In a recent post I commented on the byproducts of dehydrogenative decarbonylation, namely H₂ and CO.

I wondered whether this gas mixture, syngas, could be used in a subsequent reaction. This would improve the atom efficiency of the reaction and potentially also improve the safety (of both the syngas-producing and syngas-using reactions). Both are goals of green chemistry and I especially appreciate avoiding rolling cylinders of toxic and/or flammable gases around the lab.

After some digging I found some cool stuff, the most recent being the work from the Andersson group at Upsalla wherein the syngas produced from polyol deoxygenation is used in a subsequent hydroformylation of styrene. In addition, I found Madsen’s excellent paper, which includes lots of interesting references (such as this, where the CO liberated from decarbonylation of cinnamyl alcohol is used in the Pauson-Khand reaction. Note that cinnamyl alcohol can trap the evolved H₂! Neat!).

In Andersson’s work the syngas-producing step is catalyzed by [Ir(cod)Cl]₂ with (S)-BINAP in Reactor A (similar to Madsen’s work). The hydroformylation is simultaneously catalyzed by a different molecule, Rh(H)(CO)(PPh₃)₃, in Reactor B.

Since both reactions are known reactions, the major contribution of this work is a proof-of-principle – ie, that subatmospheric syngas produced in one reaction can be utilized for a subsequent reaction. Moreover, I think the reactor design is straightforward and something any glass shop could handle.

And for the big picture, I find the atom efficiency of the dual reaction appealing, especially in the context of conversion of biomass to fuel and commodity chemicals where evolving a portion of your ‘fuel equivalents’ for every molecule of biomass would be wasteful. Since the decarbonylation reaction tends to require > 150 C, I hope to see better catalysts in the future capable of mediating this reaction at lower temperature. With the number of potential H-bonds present on a polyol, perhaps secondary coordination sphere effects would be a possible research direction.

We here at GreenChemBlog have not posted in a while, but still hope to post and are still looking for contributors to the blog. Posts might be a bit shorter going forward, though, in order for us to post more frequently.

I’ve expanded my reading recently to include a new ACS journal, ACS Sustainable Chemistry & Engineering. In the first issue is the above article, which highlights a few recent additions to the smart phone/tablet world that utilize green chemistry!

The focus of the above paper is mostly on “Green Solvents”, which the authors developed based on the ACS GCI Pharmaceutical Roundtable Solvent Selection Guide. It is freely available for iPhone, iPod touch and iPad. (I personally use the Android platform for which the very similar app “Lab Solvents” is available) The authors also cover a few other green chemistry apps, so read the paper for more info! I especially like the Process Mass Intensity feature of the Yield101 app, though this app is $5.

For Green Solvents, the app entry page features a list of common solvents displayed as their chemical structures. More desirable solvents are color coded with a green background; less desirable solvents are color coded with a brown background. Selecting a solvent molecule brings up a box that lists the chemical’s name, CAS number, as well as scores on a scale of 1 – 10 for each of the five following categories: safety, health, environment (air), environment (water), and environment (waste). The lower the number, the greener the solvent. In addition, the numbers are color coded with 1-3 displayed as green, 4-7 displayed as yellow and 8-10 displayed as red. Furthermore, the selection box includes easily chosen links to the ChemSpider Web site, the Mobile Reagents app, and the Mobile Molecular DataSheet for more information on the selected solvent.

The power of the app is how quickly one can obtain info that should impact solvent choices in lab or in the field (without having to retreat to your desk or a nearby computer). And it’s free!

In the future I hope to see these apps provide suggestions for alternative solvent choices (for example, this editorial provides an example of a table Pfizer uses for solvent selection). For instance, I already know dichloromethane is a poor choice of solvent from a green chemistry perspective. How about an alternative choice, Green Solvents app? Trifluorotoluene?

The use of biorenewables as feedstock chemicals for commodity chemicals as well as fuels requires mild, selective removal of oxygen-containing functional groups. This is in direct contrast to the production of these chemicals from petroleum products, which, at least for highly functionalized target molecules, necessarily involves oxygenation of hydrocarbons.

There are a large amount of methods development currently underway and I highlight the recent report from the Sadow group on the decarbonylation of alcohols under Rh catalysis. I think the described reaction is a good example of green chemistry, as the reaction is high-yielding, selective, and performed at room temperature under photocatalytic conditions. One serious drawback is the use of benzene as the solvent, although toluene works as a solvent in at least some cases.

Reasoning that photolysis would prevent catalyst inhibition by CO binding, the researchers first screened Rh(I) catalysts under photocatalytic conditions with the test substrate cyclohexanemethanol. Unfortunately, no cyclohexane was observed under these reaction conditions. The group then tested Rh and Ir compounds known for C-H activation, such as Cp*Ir(CO)₂ and Tp*Rh(CO)₂, and did observe cyclohexane for one of the tested catalysts, albeit in low yield (36 % NMR yield with Tp*Rh(CO)₂). CO and H₂ were also observed, consistent with the targeted alcohol decarbonylation reaction. Interestingly, using their previously reported rhodium tris(oxazolinyl)borate complex To^MRh(CO)₂ (1) improved the yield to > 95%. Furthermore, the related dihydride, To^MRh(H)₂CO (2) was roughly three times slower and the Ir complex To^MIr(CO)₂ was inactive for this reaction.

The reaction was successful for a range of aliphatic (e.g., methanol and ethanol) and benzylic primary alcohols and tolerated the presence of aryl flouro and ether groups as well as silyl groups. The authors note that benzylic alcohols are a central component of lignin. The reaction did not tolerate aryl nitro, chloro or ester groups.

The authors then explored the mechanistic details of the reaction, noting that small amounts of the aldehyde were detected during the reaction and that using the aldehydes directly provided the hydrocarbon under the catalytic conditions using either 1 or 2. Furthermore, the presence of CO  did not inhibit the reaction at 1 atm, but did inhibit the reaction at 5 atm. No aldehyde or hydrocarbon was detected in the absence of light.

The observation that catalysis with 1 and 2 proceeds at different rates suggests distinct intermediates, although no catalytic intermediates were detected in the authors’ present study. The catalytically active intermediate from 1 is likely To^MRh(CO) (3), provided by photolytic dissociation of CO. In contrast, photolytic CO dissociation from 2 would afford the dihydride To^MRhH₂ (4).

The hypothesis that CO dissociation is required for catalysis is supported by two observations: (i) the lack of thermal reactivity of 1 with alcohols and (ii) the reaction of To^MRh(H)₂(NCCH₃) with cyclohexanemethanol, which provides cyclohexane as well as 2 (likely via acetonitrile loss to give the proposed dihydride 4).

With this information, the authors propose the following, simultaneously-operative catalytic cycles. It is interesting to me that catalysis from 1 requires two photons per cycle, while just a single photon is required for 2. I’m not sure how you could experimentally confirm this, though.

In summary, the Sadow group describes a new application for their rhodium tris(oxazolinyl)borate complex as a photocatalyst for the decarbonylation of alcohols. Although I didn’t cover it, they also apply this catalyst to the photocatalytic conversion of amines to imines, a reaction which follows a different route than alcohol decarbonylation – namely, dehydrogenation of the primary amine to form an aldimine followed by reaction with a second equivalent of amine to form the observed imine product with the presumed loss of ammonia. It is thus noteworthy that their catalyst can handle both oxygen and nitrogen functionality. Going forward, I hope to see this work extended to diols and polyols.

Do you want to share any recent papers you think are cool examples of green chemistry, whether they are “marketed” as such or not?

Do you want to help promote and shape the national discourse around green chemistry?

Do you want to help make sure the awesome science behind green chemistry doesn’t get lost in all the hype?

If you said “Yes” “Yes” or “Yes,” you have come to the right place! GreenChemBlog is opening its doors to new contributors!

If you are interested, shoot us an e-mail at greenchemblog@gmail.com and we’ll help get you started.

In my very un-scientific survey of the green chemistry-branded journals, I see way more new methodologies than I see total syntheses. I hope to single-handedly change this, and show how green a total synthesis can be by writing about the awesome recent synthesis of prostaglandin PGF_2α by Aggarwal and coworkers. First, a few words on the target molecule. Being hormones, prostaglandins such as PGF_2α are involved in tons of biological processes. Interestingly, instead of being synthesized by some important gland and acting in far-off regions of the body as are endocrine hormones, they are autocrine or paracrine hormones and are synthesized “on-site.” The first structural characterizations of prostaglandins came in the 1960s, some 30 years after their initial discovery. Soon after, they became the subject of numerous syntheses, the first of which was achieved by E. J. Corey in 1969. A series of syntheses followed, but even 40 years later, the structurally-related glaucoma drug latanoprost is synthesized in 20 steps using Corey’s 1969 prostaglandin strategy.

That’s right, the prostglandin structural motif is medicinally relevant. So, not only would an improved synthesis be cool from a fundamental science perspective, it might actually be moved into industrial production and have an immediate impact!

Aggarwal and coworkers’ strategy hinges on the conjugate addition of a functionalized vinyl side chain to enal 1. The stereochemistry of the addition would  be controlled by the cis orientation of the bicyclic lactol. This bond disconnection is what distinguishes the Aggarwal strategy from all others. Awesome idea, but first they had to synthesize enal 1. This, rightfully so, seems to have occupied the majority of the authors’ efforts.

After initial failures, a few model systems, and probably lots of bad nights in lab, they developed an efficient asymmetric synthesis of 1 via the proline-catalyzed aldol dimerization of succinaldehyde (2).

The fact that their initial “let’s just try this thing” attempts yielded only succinaldehyde oligomers is no surprise when you look at the crazy-sensitive nature of some of the reaction intermediates and the insane variety of undesired reaction pathways available to them. As shown below, initial aldol reaction of succinaldehyde (2) yields trialdehyde 3, which can only yield 1 if it forms the correct hemiacetal and undergoes a second aldol reaction followed by dehydration without reacting with another succinaldehyde molecule. Quite a tall order.

Model studies suggested that the initial aldol condensation was proceeding without a problem. They then synthesized another model system to investigate the second aldol reaction and dehydration steps. Sure enough, treatment of the lactone shown below with proline yielded only trace amounts of product, suggesting that the final two steps leading to the enal were indeed the problem. They eventually discovered an ammonium catalyst that affects the desired reaction, as shown below. Why this catalyst works is a mystery to me and is not commented on by the authors.

Heading back to the actual system, the authors subjected succinaldehyde to proline and ammonium mixed catalysis. This is where things get a bit crazy. They performed the reaction in the presence of proline for a certain amount of time to affect the initial asymmetric aldol reaction, after which they threw in the ammonium catalyst to affect the second aldol reaction and dehydration.  After optimizing for all sorts of variables (time for proline catalysts, time for mixed catalysis, etc), they came upon these conditions (e.r. determined on the methyl acetal).

Okay, 20% yield sounds bad, but to get to their key intermediate in one step from commercially available starting materials, it’s not bad! I’m really surprised at the high concentrations used here, especially when their problematic step was an intramolecular reaction that was competitive with further intermolecular reaction with more starting material. The authors do state that oligomeric byproducts were indeed formed but could be removed by filtration (awesome!). Also, they stated that the high concentration conditions “greatly facilitated scale-up of the reaction.”

So now that they’ve got their key intermediate in large quantities, I’ll just skip straight to the complete synthesis as shown below. Exactly according to plan, the installation of the vinyl side chain as well as the ketone reduction both proceed with complete stereoselectivity. Huzzah!! Somehow I had never seen a mixed vinyl cuprate like this one used before, but the Bruce Lipshutz group discovered way back in 1984 that the 2-thienyl group in such complexes could act as a “dummy” ligand and not competitively add to electrophiles. Cool! The final step also serves as an example of old chemistry being used and refined in the total synthesis arena, as a similar Wittig olefination was used by Corey way back in 1969.

Notice they were able to make 1.9 grams of this stuff, in seven linear steps. From what I hear, syntheses that can yield grams of product are very likely to be amenable to making kilograms of product. So, I will be on the lookout for this strategy actually being put to use in the large-scale manufacture of latanoprost and other prostaglandin-like drugs.

We’ve posted before on iron-catalyzed reactions (see here for a recent post) as greener alternatives to more traditional platinum group catalyzed reactions. However, even iron has toxicity concerns as described in this paper from National Taiwan University on the toxicity in medaka fish of  zerovalent iron (nZVI) nanoparticles (NPs). This is particularly pertinent research in light of the increased usage of iron(0) nanomaterials in remediation.

The study investigates the effects of four different iron dosing ‘solutions’ on the molecular, cellular and organismal health of medaka larvae: (i) carboxymethylcellulose stabilized nZVI (CMC-nZVI), (ii) non-stabilized nZVI (nZVI), (iii) magnetite NPs (nFe₃O₄), and (iv) soluble Fe(II).

They first characterize the dosing solutions. The sizes of their nanoparticles are 75 nm, 25-75 nm, and 27 nm for CMC-nZVI, nZVI, and nFe₃O₄ respectively. The zeta potentials were measured to show, not surprisingly, that the CMC-stabilized particles are much more stable to aggregation than the non-stabilized nZVI.

Interestingly, of the four iron dosing solutions, CMC-nZVI has the most significant impact on the level of dissolved oxygen, decreasing it to zero where it remained for 12 hours. Furthermore, this aerobic oxidation of CMC-nZVI leads to a release of 45 mg/L of soluble Fe(II) in 10 min from an initial concentration of 100 mg/L CMC-nZVI as well as an increase in reactive oxygen species (ROS). In contrast, nZVI and nFe₃O₄ are 20 – 40 % aggregated within 10 min and release less than 20 mg/L of Fe(II) during this time. Only nZVI induces the production of ROS with nFe₃O₄ and soluble Fe(II) showing no increase in ROS relative to the control. The following figure details these findings for CMC-nZVI; analogous graphs are found in the supplementary information for the other solutions.

As for toxicity, the most toxic solution for medaka larvae (as judged by % mortality over the course of several days exposure to the solutions normalized by the concentration of Fe atoms) is the soluble Fe(II) solution. The second- and third-most are the CMC-nZVI and nZVI solutions, respectively, with nFe₃O₄ showing little toxicity. It is noteworthy that the ‘stabilized’ particles are more toxic than the ‘non-stabilized’ particles to the larvae.

In contrast, the order of bioaccumulation was nFe₃O₄ > nZVI > CMC-nZVI ~ Fe(II) (as measured by BCF or bioconcentration factor = iron concentrated in larvae / measured total iron concentration of the dosing solution). This order, interestingly, is the opposite of the toxicity order displayed above.

With these results, the researchers then investigated the molecular effects of the dosing solutions. Though the study is very detailed and worth reading more closely, the most dramatic effect was of CMC-nZVI on the expression of superoxide dismutase (SOD), one of nature’s enzymatic responses to oxidative stress, which increases approximately four-fold in the presence of CMC-nZVI. They also studied the effects of the dosing solutions on the gene expression of the antioxidant enzymes catalase (CAT, which catalyzes the decomposition of hydrogen peroxide) and glutathione S transferase (GST, which catalyzes the transfer of reduced glutathione groups to electrophiles), though the changes in expression were less significant than for SOD. The source of the differing impact on expression of the different enzymes is not clear to me, but I’m not shocked the genetic response is more complicated than a simple increase in gene expression in the presence of oxidative stress.

The researchers next studied the effects of the solutions on the activities of the enzymes. For CMC-nZVI there is little effect on the actual activity of SOD over seven days of exposure. This indicates that while CMC-nZVI has dramatic effects on gene expression it has no effect on the activity of the proteins coded by those genes. Furthermore, the intracellular ROS also is not affected by exposure to either CMC-nZVI or nZVI, suggesting the balance of active antioxidants can handle the oxidative stress from ROS at least over the course of several days.

In summary then, the NP toxicity appears due to a combination of hypoxia, the release of Fe(II),  and other “NP-specific toxicity”. If true, the context of the application of these NPs – which are used for remediation in both aquifers and soils – then becomes important. For instance,  the release of soluble Fe(II) would probably be of higher concern in an aqueous environment than in sediment. Moreover, the transport of particles between aquifers and sediments, especially in light of the dynamic nature of these solutions, needs to be investigated.

Going forward, I am interested in the identification and description of the NP-specific modes of toxicity and how that differs with stabilization, as the CMC-nZVI solutions are more toxic than the non-stabilized ones to medaka larvae. NP stabilization with surface ligands is a common method to prevent aggregation, but, at least in this case, also increase their toxicity via the release and dissolution of surface accessible iron atoms upon oxidation.

I occasionally run reactions in water, and it is awesome. I LOVE not worrying as much about cancer. Unfortunately many interesting chemicals are simply too hydrophobic to allow reactions to be run in aqueous solution. As anyone who has ever scrubbed a greasy pan will know, one way to get around the solubility problem is with soap. Also called emulsifiers, surfactants, amphiphiles, a soap by any name is pretty much the same thing in my mind (though others will disagree I’m sure). Molecules with hydrophilic and hydrophobic ends can form micelles in water, creating variously-shaped and sized particles with hydrophobic cores. Using those hydrophobic cores as reaction media is a concept known as micellar catalysis. The Lipshutz group was not the first player in this arena, but they have been at it for some time. They recently teamed up with the medicinal chemistry company Kalexsyn and came out with a new amphiphile that caught my eye, dubbed TPGS-750-M. 

I came across this paper while perusing the most recent issue of “Green Chemistry Articles of Interest to the Pharmaceutical Industry” in Organic Process Research & Development, which I highly recommend checking out.

TPGS-750-M is a next generation version of a class of amphiphiles initially synthesized to solubilize the hydrophobic dietary supplement coenzyme Q10. The history of this amphiphile class can be found here, but for now I will simply focus on TPGS-750-M. This amphiphile is cool for three main reasons. First, it seems unlikely to be toxic–it’s hydrophilic portion is comprised of poly(ethylene glycol), which is used commonly in medicine, and its hydrophobic portion is comprised of α-tocopherol, an E vitamin. Awesome. Second, its synthesis is straightforward and high-yielding. Awesomer. Third, its use allows for numerous reactions to be carried out in water at room temperature. Awesomest.

I won’t dwell on the synthesis because it is so straightforward, but check it out:

We’ve already touched on why amphiphiles like TPGS-750-M enable organic reactions to be carried out in water by providing hydrophobic pockets within which those reactions can take place, but we haven’t discussed why they can enable those reactions to occur under lower temperatures than when they are carried out in traditional organic solvents. This part is super neat. The thinking goes that the hydrophobic reactants preferentially partition into the hydrophobic micelle, where the local concentration of reactants can be HUGE, greatly increasing the reaction rates. Thus the term micellar catalysis. This effect is not unique to TPGS-750-M, but as far as I can tell it is the most impressive micellar catalyst out there, as you will soon see.

Check out all these Pd-catalyzed couplings!

All reactions were carried out in H₂O with 2-5 wt % TPGS-750-M at room temperature. All reactions were complete in 4-25 h. Pd sources and ligands varied.

In addition to facilitating these Pd-catalyzed reactions, the amphiphile enables Ru-catalyzed olefin metathesis reactions to be carried out in water using Grubbs 2nd generation catalyst.

Definitely the awesomest thing this amphiphile can do is enable these Negishi-like couplings to occur in water.

This is cool because these reactions proceed via highly water-sensitive organozinc reagents formed in situ from alkyl or vinyl halides and metallic zinc. I’m going to quote the article authors here, because this sentence is awesome:

Crucial to success are the relative rates of organozinc halide formation, transmetalation to palladium, and aqueous protonation of RZnX, all controlled such that RZnX is not formed in situ too rapidly so as to avoid quenching by eventual exposure to the nanoparticle-surrounding water.

The paper goes on to use this amphiphile to facilitate a number of other reactions that you can go check out yourself.

The authors also compare  TPGS-750-M to other similar amphiphiles that possess different hydrophobic and hydrophilic portions and different linkers, and  TPGS-750-M generally matches or outperforms all these amphiphiles.

Why the different performance from such similar-looking amphiphiles? The answer to that question is not at all clear.

The authors performed some cryo-TEM experiments to try to figure out if differing performance is caused by differing micelle sizes or shapes, and they conclude that TPGS-750-M outperforms the other amphiphiles due to the larger (50+ nm) micelles of the former. I’ll throw these images up because pictures are awesome, but I concur with the authors that this is an area ripe for mechanistic study.

Cryo-TEM image of (A) aqueous PTS, (B) TPGS-750-M, and (C) TPGS-1000

In addition to micelle size, it’s not clear to me (or anyone else I think??) how the structure of the hydrophobic moiety can affect the reaction outcome. After all, it essentially provides the reaction solvent, so should have a significant influence. I hope to see lots more work in this area to satisfy my mechanistically-oriented brain.

In terms of scale-up and greenness one huge concern with running reactions in water is purification and recycling/disposal of the water. The authors address this issue by isolating the reaction products and catalysts via an aqueous/organic extraction. TPGS-750-M is preferentially soluble in water, so remains in the aqueous phase after this extraction. This allows the water from the previous reaction to be reused up to 8 times with little to no effect on the reaction outcome.

Neato!

As a synthetic chemist with little (actually zero) training in toxicology, it’s difficult for me to imagine how to design safer chemicals at the start of a project. I can avoid nasty solvents, use safer reagents, but when designing a new molecule I haven’t a clue of its potential toxicological impact. This is frustrating and as the authors of the above paper in Green Chemistry point out, “with the growing number of new chemicals being introduced into the market, it is not economically or ethically reasonable to assume that each can undergo systematic toxicological testing [...]“. Thus, possessing a set of easy-to-implement synthetic guidelines to reduce the toxicity of a synthetic target during the design stage, while maintaining (or better yet, augmenting) its function, is of high importance.

Recently, the Zimmerman group reported on guidelines for reducing acute aquatic toxicity and have now extended their work to chronic aquatic toxicity. This is an important next step because chronic toxicity studies are necessarily longer-term (and thus more resource intensive) than acute toxicity studies.

In the current work, they explore the relationships between 38 physicochemical properties of 865 chemicals with chronic aquatic toxicity toward three model organisms: the Japanese medaka, a cladoceran, and a green algae. The 38 properties include, for example, molecular weight, number of freely rotatable bonds, aqueous solubility, and number of hydrogen bond donors and acceptors. They first measure the correlation of single properties to toxicity  and find the highest correlation (as judged by the correlation coefficient; see table below) between toxicity and the octanol-water partition coefficient (log P_o-w) for both the medaka (O. latipes) and cladoceran  D. magna), with a less strong correlation for the algae (P. subcapitata). Additionally, the authors found a correlation between toxicity and the HOMO/LUMO gap (ΔE) for the algae. High correlations also exist between toxicity and the energies of the frontier molecular orbitals of the compounds (HOMO/LUMO).

Univariate correlation coefficients between properties and standardized chronic toxicity endpoint

These correlations are not too surprising, as the log P_o-w relates to how well a given molecule can partition from water to an organic medium (or across lipophilic biological membranes into an organism). The frontier molecular orbital energies, of course, relate to how reactive a molecule is (for example, a compound with a low-lying LUMO would be expected to react with nucleophilic amino acid side chains).

They next perform a multivariate analysis and find that for a 2-property model, the two most important properties are log P_o-w and either ΔE  or the HOMO energy for all three species. Interestingly, for a 3 property model, the third property depends on the species. For the medaka and cladoceran, the third most important property for toxicity prediction is the number of H-bond acceptors, while for the algae it is the LUMO energy.

To see how the four key properties impacted aquatic toxicity, the researchers explore their distribution (see the box plot below) against compounds grouped according to the EPA’s three levels of toxicity concern (high in red, medium in black and low in green). Consistent with the correlations described above, both log P_o-w and ΔE exhibit strong trends, with lower partition coefficients and larger ΔEs corresponding to lower levels of toxicity concern. In contrast, the HOMO and LUMO energies show a much less significant trend, suggesting ΔE is a better guideline than either the HOMO or LUMO energies when designing a molecule.

More powerfully,  scatter plots of log P_o-w vs ΔE provide threshold values for chronic aquatic toxicity. For all three species, the compounds of low chronic toxicity lie within one quadrant of the plot with boundary values of log P_o-w of less than 2 and ΔE values greater than 9 eV. For the three species, applying these guidelines eliminates 90 % of the investigated compounds classified in the highest category of concern by the EPA!

Lastly, the authors explore the outliers – for example, highly chronically toxic compounds that are wrongly classified by these new guidelines as ‘safe’. The discussion of the outliers is very detailed and worth reading (shown below are the results for the D. magna 504 hr assay), but one compound that caught my eye was hydrazine.

The guidelines place hydrazine in the ‘safe’ region of the plot, whereas the EPA level of concern for this compound is high (hydrazine is a known carcinogen in mammals). As the authors point out, there are a few possible explanations for any incorrect classification: (i) the computed log P_o-w or ΔE values are incorrect, (ii) the experimental toxicity threshold is inaccurate, or (iii) the design guidelines are inadequate for certain modes of action (MOAs) of certain toxicants. In the case of hydrazine, according to the authors, the MOAs are poorly understood in aquatic organisms making the last explanation potentially the correct one. They also note that the acute toxicity of hydrazine makes measuring chronic toxicity difficult.

These guidelines are obviously very exciting and easy to use (I think I could handle measuring log P_o-w and computing ΔE!). Going forward, I’m interested to see how the authors change the guidelines to include the outliers. In particular, I’m excited to see how they incorporate the biodegradation products of the compounds and whether that impacts the number of outliers using their guidelines.

“Continuous-Flow Synthesis of the Anti-Malaria Drug Artemisinin.” Lévesque, F.; Seeberger, P. H. Angew. Chem. Int. Ed.. 2012, 51, 1706-1709. DOI: 10.1002/anie.201107446

“Monitoring and Control of a Continuous Grignard Reaction for the Synthesis of an Active Pharmaceutical Ingredient Intermediate Using Inline NIR spectroscopy” Cervera-Padrell, A. E.; Nielsen, J. P.; Pedersen, M. J.; Christensen, K. M.; Mortensen, A. R.; Skovby, Dam-Johansen, T. K.; Kiil, S.; Gernaey, K. V. Org. Proc. Res. Dev. 2012, ASAP. DOI: 10.1021/op2002563

A little while back I wrote about an aerobic oxidation which was greatly improved by switching from a traditional round bottom flask setup to a continuous flow reactor – basically, continuous flow reactors are much better at handling oxygen, especially on scale.  But most of the advantages of the flow reactor were specific to that reaction, and it wasn’t clear to me how a flow process would improve a reaction that doesn’t use oxygen, or some other gas.  Fortunately, a lot has been published since then to help me get a handle on how continuous flow reactions can contribute towards greener processes.  In particular, this review covers continuous processing within a green chemistry context, and Organic Process Research and Developement has a continuous flow themed issue in their ASAP section, including this process-oriented review (speaking of OPRD, check out this recent editorial concerning solvent selection and green chemistry).  It turns out that flow chemistry can improve processes in a bunch of different ways, and it’s hard to get a sense for how this can work by just looking at one reaction.  So I’ll cover a few different reactions that illustrate different green aspects of continuous flow reactors.

One benefit of flow reactors is improved control over reaction temperature, due to reduced reaction volume at a given time, higher surface area, and the movement of the reaction mixture.  This is particularly helpful for very exothermic reactions, which often require cryogenic cooling to prevent runaway reactions – this type of cooling is very expensive and resource-intensive on a large scale.  One such reaction is described in a recent paper from AstraZeneca, in which a phosphinate anion adds into a glycine derivative.  The product of this reaction is an intermediate in the synthesis of a gastroesophageal reflux inhibitor drug candidate called AZD6906.

When the reaction was conducted in a batch reactor (equivalent of a large round bottom flask), the reaction had to be conducted at -78 °C to prevent side reactions, and two equivalents of the phosphinate anion were used because one equivalent was consumed by deprotonating the more acidic product.  The extra equivalent of phosphinate had to be separated by chromatography.  By conducting this reaction in a flow reactor, they were able to use one equivalent of the phosphinate (although they did use excess LDA to keep the phosphinate deprotonated)  and conduct the reaction at a slightly elevated temperature instead of -78 °C. What’s interesting is that they didn’t get more side products by conducting the reaction at higher temperature – the authors attribute this to the short reaction time, which is immediately followed by quenching.  I think it’s not just that the flow reactor dissipates heat better, but that it eliminates the local extremes in temperature and concentration that you get in a large batch reactor.  So if you have a hot spot in a reactor, the desired reaction will be complete and a slower side reaction could begin, while in a cold spot in the reactor the desired reaction hasn’t finished yet.  Even though the flow reactor is at a higher temperature, the temperature control is better, so the side reactions don’t have a chance to get started before the reaction is quenched.  Hand-wavey explanations aside, they were able to reduce the energy requirements for this reaction by replacing the original cryogenic conditions with mild heating, and the product was sufficiently clean that no purification was necessary.

The second case is the oxidative photochemical conversion of dihydroartemisinic acid into artemisinin.  Amy just wrote about an efficient semi-synthetic approach to dihydroartemisinic acid, so this is the final necessary transformation to make semi-synthetic artemisinin.  The first step in this sequence is the Ene reaction of dihydroartemisinic acid with photochemically-generated singlet oxygen to form an allylic hydroperoxide.  The acid-catalyzed Hock rearrangement  forms a transient enol, which reacts with triplet oxygen to form another hydroperoxide.  Artemisinin is formed through an acid-catalyzed series of acetal closures and a final condensation.

The problem is that existing methods are pretty lousy – this two step method proceeds in 17% yield, requires cryogenic temperature, and takes four days to complete.  Higher yields (34%) are obtained in this three-step conversion of a similar molecule, but each intermediate is isolated and purified along the way, generating extra waste.  Another problem is that photochemical reactions are difficult to scale up because light doesn’t penetrate well in large reactors.  This, in addition to problems associated with using oxygen, is why the Seeberger lab chose to develop a continuous flow process for this transformation.  Their setup actually sounds pretty simple – they pump their oxygen-pressurized reaction mixture through clear FEP tubing, which is wrapped around a water-jacketed mercury lamp.  They didn’t include a picture in their paper, but it sounds similar to this one that I found in a different paper, which uses blue LEDs inside a reflux condenser as the light source.

Passage through the photoreactor gives them the first hydroperoxide intermediate, and treatment of that intermediate with trifluoroacetic acid and oxygen under thermal conditions produces artemisinin.  Both of these steps were optimized individually, and then they put the two processes together into a single flow reactor:  first the reactant, triplet sensitizer (tetraphenylporphyrin), and oxygen are pumped through the photoreactor at room temperature.  Second, TFA is introduced to the reaction flow, which is fed into another reactor at 60 °C, where the Hock rearrangement, the second reaction with oxygen, and the final rearrangements occur.  The yield for the entire process is 39%, and their setup can produce 200 mg of artemisinin per day.  Compared to the older methods, this reaction is higher yielding, has fewer purification steps, and doesn’t require cryogenic cooling – big improvement.  Flow chemistry is easier to scale up than batch processes, so hopefully this work, combined with the semi-synthesis of dihydroartemisinic acid, represents a reasonable starting point for artemisinin production.  One disappointing detail of this work is that their reaction solvent is dichloromethane, which was necessary for the photochemistry to reduce the risk of fire, and was identified as the optimal solvent for the thermal reaction as well.  Given these constraints, I suppose it’s lucky that they weren’t forced to switch solvents, but there is always room for improvement.

The final reaction is a simple Grignard addition to a thioxanthone, forming an intermediate in the synthesis of the antipsychotic drug zuclopenthixol.  The authors had already figured out a continuous flow version of this reaction, and in their process optimization they discovered that impurities formed when excess Grignard reagent was present.

If I was running this reaction I would simply use excess thioxanthone, but the authors didn’t want to take that hit in mass efficiency, so instead their aim in this recent paper was to achieve a 1:1 stoichiometry as accurately as possible.  They achieved this by monitoring the reaction by inline NIR spectroscopy, which provides nearly real-time analysis of the reaction components.  Engineering this inline monitoring took a lot of calibration and validation that I didn’t really follow, but the upshot is that they were able to detect any excess reactant (either the Grignard or the thioxanthone) at the end of the flow reactor, and use that real-time feedback to adjust the flow rates until they were optimal.  This type of self-regulation allows the reactor to adjust to different concentrations of the Grignard reagent, which typically vary from batch to batch.  It wasn’t really made clear if their yield or product purity were improved by the continuous monitoring (a lot of work for nothing if it didn’t), but I thought it was a cool paper nonetheless.

Ok, so lots of flow chemistry, and lots of reaction-specific ways that it can improve a process. Two things stuck out to me: that flow chemistry is really nice for handling reagents that don’t diffuse well through large volumes (light and gases), and that flow reactors are good at minimizing local changes in certain conditions (temperature, concentration) that lead to side reactions.  I’d be interested to know whether there are some situations where a batch reactor would be preferred to a flow reactor – none are mentioned in anything that I read.  In any case, flow chemistry seems to be getting more popular every day, and I’m sure there will be many more green continuous flow processes developed in the near future.

Malaria, caused mainly by the parasite Plasmodium falciparum, leads to nearly a million deaths and 250 million new infections each year. The sesquiterpene lactone endoperoxide artemisinin, derived from Artemisia annua, is very effective as an antimalarial drug, and widespread resistance hasn’t yet developed. Artemisinin is the only high-volume drug that is still isolated by extraction from its native plant producer in a low-yielding (around 10 μg per g plant material), resource-intensive process that uses volatile solvents (most commonly hexane).

Artemisia annua. Photo credit: Jorge Ferreira via Wikimedia Commons.

As a result, supplies of the drug are short, and those who need it often can’t afford it. The development of new processes for artemisinin production would therefore advance both public health and green chemistry interests. Total synthesis of the drug hasn’t been considered as a viable alternative because of low yields, but a lot of effort has been directed toward developing semi-synthetic sources of artemisinin using a combination of microbial fermentation and chemical synthesis. Toward this end, the Keasling lab reported a few years ago that they had constructed a biosynthetic pathway for the artemisinin precursor amorpha-4,11-diene in yeast with yields of ~200 mg/L—already impressive given the complexity of the molecule. Amorphadiene synthase (ADS) comes from Artemisia annua; the rest of the genes are from yeast. Here is the existing pathway:

A recent paper from Amyris Biotechnologies reports optimization of this pathway for industrial production to yield 40 g/L of amorpha-4,11-diene along with a chemical process that completes the production of semi-synthetic artemisinin.

The first step was to increase flux through the pathway by tuning gene expression. While previous strains had only overexpressed HMG-CoA reductase from the mevalonate pathway that leads to the amorpha-4,11-diene precursor farnesyl pyrophosphate, the researchers at Amyris elected to construct new strains that instead overexpress every gene leading to farnesyl pyrophosphate by placing them under strong, galactose-regulated promoters. Because expression of HMG-CoA reductase (tHMG1 in the above figure), which catalyzes the first committed step in the pathway, had previously been shown to be limiting, they also added two additional copies of the gene. These adjustments in gene expression led to a fivefold increase in amorpha-4,11-diene production.

Low yields are not the only limiting factor for commercialization of pathways from academic labs. Some of the strategies that are often used during initial pathway construction to allow high-level production of foreign proteins in microbial hosts add extra expense and complexity that is undesirable for economical large-scale production. One such problem that Amyris needed to overcome was that the original pathway was set up such that the presence of galactose was required for expression of the pathway genes. The galactose-requiring expression system adds a specific DNA sequence in front of foreign genes that is normally found in front of the genes required for yeast to grow on galactose. This sequence is recognized by a DNA-binding protein, Gal4, which also binds to some of the cellular machinery needed for transcription, ultimately leading to high levels of protein production. In the absence of galactose, a galactose-sensing protein, Gal80, binds to Gal4, blocking its interaction with the transcription machinery and shutting down protein production. This system is popular in metabolic engineering because it provides a convenient way to trick yeast into expressing any protein you want by co-opting the DNA sequence that controls expression of galactose breakdown genes. The drawback, however, is it requires the use of galactose as a carbon source, and that’s too expensive to be used in industrial-scale fermentations, even when yields are higher.

To get around the need to use galactose as a carbon feedstock while still taking advantage of the high levels of gene expression that Gal4 can promote, the scientists at Amyris deleted Gal80, the protein that blocks interaction of Gal4 with the transcription machinery in the absence of galactose. Without Gal80 to interfere with transcription, HMG-CoA reductase and amorphadiene synthase are expressed at high levels regardless of the carbon source. This modification produced a yeast strain that made amorpha-4,11-diene at similar levels to the original, but using glucose, which is cheap enough for large-scale applications, as a carbon source.

Fermentation conditions were then optimized for this strain. To direct carbon flux away from biomass accumulation (i.e., cell growth) and toward synthesis of amorpha-4,11-diene, phosphate levels in the yeast media were limited, increasing titers to 5.5 g/L. Based on the ability of yeast to convert ethanol to acetyl-CoA, the yeast were fed with mixtures of ethanol and glucose in an attempt to increase the cytosolic concentration of acetyl-CoA. This indeed increased yields significantly, to 16.5 g/L. Even better results were obtained by feeding with pure ethanol, which increased yields to the reported 40 g/L, corresponding to an impressive 20% yield from carbon.

Finally, to bridge the gap between amorpha-4,11-diene and artemisinin, they developed a synthetic strategy to convert amorpha-4,11-diene to dihydroartemisinic acid, the precursor of artemisinin.

Use of 9-borabicyclo[3.3.1]nonane (9-BBN) allowed selective hydroboration of the exocyclic double bond, giving dihydroartemisinic alcohol as an 85:15 mixture of epimers, with the desired epimer making up the bigger portion of the mixture. The researchers opted for a two-step oxidation to make the carboxylic acid, avoiding the use of toxic chromium-based oxidants and undesired reactivity of the double bond. Dihydroartemisinic alcohol was therefore oxidized to dihydroartemisinic aldehyde with sulfur trioxide-pyridine, and the aldehyde was taken to the acid with sodium chlorite in DMSO to give dihydroartemisinic acid in 48% overall yield from amorpha-4,11-diene.  Luckily, the step that looks the most difficult to achieve synthetically, conversion of dihydroartemisinic acid to artemisinin, is believed to occur non-enzymatically in nature, requiring only molecular oxygen and light, so it seems very feasible to produce artemisinin itself using this synthesis, although, disappointingly, this was not explored in the paper. It will be exciting to see how the pathway is completed for industrial production.

As Courtney pointed out in his post on microbial butanediol production, a lot of papers from academic labs on microbial chemical production seem too low-yielding to ever be commercialized. This paper provides a great example in which what started out as a low-yielding pathway is now poised to change the way an important pharmaceutical is produced. The first doses of semi-synthetic artemisinin, which presumably will be produced by a pathway similar to the one developed in this paper, are slated to become available later this year through a partnership between Amyris, The Institute for OneWorld Health, and Sanofi-Aventis. In addition to addressing a public health concern and providing a platform for greener production of artemisinin, the yeast strains and methods developed in this paper may also have an impact on the production of many other chemicals. Only the final enzymatic step is specific to production of amorpha-4,11-diene, so it’s possible that the same yeast strain could easily be engineered for production of other terpene-derived molecules like flavors, fragrances, and advanced biofuels.