Development of a Cryogenic Flow Reactor to Optimize Glycosylation Reactions Based on the Active Donor Intermediate

Abstract

The development of a continuous flow reactor for stereospecific glycosylation reactions with deoxy sugars is described. This apparatus that permits optimizing the selectivity of glycosylation reactions based on the stability of the activated intermediate is described. By coupling a flow apparatus with HPLC analysis, we can optimize the yield of TsCl-mediated β-linked deoxy sugar construction in a matter of hours. In all cases, results from continuous flow processing translate into improved results in batch-scale reactions, as demonstrated by competition experiments. This is the result of carrying out optimization to identify the ideal temperature for the reaction of the activated intermediate, as opposed to the initial activation conditions. Such an approach allows for the rapid development of highly selective glycosylation reactions in cases in which classical neighboring group participation is not possible.

1. INTRODUCTION

Given the pressing need for methods for efficient and stereoselective production of homogeneous glycans, the development of new approaches to the glycosylation reaction has been a subject of considerable interest over the past several years.^1,2 In general, glycosylation reactions are often described as taking place along the S_N1-S_N2 continuum; however operationally, the process is typically more complex (Figure 1).³ The reaction begins with activation of an anomeric leaving group in 1 (Step 2) with a chemical promoter (X-Y), which is often generated in situ (Step 1). The resulting activated species (2) can either react with the counterion X⁻ generated during the initial activation to afford 3α and 3β or dissociate to form the oxocarbenium cation 4 (Step 3). These intermediates can be in equilibrium with each other prior to the actual glycosylation step (Step 4) with a nucleophile (Nu), leading to mixtures of products.^4–8

Figure 1.

Generalized mechanism of a glycosylation reaction.

A goal of modern research on glycosylation chemistry is to develop reactions that proceed through S_N2-like manifolds. This aim is challenging because the selectivity in these reactions is dictated by a number of factors, including temperature, concentration, method of activation, and the reactivity of coupling partners.⁹ While S_N2-like kinetics for chemical glycosylation reactions have been reported by several research groups,^8,10–14 obtaining highly selective glycosylation reactions that can be generalized to a range of coupling partners remains difficult. In many cases, these reactions are initiated at cryogenic temperatures and are then allowed to slowly warm to ambient temperature.¹⁵ Regardless of which approach is used, optimal conditions for suppressing the anomerization and ionization of the active glycosyl intermediate (3α or 3β, Figure 1) are generally not considered.

This approach can lead to issues with reproducibility in these systems, especially in cases where the reaction is initiated at cryogenic temperatures and allowed to warm to ambient temperature. To address this issue, several groups have turned to automated continuous flow processing to optimize the reactions. However, these studies have focused primarily on finding the optimal temperature for activation as measured by yield, donor reactivity, and acceptor reactivity (Figure 1, Steps 2 and 3).^15–19 In contrast, a systemic study to optimize the conditions for the reaction of the active intermediate with a nucleophile (step 4, Figure 1) has not been explored. Here, we report the development of an automated continuous flow processing reactor to optimize the selectivity and yield in a reaction of an active intermediate with a nucleophile. In the process, we demonstrate that such knowledge can be translated back to reproducible batch-based synthesis.

For the initial study, we elected to optimize our previously reported p-toluenesulfonyl chloride (TsCl)-mediated dehydrative glycosylation for β-selective deoxy sugar synthesis.²⁰ We selected this reaction because our lab and others have demonstrated the utility of this chemistry in complex deoxy sugar oligosaccharide synthesis.^21–25 Importantly, we have demonstrated that activation of hemiacetals as glycosyl tosylates occurs rapidly (<5 min) and cleanly at cryogenic temperatures where the only byproducts of activation were KCl and hexamethyldisilazane, neither of which react with the glycosyl sulfonate. As a result, aside from potential ionization or leaving group anomerization, side reactions that could lead to the formation of different reactive species are minimized.

2. RESULTS AND DISCUSSION

2.1. Reactor Design.

We chose to examine the reaction between 3,4-di-O-benzyl-L-olivose donor 6 and the aryl acceptor p-methoxyphenol 8 as a model system for reactor design because it was highly efficient in batch. Our initial reactor design utilized two Harvard syringe pumps. In our previously described batch reaction, TsCl is added after metalating 6. In the flow setup, we chose to combine TsCl with 6 to minimize the number of pumps used (Scheme 1, Figure 2). To this end, the reaction would require four different input solutions. Solution 1 included donor 6 (0.13 M), the proton scavenger 2,4,6-tritert-butylpyrimidine (TTBP) (0.13 M), and TsCl (0.13 M) dissolved in THF. Solution 2 consisted of a 0.1 M solution of potassium bis(trimethylsilyl)amide (KHMDS) in THF for the deprotonation of donor 6. The third solution contained a 0.06 M solution of 8 in THF. Solution 4 consisted of a 0.06 M solution of KHMDS in THF for the deprotonation of acceptor 8. For this initial setup, plastic syringes, polyether ether ketone (PEEK) joints, and perfluoroalkoxy (PFA) tubing were used to construct the reactor. Three different dry ice/acetone baths were used to contain all seven reactor coils as they did not all fit in one bath. The flow rate and residence time were screened, and the results observed were either trace yield of the desired product or issues with reactor coil clogging (Table 1). For the runs that did not clog, the reagent mixtures changed to a deep brown color when transferred between dry ice baths. Since this brown color was observed when transferred between dry ice baths, we elected to redesign the reactor, so all of the reactor coils were in the same bath. For runs that clogged, this decomposition was observed to begin in the KHMDS syringes.

Scheme 1.

Reactor Design 1: Combining TsCl with Donor 6

Figure 2.

Flow setup for the reaction described in Scheme 1.

Table 1.

Conditions Tested for Reactor Design 1

entry	flow rate (mL/min)	residence time (min)	yield	notes
1	0.5	30	trace	decompositionobserved
2	0.5	30	2%	decompositionobserved
3	0.25	60	n/a	clog
4	0.25	60	n/a	clog
5	0.25	60	trace	decompositionobserved
6	0.25	60	n/a	clog

Reasoning that the low yields we observed could be related to decomposition of one or more reagents, we chose to reevaluate our reactor design. Some trials displayed this decomposition to begin in the syringes containing KHMDS, so we concluded that the syringes were not airtight. To address this issue, we opted to pump the KHMDS solution directly from a flask under argon using an HPLC pump. To accommodate the need for KHMDS at two points in the reaction, the reactor design was altered to contain a Y-splitter to pump the base into both the donor and acceptor solutions. Aside from the Y-splitter and combining all the reactor coils in the same, larger bath, the composition of all reactant solutions remained the same as in our first-generation reactor (Scheme 2, Figure 3). For trials run with this reactor, the flow rate was kept constant, while different types of syringes (glass and plastic), use of a back pressure regulator, and concentration were altered (Table 2). In all cases, reagents turned into deep brown at the joints between reactors. These joints were suspended above the cryogenic bath as they can be observed to leak at temperatures below −60 °C and thus were likely responsible for the poor yields. To circumvent the need to have joints above cryogenic temperatures for the next design, we turned our attention to commercially available cryogenic reactors for inspiration.

Scheme 2.

Reactor Design 2: An HPLC Pump Used for KHMDS Input

Figure 3.

Picture of reactor design 2.

Table 2.

Conditions Tested for Reactor Design 2

entry	syringes	back pressure regulator	yield	notes
1	plastic	n/a	n/a	decompositionobserved
2	plastic	n/a	n/a	clog
3	glass (notgastight)	n/a	n/a	clog
4	plastic	yes	trace	decompositionobserved

Reasoning that the decomposition of the reagents may be due to the incompatibility of the PFA tubing and PEEK adapters to cryogenic temperatures, we decided to completely redesign our setup. Inspired by reports from the Ley and Browne groups,^26–28 we elected to construct a system where the tubing was wrapped around the outside of the cryogenic bath. To this end, the reactor consisted of an inner chamber containing cryogen (dry ice/acetone) around which the reaction coil was wrapped. This inner chamber was placed in a larger chamber packed with dry ice to help maintain the temperature (Figure 4). A thermocouple was placed between the tubing coils to monitor the temperature. The same basic flow design was used as the reactor in Scheme 2; however, two Harvard syringe pumps total were used instead of an HPLC pump and a Harvard syringe pump for the later trials (Scheme 3). Solution 1 included donor 6, TTBP, and TsCl dissolved in THF (0.12 M). Solution 2 included KHMDS diluted in THF (0.12 M). Solution 3 included aryl acceptor 8 dissolved in THF (0.12 M). There were a variety of different conditions tested with this design, including pump types, syringe types, concentrations, and residence times (Table 3 and Table 4). Optimization of all these parameters provided a yield of 22% (Table 4, entry 13) providing an improvement from the previous reactor design. These low yields were attributed to leaking between the PFA/PEEK tubing and joints, which we suspected could have led to exposure of the reaction to air.

Figure 4.

Picture of reactor design 3.

Scheme 3.

Reactor Design 3: Tubing Wrapped around a Metal Shipping Container

Table 3.

Preliminary Conditions Tested for Reactor Design 3

entry	residence time (min)	pumps	syringes	yield	notes
1	60	1 syringe 1 HPLC	plastic	n/a
2	15	1 syringe 1 HPLC	plastic	n/a
3	60	1 syringe 1 HPLC	plastic	5%	α:β 1:2
4	60	2 syringe	gastightKHMDS	5%	α:β 1:2
5	60	2 syringe	gastightKHMDS	12%	α:β 1:2

Table 4.

Further Conditions Tested for Reactor Design 3

entry	donor (M)	acceptor (M)	KHMDS (M)	residence time (min)	syringes	yield	notes
6	0.06	0.06	0.02	60	gastight KHMDS	trace
7	0.12	0.12	0.24	60	gastight KHMDS	11%	β only
8	0.12	0.12	0.24	30	gastight KHMDS	10%	β only
9a	0.12	0.12	0.24	15	gastight KHMDS	n/a
10	0.24	0.24	0.48	15	gastight KHMDS	14%	β only
11b	0.24	0.24	0.48	15	gastight KHMDS	17%	β only
12b	0.60	0.60	0.50	15	all gastight	12%	β only
13b	0.24	0.12	0.48	15	all gastight	22%	β only

^a−40 °C.

^bNo back pressure regulator.

Based on the lessons learned from our previous designs, we made two major changes to the reactor moving forward. To eliminate the problem of leaking, we opted to use coiled stainless-steel tubing in all reactor designs moving forward, as it was anticipated that this material would have better compatibility with the cryogenic temperatures. We also decided that we should make the order of addition of the reagents more similar to that of our batch reaction. To this end, we opted to carry out metalation of 6 prior to addition of TsCl. To better control the concentration of the base, we also opted to introduce KHMDS using two different syringes; one for the base that would metalate 6 and a separate one to metalate 8 (Figure 5). In this new setup, the reactor coils were placed in an insulated bath cooled using a cryostat. With this new reactor, we could better control temperature as compared to previous reactor designs. We initially examined running the reaction at −78 °C, in analogy with the previously reported batch reactions. The product was only produced in 20% yield, analogous with reactor design 3. We suspected that the temperature could impact the formation and/or reactivity of the active intermediate and thus could have been responsible for our low yield, and therefore, we opted to examine this variable. However, the large size of the bath required an excessive amount of time (~4 h) to equilibrate, which we deemed impractical for high-throughput optimization of the reaction temperature. We therefore set about trying to develop a more compact reactor design.

Figure 5.

Picture of reactor design 4. The bath is a plastic container wrapped with glass wool and tin foil.

To minimize the footprint and amount of cryogen, we next opted to stack the coils in an insulated dewar (Figure 6). This design followed the setup of our previous reactor (Scheme 4). A cryostat was placed in the middle of the tubing for uniform cooling (Figure 6). This more compact design permitted much more efficient cooling, and the temperature was able to reach equilibrium in under an hour. This design was used for all of the further studies. Moving forward, we opted to automate reagent delivery using the Python-based Mechwolf program (Figure 6).^29–31

Figure 6.

Pictures of reactor design 5 with MechWolf automation.

Scheme 4.

Reactor Design 4: Using Stainless Steel Tubing

2.2. Temperature Screen.

We elected to screen reaction temperatures in flow for this donor–acceptor pair coupled with high-pressure liquid chromatography (HPLC) for rapid analysis of the crude reactions. We were able to rapidly examine several parameters as the reaction run in flow is much shorter than the 4 h batch reaction. With the flow reactor, we could collect a fraction for analysis, halt the flow to equilibrate the system at a higher temperature, and then collect additional data. In this way, we would also eliminate potential variability arising from the need to measure out multiple samples for multiple reactions. From an operational standpoint, temperatures were examined in 5° increments ranging from −75 °C to −50 °C. Based on our initial HPLC analysis (Table 5), we concluded that the ideal temperature for the reaction of the glycosyl tosylate with acceptor 8 was between −65 and −55 °C.

Table 5.

% of Each Compound Present at Each Reaction Temperature

		Compound Area (Average %)
Entry	Temperature (°C)	A	B	C	D	E
1	−50 to −55	2.8 ± 1.9	22.29 ± 4.9	38.4 ± 1.8	8.4 ± 2.2	8.1 ± 4.1
2	−55 to −60	2.3 ± 1.6	7.5 ± 2.2	73.9 ± 0.8	1.a9 ± 1.5	5.2 ± 1.4
3	−60 to −65	4.0 ± 0.7	8.8 ± 5.3	73.4 ± 2.7	3.8 ± 1.3	1.3 ± 0.7
4	−65 to −70	9.6 ± 2.8	8.0 ± 0.8	57.8 ± 1.7	7.1 ± 1.5	5.4 ± 1.6
5	−70 to −75	6.5 ± 1.7	21.7± 2.7	47.7 ± 5.6	7.4 ± 1.8	1.0 ±0.4

2.3. Flow Rate Screens at −57 °C.

Once the temperature of the reaction mixture was optimized, we elected to run a continuous flow rate screen to find the optimal flow rate and residence time (Table 6). We initially tested a flow rate of 0.25 mL/min for reactor coils 1 and 2 and 0.5 mL/min in reactor coils 3, 4, and 5. HPLC analysis of the crude reaction mixture demonstrated that the product was formed with a significant amount of side products (Table 6, entry 1). Based on of the side-product profile, we reasoned that the active intermediate was decomposing over the long residence time. We therefore opted to increase the flow rates of solutions 1 and 2 to 0.5 mL/min and solutions 3, 4, and 5 to 1 mL/min (Table 6, entry 2). These conditions led to an increase in the production of product with a concomitant decrease in the amount of byproduct production. Further increasing the flow rate to 1.0 and 1.0 mL/min led to a further increase in yield (Table 6). At higher flow rates, the yield began to decrease again, which we attributed to an incomplete reaction in the activation coil.

Table 6.

% of Each Compound Present at Each Flow Rate

		Compound Area (%)
Entry	Flow Rate (mL/min)	A	B	C	D	E
1	0.25	17.1	20.0	41.3	5.5	7.3
2	0.5	15.3	1.0	62.5	0.9	1.1
3	1.0	16.6	1.0	72.2	1.1	1.1
4	1.5	1.0	15.5	58.5	8.2	6.7
5	2	2.6	18.8	53.9	1.6	8.5

Based on these results, we elected to narrow our screen of flow rates to between 0.5 and 1 mL/min for further optimization (Table 7). To this end, we initially examined a flow rate screen beginning with a flow rate of 0.4 mL/min (Table 7, entry 1). After the reactor was washed with THF, the flow rate was increased to 0.5 mL/min (Table 7, entry 2), and the product was collected in a separate receiving flask for HPLC analysis. This process was repeated for flow rates of 0.6, 0.7, and 0.8 mL/min (Table 7, entries 3–5). HPLC analysis of the crude fractions collected at the different flow rates revealed that the optimal flow rate was 0.7 mL/min (Table 7, entry 4), which produced the product in 88% yield in a gram scale as a single diastereomer, which is comparable to batch yield.

Table 7.

% of Each Compound Present at Each Flow Rate

		Compound Area (%)
Entry	Flow Rate (mL/min)	A	B	C	D	E
1	0.4	34.0	14.0	34.0	6.9	4.0
2	0.5	14.0	2.7	71.0	6.7	2.5
3	0.6	14.0	0.5	82.0	0.3	1.3
4	0.7	8.0	1.1	88.1	0.4	0.3
5	0.8	39.8	0.5	52.7	3.3	0.3

2.4. Temperature Screen with an Aliphatic Acceptor.

Having established the optimal temperature in flow for the reaction between 6 and 8, we turned our attention to glycosidic acceptors. The glycosylation between donor 6 and primary alcohol acceptor 11 in flow at −63 °C provided the product in only a 38% yield (Table 8, entry 2). Attempts to extend the chemistry to other sugar acceptors also led to less than satisfactory results (Table 8, entries 3–6). We reasoned that the lower yields were due to the reduced nucleophilicity of aliphatic alkoxides (N = 16.73) relative to 4-methoxyphenoxide (N = 20.62).^32–34 As such, it was possible that the nucleophile was not sufficiently reactive to undergo productive reactions with the glycosyl tosylate at this temperature. To address this, we returned to a continuous flow temperature screen with an aliphatic acceptor.

Table 8.

Screen of Different Glycosylation Reactions at Previously Optimized Temperatures in the Flow

Entry	Donor	Acceptor	Product	Yield	Selectivity
1				88%	β only
2				38%	β only
3				25%	β only
4				30%	β only
5				24%	β only
6				27%	β only

For the next temperature screen, 11 was used as a model acceptor (Table 9). Using the same reactor setup from our previous screen, we initially examined temperatures ranging from −73 to −43 °C, over 5° intervals and again followed the reaction by HPLC. From this screen, we found a slightly higher ideal temperature range for aliphatic acceptors, −52 to −57 °C (Table 9, entry 3). Notably, at these temperatures, the aromatic acceptor 8 still provided the desired product in excellent yield as single diastereomers (Table 5, C, entry 2, 74% yield).

Table 9.

% of Each Compound Present at Each Reaction Temperature

		Compound Area (%)
Entry	Temperature (°C)	A	B	C	D	E
1	−42 to −47	71.4	1.6	5.7	4.3	2.3
2	−47 to −52	15.3	2.2	55.5	7.9	2.2
3	−52 to −57	6.4	3.3	66.0	7.1	3.3
4	−57 to −62	54.8	3.0	17.3	6.1	4.6
5	−62 to −67	77.3	1.0	15.0	1.6	0.6
6	−67 to −72	1.4	1.5	9.0	69.9	8.6

2.5. Application to Batch Synthesis.

With this new optimized procedure in hand, we turned our attention to testing if the results could be translated into batch. To ensure that the improvements we were observing were reproducible, each reaction was conducted in triplicate and a side-by-side competition was set up where one reaction was run from −78 °C to RT while the other was maintained at a consistent temperature of −57 °C as measured by an internal thermocouple (Table 10). Pleasingly, both acceptors 8 and 11 displayed superior reactivity at −57 °C relative to slow warming to ambient temperature (Table 10, entries 1 and 2). Using D-configured donors, 13 and 23 again provided better yields and selectivity at a constant temperature of −57 °C, indicating that improved selectivity was not the result of stereochemical match/mismatch (Table 10, entries 3–6).³⁴ Importantly, the reaction between 13 and 8 could be run on a 1 mmol scale to produce 22 with consistent results (Table 10, entry 4). Reactions of L-configured donor 6 with acceptors 14 and 18 also performed significantly better at −57 °C than those under the previously reported cryogenic to RT conditions (Table 10, entries 7 and 8). Collectively, these results illustrate how subtle temperature changes can influence the outcome of chemical glycosylation reactions, even when strong nucleophiles such as alkoxides are used.

Table 10.

Screen of Different Glycosylation Reactions at Optimized Temperature in Flow and Run in Batch

Entry	Donor	Acceptor	Product	Yield at −78 °C	Yield at −57 °C	Selectivity
1				73%	95%	β only
2				44%	69%	β only
3				65%	90%	β only
4a				56%	68%	β only
5				67%	96%	β only
6				65%	82%	β only
7				41%	94%	β only
8				43%	76%	β only
9				12%	30%	β only
10				70%	86%	−57 °C: 1:6 α:β
						−78 °C: 1:3 α:β
11				19%	59%	−57 °C: 2:1 α:β
						−78 °C: 2:1 α:β

^a1 mmol scale.

Having established that the yields of glycosylation reactions with dideoxy sugar donors could be improved under the conditions optimized in our flow reactor, we next examined substrates that we had not previously examined or had proven problematic with our chemistry in the past. To this end, we first examined the use of thioglycoside acceptor 26 in the reaction (Table 10, entry 9). Activation under our previously described conditions led to a complex mixture of products, from which we were able to obtain 27 as a single β-linked isomer in a disappointing 12% yield. Using the new temperature, we were able to improve the yield, albeit only to 30%. In both cases, a significant amount of intermolecular aglycon transfer was observed, indicating a potential limitation of sulfonate-mediated glycosylation.³⁵

Previously, we had found that activating 2,3,6-trideoxy sugars such as amectiose and rhodinose using TsCl did not lead to the β-specificity observed with 2,6-dideoxy sugars.²⁰ In an effort to see if holding the reaction at a constant temperature could lead to improvements in selectivity, we next examined donors 28 and 30 in the reaction under the conditions obtained from our flow screen (Table 10, entries 10 and 11). In both cases, the elevated temperatures, identified using the continuous flow-based optimization, led to higher yields of the respective products. Notably, the selectivity of 29 was improved from 3:1 to 6:1, α:β (Table 10, entry 10). These lower selectivities point to the fact that a different sulfonyl chloride promoter is likely needed to obtain β-selectivities with our substrates, by analogy with observations in studies on C2 substituted sugars.¹²

3. CONCLUSIONS

Through the use of continuous flow processing, for the first time, it is possible to optimize a chemical glycosylation reaction based on the reactivity/stability of the actual reaction intermediate. Through this approach, it is possible to screen multiple conditions, a process that may take days in batches in a matter of hours. Previous studies that optimized chemical glycosylation reactions using automated platforms have been focused on the activation step of chemical glycosylation. While this information is very helpful for increasing the yield of a particular reaction, the optimal activation temperature may not be the optimal temperature for the reaction between an active intermediate and a nucleophile.

In the absence of any understanding of the reactivity/stability of the active glycosylation intermediate, multiple reaction pathways can take place that can lead to mixtures of diastereomers, and most optimizations for selectivity must be carried out empirically. As a consequence, previous studies in flow have relied on classical methods, such as neighboring group participation, to control the selectivity, which cannot provide access to many glycosidic linkages with the deoxy sugars examined in this study. Understanding the conditions under which a covalent intermediate is stable enough to maintain its configuration, while reactive enough to undergo productive bond forming processes, is critical to achieve the rapid production of glycosides beyond 1,2-trans-glycosides.

By focusing on the reactivity of the active intermediate, we are able to obtain high yielding reactions where selectivity is not reliant on the protecting group patterns of the donor or the nucleophilicity of the alcohol acceptor. The power of this approach is demonstrated through optimization of the construction of what is considered one of the most difficult glycosylation reactions to control, β-linked deoxy sugars. Notably, optimization done on a single set of coupling partners in flow can be extended to the scalable and reproducible construction of sugars derived from several donor/acceptor pairs. Certain substrates, such as trideoxy sugars, did not lead to β-specific reactions; however, this result is likely due to the enhanced reactivity of the donor rendering the α-glycosyl tosylate unstable under the conditions we examined. While the current study is focused on deoxy sugar tosylates, the continuous flow approach to optimization based on active intermediates should be applicable to a wide range of donors, such as triflates, provided that the active intermediate can be generated quickly and cleanly through a preactivation protocol. The extension of this approach to optimizing other classes of glycosylation reactions, including trideoxysulfates, will be reported in due course.