Monday, August 22, 2016

Cross-Couple While the Iron is Hot

We are delighted to announce that our recent work on Fe-catalyzed cross-couplings of redox-active esters (RAEs) has recently been published in JACSWe'd like to share with you how we started out, developed this project, and how things evolved from the first discovery.

During the past year our group has been involved in using RAEs as coupling partners in cross-coupling. We have developed a variety of coupling reactions to make the route to C–C bonds easy and practical to anyone regardless of skill or level of funding. The reactions with RAEs allowed for the first time the use of any alkyl carboxylic acid as coupling partner. The acid is primed for reaction similar to the activation for amide coupling using simple phthalimide derivatives (NHPI or TCNHPI) or peptide coupling agents (HATU or HBTU). The use of a cheap and abundant Ni catalyst does the rest of the job. These reactions revealed very interesting features which were already described in previous blog posts here and here at Openflask.

We were very aware that our initial reports on Ni still presented some limitations which needed to be addressed. It's always fun to read the comments and concerns on Derek Lowe's blog :) The story with Fe started by trying to address those drawbacks and provide solutions to those chemists out there, which need molecules synthesized easily and in high yield during a coffee break.

Up until now, all RAE couplings we developed were restricted to the use of 10 or 20 mol% loading of Ni (sometimes as low as 5%). FDA considers Ni as class 2A toxic metal impurity, which makes its use in pharma industries less attractive (at least on process scale). In addition, these methodologies posed other issues: occasionally long reaction times for the cross-coupling, the use of isolated RAEs to obtain the very best yields, and in our Suzuki coupling, diluted conditions were also required.

All these issues were then considered in the lab and we decided to tackle these challenges by taking a closer look into the mechanistic aspects of the reaction. We knew this would not be an easy task but we were lucky to gather a team with different backgrounds and expertise to put together this nice story.

Everything started with a simple question, which would solve most of the drawbacks in our Ni-catalyzed reactions: What about Fe? Our lab has been playing around with Fe for quite some time now and we all know the advantages of using Fe over Ni.
What about Fe??
In the past decades chemists have been trying to identify what are the operating mechanisms in Fe-catalyzed cross-couplings. The literature is brimming with papers trying to catch the right intermediate… However, what it seems to emerge as a common theme is that oxidative addition of alkyl bromides and iodides proceeds via SET of a low-valent Fe complex. I believe you all know where I am going here… We simply wondered whether Fe could actually undergo SET to our RAEs, inducing radical fragmentation. If this was possible, a plethora of opportunities would open up by combining the RAEs and known reactivities for Fe catalysis.

Indeed, the answer to the question “What about Fe?” seemed to be “Why not!”. Jacob  tried this reaction using TMEDA and obtained around 30% yield. He graciously shared those results with us before moving to another exciting project (that you can read about in a few months). Fumi (a tremendously talented visiting scientist from Daiichi Sankyo) set up the first reaction using phosphine ligands and 60% yield of product was obtained!! But there is more… it not only afforded good yield of cross-coupling product, but it was extremely fast and proceeded in less than 1 hour! A little bit of tuning identified Bedford’ and Nakamura’s Fe system (Fe(acac)3/dppBz) as optimal for these reactions.

However, it was clear since the very beginning of the project that we did not want to come up with “just the same but with Fe”. What we wanted is to improve our previous work by addressing all the issues we had until now with the Ni chemistry. To make this point very clear, we decided to directly compare Ni and Fe in every compound reported in the paper. In this manner we wanted the reader to contextualize the research and simplify the choice of the appropriate conditions. 

We were really thrilled to observe that a simple activation of the acid by HATU at rt afforded comparable yields as when using the isolated RAE. Unlike our experience with Ni-catalysis, HATU, TCNHPI, NHPI or HBTU all performed similarly. A very interesting feature of this reaction is that you can run the Negishi coupling with pretty much any solvent at reach in the lab. We used toluene, benzene, THF, 1,4-dioxane, DCM, DMF, hexanes, Et2O… and all of them afforded the coupling product in good yields. The difficulty came to select the solvent to carry out the substrate scope. The source of Fe did not matter either… FeCl3, FeCl3 hydrated or Fe(acac)3, valyrian steel, they all performed equally. We bet the reaction would work as well with a piece of the iron throne.

Another interesting feature of this reaction is how fast these couplings are... even at –20 ºC. It is incredible to observe that once you add the zinc reagent into the Fe/ligand/substrate mixture you just have to wait only if you like waiting. The reaction is finished by the time you dispose of the syringe. Simple aqueous workup, evaporation and short chromatography will give you the pure cross-coupling product.
Selected scope of >40 examples

We were very pleased to see that these conditions were very general across a wide range of substrates. Primary and secondary acids performed extremely well. Natural products and drugs were all arylated easily. As in any Baran lab project, pyridines are mandatory :)
During the exploratory scope of the reaction, our “human molecular machine synthesizer” Tie-Gen, also managed to obtain secondary-primary couplings with alkyl zincates. However, we want to warn practitioners that the primary alkyl-alkyl coupling is still beyond the scope of these conditions and more screening is needed. The main byproducts observed when attempting the primary alkyl-alkyl coupling were beta-hydride elimination of both partners in addition to the decarboxylated Barton-type product from the acid. If one wants to perform such coupling, we recommend having a look at our previous method using Ni. It is extremely efficient.

Another important feature of these particular Fe conditions is the possibility of arylating tertiary carboxylic acids. Previously, arylation of such acids was attempted for long time albeit unsuccessfully… We were really happy in the lab when we managed to arylate a [1.1.1] bicyclic system bearing a carboxylic acid in 35% yield. Notably, this is a bond formation that our collaborators at Bristol-Myers Squibb have been eying for years now.

We were thrilled to find that these conditions now allow another type of nucleophile to be introduced in our repertoire: welcome aryl Grignards. We had the idea of using Grignards for the cross-coupling for quite some time now but using Ni, ketone formation with the RAE was always problematic. And we kept failing dramatically… With Fe however, the extremely fast reaction rates observed allow for the integration of aryl Grignards without any ketone formation. It is interesting to mention that Grignards react at faster rates with the Fe catalyst than with the starting RAE. This result is consistent with Fürstner’s cross-coupling with low-valent Fe species where even isocyanates are tolerated.

Grignards work!
At this point, our Austrian postdoc Laurin took a step forward and decide to explore the limits of this system. He decided to subject the terrible beast of cubane to our arylation conditions; cubane is known to be an extremely sensitive material in the presence of transition metals. We were extremely delighted when we saw the GCMS of that reaction… a peak for phenylated product was there! Indeed, Laurin managed to put a phenyl ring into a cubane!!!!

To contextualize this result, a deep search in the literature revealed seldom methods for the synthesis of arylated cubanes. They seem to be restricted to rather bizarre reactions, such as Pb-promoted oxidative Minisci, where the regioselectivity is not controlled. 

Remarkable cubane examples
This modular protocol now allows the introduction of a wide variety of aromatic rings as we demonstrated in a 3-step double arylation of cubane scaffolds. These structures were obtained in reasonable yields and they turned out to be crystalline. We are so proud of these bis-arylated bioisosteres that the X-ray pictures are hanging on the walls in our offices as one of our best achievements :) (Proof of it just below).

Beautiful X-rays
Methods at the Baran lab are conceived with the ultimate goal of being applied at the industrial level. For this reason, the Fe team tackled a question that has been around since we started developing the chemistry of RAEs: can we apply this method in the context of process chemistry (it's already used in medicinal chemistry)?

One of the Figures in the manuscript resumes the work involved in the development of both a medicinal and a process chemistry approach. Since this protocol is easily scalable, no problem was found for the medical chemistry approach when using 1.4 g of starting carboxylic acid. Simple aqueous work-up followed by quick chromatography afforded pure cross-coupling material.

Real scalability
A critique we heard over and over again from process chemists associated with this chemistry was the high catalyst loading of toxic Ni salts, difficulties associated to its removal and dilute conditions for the Ni-catalyzed Suzuki. Our visiting process chemist Fumi gathered the team and said: “I think we can do better”.

Fumi ran the Fe cross-coupling reaction with just 1 mol% Fe and after a series of solvent exchanges and washes, followed by a crystallization, gave us a 61% yield of product. To our excitement, the purity of the compound was 99%. And since we are in the realm of non-toxic Fe, there is no regulatory limit for Fe impurities according to the FDA.

For those process chemistry readers, another feature of this chemistry is that a very small exotherm was obtained when running the reaction at 1.4 g. The internal temperature only rose from 1.5 to 19 ºC (note that most of the temperature increase came from addition of room temperature solution)! So we hope such small temperature change will pose no problem for scaling up cross-couplings with Fe and RAEs in the near future.  We also got the seal of approval from our process chemistry collaborators at BMS (thanks Gardner and Darryl!) who performed several examples in the table.

We hope people will find immediate applications for this interesting cross coupling. We hope you enjoy reading the manuscript and as always, if you have any comments/suggestions/criticism/questions feel free to contact us here directly (anonymous posts are welcome).

Pep and the Fe team

p.s. Check out the SI with photos, FAQ, and even some flow-charts to help practitioners :)


  1. I love Baran group substrate tables. So big, so colourful, so busy! Congrats :)

  2. I realize that criticizing the practicality of methodologies in JACS is often unfair, as that’s not really a criterion for publication in that journal. However, when a paper uses its practicality as a major selling point, then criticism along those lines is justified. Compared to earlier iterations of this reaction, certain aspects of the chemistry have been improved, and it is laudable that the group has invested time in trying to beat this chemistry into a more practical form. However, I still have a few concerns about how process-friendly this reaction really is:

    1. A recent analysis of the FDA database of approved unique small-molecule drugs by the Njardarson group found that 84% of these compounds contain nitrogen, and 59% include a nitrogen-containing heterocycle (J. Med. Chem., 2014, 57, 10257–10274). Of the compounds in your substrate table (Parts A and B), approximately 24% contain nitrogen (9/38) and just 3% (1/38) contain nitrogen-bearing heterocycles. These are not bad numbers for a methodology paper per se, but for a reaction that you’re explicitly marketing to industrial chemists, a few more heterocycles might not go amiss. Unless this method really is inapplicable for ~60% of small drug molecules.

    2. Iron does have certain advantages over nickel and palladium in terms of reactivity and toxicity, but it is hard to make a compelling argument that using it offers big cost savings here. A back-of-the-envelope calculation shows that Pd(PPh3)4 and Pd(OAc)2 are about $20/mmol.* Although the cost of Fe(acac)3 is negligible (0.4 $/mmol, probably less for the other Fe sources that also work), the dppBz ligand is 19$/mmol---or about the same cost as the above palladium reagents. Do you still feel good about that 6-48% loading? Would you on kilo-scale? This problem of ligand cost is a common one---the major contributor to the cost of even precious metal catalysts is often the ligand. For the Buchwald ligands, the disparity is staggering.

    3. Most in situ examples use HATU or HBTU, and these are not practical reagents. For one thing, HATU is a non-negligible $20/mmol---that’s roughly comparable to Pd(PPh3)4, except it’s obviously used stoichiometrically. These are not process-friendly numbers. Yes, I know you said other slightly-less-impractical activating groups work, and HBTU/HATU’s just for cases where the RAE isn’t isolated, but those are the conditions that you recommend, and the ones that process chemists will use the most. They hate isolations almost as much as class 1.4 explosives, which, incidentally, the HOAt/HOBt byproducts are.

    4. Diaryl and dialkyl zincs are impractical nucleophiles on almost any scale. Few are commercially available, and the ones you can buy are damn expensive. Also, they’re a total pain to make on any kind of scale. Yes, a few of your examples do work with Grignards, but if we’re honest, process chemists strongly averse to even making those.

    5. The stoichiometry. Let’s imagine that I have an alkyl halide that I want to couple with an acid using your reaction. Depening on if I pre-form my RAE or not, I’ll need 1.5-2.5 equivalents of zinc nucleophile. The chances are slim that organozinc I need is commercially available, so I’ll have to make it. In order to prepare 1.5-2.5 equivalents of R2Zn, I’ll need 3-5 equivalents of my halide, 2-4 equivalents of which are presumably consigned to oblivion.

    *Based on currently Aldrich prices per gram in USD. Obviously, prices vary on scale and by supplier, but I wanted some quick ballpark numbers so I went with the easy option here.

  3. Also:-

    6. The solvent. THF is not an ideal process solvent, but I’ll give you a pass on that as it’s used often enough, and it isn’t the choice of solvent here that’s most problematic. Instead, let’s talk volume(s). A good target for solvent usage in a process setting in order to keep one’s PMI down, and help your friends at BMS retain their A-grade under the Pacific Sustainability Index, is 5 volumes---process-speak for 5 mL solvent per gram of starting material. Your process demonstration example uses 44 volumes of solvent---just for the reaction---before we even talk about workup and recrystallization. So the PMI thing doesn't look so good.

    7. There are two bona fide process chemists among the authors and yet you “demonstrate the potential of a process-scale application” on 1.4 grams of starting material. Heck, that would be small scale for a Baran group total synthesis.

    8. The data for the exotherm are close to meaningless. All I can take away from this is that there might be one.

    Look, this isn’t bad methodology for certain applications, and I’ve no doubt people will use it. I realize that there is no practical solution to the decarboxylative-coupling problem right now. I don’t think the MacMillan decarboxylative couplings, which come with slews of their own problems, are any better suited to process applicatons. But to me, this paper invites criticism with largely indefensible claims of practicality and process-readiness.

    All that being said, I would like to add, unrelatedly, that you guys probably write the best SI in the world right now. I do believe that if I were to run this, I would get it to work as you say, and that’s highly commendable.

    1. Thank you for your comment.
      where is your comment 1-5. Unfortunately I can't see it. I'd like to read it.
      Thanks, Fumi

    2. I'm not sure. It was up before I posted 5-8 (the character limit for comments is a bit short).

      I've posted it again below. I hope it sticks this time.

    3. Nope, it's gone again. It must be getting picked up as spam. I guess regular people don't write comments that close to the character limit. Can you get the blog to spit one of them back out?

    4. Fixed. Thank you for your comments. Fumi and the team will be responding shortly.

    5. Hi Pharma Refuted,

      Thank you for your comments! My name is Fumi - and I have 16 years of experience as a process chemist in pharma industry. In the name of the Fe-team I would like to address your comments point-by-point.

      1, Our scope table was designed to showcase the diversity of structural motifs tolerated in the reaction. The presence of nitrogen in the molecule is only one among many relevant aspects. We have shown that several nitrogen-containing compounds like amino acid, heteroaromatic, and protected alkylamines are well tolerated in the reaction. Without being able to disclose the structures, I would also like to mention that BMS has tested the reaction on many more nitrogen-containing compounds.

      2, As you pointed out, Fe(acac)3 is a very cheap metal source, and the even cheaper iron(III) chloride can be used as well. As you also mentioned prices for ligands in industry are typically way lower than at a research chemical supplier like Aldrich.
      In our mock process chemistry approach, we succeeded at decreasing the catalyst and ligand loading to 1.0 and 1.2 mol%, respectively. After a rigorous optimization process for a real-life industry application loadings will be adjusted considering the overall-balance to achieve an optimum process.
      If you want, you can even recover ligand from the organic layer after quenching the reaction. Yes, the ligand contributes to the overall cost of the reaction but this does not make the reaction any less suitable for process applications.

      3, HATU/HBTU are expensive but they are acceptable for process chemistry - particularly in pharma industry where products are often high-value compounds. If cost is a critical factor for your process, NHPI/TCNHPI esters are a good alternative. Also, isolation of highly crystalline products such as NHPI esters is not an issue on scale and can even be used to improve the purity.
      The HOAt/HOBt side products can be efficiently removed by aqueous extraction which alleviates stability/safety issues.

      4, Unfortunately, not all reactions run on love and air, or boronic acids ;) Using arylzinc reagents basically means adding zinc chloride to your Grignard solution and you are there. This is well established in industry and easily done in the lab if you follow our graphical guide. By the way, why Prof. Negishi get a Nobel prize? Zinc reagents can be kind of useful ;-)

      5, We agree with you on this one - the valuable starting material should be the RAE coupling partner. However, using an excess of e.g. diphenyl zinc is not such a big of a deal - especially if it gets you to your desired product faster or through a better route. If optimized in a more targeted approach it might also be possible to improve stoichiometry. We are not claiming that this reaction is always the best choice - it has to be applied to the right problem.

    6. 6, This reaction worked well in most solvents we tried and under more concentrated condition (see SI). Again, the current conditions and all the information which is supplied in the SI are there to provide practitioners in medicinal and process chemistry with the best information basis. Applied to a reaction on scale targeted at a single product, process chemists will be able tune conditions - and most probably also increase the concentration.

      7, Three process chemists! Please count me as well. We scaled-up from 0.1 mmol to 5 mmol in one step without changing the conditions. And the reaction works reliably… But if you want to run it on 1000kg is will clearly require some work.

      8, I don’t think it is meaningless. The high reaction rate also leads to the formation of heat in a short period of time. Loss of heat to the cooling bath during that time (approx. 1min) is almost negligible. A jump of reaction temperature of less than 20 °C under quasi-adiabatic conditions is a good indication that the exotherm can be manageable on scale - of course after appropriate safety assessment.

      Fumi and the Fe team

  4. With all of these different methods you guys are publishing, do you ever see homocoupling of the redox active esters?

    1. Hi chemist,

      Thanks for your comment! In this particular reaction, we didn’t observe the homocoupling product.
      However in another project going on in this lab (yet to be published), homocoupling product of the redox active esters was observed as a major byproduct, especially using primary RAEs.

      Fumi and the team

  5. I really like this - it looks sensible from the process point of view, especially since it does not require high dilution or a narrow temperature window. Someone has raised the cost of the phosphine ligand but I think it is a non-issue: this one is extremely easy to make (If I remember correctly it was just single step preparation) and the price will come down once this becomes more generally used.

    Just curious - please have you also looked at Co(acac)2 ? {A long time ago, I did some Et2Al-acetylide conjugate additions catalyzed with Ni(I) generated from Ni(acac)2+DIBAL and just switching to Co(acac)2 improved the yield - without necessity to pre-reduce the metal catalyst with DIBAL; enantiopure Co-Salene as a catalyst worked too but with zero ee}

    1. Hey milkshake, unfortunately I can't find any contact address, so maybe it works like this. I would like to do a similar conjugate addition under Co catalysis. Could you maybe contact me under m4f at hotmail dot de so I can disclose the details? Looking forward to hearing from you

  6. one more comment: from the SI table of ligands tried, the best results were with dppBz, and then dppe: both of them diphosphines with a small bite angle. Please have you tried Ph2PCH2PPh2 aka dppm? That one is quite cheap but not as frequently used (because Pd couplings like ligands with a large bite angle)

    1. Hi milkshake,
      Thanks for your comment! Yes, we have tried Co(acac)2 in both Negishi and Kumada couplings. They worked with lower yields than Fe catalyst so far. This is something we want to look at again soon and possibly optimize.
      On the other hand, we haven’t tried dppm ligand yet. Thanks for the suggestion, we will look into that!

    2. Hi Fumi -thank you for your reply - Aldrich even sells (Ph2P)2CHPPh2 tris-(diphenylphosphino)-methane

      It is possible that Aryl-Co(L)n species might have a problem with side-reaction, who knows, acting more as a nucleophile rather than a single electron donor. (I remember preparing some alkyl and allyl manganese from Grignards by transmetalation with MnCl2.2LiCl in THF, and these organomanganese intermediates were pretty stable in solution and behaved in unremarkable Grignard-like fashion, so Mn might not be such a good candidate either)

  7. Fumi, hi,

    Fürstner has highly informative outlook in ACS Central Sci on Fe-based catalysis, I am sure you have seen it.

    The one thing that caught my attention there is his emphasis that small bite angle bidentate and tridentate ligands enforce orbital degeneracy and thus strongly modulate the chemical character of Fe central atom, something which apparently had not been appreciated enough

  8. I saw tetramethyl substituted cyclopropyl but not simple cyclopropyl in the examples. Did you try unsubustituted cyclopropyl for this reaction or this specific radical is not stable under the reaction condition?

  9. The soleplate of an iron can be made of aluminum or hardened steel and can likewise have an artistic (smooth skim), titanium (non-stick surface) or even palladium covering (self-cleaning).Best steam irons