In batch solid-phase peptide synthesis the coupling clock is set by diffusion into the bead. You shake a vessel, you carry a large reagent excess so the interior concentration stays high, and you stop after a fixed interval chosen in advance and never observed. Continuous-flow SPPS packs the same resin into a column and pumps activated amino acid through it, so concentration at the reactive site is sustained by convection rather than by excess. Simon and co-workers built the modern form of this around a heat exchanger feeding a low-volume, low-back-pressure reaction vessel with a UV detector immediately downstream, and reported incorporation of one residue every 1.8 min under automatic control or every 3 min under manual control [1]. Their stated design logic explains most of what follows: continuous delivery of heated solvents and reagents to the solid support at high flow rate maintains maximal reagent concentration in the reaction vessel, exchanges reagents quickly, and eliminates the need to heat reagents after they have been added to the vessel [1].
Pushed harder, the same architecture gets fast. Mijalis and co-workers reported a fully automated flow-based approach with amide bond formation in 7 s and total synthesis times of 40 s per amino acid residue, with crude peptide purities and isolated yields comparable to those for standard-batch SPPS [2]. Accessible chain length moved with it: Hartrampf and co-workers used an automated fast-flow instrument for direct manufacture of peptide chains up to 164 amino acids long over 327 consecutive reactions, with chain elongation complete in hours, and made nine different protein chains that after purification and folding displayed biophysical and enzymatic properties comparable to the biologically expressed proteins [3].
Residence time and reagent concentration replace coupling time
You do not set a coupling duration in a packed bed. You set a bed geometry and a volumetric flow rate, and residence time falls out of the two. Varró and co-workers make the dependency explicit in a scale-up study: moving to larger column geometries required optimising resin load and flow rate together to reach the residence times they wanted, and on that basis they ran syntheses on a greater-than-4 g scale [4]. The same arithmetic constrains the small end. Mándity and co-workers built a mesoscale reactor that permits high temperature and pressure, carried out a complete reaction parameter optimisation, and coupled all 20 proteinogenic amino acids with only 1.5 equivalents of amino acid at quantitative conversion [5]. The concentration mechanism, rather than any change in stoichiometry alone, is the plausible reading, and the mechanism is the one Simon and co-workers set out: continuous delivery at high flow rate holds reagent concentration at the resin at its maximum instead of letting a spent bulk phase dilute it [1].
The practical consequence is that two knobs now do the work one used to. Halving the flow rate doubles residence time but also halves the rate at which spent reagent and cleavage by-product are swept out of the bed, and it changes the pressure drop across the column. Neither effect is visible in a coupling-time entry on a batch protocol, which is why a flow method transferred from a batch protocol by matching "time" alone rarely reproduces.
In-line UV gives you one trace per cycle
Because the effluent leaves the bed continuously, a detector placed downstream sees the Fmoc-deprotection product as it elutes. Mohapatra and co-workers describe the arrangement plainly: on an automated fast-flow synthesiser, Fmoc deprotection is generally quantitative and the resulting dibenzofulvene by-product is detected with an in-line UV–vis detector at 310 nm, and those data can be used to obtain information indirectly on the individual stepwise coupling cycles and on overall synthesis performance [6]. The UV detector in the original fast-flow design served exactly this purpose — continuous monitoring of the process [1]. The same trick is available without custom hardware: Romanens and co-workers repurposed a standard analytical HPLC by packing a short guard column with resin and programming reagent injections through the autosampler, exploiting the precise flow control, heating and in-line UV detection intrinsic to HPLC systems to monitor Fmoc deprotection in real time, and obtained peptides up to 30 residues, and separately a 58-residue miniprotein in under 7 h on milligram scale [7].
What makes the trace informative is its shape, not just its area. The integral, height and width of the time-resolved UV–vis deprotection traces are the analytical handles [6]. A failed or struggling cycle does not simply give a small peak. Sequence-dependent events commonly described as aggregation produce mass-transfer problems and slow reaction kinetics, and these are reflected in flattened, wider UV–vis deprotection peaks [6]. In other words: a sharp, tall, early-eluting peak means the deprotection reagent reached every chain quickly; a low, broad, tailing peak of similar area means the reagent had to work its way into a poorly solvated bead, and the coupling that preceded it was almost certainly running under the same transport limitation. That data set is large enough to model — 35 427 individual, highly reproducible deprotection steps, on which a deep-learning model predicted deprotection outcome with less than 6% error [6].
Be precise about what the trace licenses. It reports removal of the Fmoc group installed by the previous coupling, so it is a retrospective proxy for that coupling, and Mohapatra and co-workers describe the information as indirect for exactly this reason [6]. It tells you nothing about stereochemistry, nothing about aspartimide, and nothing about which chains failed — a small peak is consistent with an incomplete deprotection and with a chain population that never got acylated in the first place. An orthogonal in-line signal helps: Sletten and co-workers incorporated a pressure-based variable bed flow reactor into an automated solid-phase peptide synthesiser, which permitted real-time monitoring of resin swelling to determine amino acid coupling efficiency and on-resin aggregation [8].
Swelling and pressure limits
The bed is not a static object. Its volume is set by solvation of the gel-like polymer network, which changes as the chain grows. Pinzón-López and co-workers analysed 1% cross-linked polystyrene–divinylbenzene resin in dichloromethane and DMF using the Flory–Rehner framework with Hansen solubility parameters and found that attached glycans increase polymer–solvent affinity toward DCM whereas attached peptides reduce it [9]. Equilibrium swelling degree showed only a weak temperature dependence, but swelling kinetics in packed-bed geometries were strongly temperature dependent, with relaxation times in DCM rising from 18 s at 20 °C to 35 s at −30 °C [9].
Two consequences follow for a flow reactor. First, a fixed-geometry column cannot hold packing density and residence time constant across a synthesis if the bed volume drifts; something has to give, and it gives as a change in pressure drop. Second, because swelling relaxation has a measurable time scale, a solvent switch is not instantaneous, and a wash short enough to satisfy a cycle-time target may leave the bed out of equilibrium. The variable bed flow reactor addresses the first point by making bed volume a measured variable rather than an assumption, which is what turns resin-volume change into a real-time indicator of reaction progress [8, 9].
Heat is the trade you are making
Flow buys its cycle times with temperature, and the trade is well stated by Wettler and co-workers: higher temperatures would reduce aggregation but aggravate side reactions [10]. Their response was to confine the basic environment to the amino acid activation step using a flow system with a reusable, immobilised amine base; having identified effective coupling conditions with minimal epimerisation for each canonical amino acid, they confirmed the viability of higher temperatures in the coupling step in the absence of excess base across four test peptides, and on aggregation-prone sequences the method was on par with or superior to standard protocols [10].
Epimerisation is the first thing to budget for. In batch microwave SPPS, Palasek and co-workers found cysteine, histidine and aspartic acid susceptible to racemisation in a model 20-mer containing all 20 natural amino acids, and lowering the microwave coupling temperature from 80 °C to 50 °C limited racemisation of histidine and cysteine; racemisation during the coupling reaction was limited to the activated ester state of the amino acids up to 80 °C, and use of the hindered amine collidine also minimised formation of D-cysteine [11]. Flow studies converge on the same residues as the pressure points and reach low numbers by tuning flow rather than by dropping temperature. Ferentzi and co-workers, in the study that selected an anisole/DMSO (17:3) mixture as the coupling medium, reduced racemisation to below 2% for histidine and below 1% for cysteine by adjusting the flow parameters, and with their final parameters ran 10-, 30- and 58-residue targets at 12 min per cycle [12]. Farkas and co-workers, using PyAOP and DIC/HOBt with a 1.7 min coupling reaction, reported virtually racemisation-free peptides, quoted as L-His/D-His <1–5%, even at a column temperature of 70 °C [13]. Romanens and co-workers likewise observed no significant epimerisation under their optimised HPLC-based flow conditions [7].
Aspartimide is the second. Mong and co-workers used the fast-flow platform to explore assembly conditions for long polypeptides rapidly and were able to mitigate common side reactions including deletion and aspartimide products, producing fragments good enough for convergent native chemical ligation [14]. But heat and pressure can also switch on side reactions that batch practice teaches you to expect protection from. Pseudoproline derivatives are commonly used precisely to reduce the likelihood of aggregation and to prevent aspartimide formation; investigating flow peptide chemistry with Thr(ΨPro) derivatives, Szaniszló and co-workers observed both aspartimide and an imine derivative of the Thr(ΨPro) moiety, demonstrated the oxazolidine character of the pseudoproline, and presented evidence that aspartimide formation appears to be catalysed by the presence of the pseudoproline moiety, attributing the side reactions to elevated temperature and pressure and advising caution under such harsh conditions [15]. They propose that thermodynamically controlled aspartimide formation can be kinetically prevented [15]. Regard a pseudoproline insertion as a variable to re-validate in flow, not as a fix carried over intact from batch.
The washes are reactions too, and flow moves a great deal of hot solvent. Laude and co-workers quantified unintended Fmoc removal using continuous-flow methods and found that for every millilitre of DMF used at 80 °C, up to 1 µmol of Fmoc-protected peptide is deprotected, generating additional impurities; adding HOBt reduced that unwanted deprotection to 0.1 µmol per millilitre [16]. That is a per-millilitre figure, so it scales with the wash volume your flow rate implies, and it is a direct argument for accounting for solvent volume at temperature rather than only for reagent equivalents.
Reagent and solvent consumption
The consumption case for flow is the strongest quantitative argument in its favour, and it comes from the same mechanism as the speed. Mándity and co-workers achieved quantitative conversions for all 20 proteinogenic amino acids at 1.5 equivalents, incorporated expensive artificial amino acids at a reasonable number of equivalents, and ran from microgram to gram quantities under automation [5]. Farkas and co-workers report an economic and environmentally friendly procedure using no more than 3 equivalents of activated amino acid and generating 6 mL of organic waste per cycle, with crude purity above 80% for sequences up to 30 residues and for difficult sequences [13]. Varró and co-workers replaced DMF with propylene carbonate used in minimal amounts, reporting low solvent consumption and low process mass intensity for the method, which they also operated at greater-than-4 g scale [4]. Reagent immobilisation extends the same idea to the base: a reusable immobilised amine base keeps the reagent in the reactor while the substrate stream passes through [10].
Two cautions on those figures. They are per-cycle numbers tied to one bed size and one flow rate; scaling the column changes the solvent-to-resin ratio, and a consumption figure quoted without the bed volume and flow rate it was measured at cannot be transferred. And low equivalents depend on the bed behaving as a plug: channelling through a compacted or shrunken bed defeats the concentration argument that makes 1.5 equivalents work in the first place [5, 9].
Why this is not microwave-assisted batch
Microwave heating accelerates a batch reaction; it does not replace it. In microwave-enhanced Fmoc SPPS the vessel, the reagent excess and the fixed, unobserved coupling interval all remain, and the acceleration brings its own side-reaction load that has to be controlled by choosing temperature, base and deprotection additive — lowering the coupling temperature, substituting collidine, adding HOBt to the deprotection solution or replacing piperidine with piperazine [11]. Flow changes the reactor rather than the heating. Reagent is delivered continuously and exchanged quickly instead of sitting in a batch [1], the effluent is available for measurement on every cycle [6], and the bed itself becomes an instrument you can read [8]. In exchange you inherit two variables microwave batch does not have — volumetric flow rate and back pressure — and a solid support whose volume is a function of solvent, temperature and the chain you have built on it [9]. The two approaches therefore fail differently in what they let you see: a microwave batch run is assessed after the fact, whereas in flow a struggling cycle can present itself while the synthesis is still running, as a broadened deprotection peak [6] or a change in bed volume [8].
What to record for every run
Record bed volume and volumetric flow rate as a pair, because residence time is derived from them and neither is meaningful alone [4]. Record inlet pressure as a trace, not a single reading, so bed compaction and swelling are separable from a partially blocked frit [8, 9]. Record the integral, height and width of each Fmoc-deprotection peak rather than integral alone, because peak broadening is the aggregation signal and area may be preserved while shape degrades [6]. Record the coupling temperature and the activation conditions together, since in microwave batch SPPS racemisation during coupling was confined to the activated-ester state of the amino acid up to 80 °C [11]. Record equivalents and the solvent volume passed at temperature, because both drive cost and the second drives premature Fmoc loss [13, 16].
Finally, do not let a clean UV trace stand in for analysis. None of the in-line signals identifies a by-product: a normal-looking deprotection peak is fully consistent with a quantitative coupling that also epimerised a histidine or formed an aspartimide. Those have to be measured off-line, as the racemisation and aspartimide studies cited above did [11, 12, 13, 15]. The in-line trace tells you which cycle to go and look at; it does not tell you what went wrong there.