Automated Quantification of Hydroxyl Reactivities: Prediction of Glycosylation Reactions

Article abstract of DOI:10.1002/anie.202013909

The stereoselectivity and yield in glycosylation reactions are paramount but unpredictable. We have developed a database of acceptor nucleophilic constants (Aka) to quantify the nucleophilicity of hydroxyl groups in glycosylation influenced by the steric, electronic and structural effects, providing a connection between experiments and computer algorithms. The subtle reactivity differences among the hydroxyl groups on various carbohydrate molecules can be defined by Aka, which is easily accessible by a simple and convenient automation system to assure high reproducibility and accuracy. A diverse range of glycosylation donors and acceptors with well-defined reactivity and promoters were organized and processed by the designed software program “GlycoComputer” for prediction of glycosylation reactions without involving sophisticated computational processing. The importance of Aka was further verified by random forest algorithm, and the applicability was tested by the synthesis of a Lewis A skeleton to show that the stereoselectivity and yield can be accurately estimated.

Full text of DOI:10.1002/anie.202013909

Angewandte
Research Articles
Chemie
How to cite:
International Edition: doi.org/10.1002/anie.202013909
German Edition: doi.org/10.1002/ange.202013909
Glycoscience
Automated Quantification of Hydroxyl Reactivities: Prediction of
Glycosylation Reactions
Chun-Wei Chang, Mei-Huei Lin, Chieh-Kai Chan, Kuan-Yu Su, Chia-Hui Wu, Wei-Chih Lo,
Sarah Lam, Yu-Ting Cheng, Pin-Hsuan Liao, Chi-Huey Wong,* and Cheng-Chung Wang*
Abstract: The stereoselectivity and yield in glycosylation
reactions are paramount but unpredictable. We have developed
a database of acceptor nucleophilic constants (Aka) to quantify
the nucleophilicity of hydroxyl groups in glycosylation influ-
enced by the steric, electronic and structural effects, providing
a connection between experiments and computer algorithms.
The subtle reactivity differences among the hydroxyl groups on
various carbohydrate molecules can be defined by Aka, which
is easily accessible by a simple and convenient automation
system to assure high reproducibility and accuracy. A diverse
range of glycosylation donors and acceptors with well-defined
reactivity and promoters were organized and processed by the
designed software program “GlycoComputer” for prediction
of glycosylation reactions without involving sophisticated
computational processing. The importance of Aka was further
verified by random forest algorithm, and the applicability was
tested by the synthesis of a Lewis A skeleton to show that the
stereoselectivity and yield can be accurately estimated.
provide more plausible synthetic routes.[2] However, their
applications to diastereoselective reactions are still rare.
Carbohydrates are ubiquitous biomolecules that mediate
numerous biological processes and exhibit important patho-
genic effects.[3] However, obtaining carbohydrates by extrac-
tion from natural sources is impractical and chemical syn-
thesis of glycoconjugates has been hampered by unpredict-
able glycosylation. Currently, there is no clear rule to direct
the configuration of glycosidic linkage as glycosylation
reaction often proceeds through a range of mechanisms
involving different intermediates, leading to a mixture of a-
and b-glycosides.[4] In addition to the reactivities of glycosyl
donor 1 and acceptor 2, the steric effects and use of promoters
and solvents also affect the outcome of glycosylation reac-
tions[5] (Figure 1A). It is known that solvents and promoters
will influence the counter ion coordination,[4c,5,6] and the
anomeric selectivity and yield are affected by a combination
of all these unquantifiable factors. Tedious trial and error are
thus required to optimize each glycosyl linkage, making the
chemical synthesis of complex oligosaccharides a major
challenge.
Introduction
The advances in synthetic chemistry have made possible
access to complex natural or unnatural molecules. These
advances are mainly attributed to the development of new
chemical reactions and strategies, and the improvement of
reaction yield and selectivity. Recently, computer-aided
organic synthesis[1] and synthesis planning software with
algorithm and machine-learning have been developed to
In a glycosylation reaction with covalent intermediate, the
acceptor tends to attack the anomeric carbon (C1) via the
SN2-like pathway, while for the reaction via oxocarbenium ion
the acceptor attachment generally undergoes the SN1-like
mechanism either from the top (b) or the bottom (a). The
reaction results in a spectrum from SN2- to SN1-type reaction
pathways among various coupling partners[7] (Figure 1B). In
general, donors with higher reactivity favor a-glycosylation,
while acceptors with higher reactivity tend to form b-glyco-
sides.[5,8] In the absence of C2 neighboring group participa-
tion, a less reactive donor gives a-triflate intermediate 4 of
higher stability, and a stronger nucleophile favors the SN2-like
substitution to yield b-glycoside from the a-triflate inter-
mediate 4. In addition, previous studies by Codꢀe, Bennett
and us have shown that increasing the reactivity of donor
gives more a-glycosylation,[8a,c–g] probably due to the presence
of oxacarbenium ion-like intermediate (e.g., 4-SSIP) which
leads to the a-selective reaction through a unimolecular SN1
[*] C.-W. Chang, M.-H. Lin, C.-K. Chan, K.-Y. Su, C.-H. Wu, W.-C. Lo,
S. Lam, Y.-T. Cheng, P.-H. Liao, C.-C. Wang
Institute of Chemistry, Academia Sinica
Taipei 115 (Taiwan)
E-mail: wangcc@chem.sinica.edu.tw
C.-H. Wong
The Genomics Research Center, Academia Sinica
Taipei 115 (Taiwan)
and
Department of Chemistry, The Scripps Research Institute
10550 N Torrey Pines Road, La Jolla, 92037 (USA)
E-mail: chwong@gate.sinica.edu.tw
3
pathway.[7] These oxacarbenium ions favor E/3H4-like con-
formations allowing hyperconjugation between O5 lone pairs
and the forming s* orbital of a-glycoside in the transition
state.[9] Codꢀe et al. further described the conformation of
oxocarbenium ion which is highly associated with its func-
tional groups, and these conformational changes directly
affect the stereoselectivity of glycosylation reactions. De-
pending on the shape of ion and the distortion of sugar ring,
both the 1,2-cis and 1,2-trans products can be obtained under
C.-C. Wang
Chemical Biology and Molecular Biophysics Program, Taiwan
International Graduate Program (TIGP), Academia Sinica
Taipei 115 (Taiwan)
Supporting information and the ORCID identification number(s) for
the author(s) of this article can be found under:
https://doi.org/10.1002/anie.202013909.
&&&&
ꢀ 2021 Wiley-VCH GmbH
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
These are not the final page numbers!

Angewandte
Research Articles
Chemie
Figure 1. A) The formation of glycosidic bonds via pathways with various transition states and intermediates. B) Reactivities of glycosyl donors
and nucleophilic acceptors related to the glycosylation reactions at the SN1–SN2 interface. C) Acceptor reactivity in glycosylation. D) The diagram
represents differences in the rate-determining step.
kinetic conditions. The formation of the glycosidic linkages
proceeds under kinetic conditions, and a high energy confor-
mer may be formed initially, after which a ring conformation
change can take place in the subsequent step.[10] The SN1/SN2
boundary for a given donor and acceptor pair is typically not
well understood, leading to a poor stereoselectivity and
reduced efficiency of oligosaccharide synthesis.
Based on the basic mechanistic principles, different
strategies of sequential glycosylations were developed to
facilitate the synthesis of oligosaccharides. Inspired by the
armed-disarmed concept,[11] the Ley group reported the use of
protecting groups to tune the anomeric reactivity of glycosyl
donors,[12] and we developed the programmable one-pot
synthesis of oligosaccharides using the relative reactivity
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
ꢀ 2021 Wiley-VCH GmbH
www.angewandte.org
&&&&
These are not the final page numbers!

Angewandte
Research Articles
Chemie
values (RRVs) of thioglycoside building blocks measured by
experiments or by prediction through machine learning.[13]
However, we noticed that although RRV provided a useful
indicator for assessing the reactivity of glycosyl donors, this
strategy could not work smoothly using acceptors with poor
reactivity. For example, Manabe and Fukase et al. reported
the one-pot synthesis of a trisaccharide[14] (Figure 1C) using
state TS-2b with a much higher energy barrier and lower
yield.[16] The reactivity of acceptor is highly sensitive to its
structural features and protecting groups, so development of
a digital index of the acceptor reactivity will make the
glycosylation reaction more predictable.
glucosaminyl acceptor 10 to give the desired product 12 in Results and Discussion
86% yield in 5 min, but with an unreactive acceptor 11 the
yield was significantly reduced to 25% after 3 h. Though the
formation of oxacarbenium ion is often the rate-limiting step,
which can be quantified by RRV, this result indicated possible
existence of two distinct pathways via transition state TS-2a
or TS-2b in glycosylation reactions[11a,15] (Figure 1D). With
a strong acceptor, the glycosylated product 7 is predominantly
formed in a higher yield since the reaction goes through
a rate-limiting transition state 1 (TS1) and high-energy
oxacarbenium intermediate I which reacts with the acceptor
quickly to form the product via transition state TS-2a. In
contrast, with a weak acceptor, the high-energy intermediate I
reacts with the acceptor through a rate-limiting transition
Herein, we developed a GlycoComputer program[17]
based on the properties of various donors, acceptors, activa-
tion systems and solvents to foresee and predict the yield and
stereoselectivity of a glycosylation reaction[17] (Figure 2A).
The key success of this program is to establish a general
comparing system to characterize the reactivity of donor
1 and acceptor 2 by using the RRV[8a,c,13] and the acceptor
nucleophilic constant (Aka) for statistical analysis. Although
researchers have been trying to use high-level quantum
chemistry to predict chemical glycosylations,[10a,b] our param-
eters obtained from the competition experiments took both
the structural and diverse protecting group effects present on
Figure 2. A) The GlycoComputer concept. B) Determination of Aka values. C) Automated system for data collection via upgrading of
a commercially available HPLC autosampler.
&&&&
www.angewandte.org
ꢀ 2021 Wiley-VCH GmbH
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
These are not the final page numbers!

Angewandte
Research Articles
Chemie
numerous glycosyl donors and nucleophilic acceptors into
sampler syringe is 0.1 mL, we can significantly reduce the
account to avoid complicated computational calculation.
Next, we programmed a random forest algorithm to test the
predictability of our GlycoComputer.[17] The variables present
in glycosylation reactions were well defined, including
substrate factors (donors and acceptors) and environmental
factors (solvent and promoter). This advanced programmable
system provides potential synthetic solutions for complex
glycans as demonstrated in the four successful examples
reported here.
Since a statistical measurement of hydroxyl reactivity
under an acidic glycosylation condition is absent in literature,
we introduced the acceptor reactivity value Aka (Figure 2B;
Supporting Information, Figures S2–S7). The acid-catalyzed
reaction between a hydroxyl and an achiral electrophile, such
as 3,4-dihydropyran (DHP)[18] which forms an tetrahydro-
pyrylium ion in the presence of catalytic triflic acid (TfOH) to
react with the hydroxyl group, was chosen as a model to
simulate common glycosyl reaction. Aka is defined as the
relative reactivity difference towards the tetrahydropyrylium
ion between the two hydroxyls, R1OH and R2OH, and is
determined by using HPLC. By comparing the HPLC
patterns of R1OH and R2OH before and after the reaction,
the Aka values of R1OH and R2OH were obtained using the
second-order rate equation established by our previous
works.[13]
substrate consumption to only 0.5 mmol ( ꢂ 0.23 mg) for each
trial. Although tetrahydropyran (THP) migration could
happen from one THP hemiacetal to the other alcohol, based
on our experiment (Figure S14), this thermodynamic trans-
formation is quite slow. Exemplified by mixing phenol (29,
Aka = 1.36) and the THP-hemiacetal of benzyl alcohol (25,
Aka = 5.76) in the presence of TfOH under the same
concentration of Aka test, the migration of THP from 25 to
29 was only 0.3% in 1 h and 0.6% in 4 h, suggesting a minor
influence on the Aka measurement.
The Aka values of various hydroxyls 14–51 were system-
atically determined for the first time using the least reactive
axial 4-OH of galactoside 44 as the reference, 1.00 (Figure 3).
In general, small molecules showed much higher reactivities,
but surprisingly, the adamantanols 14 and 16 showed the
highest Aka of 126 and 80 respectively, despite their
commonly known steric hindrance and rigid configuration
(Figure 3A). For the positional effect of thioglucosyl deriv-
atives (Figure 3B), the decreasing reactivity is in the order of
6-OH 32 (Aka = 5.86) > 2-OH 36 (Aka = 3.91) > 4-OH 38
(Aka = 2.68) > 3-OH 40 (Aka = 1.62). This is in agreement
with the results established by Codꢀe and co-workers,[8b,d–f,h]
where the primary 6-OH of 32 is the most reactive; but
surprisingly, the 2-OH of 36 is the most reactive among the
three secondary hydroxyl groups. A similar trend was noticed
for methyl glucosides 31, 34, 37 and 39. We also extended our
scope to understand the reactivity of acceptors having
different functional groups.
ka
kref
ln½Aꢀtꢁln½Aꢀ0
Aka ¼
¼
ln½Refꢀtꢁln½Refꢀ0
When compared to thioglucoside 32, compounds 33 and
35 exhibited reduced reactivity of 6-OH when increasing the
number of electron withdrawing group (OBz). It is also
interesting that the primary alcohol 35 is less reactive than the
secondary alcohol 34, suggesting the di-benzoyl (Bz) modi-
fication on 35 significantly reduced its nucleophilicity due to
the electronic effect. A reduced reactivity was also noted
when the C2-position was replaced by a more electron
withdrawing substituent such as 2-azidoglucoside 45. The
preliminary trend showed that the deactivating ability of
electron withdrawing group was in the order of -OBz > -N3 >
-OBn.
With regard to galactosides (Figure 3C), it became
apparent that the 6-OH groups of 41 and 42 show very high
Aka of 10.43 and 6.53 individually. The 3-OH of 43 still
revealed a very high reactivity, as the corresponding Aka was
determined to be 6.00. However, the axial 4-OH of galacto-
side 44 was the least reactive among all the hydroxyls tested
(Aka = 1.00). For the mannose 2-OH acceptors (49–51), their
reactivities also depended on the identity of the anomeric
group. a-Thiomannoside 51 (Aka = 2.90) was found to be less
reactive than a-methyl mannoside 49 (Aka = 4.77). When the
mannose was transformed to 1,6-anhydromannoside 50 to
place the 2-OH in the equatorial instead of the usual axial
position, the reactivity of 50 (Aka = 3.31) became intermedi-
ate, between mannoside 49 and 51. However, not all of these
trends can be explained by basic steric and electronic effects.
Therefore, there must be other factors that influence the
reactivity of alcohols in glycosylation reactions, such as
intramolecular hydrogen bonding,[8b,19] conformation[11c–g,20]
Since DHP cation is prochiral, the formation of the
hemiacetal product would generate a new chiral center.
Indeed, both the R/S diastereoisomers can be formed and
were further confirmed by NMR (Figures S11–S13). How-
ever, according to our previous RRV concept[13] the Aka, that
quantifies the relative acceptor reactivity, was based on the
disappearance of R1OH and R2OH rather than the appear-
ance of THP product, and therefore the R/S chirality of the
corresponding THP hemiacetal product was not taken into
account. Therefore it is reasonable to directly measure
a relative first-order rate constant by comparing the HPLC
signals of R1OH and R2OH before and after the reaction
(Figures S2–S7). The equations (shown in Figure 2B) repre-
sented in the relationship between relative rate constant
ratios (k) and acceptor concentrations [A] towards the
tetrahydropyrylium ion.
The Aka values were acquired from a fully automated
system by upgrading a commercially available HPLC auto-
sampler (Figure 2C), which assured high reproducibility and
was capable of screening 21 nucleophilic hydroxyls automati-
cally within 24 h. The two required reagents and two
substrates were automatically introduced into the sample
loop (100 mL) with HPLC pump. The autosampler mixed the
solutions by bubble movement in the loop. The combination
allowed the test reaction to occur in the sample loop for 1 h
and directly injected the crude solution into the analytical
HPLC column. The Aka of each hydroxyl was measured for
three independent times and determined after taking their
average. Since the minimum injection volume of the auto-
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
ꢀ 2021 Wiley-VCH GmbH
www.angewandte.org
&&&&
These are not the final page numbers!

Angewandte
Research Articles
Chemie
Figure 3. Quantitation of hydroxyl reactivity under acidic condition (Akas). A) Nucleophilic linker. B) Glucosyl acceptors. C) Galactosyl acceptors.
D) C-2 Modified glucosyl acceptors. E) 2-OH Mannosyl acceptors. Decreasing trends are shown.
and bond rotation.[7a,e,21] With the Aka index, we noticed
a-methyl glycoside usually gives higher nucleophilicity than
b-tolyl thioglycoside [for example, 6-OH glucoside: 31 (7.16)
vs. 32 (5.86); 4-OH glucoside: 37 (3.51) vs. 38 (2.68); 2-OH
mannoside: 49 (4.77) vs. 51 (2.90)]. Analyzing glycosyl
acceptor 37 using 500 MHz NMR gave a triplet (t) pattern
on H-4 with a J1,2 value of 9.1 Hz (Figure S15). Compound 38,
on the other hand, showed a triplet (t) peak on H-4 with a J1,2
value of 9.7. This clearly indicated that anomeric functionality
(b-STol or a-OMe) significantly impacts the conformation of
sugar ring and further influences the reactivity of the alcohols.
Besides, it is interesting that different N-substituents on serine
changes the Aka values [for example, 18 (16.9); 24 (6.71)].
The distinct reactivities were probably driven by different
bond rotation on the primary alcohol.[7a,21] This hypothesis
was further supported by our NMR analysis (Figure S16). For
18,[22] the J values are 10.8 and 5.8 Hz on H-b, and 10.8 and
4.3 Hz on H-b’. Compound 24,[23] on the other hands, shows
the J values of 11.1 and 2.9 Hz on H-b and J values of 11.1 and
3.3 Hz on H-b’. A detail investigation to explain these trends
is currently undergoing in our group.
Moreover, recently, fluoroalcohols, such as 2-fluoro-
ethanol, 2,2-difuoroethanol, 2,2,2-trifuoroethanol and hexa-
fluoroisopropanol, have been widely used in the modeling
studies of glycosylation reactions due to their low reactivi-
ties,[8b,d–f,h,10c] but we are so far unable to measure their Aka.
These molecules are not detectable in HPLC using UV and
ELSD (evaporative light scattering detectors) detectors due
to their lack of UV chromophore and low boiling points, and
the determination using NMR was hampered by the overlap
of peaks between the substrate and their THP product
(Figures S17–S21). Although gas chromatography-mass spec-
trophotometer (GC-MS) can be applied to detect and
quantify compounds of low polarity (such as menthol,
adamantanol; Figures S8–S10), these fluoro-modified mole-
cules are not detectable in GC-MS because of their high
polarity. We are currently trying to find a solution for this
difficulty.
To predict the stereoselectivity of glycosylation reactions,
a database of single-step glycosylation reaction was necessary.
We chose the glycosylation in dichloromethane (DCM), as it
is the most commonly used solvent in glycosylation. Based on
&&&&
www.angewandte.org
ꢀ 2021 Wiley-VCH GmbH
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
These are not the final page numbers!

Angewandte
Research Articles
Chemie
the numbers of RRV and Aka, we analyzed our data in the
stereoselectivity was then indicated with colors. The selective
previous work to investigate the stereoselective glycosylation
with 12 thioglycoside donors and 4 types of hydroxyls in DCM
at ꢁ408C under NIS/TfOH activation[8a] (Figure 4). After
combining the statistics of 48 glycosylation examples, we
found that in the absence of a C2 participation, the
stereoselectivity is highly associated with RRV and Aka,
and both a- and b-selectivities can be clearly estimated. The
b-glycosylations are shown in blue, while the selective
a-glycosylation reactions are highlighted in orange. The
higher RRV and lower Aka favor a-selectivity, but lower
RRV and higher Aka favor b-selectivity. A transition from
a-selectivity to b-selectivity was observed from the bottom-
right to the top-left corner in the scaffold with RRV values of
2656–4000 and Aka of 3.51–7.16.
Neighboring group effect[15d,24] and long-range participa-
tion[10c,21b,25] are known to significantly influence the stereo-
selectivity in glycosylation, although the real mechanism and
participating level of carbonyl group are at present still
unclear. To precisely analyze the correlation between the
stereoselectivity and reactivity differences of donor/acceptor
counterparts, we studied the trends mathematically on donors
A-M functionalized with only non-participating groups, and
the RRV and Aka were introduced and emphasized as the
main parameters. Previously, our laboratory discovered that
the a-selectivity showed a roughly linear relationship with
log(RRV) on different acceptors in a NIS/TfOH system.[8a,c]
The a-selectivity was found to increase with increasing
reactivity of donor (RRV), while the b-selectivity was
increased with improving nucleophilicity of acceptor (Aka).
Capitalizing on this observation, we processed our database
by defining log (RRV/Aka) as the X-axis, while the
a-selectivity as the Y-axis (Figure 5). Interestingly, we initially
analyzed 37 examples of glycosylation reactions using
a-glycosyl imidates and the most commonly used TMSOTf
catalyst system in DCM at temperatures ranging from ꢁ788C
to 258C (Figure 5A; Supporting Information, Table S1).
Linear fitting of data led to an Equation (1) (Table 1).
a-ratio ð%Þ ¼ 17:3 ꢃ logðRRV=AkaÞ; RMSE ¼ 15 %
ð1Þ
a-ratio ð%Þ ð¼ 100 % ꢁ b-ratioÞ, R2 ¼ 0:76; Pearson0 R ¼ 0:87
The predicted a-ratio (%) can only range from 0% to 100%.
Therefore, a predicted a-ratio (%) below 0 (a negative
number) is equivalent to exclusive b-selectivity (100% b).
Similarly, a predicted a-ratio (%) exceeding 100 represents
complete a-selectivity (100% a). With the dataset, the RMSE
was 15–16%.
Encouraged by these results, we further extended our
study to a diverse combination of thioglycoside donors and
acceptors in DCM at temperatures ranging from ꢁ788C to
258C (Figure 5B,C; Table S1). We included the examples
reported previously which were performed using the same
solvent (DCM), promoter (TolSCl/AgOTf, NIS/TfOH and
NIS/TMSOTf) and acceptor, and thioglycoside donors with
the same sugar type and protecting group patterns but had
different anomeric thio-functionalities (tolyl thioglycoside,
ethyl thioglycoside, phenyl thioglycoside; the RRVs were
adapted from their corresponding tolyl thioglycosides; Fig-
ure S1) despite the fact that conditions such as concentration
and temperature were slightly different. As our expectations,
this trend is not limited to the a-imidate/TMSOTf system.
Extended glycosylation studies of thioglycosides also clearly
exhibit very similar trends. The RRV/Aka ratio served as
a useful indicator to finalize a universal equation for a diverse
combination of coupling partners. When we analyzed 176
Figure 4. Comparison of the stereoselective glycosylation using 12
types of thioglycoside donors (the reactivity was defined by RRV) and
four types of hydroxyls (the reactivity was defined by Aka). The b-sel-
ectivity is shown in blue, while the a-selectivity is labeled in orange.
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
ꢀ 2021 Wiley-VCH GmbH
www.angewandte.org
&&&&
These are not the final page numbers!

Angewandte
Research Articles
Chemie
Figure 5. Relationship between glycosylation stereoselectivity and RRV/Aka. A) TMSOTf activated a-glycosyl imidate system in DCM. B) TolSCl/
AgOTf activated thioglycoside system in DCM. C) NIS/TfOH activated thioglycoside system in DCM. D) Stereoselective glycosylation through
a fast equilibrium among covalent triflate intermediate, contact ion pair, solvent-separated ion pair and oxocarbenium ion. E) GlycoComputer
program navigating chemical glycosylation.[17] F) Validating the synthesis of Lewis A.
&&&&
www.angewandte.org
ꢀ 2021 Wiley-VCH GmbH
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
These are not the final page numbers!

Angewandte
Research Articles
Chemie
Table 1: Linear fitting of the data using RRV/Aka in the analysis of
stereoselective glycosylation.
when ether (Et2O) or nitrile solvent (ACN or EtCN) is used.
For glucosylation and galactosylation, in the presence of
nitrile co-solvent the glycosylation was expectedly more
b-selective than that in pure DCM. Moreover, b-glycosyla-
tions (b-selectivity > 80%) were observed when log(RRV/
Aka) was lower than 2.5 and a-glycosylation was gradually
increased with increasing log(RRV/Aka) from 2.5 to 6. In
contrast, with Et2O as a co-solvent, the a-selectivity was
significantly elevated as compared to that in pure DCM when
log(RRV/Aka) was lower than 2.5.
Donor, promoter
Equation[a]
a-Glycosyl imidate,
TMSOTf
Thioglycoside,
TolSCl/ AgOTf
Thioglycoside,
NIS/ TfOH
(1) a-ratio (%)=17.3 ꢀ log(RRV/Aka);
R2 =0.76; Pearson’s R=0.87; RMSE=15%
(2) a-ratio (%)=19.5 ꢀ log(RRV/Aka) ꢁ3;
R2 =0.79; Pearson’s R=0.89; RMSE=16%
(3) a-ratio (%)=17.4 ꢀ log(RRV/Aka) + 5;
R2 =0.73; Pearson’s R=0.85; RMSE=15%
[a] a-Ratio [%]=100%ꢁb-ratio [%].
The remote participation effect[10c,21b,25] related to stereo-
selective glycosylation can be preliminarily analyzed in
a statistical manner (Figure S24, Table S3), although more
data are necessary to elucidate a detailed trend. The results of
glycosylation reactions, of which 53 examples were acquired
directly from the literature (Table S1), consistent results
related to the changes in stereoselectivity were observed,
including the reactions with TolSCl/AgOTf promoted thio-
glycosides (Figure 5B) and NIS/TfOH activated thioglyco-
sides (Figure 5C) in DCM. The trend of distribution among
the three regression lines gave similar root-mean-square error
(RMSE) of about 15–16% and the Pearsonꢁs R of about 0.85–
0.89. In addition, the corresponding R2 is in the range between
0.73 and 0.79, implying a general distribution existed, despite
the examples from the literature were carried out in slightly
different conditions. Therefore, Equation (2) (TolSCl/
AgOTf) and Equation (3) (NIS/TfOH) can be introduced,
respectively to summarize the stereoselectivity in the thio-
glycoside activation system (Table 1, entries 2 and 3).
a
series of acetylated 2-azido-2-deoxy-thioglucoside
(GlcN3)[25a] and 2-azido-2-deoxygalactosyl (GalN3)[25b] donors
were obtained from literature. It was observed that the
remote acyl participation effect did cause significant devia-
tions from the standard trend, in agreement with the
mechanistic study by Pagel, Codꢀe, Boltje and Li.[10c,25b,c]
The deviation from the non-participation slope can indicate
the level of long range participation, and our preliminary
statistical analysis showed that so far the C4 long-range
participation on GalN3 enhances the a-glycosylation most
significantly.
The GlycoComputer program[17] was designed to include
comments from users (Figure 5E). The customer feedback
and rating scale can help us understand the deficiency in this
RRV/Aka platform. Moreover, human variables from differ-
ent users are considered, and we would adopt the suggestions
and repeat the reaction to further optimize the equation, so
the precision and accuracy of the program could be further
improved and the data base could be enlarged.
a-ratio ð%Þ ¼ 19:5 ꢃ logðRRV=AkaÞ ꢁ3; RMSE ¼ 16 %
ð2Þ
a-ratio ð%Þ ð¼ 100 %ꢁb-ratioÞ, R2 ¼ 0:79; Pearson0 R ¼ 0:89;
a-ratio ð%Þ ¼ 17:4 ꢃ logðRRV=AkaÞ þ 5; RMSE ¼ 15 %
ð3Þ
a-ratio ð%Þ ð¼ 100 %ꢁb-ratioÞ, R2 ¼ 0:73; Pearson0 R ¼ 0:85;
We also verified the GlycoComputer program by synthe-
sizing a Lewis A skeleton 57 which contains a branched core
of a-Fuc-(1!4)-b-GlcNAc and b-Gal-(1!3)-b-GlcNAc with
a b-pentyl amino linker (Figure 5F). The predicted stereose-
lectivities and yields were compared with practical experi-
ments. Our GlycoComputer evaluated each building block by
foreseeing the results and obviating trial and error process.
According to the GlycoComputer,[17] to obtain a b-glucos-
amine glycoside in high selectivity, the donor should have
a low RRV; thus, semi-protected glucosamine donor 52
(RRV= 75) was chosen for the first glycosylation with linker
22 (Aka = 12). Indeed, this NIS/TfOH promoted glycosyla-
tion in DCM furnished glucosamine 53 up to 81% with an a/b
of 22/78, which agreed well with our prediction (predicted:
65–80% yield and a/b = 20/80 ꢄ 15). Next, the subsequent
a-fucosylation at O4 of 53 (Aka = 2.79) required a donor with
Despite the differences in promoter systems, similar
trends were still observed, but with slightly different sensitiv-
ities (slopes) related to the change in the reactant reactivity. It
suggested that the promoter could potentially influence the
distributions among the covalent intermediate (glycosyl
triflate) (4), contact-ion pairs (4-CIP), solvent-separated ion
pair (4-SSIP) and oxocarbenium ion (Figure 1B),[5,26] which
could shift the SN1/SN2 interface and leads to the stereoselec-
tivity changes[5,7,8] (Figure 5D).
It was observed that both RRVs of donors and Akas of
acceptors served as simple and successful indicators that
could correlate the saccharide reactivity and stereoselectivity
(Figure 5D). A clear transition from the a-selectivity to the
b-selectivity was noted in the scaffold ranging from log(RRV/
Aka) of ꢁ1 to 6. The most favorable a-glycosylation reaction
(a-selectivity more than 80%) was performed when log-
(RRV/Aka) was higher than 4.0.
We then extended our study to the ACN/DCM (v/v = 3/1)
and Et2O/DCM (v/v = 3/1) systems to analyze the solvent
effect on the formation of glycosides, galactosides and
2-deoxy glucosides (44 examples, Figure S22, Table S2).
The preliminary data are aligned with previous work on the
study of stereoselective glycosylation affected by sol-
vents.[6c,27] But, more data are needed to conclude a linear
relationship between stereoselectivity and log(RRV/Aka)
high RRV;
a simple per-O-benzylated thiofucoside 54
(RRV= 72000) was used directly. The NIS/TfOH-promoted
fucosylation resulted in an a-disaccharide 55 at 80%
(a/b = 82:18) and again matched with the GlycoComputer
predicted value: 66–80%, a/b = 82/18 ꢄ 15. After removal
of the C3 acetyl group to give the 3-OH acceptor 56
(Aka = 1.51), the product was reacted with per-O-benzylated
donor I (RRV= 17000) to give the galactosyl linkage. The
GlycoComputer suggested a moderate yield (prediction:
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
ꢀ 2021 Wiley-VCH GmbH
www.angewandte.org
&&&&
These are not the final page numbers!

Angewandte
Research Articles
Chemie
57–69%, a/b = 76/24 ꢄ 15) and indeed, the target trisacchar-
ide 57 was isolated in 67% with a/b = 89/11.
forest algorithm yielded a similar result, giving R2 of 0.81 and
RMSE of 15.8%.
Next, we carried out a random forest algorithm to
understand the role of reaction counterparts (glycosyl donor
and nucleophilic acceptor) involved in controlling the stereo-
selectivity (Figure 6).[28] RRV and Aka were obtained in
a competition experiment using the model reactions (with
MeOH and DHP respectively) and clear statistical trends
could already be observed. Since machine learning algorithm
can deal with numerous parameters simultaneously to analyze
a set of complicated data, we introduced a number of
“descriptors” that took the structural characterization of
sugar substrates of both donor and acceptor into account. We
applied 20 descriptors to analyze the stereochemistry, func-
tionality and alcohol types (such as primary, secondary,
tertiary, benzyl and aromatic) as reported (Figure S25).[29]
Several additional potential factors were also applied, includ-
ing selection of promoter and temperature, and a Table
summarizing all these parameters is provided. The training set
Regarding the importance obtained from random forest
algorithm (Figure 6B; Figure S25), both RRV (33%) and
Aka (13%) showed the highest impact on donor and acceptor
individually, indicating quantitative assessment of reactivity
can successfully represent most of the potential factors on the
structural character of building block, such as the random
contribution of steric, inductive and structural effects. In
addition, the sum of donor effect (59%) gave a higher impact
than the acceptor (24%). According to the work by Seeberger
et al.,[5] a significant impact of promoter and temperature
(ꢁ508C to + 308C), on the stereoselectivity of glycosylation
in the automated flow reactor was observed. However, based
on our RRV/Aka and random forest analysis, the influence of
promoter (2.0%) and temperature (0.9%) was minor at low
temperatures ranging from ꢁ708C to ꢁ408C.
was established by adopting our experimental data (139 Conclusion
examples) in DCM at the temperatures ranging from ꢁ708C
to ꢁ408C, and the result showed an RMSE of 5.4% and an R2
of 0.97. We also applied five-fold cross validation,[30] which is
a common statistical approach, to test the performance. The
test error rate of the training set was relatively low, of which
the a-selectivity showed an RMSE of 15.1% and an R2 of 0.82
(Figure 6A). Next, a separate test set (29 examples) was
compiled from recent literature to validate our system. The
NIS/TMSOTf promoter system was adopted into our stan-
dard promoter system of NIS/TfOH. Gladly, the random
In summary, chemical synthesis is one of the best tools to
access homogeneous oligosaccharides with well-defined con-
figurations. However, optimizing the synthesis is complicated
and highly time consuming due to our limited ability to
control the glycosylation reactions, for which the stereoselec-
tivity and yield is influenced by numerous factors. Tedious
trial and error experimentation in the synthesis of oligosac-
charides have long been a rate-limiting advancement in
glycoscience. We have established a new index (Aka) to
define the reactivity of hydroxyl
groups under acidic conditions and this
useful indicator can be acquired/deter-
mined through competition experi-
ments using an automated system.
With our reactivity database (RRV of
donor and Aka of acceptor) as two
common numeric parameters, we can
now predict the glycosylation reaction
for donors bearing non-participating
groups (such as benzyl or benzylidene
protection). In addition, the assess-
ment of building-block reactivity
(RRV of donor and Aka of acceptor)
is relatively easy through simple and
fast competition experiments, and data
regression does not require complicat-
ed computational calculations. The
RRV and Aka take both the inductive
effect and steric effect into account
and systematically organize 237 glyco-
sylation reactions in the same compar-
ison system, and the stereoselectivity
of glycosylation reaction can be further
confirmed by random forest model. In
the donor system without participating
Figure 6. A) The random forest model was built with 20 descriptors, including donor effect (RRV,
C2-funtionalities and stereochemistry, etc.), acceptor effect (Aka, types of hydroxyls and their
position etc.) and environmental effects (promoter and temperature). B) Importance related to
a-selectivity.
protecting groups, we try to include as
many factors that affect the stereo-
selectivity and yield in the synthesis of
&&&&
www.angewandte.org
ꢀ 2021 Wiley-VCH GmbH
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
These are not the final page numbers!

Angewandte
Research Articles
Chemie
padhyay, ChemistryOpen 2016, 5, 401 – 433; g) Y. Yang, X.
oligosaccharides, including benzyl/benzylidene donors,
acceptors, activators and solvents, as possible in the predicting
system. The building blocks for oligosaccharide synthesis can
be more accurately selected using the RRV-Aka database in
the GlycoComputer program. Since neighboring group ef-
fects,[15d,24] long-range participation[10c,21b,25] and temperature
effects[5,29] are all important factors to control the stereo-
selectivity of glycosylation reactions, the numeric analysis of
stereoselectivity change on acylated donors is now underway
at various temperatures. We are also trying to connect the
relationship of stereoselectivity and intermediate change
using linear free energy relationships (LFER).[31]
Zhang, B. Yu, Nat. Prod. Rep. 2015, 32, 1331 – 1355.
[5] S. Chatterjee, S. Moon, F. Hentschel, K. Gilmore, P. H.
Seeberger, J. Am. Chem. Soc. 2018, 140, 11942 – 11953.
[6] a) S. K. Mulani, W.-C. Hung, A. B. Ingle, K.-S. Shiau, K.-K. T.
Mong, Org. Biomol. Chem. 2014, 12, 1184 – 1197; b) C.-S. Chao,
C.-W. Li, M.-C. Chen, S.-S. Chang, K.-K. T. Mong, Chem. Eur. J.
2009, 15, 10972 – 10982; c) K.-K. T. Mong, T. Nokami, N.
Thi Thanh Tran, P. Be Nhi, Selective Glycosylations: Synthetic
Methods and Catalysts (Ed.: C. S. Bennett), Wiley, Hoboken,
2017, chap. 3, pp. 59 – 77.
[7] a) P. O. Adero, H. Amarasekara, P. Wen, L. Bohꢀ, D. Crich,
Chem. Rev. 2018, 118, 8242 – 8284; b) M. Huang, G. E. Garrett,
N. Birlirakis, L. Bohꢀ, D. A. Pratt, D. Crich, Nat. Chem. 2012, 4,
663 – 667; c) T. G. Frihed, M. Bols, C. M. Pedersen, Chem. Rev.
2015, 115, 4963 – 5013; d) P. O. Adero, T. Furukawa, M. Huang,
D. Mukherjee, P. Retailleau, L. Bohꢀ, D. Crich, J. Am. Chem.
Soc. 2015, 137, 10336 – 10345; e) L. Bohꢀ, D. Crich, Carbohydr.
Res. 2015, 403, 48 – 59; f) A. G. Santana, L. Montalvillo-Jimenez,
L. Diaz-Casado, F. Corzana, P. Merino, F. J. Canada, G. Jimenez-
Oses, J. Jimenez-Barbero, A. M. Gomez, J. L. Asensio, J. Am.
Chem. Soc. 2020, 142, 12501 – 12514; g) J. L. Asensio, A. G.
Santana, L. Montalvillo-Jimenez, L. Diaz-Casado, E. Mann, J.
Jimenez-Barbero, A. M. Gomez, Chem. Eur. J. 2020, 27, 2032 –
2042; h) M. Huang, P. Retailleau, L. Bohꢀ, D. Crich, J. Am.
Chem. Soc. 2012, 134, 14746 – 14749.
[8] a) C.-W. Chang, C.-H. Wu, M.-H. Lin, P.-H. Liao, C.-C. Chang,
H.-H. Chuang, S.-C. Lin, S. Lam, V. P. Verma, C.-P. Hsu, C.-C.
Wang, Angew. Chem. Int. Ed. 2019, 58, 16775 – 16779; Angew.
Chem. 2019, 131, 16931 – 16935; b) S. van der Vorm, T. Hansen,
J. M. A. van Hengst, H. S. Overkleeft, G. A. van der Marel,
J. D. C. Codꢀe, Chem. Soc. Rev. 2019, 48, 4688 – 4706; c) C.-W.
Chang, M.-H. Lin, C.-C. Wang, Chem. Eur. J. 2021, 27, 2556 –
2568; d) S. van der Vorm, T. Hansen, H. S. Overkleeft, G. A.
van der Marel, J. D. C. Codꢀe, Chem. Sci. 2017, 8, 1867 – 1875;
e) S. van der Vorm, H. S. Overkleeft, G. A. van der Marel,
J. D. C. Codꢀe, J. Org. Chem. 2017, 82, 4793 – 4811; f) S.
van der Vorm, J. M. A. van Hengst, M. Bakker, H. S. Overkleeft,
G. A. van der Marel, J. D. C. Codꢀe, Angew. Chem. Int. Ed. 2018,
57, 8240 – 8244; Angew. Chem. 2018, 130, 8372 – 8376; g) M. H.
Zhuo, D. J. Wilbur, E. E. Kwan, C. S. Bennett, J. Am. Chem. Soc.
2019, 141, 16743 – 16754; h) T. Hansen, T. P. Ofman, J. G. C.
Vlaming, I. A. Gagarinov, J. van Beek, T. A. Gote, J. M. Tichem,
G. Ruijgrok, H. S. Overkleeft, D. V. Filippov, G. A. van der Mar-
el, J. D. C. Codꢀe, Angew. Chem. Int. Ed. 2021, 60, 937 – 945;
Angew. Chem. 2021, 133, 950 – 958.
[9] a) O. Takahashi, K. Yamasaki, Y. Kohno, R. Ohtaki, K. Ueda, H.
Suezawa, Y. Umezawa, M. Nishio, Carbohydr. Res. 2007, 342,
1202 – 1209; b) I. Cumpstey, Org. Biomol. Chem. 2012, 10, 2503 –
2508; c) R. J. Meredith, R. J. Woods, I. Carmichael, A. S.
Serianni, Phys. Chem. Chem. Phys. 2020, 22, 14454 – 14457.
[10] a) T. Hansen, L. Lebedel, W. A. Remmerswaal, S. van der Vorm,
D. P. A. Wander, M. Somers, H. S. Overkleeft, D. V. Filippov, J.
Desire, A. Mingot, Y. Bleriot, G. A. van der Marel, S. Thibau-
deau, J. D. C. Codꢀe, ACS Cent. Sci. 2019, 5, 781 – 788; b) S.
van der Vorm, T. Hansen, E. R. van Rijssel, R. Dekkers, J. M.
Madern, H. S. Overkleeft, D. V. Filippov, G. A. van der Marel,
J. D. C. Codee, Chem. Eur. J. 2019, 25, 7149 – 7157; c) T. Hansen,
H. Elferink, J. M. A. van Hengst, K. J. Houthuijs, W. A. Re-
mmerswaal, A. Kromm, G. Berden, S. van der Vorm, A. M. Rijs,
H. S. Overkleeft, D. V. Filippov, F. Rutjes, G. A. van der Marel, J.
Martens, J. Oomens, J. D. C. Codꢀe, T. J. Boltje, Nat. Commun.
2020, 11, 2664.
Acknowledgements
We thank Ping-Yu Lin (IOC, Academia Sinica) for mass
measurement and Chun-Lu Cheng (IOC, Academia Sinica)
for establishing the GlycoComputer. This work was supported
by the Ministry of Science and Technology of Taiwan (MOST
108-2113-M-001-019-; MOST 106-2113-M-001-009-MY2) and
Academia Sinica (MOST 108-3114-Y-001-002; AS-SUMMIT-
109). C.-W.C. thanks MOST, Taiwan for financial support
(MOST 109-2811-M-001-604-).
Conflict of interest
The authors declare no conflict of interest.
Keywords: carbohydrates · diastereoselectivity · hydroxyl ·
glycosylation · predictive algorithms
[1] a) S. Steiner, J. Wolf, S. Glatzel, A. Andreou, J. M. Granda, G.
Keenan, T. Hinkley, G. Aragon-Camarasa, P. J. Kitson, D.
Angelone, L. Cronin, Science 2019, 363, eaav2211; b) J. M.
Granda, L. Donina, V. Dragone, D. L. Long, L. Cronin, Nature
2018, 559, 377 – 381; c) M. Panza, K. J. Stine, A. V. Demchenko,
Chem. Commun. 2020, 56, 1333 – 1336; d) M. Panza, S. G.
Pistorio, K. J. Stine, A. V. Demchenko, Chem. Rev. 2018, 118,
8105 – 8150.
[2] a) D. Caramelli, D. Salley, A. Henson, G. A. Camarasa, S.
Sharabi, G. Keenan, L. Cronin, Nat. Commun. 2018, 9, 3406;
b) C. W. Coley, W. H. Green, K. F. Jensen, Acc. Chem. Res. 2018,
51, 1281 – 1289; c) K. T. Butler, D. W. Davies, H. Cartwright, O.
Isayev, A. Walsh, Nature 2018, 559, 547 – 555; d) I. W. Davies,
Nature 2019, 570, 175 – 181.
[3] a) S. S. Pinho, C. A. Reis, Nat. Rev. Cancer 2015, 15, 540 – 555;
b) T. J. Boltje, T. Buskas, G.-J. Boons, Nat. Chem. 2009, 1, 611;
c) J. Munkley, Endocr.-Relat. Cancer 2017, 24, R49 – R64; d) A.
Varki, Glycobiology 2017, 27, 3 – 49; e) L. Krasnova, C.-H.
Wong, Annu. Rev. Biochem. 2016, 85, 599 – 630; f) S. J. Dani-
shefsky, Y.-K. Shue, M. N. Chang, C.-H. Wong, Acc. Chem. Res.
2015, 48, 643 – 652.
[4] a) W.-L. Leng, H. Yao, J.-X. He, X.-W. Liu, Acc. Chem. Res.
2018, 51, 628 – 639; b) S. S. Kulkarni, C.-C. Wang, N. M. Sabba-
varapu, A. R. Podilapu, P.-H. Liao, S.-C. Hung, Chem. Rev. 2018,
118, 8025 – 8104; c) M. M. Nielsen, C. M. Pedersen, Chem. Rev.
2018, 118, 8285 – 8358; d) H. Yao, M. D. Vu, X.-W. Liu,
Carbohydr. Res. 2019, 473, 72 – 81; e) S. S. Park, H. W. Hsieh, J.
Gervay-Hague, Molecules 2018, 23, 1742; f) R. Das, B. Mukho-
[11] a) L. K. Mydock, A. V. Demchenko, Org. Biomol. Chem. 2010,
8, 497 – 510; b) P. Konradsson, U. E. Udodong, B. Fraser-Reid,
Tetrahedron Lett. 1990, 31, 4313 – 4316; c) H. H. Jensen, C. M.
Pedersen, M. Bols, Chem. Eur. J. 2007, 13, 7576 – 7582; d) C. M.
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
ꢀ 2021 Wiley-VCH GmbH
www.angewandte.org
&&&&
These are not the final page numbers!

Angewandte
Research Articles
Chemie
Pedersen, L. U. Nordstrøm, M. Bols, J. Am. Chem. Soc. 2007,
129, 9222 – 9235; e) M. Heuckendorff, H. D. Premathilake, P.
Pornsuriyasak, A. Ø. Madsen, C. M. Pedersen, M. Bols, A. V.
Demchenko, Org. Lett. 2013, 15, 4904 – 4907; f) M. Bols, C. M.
Pedersen, Beilstein J. Org. Chem. 2017, 13, 93 – 105; g) M. J.
Pedersen, C. M. Pedersen, Angew. Chem. Int. Ed. 2021, 60,
2689 – 2693; Angew. Chem. 2021, 133, 2721 – 2725.
[22] a) M. Brandstꢃtter, F. Roth, N. W. Luedtke, J. Org. Chem. 2015,
80, 40 – 51; b) J. Brekalo, G. Despras, T. K. Lindhorst, Org.
Biomol. Chem. 2019, 17, 5929 – 5942; c) K. Fujiki, K. Tanaka,
Eur. J. Org. Chem. 2020, 4616 – 4620.
[23] a) H. L. Stewart, A. R. Hanby, T. A. King, A. D. Bond, T. A.
Moss, H. F. Sore, D. R. Spring, Chem. Commun. 2020, 56, 6818 –
6821; b) P. M. E. Hawkins, A. M. Giltrap, G. Nagalingam, W. J.
Britton, R. J. Payne, Org. Lett. 2018, 20, 1019 – 1022.
[24] a) J. Guo, X. S. Ye, Molecules 2010, 15, 7235 – 7265; b) E. Mucha,
M. Marianski, F. F. Xu, D. A. Thomas, G. Meijer, G. von Helden,
P. H. Seeberger, K. Pagel, Nat. Commun. 2018, 9, 4174.
[25] a) G. Ngoje, Z. Li, Org. Biomol. Chem. 2013, 11, 1879 – 1886;
b) J. Kalikanda, Z. Li, J. Org. Chem. 2011, 76, 5207 – 5218; c) M.
Marianski, E. Mucha, K. Greis, S. Moon, A. Pardo, C.
Kirschbaum, D. A. Thomas, G. Meijer, G. von Helden, K.
Gilmore, P. H. Seeberger, K. Pagel, Angew. Chem. Int. Ed.
2020, 59, 6166 – 6171; Angew. Chem. 2020, 132, 6224 – 6229;
d) A. V. Demchenko, E. Rousson, G.-J. Boons, Tetrahedron Lett.
1999, 40, 6523 – 6526; e) D. Crich, J. Am. Chem. Soc. 2020, 142,
7760 – 7764; f) A. Khanam, A. Tiwari, P. K. Mandal, Carbohydr.
Res. 2020, 495, 108045; g) A. A. Hettikankanamalage, R. Lass-
folk, F. S. Ekholm, R. Leino, D. Crich, Chem. Rev. 2020, 120,
7104 – 7151.
[12] a) N. L. Douglas, S. V. Ley, U. Lꢂcking, S. L. Warriner, J. Chem.
Soc. Perkin Trans. 1 1998, 51 – 66; b) P. Grice, S. V. Ley, J.
Pietruszka, H. W. M. Priepke, E. P. E. Walther, Synlett 1995,
781 – 784; c) M.-K. Cheung, N. L. Douglas, B. Hinzen, S. V. Ley,
X. Pannecoucke, Synlett 1997, 257 – 260; d) P. Grice, S. V. Ley, J.
Pietruszka, H. M. I. Osborn, H. W. M. Priepke, S. L. Warriner,
Chem. Eur. J. 1997, 3, 431 – 440.
[13] a) C.-W. Cheng, Y. Zhou, W.-H. Pan, S. Dey, C.-Y. Wu, W.-L.
Hsu, C.-H. Wong, Nat. Commun. 2018, 9, 5202; b) C. W. Cheng,
C. Y. Wu, W. L. Hsu, C. H. Wong, Biochemistry 2020, 59, 3078 –
3088; c) H. J. Lo, L. Krasnova, S. Dey, T. Cheng, H. Liu, T. I.
Tsai, K. B. Wu, C. Y. Wu, C. H. Wong, J. Am. Chem. Soc. 2019,
141, 6484 – 6488; d) S. Dey, C.-H. Wong, Chem. Sci. 2018, 9,
6685 – 6691; e) K.-K. T. Mong, C.-H. Wong, Angew. Chem. Int.
Ed. 2002, 41, 4087 – 4090; Angew. Chem. 2002, 114, 4261 – 4264;
f) Z. Zhang, I. R. Ollmann, X.-S. Ye, R. Wischnat, T. Baasov, C.-
H. Wong, J. Am. Chem. Soc. 1999, 121, 734 – 753.
[26] Y. Fu, L. Bernasconi, P. Liu, J. Am. Chem. Soc. 2021, 143, 1577 –
1589.
[14] M. Tsutsui, J. Sianturi, S. Masui, K. Tokunaga, Y. Manabe, K.
Fukase, Eur. J. Org. Chem. 2020, 1802 – 1810.
[27] a) A. Demchenko, T. Stauch, G.-J. Boons, Synlett 1997, 818 – 820;
b) H. Satoh, H. S. Hansen, S. Manabe, W. F. van Gunsteren, P. H.
Hꢂnenberger, J. Chem. Theory Comput. 2010, 6, 1783 – 1797;
c) C. S. Chao, C. Y. Lin, S. Mulani, W. C. Hung, K.-K. T. Mong,
Chem. Eur. J. 2011, 17, 12193 – 12202.
[28] a) M. K. Nielsen, D. T. Ahneman, O. Riera, A. G. Doyle, J. Am.
Chem. Soc. 2018, 140, 5004 – 5008; b) D. T. Ahneman, J. G.
Estrada, S. Lin, S. D. Dreher, A. G. Doyle, Science 2018, 360,
186 – 190.
[15] a) T. Nukada, A. Bꢀrces, D. M. Whitfield, Carbohydr. Res. 2002,
337, 765 – 774; b) W. Yang, B. Yang, S. Ramadan, X. Huang,
Beilstein J. Org. Chem. 2017, 13, 2094 – 2114; c) C. M. Pedersen,
L. G. Marinescu, M. Bols, Chem. Commun. 2008, 0, 2465 – 2467;
d) Y. Zeng, Z. Wang, D. Whitfield, X. Huang, J. Org. Chem.
2008, 73, 7952 – 7962.
[16] a) H. M. Christensen, S. Oscarson, H. H. Jensen, Carbohydr.
Res. 2015, 408, 51 – 95; b) C. W. Chang, M. H. Lin, C. H. Wu,
T. Y. Chiang, C. C. Wang, J. Org. Chem. 2020, 85, 15945 – 15963.
[29] S.-Y. Moon, S. Chatterjee, P. Seeberger, K. Gilmore, Chem. Sci.
2021, https://doi.org/10.1039/d0sc06222g.
[17] GlycoComputer:
http://chemwww.chem.sinica.edu.tw/
ChemicalGlycosylation/index.php.
[18] K. F. Bernady, M. B. Floyd, J. F. Poletto, M. J. Weiss, J. Org.
Chem. 1979, 44, 1438 – 1447.
[19] W. Du, S. S. Kulkarni, J. Gervay-Hague, Chem. Commun. 2007,
2336 – 2338.
[30] S. Boyd, L. Vandenberghe An Introduction to Statistical Learn-
ing: with Applications in R, Springer, Heidelberg, 2013, p. 184.
[31] a) J. K. Lassila, J. G. Zalatan, D. Herschlag, Annu. Rev. Biochem.
2011, 80, 669 – 702; b) J. C. Bradley, M. H. Abraham, W. E.
Acree, Jr., A. S. Lang, Chem. Cent. J. 2015, 9, 12.
[20] a) M. Heuckendorff, J. Bendix, C. M. Pedersen, M. Bols, Org.
Lett. 2014, 16, 1116 – 1119; b) T. G. Frihed, M. T. C. Walvoort,
J. D. C. Codꢀe, G. A. van der Marel, M. Bols, C. M. Pedersen, J.
Org. Chem. 2013, 78, 2191 – 2205; c) M. Heuckendorff, C. M.
Pedersen, M. Bols, J. Org. Chem. 2013, 78, 7234 – 7248.
[21] a) M. Moumꢀ-Pymbock, T. Furukawa, S. Mondal, D. Crich, J.
Am. Chem. Soc. 2013, 135, 14249 – 14255; b) S. Dharuman, H.
Amarasekara, D. Crich, J. Org. Chem. 2018, 83, 10334 – 10351.
Manuscript received: October 16, 2020
Revised manuscript received: February 7, 2021
Accepted manuscript online: February 26, 2021
Version of record online: && &&
,
&&&&
&&&&
www.angewandte.org
ꢀ 2021 Wiley-VCH GmbH
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
These are not the final page numbers!

Angewandte
Research Articles
Chemie
Research Articles
Glycoscience
A so-called “GlycoComputer” program
has been developed to foresee and pre-
dict the yield and stereoselectivity of gly-
cosylation reactions based on the prop-
erties of various donors, acceptors, acti-
vation systems and solvents. The pro-
gram statistically analyzes and compares
the relative reactivity value (RRV) of
donors and the acceptor nucleophilic
constant (Aka) of acceptors.
C.-W. Chang, M.-H. Lin, C.-K. Chan,
K.-Y. Su, C.-H. Wu, W.-C. Lo, S. Lam,
Y.-T. Cheng, P.-H. Liao, C.-H. Wong,*
C.-C. Wang*
&&&& — &&&&
Automated Quantification of Hydroxyl
Reactivities: Prediction of Glycosylation
Reactions
Angew. Chem. Int. Ed. 2021, 60, 2 – 13
ꢀ 2021 Wiley-VCH GmbH
www.angewandte.org
&&&&
These are not the final page numbers!