Synthesis Strategies — Cross-Cutting Comparison

This note compares the strategies a synthetic chemist actually chooses between when planning a target — convergent vs linear, choice of protecting groups, retrosynthesis tool, catalysis platform (transition-metal vs organo vs photoredox vs enzymatic vs electro), reactor mode (batch vs flow vs SPPS), and synthesis era exemplar — across every Chemistry library note that touches synthesis. Read each section’s table to see how a given dimension breaks down across topics; the final decision tree is the practical “which strategy do I pick for this target” view.

See also

1. The two architectural questions

Every total synthesis answers two architectural questions before a flask is touched.

  1. Convergent vs linear? A linear synthesis adds one piece at a time; total step count goes up linearly with size and yield decays geometrically (0.85^n at 85% per step → 20% at 10 steps, 4% at 20 steps). A convergent synthesis builds fragments in parallel and joins them late; the longest linear sequence is what controls overall yield, not the total step count. For any target above ~15 steps, convergent is mandatory.
  2. Protecting-group strategy? Bn (benzyl, hydrogenolysis), TBS/TBDPS (silyl, F⁻), Boc (acid), Fmoc (base), Cbz (H2), Ac (base), Bz (base), MOM/MEM (acid), allyl (Pd), PMB (DDQ/oxidation), THP (acid). Orthogonality means each protecting group can be removed without touching the others — a four-orthogonal set (e.g., Boc + Bn + TBS + Fmoc) covers most peptide and natural-product syntheses.
Architectural choiceWhen to useExamplePenalty if wrong
Linear< 8 steps, all transformations high-yield, no large groups to add lateearly-route discovery workstep count compounds; yield collapses
Convergent> 12 steps, modular target, late-stage diversificationTaxol (Nicolaou, Holton), palytoxin (Kishi)over-engineering for tiny targets
Block-synthesis (peptide, oligo, glycan)repeating subunitSPPS Merrifield, oligo synthesislocked into available monomers
Combinatorial / DEL (DNA-encoded library)screening (no scale-up intent)hit-finding for medchemnot for resynthesis or scale
Biocatalysiswhen wild-type enzyme matches functional-group operationsitagliptin (Codexis–Merck transaminase 2010)enzyme engineering cost
Flow chemistryhazardous intermediates, photochemistry, fast exothermsdiazomethane in-situ, photoredox at scaleupfront capex; limited mixing for slow reactions

2. Retrosynthetic analysis — the strategy layer

E. J. Corey’s retrosynthesis (Nobel 1990 in Chemistry, “for his development of the theory and methodology of organic synthesis”) is the formal strategy: dissect a target into precursors by reversing known reactions until commercial materials remain. The disconnections divide into:

  • Synthon-based — Corey’s original. Identify electrophile/nucleophile pairs at strategic bonds.
  • Functional-group-based — work backward from FG manipulations.
  • Strategic-bond-based — Bertz, Hudlický measures of complexity; pick bonds whose disconnection maximally simplifies the graph.
  • Transform-based (computer-aided) — match named reactions in a database, score by feasibility.
ToolEraApproachStatus (2026)
LHASA (Corey, Harvard)1969–1980srule-based, expert-curateddiscontinued; foundational
SYNTHIA (formerly Chematica, Grzybowski)2018+ (Sigma-Aldrich Merck KGaA)rule-based + machine learning + ChemAxoncommercial, BIOVIA-grade
Reaxys + Synthia1990s+ (Elsevier acquired 2009)reaction database (>50M reactions) + similarity searchdominant in industry
ASKCOS (MIT, Coley, Jensen 2018+)open sourcetemplate-based neural retrosynthresearch/open
AiZynthFinder (AstraZeneca 2020+)open sourcetemplate-based, MCTSresearch/pharma
Synthia + IBM RoboRXN (2019+)cloudcombine retrosynthesis + robotic executionearly commercial
Chemical.AI / Innosynthstartups 2020+transformer-based reaction predictiongrowing

The state of the art in 2026 is hybrid — rule-based templates for common reactions, graph-neural-network or transformer prediction for novel disconnections, and beam-search/MCTS planning over the reaction graph. Grzybowski’s 2018 Nature paper “Efficient syntheses of diverse, medicinally relevant targets planned by computer and executed in the laboratory” demonstrated Chematica planning syntheses that human chemists then ran with no replanning. By 2025 Sigma-Aldrich’s Synthia (the productized Chematica) is integrated into Merck KGaA’s full medchem stack.

3. Total-synthesis exemplars — the historical curve

These syntheses are the field’s milestones. Each pushed at least one frontier and is taught as a strategic case study.

TargetYearLeadStepsStrategyLasting contribution
Strychnine1954R. B. Woodward (Harvard)28 (LLS)mostly linear, classic ring-constructionfirst total synthesis of a complex alkaloid; founded “Woodward school”
Cholesterol1951Woodward~36mostly linearfirst steroid total synthesis
Quinine1944 / 2001 confirmedWoodward–Doering / Storkvariedlinearresolved 1944 controversy; teaching example
Reserpine1956Woodward14convergent (early)Woodward at his peak; stereocontrol via conformational analysis
Chlorophyll a1960Woodward + Strell55linear; modular ring constructionextended porphyrin chemistry
Vitamin B121973Woodward + Eschenmoser~70 (combined effort)convergent, joint Harvard-ETHthe Mount Everest of total synthesis; ~100 chemists, ~12 years
Erythromycin A1981Woodward (posthumous)50+convergentmacrolide construction (Yamaguchi macrolactonization)
Palytoxin1994Y. Kishi (Harvard)140+hyperconvergent, 8 fragmentsmost complex natural product ever made; 64 stereocenters
Taxol1994Nicolaou (Scripps) / Holton (FSU); also Wender, Mukaiyama, Danishefskyvarieddistinct strategies — convergent, semisyntheticfive total + multiple semisynthetic routes; classic comparison study
Strychnine (modern)2011Vanderwal (UCI)6shortest known; Diels-Alder cascadeshows how RS + modern cat shrinks step counts
Avermectin1989Hanessian (Montreal)35+convergentmacrolide w/ challenging hexahydrobenzofuran
Eribulin (Halaven)2009Kishi (Eisai)62convergentmost complex marketed drug ever made; ~$200M/yr revenue
Sirolimus / rapamycin2003Nicolaou; later Wender, Smith, Danishefsky30+macrocyclic, convergentmTOR inhibitor scaffold
Discodermolide2004Smith / Paterson / Schreiber / Novartis25+convergent + scaled to 60 g (Novartis)rare example of scaled total synthesis for clinical trial
Vinblastine2009Boger (Scripps)11convergent dimerizationbio-inspired indole-indole coupling
Maoecrystal V2010Yang (HK) / Reisman / Danheiser17–25convergentbridged terpene
Welwitindolinone A2011Garg (UCLA) / Rawal15convergentindol benzofuran; combined photoredox-era thinking
Strychnos alkaloid family2020MacMillan + Samesvariedphotoredox-driven C–Henabling-step contributions

By the 2010s the field had matured to where the question was no longer “can we make X” but “can we make X in 5 steps with 30% yield from commercial materials”. Vanderwal’s 6-step strychnine (2011) crystallized this. Photoredox, organocatalysis, and C–H functionalization are why.

4. The catalysis platform spectrum — choose your bond-forming engine

The bond-forming engines that 2020-era chemists choose between, and the Nobel timeline:

EraPlatformKey chemistsNobelFootprint in synthesis (2026)
pre-1980classical (Grignard, aldol, Wittig, Diels–Alder)Grignard 1912, Diels-Alder 1950, Wittig 1979manyfoundational, still dominant for simple FG operations
1980searly TM cross-couplingNegishi, Suzuki, Miyaura, Stille, Sonogashira(2010 Heck/Negishi/Suzuki)Pd-catalyzed C–C, every medchem program
1990smetathesisChauvin, Grubbs, Schrock2005RCM for macrocycles, CM for fragment joining
1990sasymmetric catalysisKnowles, Noyori, Sharpless2001chiral hydrogenation, dihydroxylation, epoxidation
2000sorganocatalysisList, MacMillan2021iminium / enamine, Brønsted acid, NHCs
2010sC–H activationYu (Scripps), Sanford, Hartwig, Bergmanlate-stage functionalization of complex scaffolds
2010sphotoredoxMacMillan, Yoon (Wisc), Stephenson (UMich)radical chemistry under mild conditions, Ru/Ir or organic dye
2010sflow chemistryJamison (MIT), Ley (Cambridge)hazardous intermediates, photochem, on-demand
2010selectrochemistryBaran (Scripps 2018+)anodic oxidation of complex molecules; programmable potential
2020sbiocatalysis (engineered)Arnold (CalTech)2018 (directed evolution)enzymes for stereo, FG, scalable
2020sclick + bioorthogonalSharpless, Meldal, Bertozzi2022CuAAC + SPAAC for ligation, ADCs, conjugates
PlatformBest atWatch out
Pd cross-coupling (Suzuki, Negishi, Buchwald-Hartwig)sp²–sp² and Cl/N–H bonds, predictablePd residue in API (ICH Q3D limits)
RuO4/OsO4 (Sharpless dihydroxylation)cis-diol from alkeneOsO4 toxicity
Sharpless asymmetric epoxidationallylic alcohols → epoxidessubstrate scope (needs allylic OH)
Noyori BINAP hydrogenationβ-ketoester → β-hydroxy ester (>99% ee)sub costly, but proven on tonne scale (Takasago menthol 30 kt/yr)
Grubbs RCMmedium / large rings, late stageE/Z selectivity in macrocycles
MacMillan iminium / List enamineenantioselective α-functionalization of carbonylssubstrate scope
Yu C–H activationunactivated C–Hligand engineering nontrivial
MacMillan photoredoxradical chemistry under mild conditionsLED setup; lamp uniformity
Stephenson photoredoxradical cascades, late-stagereproducibility cross-lab
Baran electrochemistryreagent-free oxidation, scaleelectrode passivation, undivided vs divided cell
Codexis transaminase (engineered)β-chiral amineenzyme engineering ROI
Sharpless–Meldal CuAACbioconjugation, conjugatesCu in biological samples
Bertozzi SPAAC (DBCO-azide)living-cell labeling, ADCsDBCO cost

5. Flow vs batch — when geometry matters

Batch is the default; flow earns its keep in specific situations.

SituationFlow wins because…Examples
Hazardous intermediatesmall volume in-situ, never accumulatesdiazomethane (Ley), HN3, hydrazoic acid, F2, ozone
Fast exothermmm-scale heat transfer >100× batchnitration, lithiation, organolithium
Photochemistrythin film, uniform photon flux[2+2], photoredox cascades (Stephenson, Booker-Milburn)
Multi-step telescopingno intermediate isolationEli Lilly diphenhydramine flow (2014); Ley telescoped 1-naphtholol benzodiazepine
Continuous productionAPI on demandJanssen prezista, Vertex multiple
Heterogeneous catalysispacked-bed catalyst, no filtrationcontinuous hydrogenation (H-Cube)

Major industrial flow installations: Vertex (Boston) — first FDA-approved continuous-manufacturing API (Orkambi, 2015). Janssen (Beerse, Belgium) — continuous prezista since 2016. Eli Lilly Kinsale (Ireland) — multiple oncology APIs in flow. Snapdragon Chemistry (Boston, acquired by Cambrex) — contract flow chemistry. The FDA Emerging Technology Team (ETT) explicitly endorses continuous manufacturing as the modernization target.

6. Solid-phase and parallel synthesis

MethodOriginScaleUse
SPPS (solid-phase peptide synthesis)Bruce Merrifield 1963 (Nobel 1984)mg–g (lab), kg (CMC)every peptide drug — semaglutide, liraglutide, tirzepatide, vasopressin agonists
Solid-phase oligonucleotideCaruthers 1981 (phosphoramidite)µg–gevery ASO, siRNA, mRNA primer
Glycan SPPSSeeberger (MPI Potsdam) 2001+mgresearch; commercial via GlycoUniverse
Parallel batch (medchem)1990s combichemmg per compoundevery medchem hit-to-lead campaign
DEL (DNA-encoded library)Brenner-Lerner 1992; revived 2009 (X-Chem, GSK Encoded Library Technology)10^6–10^12 compoundsscreening, not resynthesis
Microreactor parallel (Chemspeed, Mettler EasyMax, HEL, AMTechnology)2000smg–ghigh-throughput experimentation in pharma

For peptide drugs, SPPS is so dominant that the entire workflow (Fmoc-AA-OH purchase → automated couplings → resin cleavage → preparative HPLC purification → lyophilization → fill-finish) is industrialized. Semaglutide (Ozempic / Wegovy) is made at >50 t/yr by Novo Nordisk and PolyPeptide using Fmoc-SPPS, with the cost dominated by amino acid building blocks.

7. The “step count” floor

A useful mental model: every target has an irreducible step count set by the bond-graph complexity. Hudlický’s index, Bertz’s index, and the CSI (Computed Synthetic Intuition) index of Coley et al. (2018) all attempt to quantify it. Empirically, by 2020, the median total synthesis published in JACS / Nature / Science was ~18 LLS steps; the median published in 1990 was ~35. The drop is from RDRP (controlled radical), photoredox, C–H activation, and biocatalysis — each replaces several FG-interconversion steps with a direct bond-forming step.

8. Cost, scale, IP

Synthesis decisionDrivesPenalty
Choice of chiral pool starting materialroute + eenot always available, or seasonally
Choice of resolution vs asymmetric catalysisyield (50% max for resolution unless racemization-coupled)resolution wastes the wrong enantiomer
Choice of biocatalysis vs metal catalysisresidual metal limits (Pd, Ru, Pt in API < 5 ppm per ICH Q3D)biocat ROI on tonnage
Choice of solvent (CHEM21 guide, Pfizer green-chem)E-factor, environmentalhexane and DMF banned in many EU contracts
Telescoping (no intermediate isolation)E-factor + cycle timeregulatory burden (more impurities to qualify)
Patent landscaperoute freedomcompeting pharma patent your favorite step

The CHEM21 solvent selection guide and Pfizer’s green-chemistry solvent guide are the operational tools. Ethanol, EtOAc, MIBK, acetone, water — preferred. DMSO, NMP, DMF, DMAc, DCM, CHCl3 — restricted. CCl4, benzene, Et2O — banned.

9. Modern catalysis era — 2020–2026

What’s new vs textbook organic chemistry:

  • Photoredox + nickel dual catalysis (Doyle, MacMillan, Molander 2014+) — replaces palladium for sp³-sp² couplings; opens “metallaphotoredox” as a unified platform.
  • Electrochemistry (Baran 2018+) — eliminates stoichiometric oxidants. Notable: ElectraSyn 2.0 (IKA / Scripps spinout) is in 1500+ academic labs by 2025; Baran’s papers explicitly include “make at-home” gear lists.
  • Enzyme cascades — Codexis (transaminase for sitagliptin Januvia, 2010), Codex Bio for islatravir 2020, Merck + Codexis multi-enzyme cascade for molnupiravir 2021.
  • Biocatalysis directed evolution (Arnold, Nobel 2018) — engineer the enzyme for the target, not the target for the enzyme.
  • C–H activation late-stage functionalization (Yu, Engle, Sames, MacMillan) — change a hit compound late without re-doing the route.
  • DEL (DNA-encoded libraries) — billions of compounds screened in a single afternoon; Vipergen, X-Chem, HitGen, Insilico’s DEL platforms.
  • Generative chemistry (Insilico, Exscientia, Iktos, Schrödinger, BenevolentAI) — propose new structures from disease/target hypotheses; route designed by Synthia/ASKCOS; execute robotically (RoboRXN).
  • 3D printed flow reactors (PHASTAR, ChemTrix, Vapourtec, Chemspeed) — bespoke residence times and mixing.

10. Decision tree — picking a synthesis strategy

What's the target?
├─ Single small molecule, novel scaffold (research / process)
│    ├─ < 8 steps → linear, classical, batch
│    ├─ 8–15 steps → convergent, batch with telescoping where safe
│    ├─ > 15 steps → highly convergent, late-stage diversification, RAFT/RDRP if polymer
│    └─ Complex stereocenters → asymmetric cat (BINAP, salen-Mn, organocat) or chiral pool
├─ Peptide drug
│    └─ Fmoc-SPPS (Merrifield); parallel or microwave; cleave + prep HPLC
├─ Oligonucleotide drug
│    └─ phosphoramidite SPPS (Caruthers); ASO/siRNA scale
├─ Macrocyclic natural product
│    └─ Yamaguchi macrolactonization or RCM (Grubbs); convergent fragments
├─ ADC payload + linker + antibody
│    ├─ Payload: classical or convergent synthesis
│    ├─ Linker: orthogonal protecting groups
│    └─ Conjugation: click (CuAAC or SPAAC)
├─ Bio-similar / generic API
│    └─ Reverse-engineer route from public patents; optimize on cost/PMI
├─ Hazardous reagent in chain
│    └─ Flow chemistry (diazomethane, lithiation, ozonolysis, nitration, photoredox)
├─ Continuous-manufacturing target
│    └─ Flow + telescoping; FDA Emerging Technology Team consultation
├─ Screening / hit-finding
│    └─ DEL (10^9 cmpds) or HTS (10^6 cmpds via combichem)
├─ Engineered enzymatic route exists
│    └─ Biocatalysis (Codexis, Merck, Novartis, BASF)
└─ Polymer / material target
     → See [[Sciences/Chemistry/_compare_polymerization_methods]]

Adjacent

When to pick what

The fastest narrowing: simple → classical batch; complex → convergent + modern cat (photoredox, organocat, biocat, C–H); peptide → SPPS; oligo → phosphoramidite SPPS; macrocycle → RCM or Yamaguchi; hazardous → flow; scale-up commodity API → telescoped flow under FDA ETT; screening → DEL + AI retrosynthesis (Synthia/ASKCOS). The single biggest practical lesson of 2010–2026 is that step count matters less than route convergence — a 12-step convergent synthesis at 40% overall is almost always better than a 7-step linear at 30%, because the 12-step route ships kilograms while the 7-step route stalls at grams. Choose the catalyst platform that maximizes step economy at your target’s most strategic bond, then defend the rest with classical chemistry.