lmo2/Lyl1; however, NHD13 can also promote T-ALL via an alternative pathway.

Long-Lived Charge-Transfer States of Nickel(II) Aryl Halide Complexes Facilitate Bimolecular Photoinduced Electron Transfer.

Here we investigate the photophysics and photochemistry of Ni(II) aryl halide complexes common to cross-coupling and Ni/photoredox reactions. Computational and ultrafast spectroscopic studies reveal that these complexes feature long-lived 3MLCT excited states, implicating Ni as an underexplored alternative to precious metal photocatalysts. Moreover, we show that 3MLCT Ni(II) engages in bimolecular electron transfer with ground-state Ni(II), which enables access to Ni(III) in the absence of external oxidants or photoredox catalysts. As such, it is possible to facilitate Ni-catalyzed C-O bond formation solely by visible light irradiation, thus representing an alternative strategy for catalyst activation in Ni cross-coupling reactions.

Hhex induces promyelocyte self-renewal and cooperates with growth factor independence to cause myeloid leukemia in mice.

The hematopoietically expressed homeobox (Hhex) transcription factor is overexpressed in human myeloid leukemias. Conditional knockout models of murine acute myeloid leukemia indicate that Hhex maintains leukemia stem cell self-renewal by enabling Polycomb-mediated epigenetic repression of the Cdkn2a tumor suppressor locus, encoding p16Ink4a and p19Arf However, whether Hhex overexpression also affects hematopoietic differentiation is unknown. To study this, we retrovirally overexpressed Hhex in hematopoietic progenitors. This enabled serial replating of myeloid progenitors, leading to the rapid establishment of interleukin-3 (IL-3)-dependent promyelocytic cell lines. Use of a Hhex-ERT2 fusion protein demonstrated that continuous nuclear Hhex is required for transformation, and structure function analysis demonstrated a requirement of the DNA-binding and N-terminal-repressive domains of Hhex for promyelocytic transformation. This included the N-terminal promyelocytic leukemia protein (Pml) interaction domain, although deletion of Pml failed to prevent Hhex-induced promyelocyte transformation, implying other critical partners. Furthermore, deletion of p16Ink4a or p19Arf did not promote promyelocyte transformation, indicating that repression of distinct Hhex target genes is required for this process. Indeed, transcriptome analysis showed that Hhex overexpression resulted in repression of several myeloid developmental genes. To test the potential for Hhex overexpression to contribute to leukemic transformation, Hhex-transformed promyelocyte lines were rendered growth factor-independent using a constitutively active IL-3 receptor common ß subunit (ßcV449E). The resultant cell lines resulted in a rapid promyelocytic leukemia in vivo. Thus, Hhex overexpression can contribute to myeloid leukemia via multiple mechanisms including differentiation blockade and enabling epigenetic repression of the Cdkn2a locus.

Analysis of Histone Modifications in Acute Myeloid Leukaemia Using Chromatin Immunoprecipitation.

Chromatin Immunoprecipitation (ChIP) using antibodies specific for histone modifications is a powerful technique for assessing the epigenetic states of cell populations by either quantitative PCR (ChIP-PCR) or next generation sequencing analysis (ChIP-Seq). Here we describe the procedure for ChIP of histone marks in myeloid leukaemia cell lines and the subsequent purification of genomic DNA associated with repressive and activating histone modifications for further analysis. This procedure can be widely applied to a variety of histone marks to assess both activating and repressive modifications in the context of myeloid leukaemia.

Hhex Regulates Hematopoietic Stem Cell Self-Renewal and Stress Hematopoiesis via Repression of Cdkn2a.

The hematopoietically expressed homeobox transcription factor (Hhex) is important for the maturation of definitive hematopoietic progenitors and B-cells during development. We have recently shown that in adult hematopoiesis, Hhex is dispensable for maintenance of hematopoietic stem cells (HSCs) and myeloid lineages but essential for the commitment of common lymphoid progenitors (CLPs) to lymphoid lineages. Here, we show that during serial bone marrow transplantation, Hhex-deleted HSCs are progressively lost, revealing an intrinsic defect in HSC self-renewal. Moreover, Hhex-deleted mice show markedly impaired hematopoietic recovery following myeloablation, due to a failure of progenitor expansion. In vitro, Hhex-null blast colonies were incapable of replating, implying a specific requirement for Hhex in immature progenitors. Transcriptome analysis of Hhex-null Lin- Sca+ Kit+ cells showed that Hhex deletion leads to derepression of polycomb repressive complex 2 (PRC2) and PRC1 target genes, including the Cdkn2a locus encoding the tumor suppressors p16Ink 4a and p19Arf . Indeed, loss of Cdkn2a restored the capacity of Hhex-null blast colonies to generate myeloid progenitors in vitro, as well as hematopoietic reconstitution following myeloablation in vivo. Thus, HSCs require Hhex to promote PRC2-mediated Cdkn2a repression to enable continued self-renewal and response to hematopoietic stress. Stem Cells 2017;35:1948-1957.

Mild, Redox-Neutral Formylation of Aryl Chlorides through the Photocatalytic Generation of Chlorine Radicals.

We report a redox-neutral formylation of aryl chlorides that proceeds through selective 2-functionalization of 1,3-dioxolane through nickel and photoredox catalysis. This scalable benchtop approach provides a distinct advantage over traditional reductive carbonylation in that no carbon monoxide, pressurized gas, or stoichiometric reductant is employed. The mild conditions give unprecedented scope from abundant and complex aryl chloride starting materials.

Direct C(sp3)-H Cross Coupling Enabled by Catalytic Generation of Chlorine Radicals.

Here we report the development of a C(sp3)-H cross-coupling platform enabled by the catalytic generation of chlorine radicals by nickel and photoredox catalysis. Aryl chlorides serve as both cross-coupling partners and the chlorine radical source for the α-oxy C(sp3)-H arylation of cyclic and acyclic ethers. Mechanistic studies suggest that photolysis of a Ni(III) aryl chloride intermediate, generated by photoredox-mediated single-electron oxidation, leads to elimination of a chlorine radical in what amounts to the sequential capture of two photons. Arylations of a benzylic C(sp3)-H bond of toluene and a completely unactivated C(sp3)-H bond of cyclohexane demonstrate the broad implications of this manifold for accomplishing numerous C(sp3)-H bond functionalizations under exceptionally mild conditions.

Acute myeloid leukemia requires Hhex to enable PRC2-mediated epigenetic repression of Cdkn2a.

Unlike clustered HOX genes, the role of nonclustered homeobox gene family members in hematopoiesis and leukemogenesis has not been extensively studied. Here we found that the hematopoietically expressed homeobox gene Hhex is overexpressed in acute myeloid leukemia (AML) and is essential for the initiation and propagation of MLL-ENL-induced AML but dispensable for normal myelopoiesis, indicating a specific requirement for Hhex for leukemic growth. Loss of Hhex leads to expression of the Cdkn2a-encoded tumor suppressors p16(INK4a) and p19(ARF), which are required for growth arrest and myeloid differentiation following Hhex deletion. Mechanistically, we show that Hhex binds to the Cdkn2a locus and directly interacts with the Polycomb-repressive complex 2 (PRC2) to enable H3K27me3-mediated epigenetic repression. Thus, Hhex is a potential therapeutic target that is specifically required for AML stem cells to repress tumor suppressor pathways and enable continued self-renewal.

Requirement for Lyl1 in a model of Lmo2-driven early T-cell precursor ALL.

Lmo2 is an oncogenic transcription factor that is frequently overexpressed in T-cell acute lymphoblastic leukemia (T-ALL), including early T-cell precursor ALL (ETP-ALL) cases with poor prognosis. Lmo2 must be recruited to DNA by binding to the hematopoietic basic helix-loop-helix factors Scl/Tal1 or Lyl1. However, it is unknown which of these factors can mediate the leukemic activity of Lmo2. To address this, we have generated Lmo2-transgenic mice lacking either Scl or Lyl1 in the thymus. We show that although Scl is dispensable for Lmo2-driven leukemia, Lyl1 is critical for all oncogenic functions of Lmo2, including upregulation of a stem cell-like gene signature, aberrant self-renewal of thymocytes, and subsequent generation of T-cell leukemia. Lyl1 expression is restricted to preleukemic and leukemic stem cell populations in this model, providing a molecular explanation for the stage-specific expression of the Lmo2-induced gene expression program. Moreover, LMO2 and LYL1 are coexpressed in ETP-ALL patient samples, and LYL1 is required for growth of ETP-ALL cell lines. Thus, the LMO2-LYL1 interaction is a promising therapeutic target for inhibiting self-renewing cancer stem cells in T-ALL, including poor-prognosis ETP-ALL cases.

TCPTP regulates SFK and STAT3 signaling and is lost in triple-negative breast cancers.

Tyrosine phosphorylation-dependent signaling, as mediated by members of the epidermal growth factor receptor (EGFR) family (ErbB1 to -4) of protein tyrosine kinases (PTKs), Src family PTKs (SFKs), and cytokines such as interleukin-6 (IL-6) that signal via signal transducer and activator of transcription 3 (STAT3), is critical to the development and progression of many human breast cancers. EGFR, SFKs, and STAT3 can serve as substrates for the protein tyrosine phosphatase TCPTP (PTPN2). Here we report that TCPTP protein levels are decreased in a subset of breast cancer cell lines in vitro and that TCPTP protein is absent in a large proportion of "triple-negative" primary human breast cancers. Homozygous TCPTP deficiency in murine mammary fat pads in vivo is associated with elevated SFK and STAT3 signaling, whereas TCPTP deficiency in human breast cancer cell lines enhances SFK and STAT3 signaling. On the other hand, TCPTP reconstitution in human breast cancer cell lines severely impaired cell proliferation and suppressed anchorage-independent growth in vitro and xenograft growth in vivo. These studies establish TCPTP's potential to serve as a tumor suppressor in human breast cancer.

T cell protein tyrosine phosphatase attenuates T cell signaling to maintain tolerance in mice.

Many autoimmune diseases exhibit familial aggregation, indicating that they have genetic determinants. Single nucleotide polymorphisms in PTPN2, which encodes T cell protein tyrosine phosphatase (TCPTP), have been linked with the development of several autoimmune diseases, including type 1 diabetes and Crohn's disease. In this study, we have identified TCPTP as a key negative regulator of TCR signaling, which might explain the association of PTPN2 SNPs with autoimmune disease. We found that TCPTP dephosphorylates and inactivates Src family kinases to regulate T cell responses. Using T cell-specific TCPTP-deficient mice, we established that TCPTP attenuates T cell activation and proliferation in vitro and blunts antigen-induced responses in vivo. TCPTP deficiency lowered the in vivo threshold for TCR-dependent CD8(+) T cell proliferation. Consistent with this, T cell-specific TCPTP-deficient mice developed widespread inflammation and autoimmunity that was transferable to wild-type recipient mice by CD8(+) T cells alone. This autoimmunity was associated with increased serum levels of proinflammatory cytokines and anti-nuclear antibodies, T cell infiltrates in non-lymphoid tissues, and liver disease. These data indicate that TCPTP is a critical negative regulator of TCR signaling that sets the threshold for TCR-induced naive T cell responses to prevent autoimmune and inflammatory disorders arising.