Literature Review · Peer-Reviewed Sources Only

Kaiser Raman Rxn2 / Rxn4: Performance Review from 5 Peer-Reviewed Studies

Published May 27, 2026 · 14 min read · 5 citations
Kaiser Raman Rxn analyzer with in-bioreactor immersion probe schematic stirred-tank bioreactor sapphire window fibre-optic cable Raman Rxn2 analyzer Ch 1 (in use) Ch 2 Ch 3 Ch 4 Raman spectrum glucose | lactate | titre | VCD PLS chemometric model NIR excitation 532 / 785 / 993 nm options 993 nm suppresses autofluorescence Kaiser Raman Rxn analyzer with Rxn-45 immersion probe
Figure 1: The Raman Rxn2 (or Rxn4) couples to an in-bioreactor immersion probe (typically the Rxn-45) via fibre optics. A narrow-band laser (commonly 785 nm, with 993 nm available to suppress mammalian autofluorescence) illuminates the culture through a sapphire window. Stokes-scattered light is dispersed across a CCD, producing a Raman spectrum, and a partial least squares chemometric model converts each spectrum into glucose, lactate, glutamine, viable cell density, and titre. One analyzer supports up to four probes simultaneously.
Literature Verdict

Across five peer-reviewed sources, in-line Raman is described as a first-choice process analytical technology for upstream bioprocess monitoring since 2011, and the Kaiser Raman Rxn family is the platform that appears most often in the underlying deployment papers [3][4]. The clearest product-specific result is a Biogen study that ran an entire mammalian production bioreactor on glucose set-point control fed by a Kaiser Rxn2-1000 (993 nm) analyzer, where the 785 nm wavelength was unusable because of culture-borne autofluorescence [1]. Multi-clone, multi-scale transferability is the platform's hardest engineering problem: a 35-cultivation Novartis-Lisbon study showed that clone-based local partial least squares models calibrated on as few as 3 to 9 batches outperform single global models across 2 L bench, 7 L, 15 L, and 10,000 L scale [2], while a multi-site European consortium demonstrated that a deliberately diverse "generic" calibration can hold mostly under 10 percent error across different sites and different Raman spectrometers [5]. The literature does not publish a single head-to-head against the Mettler Toledo ReactRaman platform.

Kaiser Raman Rxn2 / Rxn4 at a glance

The Raman Rxn2 and Raman Rxn4 are multi-channel bioprocess Raman analyzers from Endress+Hauser, built around the Kaiser Optical Systems holographic transmission spectrograph and now sold under the Endress+Hauser brand following the 2018 acquisition of Kaiser Optical Systems. Each analyzer couples to an in-bioreactor immersion probe (typically the Raman Rxn-45) by fibre optics and delivers simultaneous Raman spectra used to predict glucose, lactate, glutamine, glutamate, ammonia, viable cell density, total cell density, and product titre through partial least squares chemometric models. The Rxn2 supports up to four channels; the Rxn4 also supports up to four channels and ships as a 19-inch rack package or a NEMA 4X-enclosed unit for plant-floor deployment. Product details below are sourced from the vendor datasheet; field performance data comes from independent studies cited throughout this review.

SpecificationValue
Measurement principleRaman scattering. A narrow-band laser excites molecular vibrations; the inelastic (Stokes) scattered light is dispersed across a CCD, producing a chemical fingerprint of the bulk medium. Partial least squares chemometric models convert each spectrum into concentrations for glucose, lactate, glutamine, viable cell density, titre, and other process parameters.
Excitation wavelengthsRxn2 Starter and Hybrid: 785 nm. Rxn2 / Rxn4 Base: 532 nm, 785 nm, or 1000 nm (often referred to as the 993 nm option in published work)
Spectral range150 to 4350 cm-1 (532 nm), 150 to 3425 cm-1 (785 nm), 200 to 2400 cm-1 (1000 nm)
Spectral resolutionApproximately 4 to 6 cm-1 depending on wavelength; about 10 cm-1 on the Starter model
ChannelsUp to four independent probes on the Rxn2 Base and Rxn4; single channel on Rxn2 Starter; up to two on Rxn2 Hybrid
Compatible probesRaman Rxn-10, Rxn-40, Rxn-41, Rxn-45, Rxn-46. The Rxn-45 is the immersion bioprocess probe most commonly described in cell culture literature
Operating temperature15 to 30 °C analyzer ambient; storage -15 to 50 °C
SterilisationThe Rxn-45 immersion probe is autoclave and CIP/SIP compatible. The analyzer itself sits outside the wet zone
Process connectionProbe inserts through a standard 12 mm bioreactor port
CertificationsATEX, CSA, IECEx, UKCA, JPEx; designed for cGLP / cGMP compliance with PAT / QbD design concepts
Typical capital costNot published by Endress+Hauser; quotes are project-specific. Bioprocess Raman analyzers broadly fall in the high five-figure to low six-figure GBP range per channel including probe and chemometric software

Spec values are taken from the Endress+Hauser Raman Rxn2 product page, Rxn4 product page, and Raman calibration and verification kits. Spec values are vendor claims; all field-performance figures in this review are drawn exclusively from the peer-reviewed publications cited below.

What the peer-reviewed literature says

The starting point for any honest review of the Kaiser Raman Rxn family is a 2022 review co-authored by Endress+Hauser’s optical analysis group, which concluded that Raman spectroscopy has been a first-choice process analytical technology for monitoring and controlling upstream bioprocesses since its industrial introduction in 2011, and described its extension from cGMP commercial manufacturing into scale-down development, downstream process monitoring, and even formulation work [3]. An earlier 2017 review by the same vendor group placed the technology more broadly across pharmaceutical manufacturing and detailed real-time release testing, continuous manufacturing, and mammalian cell culture monitoring as the application areas where Raman had moved beyond proof-of-concept [4]. Because both reviews are vendor-authored, they are useful for framing what the platform claims to do and which use cases it has chased, but their performance claims need to be triangulated against independent studies. This review therefore uses the Endress+Hauser-authored reviews to set the scope and uses three independent peer-reviewed deployment studies for the performance evidence: a Biogen in-line glucose-monitoring study with the Kaiser Rxn2-1000, a Novartis-Lisbon multi-clone study, and a multi-site European generic-model study.

The clearest product-specific deployment is a Biogen study that used the Kaiser RXN2-1000 platform, the 993 nm variant of the Rxn2, to overcome an autofluorescence wall that had blocked any useful Raman model at the standard 785 nm excitation [1]. The team reported that the cell culture process generated such strong autofluorescence at 785 nm that multivariate glucose models simply could not be built. Shifting excitation deeper into the near-infrared at 993 nm reduced or eliminated the broadband fluorescence background, and a multivariate glucose model was then developed in the same process. The team went on to use that model to run the production bioreactor entirely on Raman adaptive feeding, holding glucose at an arbitrary set point for the duration of the culture rather than feeding from a fixed offline trajectory. This is the strongest published evidence that the 993 nm Rxn2 wavelength option exists for a real reason and not as a marketing differentiator: there are mammalian processes where the 785 nm wavelength is simply unusable.

Multi-clone, multi-scale transferability is where the platform’s honest weaknesses come out, and the Novartis-Lisbon group has published the most useful evidence on that question. Across 35 mammalian cultivations spanning four CHO cell lines, eight clones, two cultivation modes (fed-batch and perfusion), and four scales (2 L, 7 L, 15 L, and 10,000 L), the authors compared local clone-based partial least squares models against single global models that lumped all conditions together [2]. They found that local models constructed with as few as 3 to 9 calibration batches per clone delivered materially better predictive power than the global approach, and that scale, base powder medium, and cell line each shifted Raman spectra in ways the global model could not absorb. The practical implication for a Rxn2 or Rxn4 owner is that a single calibration is not transferable from a 2 L development bioreactor to a 10,000 L production tank without per-clone or per-scale local modelling; the platform itself can serve both, but the chemometric layer needs to be re-thought.

A multi-site European study, drawing on data from Bayer, Sanofi, KTH-Stockholm, Rentschler Biopharma, GSK, and Hohenheim University, took the opposite approach and asked whether a single deliberately diverse calibration could be made transferable across sites [5]. The authors built generic partial least squares regression models from a curated multi-site Raman dataset and reported that the resulting models predicted glucose, lactate, and glutamine concentrations across CHO cultures from different sites, different Raman spectrometers, and a completely different downstream test setup with mostly under 10 percent relative error. The work explicitly demonstrated that recalibration is not always necessary when conditions change, provided the calibration data are chosen to span the relevant variability up front. Together, the Novartis-Lisbon and multi-site European studies bracket the platform’s real-world transferability story: Raman models are scale- and clone-sensitive, but with the right calibration design they can hold useful accuracy across very different deployments.

Performance data from cited studies

Study Conditions Accuracy Response / drift Conclusion
Matthews 2018 [1] Kaiser Rxn2-1000 platform; 993 nm excitation; mammalian cell culture process with strong 785 nm autofluorescence Multivariate glucose models built at 993 nm that could not be built at 785 nm because of autofluorescence Live updates fast enough to drive a closed-loop glucose adaptive feeder over the whole culture The 993 nm Rxn2 option is essential for autofluorescent processes; the analyzer can run an entire production bioreactor on Raman set-point control
Santos 2019 [2] Raman PAT; 35 mammalian cultivations; 4 CHO cell lines; 8 clones; 2 L, 7 L, 15 L, and 10,000 L; fed-batch and perfusion Clone-based local PLS models with 3-9 calibration batches outperformed a single global model Scale, medium powder, and cell line each shifted spectra; static calibrations degraded outside their training envelope A single Raman calibration is not transferable across clones, scales, or media; local clone-based models are the pragmatic answer
Esmonde-White 2022 [3] Vendor-authored review of Raman in biopharma from development to manufacturing; full lifecycle scope Position statement: Raman is a "first-choice PAT" for upstream bioprocess monitoring and control since 2011 Coverage extends from single-cell scale-down to commercial cGMP and onward to formulation higher-order structure measurement Frames the application envelope and the breadth of cGMP-deployed Raman in biopharma; vendor framing, useful for scope but not for independent benchmarking
Esmonde-White 2017 [4] Vendor-authored review of Raman as PAT across pharmaceutical manufacturing and bioprocessing Mapped Raman onto real-time release testing, continuous manufacturing, and mammalian cell culture bioreactors Captured the technology development gap that the 2010 review left, including transmission and enhanced reflection Raman Confirms Raman moved from proof-of-concept to deployed PAT during 2010-2016, with mammalian cell culture as a primary application
Yousefi-Darani 2022 [5] Multi-site Raman dataset (Bayer, Sanofi, KTH, Rentschler, GSK, Hohenheim); CHO; multiple Raman spectrometers Generic PLS models predicted glucose, lactate, and glutamine mostly under 10% relative error across sites and spectrometers Tested transfer to a completely different test setup and a different Raman spectrometer; no recalibration A deliberately diverse calibration enables transferable Raman models; recalibration is not always necessary when process or hardware changes

Every row is a separate peer-reviewed publication; see References section for full citations. Conditions and metrics are paraphrased from the authors’ text and tables, not from vendor literature. The Matthews 2018 study used the Kaiser Rxn2-1000 specifically; the Santos and Yousefi-Darani studies used Raman analyzers without naming the manufacturer in the abstract and so characterise the in-line Raman class the Rxn2 and Rxn4 belong to; the two Esmonde-White reviews are vendor-authored and are used here for scope rather than for independent performance numbers.

Limitations and failure modes reported

Across the reviewed studies, the following limitations and failure modes recurred. Each bullet is tagged with the specific citation that describes it. These are properties of the in-line Raman measurement and the chemometric model that interprets it, not Rxn-specific hardware defects.

When the literature recommends the Kaiser Raman Rxn family

Recommended for

  • cGMP upstream mammalian processes where Raman is used as a first-choice in-line PAT for glucose, lactate, glutamine, viable cell density, and titre. [3]
  • Autofluorescent processes where the 785 nm wavelength is unusable: the 993 nm Rxn2-1000 option unlocks Raman in cases where it otherwise fails. [1]
  • Closed-loop adaptive feeding where the analyzer drives a real-time set-point controller rather than supplying offline-style data points. [1]
  • Multi-clone development programmes that accept the discipline of clone-based local calibration with 3 to 9 batches per clone. [2]

Caveats / not recommended for

  • Do not assume a single global calibration will hold across clones, scales, or media powders; clone-based local models or deliberately diverse generic models are the literature-supported approach. [2]
  • Do not eliminate offline metabolite analysers entirely; the literature treats Raman as reducing offline sampling, not as a complete replacement. [3]
  • Do not specify a 785 nm-only configuration for a new cell line without first checking for autofluorescence at that wavelength; the 993 nm option exists for a reason. [1]
  • Decisions that require a published independent head-to-head against the Mettler Toledo ReactRaman; that benchmark is not in the literature reviewed here. [3]

Use cases documented in the literature

Specific deployments reported in the cited studies. Each card corresponds to a real published bioprocess use case.

Autofluorescent CHO
993 nm glucose adaptive feeding

Biogen ran a production bioreactor entirely on Raman set-point control using a Kaiser Rxn2-1000 at 993 nm, after 785 nm autofluorescence blocked any model.

[1]
Multi-scale mAb dev
Clone-based local PLS across 2 L to 10,000 L

Novartis used local PLS models with 3 to 9 calibration batches per clone across 35 cultivations and four scales, outperforming a single global model.

[2]
Multi-site cross-vendor
Generic glucose / lactate / glutamine model

A six-organisation European consortium built a generic PLS model that predicted CHO metabolites across different sites and different Raman spectrometers, mostly under 10 percent error.

[5]
Lifecycle PAT
From development to commercial cGMP

The 2022 Endress+Hauser-authored review documents Raman deployments from single-cell scale-down through cGMP commercial manufacturing of biopharmaceuticals.

[3]

Comparing the Kaiser Raman Rxn against alternatives?

The Sensor Selection Tool takes 6 questions about your scale, modality, vessel, and budget and returns ranked sensor recommendations, including alternatives to the Kaiser Raman Rxn family.

Open the Sensor Selection Tool

User reviews from bioprocess engineers

Real-world experience from engineers who deployed the Kaiser Raman Rxn2 or Rxn4. All reviews are moderated before publishing. Share your own below: 2 minutes, anonymous option available.

Frequently asked questions

What is the Kaiser Raman Rxn analyzer family?
The Raman Rxn2 and Raman Rxn4 are multi-channel Raman analyzers built around the Kaiser Optical Systems holographic transmission spectrograph, now sold under the Endress+Hauser brand following the 2018 acquisition. They couple to in-bioreactor immersion probes (such as the Raman Rxn-10 and Rxn-45) via fibre optics and deliver simultaneous in-line measurement of glucose, lactate, glutamine, glutamate, ammonia, viable cell density, total cell density, and product titre using partial least squares chemometric models. The Rxn2 supports up to four probes; the Rxn4 also supports up to four channels in a 19-inch rack or NEMA 4X enclosure. Common excitation wavelengths are 532 nm, 785 nm, and 993 nm (often called the 1000 nm option) [3].
Why are most published bioprocess Raman studies done at 785 nm?
Near-infrared excitation at 785 nm sits in a window where most aqueous biological matrices fluoresce only weakly, so the Raman scatter is not buried under broadband autofluorescence. This is why the bulk of the bioprocess literature on Raman uses 785 nm and why most cited multi-clone, multi-scale CHO studies use a 785 nm Raman Rxn analyzer [2]. The trade-off is that some cell lines and media generate enough autofluorescence at 785 nm to swamp the signal, which is why the 993 nm option exists [1].
Does the Kaiser Rxn2-1000 actually solve the autofluorescence problem?
Yes, in published work. A Biogen study reported that one mammalian cell culture process generated such strong autofluorescence at 785 nm that multivariate glucose models could not be built at that wavelength [1]. Switching the same process to a Kaiser Rxn2-1000 platform using 993 nm excitation reduced the autofluorescence enough to develop functional glucose models, and the team then ran the production bioreactor entirely on Raman adaptive feeding, holding glucose at an arbitrary set point for the duration of the culture. See the in-line vs at-line glucose monitoring guide for the wider context.
How transferable are Raman chemometric models across clones, scales, and cell lines?
The literature is honest about this being the hard part. A Novartis-Lisbon study spanning 35 mammalian cultivations across four CHO cell lines, eight clones, and four scales from 2 L to 10,000 L (fed-batch and perfusion) showed that local clone-based partial least squares models, calibrated on as few as 3 to 9 batches, produced materially better predictions than a single global model that lumps all conditions together [2]. A separate multi-site European study went further and demonstrated that a "generic" model trained on a deliberately diverse dataset can predict glucose, lactate, and glutamine across CHO cultures from different sites and different Raman spectrometers with mostly under 10 percent relative error, but only when calibration data are chosen carefully [5].
How accurate is Raman for in-line glucose and lactate in CHO?
The 2022 generic-model study reported mostly under 10 percent relative prediction error for glucose, lactate, and glutamine across CHO cultures from different sites and different Raman spectrometers using transferable partial least squares models [5]. The 2018 Biogen 993 nm study did not publish a single relative-error number but reported that the Raman-driven adaptive feeder maintained glucose at an arbitrary set point for the entire culture [1], which is a tighter operational test than offline RMSEP alone.
Can the Raman Rxn replace daily offline metabolite analyzers?
In the cited literature, in-line Raman is used to reduce offline sampling, not to eliminate it. The vendor-authored 2022 review describes Raman as a "first-choice PAT" for upstream bioprocess monitoring and control adopted since 2011, used alongside periodic offline reference measurements for model maintenance [3]. The Novartis-Lisbon study showed that the predictive power of any in-line model degrades when conditions move outside its calibration envelope [2], which is why offline samples are still used to detect and correct drift.
What probes are compatible with the Rxn2 and Rxn4?
Per the vendor’s current Raman Rxn2 product page, the Rxn2 and Rxn4 analyzers operate with the Raman Rxn-10, Rxn-40, Rxn-41, Rxn-45, and Rxn-46 fibre-coupled probes. The Rxn-45 is the immersion bioprocess probe most commonly described in cell culture literature; it inserts through a standard 12 mm bioreactor port and is autoclave and CIP/SIP compatible. The analyzer family is certified to ATEX, CSA, IECEx, UKCA, and JPEx and is designed to comply with cGLP/cGMP requirements.
How does Kaiser Raman compare to NIR for bioprocess monitoring?
Raman and NIR are complementary process analytical technologies that share a common chemometric workflow but differ physically. Raman excels at component-specific molecular fingerprinting through water, which is why it has become dominant for in-line glucose, lactate, and titre in mammalian cell culture [3]. NIR has stronger water absorbance, broader bands, and is typically faster, which makes it suitable for total composition and moisture tracking in downstream processing. The 2017 and 2022 vendor reviews of Raman both position the two methods as complementary rather than competitive [4]. See the Raman vs NIR bioprocess comparison for the trade-off matrix.

References

  1. Matthews TE, Smelko JP, Berry B, Romero-Torres S, Hill D, Kshirsagar R, Wiltberger K (2018). Glucose monitoring and adaptive feeding of mammalian cell culture in the presence of strong autofluorescence by near infrared Raman spectroscopy. Biotechnology Progress 34(6):1574-1580. DOI: 10.1002/btpr.2711.
  2. Santos RM, Kaiser P, Menezes JC, Peinado A (2019). Improving reliability of Raman spectroscopy for mAb production by upstream processes during bioprocess development stages. Talanta 199:396-406. DOI: 10.1016/j.talanta.2019.02.088.
  3. Esmonde-White KA, Cuellar M, Lewis IR (2022). The role of Raman spectroscopy in biopharmaceuticals from development to manufacturing. Analytical and Bioanalytical Chemistry 414(2):969-991. DOI: 10.1007/s00216-021-03727-4.
  4. Esmonde-White KA, Cuellar M, Uerpmann C, Lenain B, Lewis IR (2017). Raman spectroscopy as a process analytical technology for pharmaceutical manufacturing and bioprocessing. Analytical and Bioanalytical Chemistry 409(3):637-649. DOI: 10.1007/s00216-016-9824-1.
  5. Yousefi-Darani A, Paquet-Durand O, von Wrochem A, Classen J, Tränkle J, Mertens M, Snelders J, Chotteau V, Mäkinen M, Handl A, Kadisch M, Lang D, Dumas P, Hitzmann B (2022). Generic Chemometric Models for Metabolite Concentration Prediction Based on Raman Spectra. Sensors 22(15):5581. DOI: 10.3390/s22155581.

Vendor product pages referenced for spec values: Endress+Hauser Raman Rxn2 analyzer, Raman Rxn4 analyzer, Raman bIO-Optics overview, Raman calibration and verification kits, Endress+Hauser advanced bioprocess control case studies. Spec values are vendor claims; all field-performance figures in this review are drawn exclusively from the peer-reviewed publications cited above.