A microfluidic device and a method of performing biophysical profiling of a sample, and a method of classifying cells
The microfluidic device with pillar configurations addresses the lack of real-time monitoring in CAR T cell manufacturing by modulating cell trajectories for precise biophysical profiling and classification, enhancing therapy consistency and efficacy.
Patent Information
- Application Number
- PCT/SG2026/050286
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-05
- Filing Date
- 2026-05-05
- Publication Date
- 2026-11-12
AI Technical Summary
Existing CAR T cell manufacturing processes lack real-time monitoring of cellular attributes, leading to significant variability in cell number, phenotype, and efficacy, and are limited by complex instrumentation unsuitable for real-time or point-of-care applications.
A microfluidic device with independent channels and pillar configurations modulates cell trajectories to generate distinct biophysical signatures, allowing rapid profiling and classification of cells based on size and deformability, using machine learning for analysis.
Enables rapid, precise biophysical profiling and classification of cells, improving consistency and efficacy of CAR T cell therapy by providing real-time monitoring and reducing variability.
Smart Images

Figure SG2026050286_12112026_PF_FP_ABST
Abstract
Description
[0001] A MICROFLUIDIC DEVICE AND A METHOD OF PERFORMING BIOPHYSICAL PROFILING OF A SAMPLE, AND A METHOD OF CLASSIFYING CELLS
[0002] TECHNICAL FIELD
[0003] The present disclosure relates broadly to a microfluidic device and a method of performing biophysical profiling of a sample, and a method of classifying cells.
[0004] BACKGROUND
[0005] Chimeric antigen receptor (CAR) T cell therapy has emerged as a breakthrough medicine for treating various hematological malignancies. However, the personalized nature of autologous CAR T cell manufacturing presents significant hurdles in scalability, cost-effectiveness, treatment efficacy and toxicity. Manufacturing processes are complex and expensive, with individual treatments costing several hundred thousand dollars, and a proportion of batches failing due to insufficient CAR T cells, low effector cell numbers or low cell viability. In some cases, manufacturing failures lead to re-production, resulting in delays that may adversely affect patient outcomes.
[0006] Existing CAR T cell manufacturing methodologies typically utilize bioreactor systems characterized by a lack of integrated, real-time monitoring of cellular attributes. Product functional phenotypes are not monitored until final product release tests. These single end-point measurements of cell quality attributes typically do not correlate well with clinical response. This inevitably results in CAR T cells that are manufactured with large variations in cell number,phenotypes, in vivo efficacy and potential side effects. Critically, manufactured CAR T potency and functional tests are only performed before the addition of cryopreservation media and are not performed during the post -thaw, pre-product infusion. Thus, the CAR T cell attributes measured at the release test can be significantly altered at pre-infusion, potentially impacting in vivo efficacy.
[0007] Such analytical gaps result in substantial variability in terms of therapeutic safety, potency, and performance. Coupled with inherent patient-to-patient heterogeneity, there is therefore a need for new technologies with more specific and rapid cell functional phenotype measurements during, at the end of the manufacturing, and at pre-infusion of CAR T cells, to effectively monitor productspecific variations that could affect the consistency and efficacy of cell therapy products.
[0008] In addition, cells exhibit remarkable complexity in function and structure, determined by a range of biophysical and biochemical properties. It has been increasingly recognized that biophysical features of cells, such as cell size, cell deformability, and morphology are informative markers for cellular phenotypes and functions. Studies on T cells suggest that the T cell architecture is a probable marker for cellular response to antigens. Existing systems for cell morphology analysis typically rely on high-resolution imaging, complex instrumentation, and substantial computational resources. Such requirements limit their accessibility, scalability, and suitability for real-time or point-of-care applications.
[0009] Accordingly, there is a need for a microfluidic device and a method of performing biophysical profiling of a sample, and a method of classifying cells that seek to address or at least alleviate one of the above problems.
[0010] SUMMARYIn one aspect, there is provided a microfluidic device for biophysical profiling of a sample, the microfluidic device comprising, a first microfluidic channel and a second microfluidic channel configured to allow the sample comprising a plurality of cells to flow through; a first cell trajectory modulation region disposed within the first microfluidic channel, the first cell trajectory modulation region comprising a first array of pillars configured to sort the cells and provide one or more first modulated biophysical signatures comprising one or more first distribution profiles of the sorted cells; and a second cell trajectory modulation region disposed within the second microfluidic channel, the second cell trajectory modulation region comprising a second array of pillars configured to sort the cells and provide one or more second modulated biophysical signatures comprising one or more second distribution profiles of the sorted cells; wherein the first modulated biophysical signatures are substantially different from the second modulated biophysical signatures of the sorted cells.
[0011] In one embodiment of the device, the first microfluidic channel and the second microfluidic channel are fluidically independent of each other such that flow through one of the microfluidic channels does not pass through the other microfluidic channel.
[0012] In one embodiment of the device, the first array of pillars comprises a first pillar configuration; and the second array of pillars comprises a second pillar configuration; wherein the first and second pillar configurations differ in at least one of pillar dimension, pillar shape, pillar structure, pillar arrangement or pillar orientation, with respect to the direction of flow of cells.
[0013] In one embodiment of the device, the device further comprises a first detection region disposed downstream of the first cell trajectory modulation region, the first detection region comprising a first set of location bins for allowing the cells that are sorted by the first array of pillars to flow into or flow through; and a second detection region disposed downstream of the second cell trajectory modulation region, the second detection region comprising a second set oflocation bins for allowing the cells that are sorted by the second array of pillars to flow into or flow through.
[0014] In one embodiment of the device, the device further comprises a first prefocusing region disposed upstream of the first cell trajectory modulation region, the first prefocusing region comprising a first modified array of pillars; and / or a second prefocusing region disposed upstream of the second cell trajectory modulation region, the second prefocusing region comprising a second modified array of pillars; wherein the first and / or second prefocusing regions are configured to pre-separate the cells in the sample prior to passing through the first and / or second cell trajectory modulation regions; and wherein the spacing between the first and / or second modified arrays of pillars in the first and / or second prefocusing regions are different from the spacing between the first and second arrays of pillars in the first and second cell trajectory modulation regions.
[0015] In one embodiment of the device, the device further comprises one or more sample inlets fluidically coupled to the first and second microfluidic channels, wherein the one or more sample inlets are configured to deliver one or more sample streams to the first and second microfluidic channels.
[0016] In one embodiment of the device, the device further comprises one or more first sheath inlets and / or one or more second sheath inlets fluidically coupled to the first microfluidic channel and / or the second microfluidic channel; wherein the one or more first sheath inlets and / or the one or more second sheath inlets are configured to introduce a first sheath stream and / or a second sheath stream to the first microfluidic channel and / or the second microfluidic channel; wherein the first sheath stream and / or the second sheath stream are configured to focus the sample stream within the respective microfluidic channel.
[0017] In one embodiment of the device, the first sheath stream, sample stream, and second sheath stream are associated with respective hydraulic resistances R1 , R2 and R3; and wherein the first microfluidic channel and / or the secondmicrofluidic channel are configured such that conditions (i) and (ii) are satisfied: (i) 0.33 < R3 / R2 < 49, and (ii) 2 < R1 / R2 < 94.
[0018] In one embodiment of the device, the device further comprises one or more outlets fluidically coupled to the first and second microfluidic channels; wherein the one or more outlets are configured to allow discharge of fluid.
[0019] In one embodiment of the device, the first and second arrays of pillars are configured to resolve differences in cell trajectories for cells over a dynamic size range of from 2 pm to 25 pm.
[0020] In one aspect, there is provided a method of performing biophysical profiling of a sample, the method comprising, passing a volume of the sample comprising a plurality of cells through a microfluidic device comprising a plurality of pillars configured to modulate flow trajectories of the plurality of cells; obtaining one or more modulated biophysical signatures of the plurality of cells generated by the microfluidic device, the modulated biophysical signatures comprising distribution profiles of the sorted plurality of cells; processing the one or more modulated biophysical signatures using one or more transformation functions configured to map and integrate the one or more modulated biophysical signatures to obtain one or more cell trajectory modulation (CTM) features; and inputting the one or more CTM features into a trained determination unit configured to correlate the one or more CTM features to one or more properties of the plurality of cells.
[0021] In one embodiment of the method, the microfluidic device comprises, a first array of pillars having a first pillar configuration; and a second array of pillars having a second pillar configuration; wherein the first array of pillars differs from the second array of pillars in at least one of pillar dimension, pillar shape, pillar structure, pillar arrangement or pillar orientation, with respect to the direction of flow of cells.In one embodiment of the method, the step of passing a volume of the sample through a microfluidic device is repeated for a plurality of times under different flow rates; and wherein the flow rates and pillars within the microfluidic device cooperatively modulate trajectories of the cells based on one or more biophysical properties of the cells to generate a plurality of modulated biophysical signatures under different flow and pillar configurations.
[0022] In one embodiment of the method, obtaining the one or more modulated biophysical signatures of the plurality of cells comprises, detecting the cells at a detection region of the microfluidic device; counting the cells that are sorted to a plurality of histogram bins at the detection region of the microfluidic device; and obtaining a set of histogram distribution of the cells.
[0023] In one embodiment of the method, the one or more transformation functions comprise a first set of transformation functions configured to transform the modulated biophysical signatures based on cell size, a second set of transformation functions configured to transform the modulated biophysical signatures based on cell deformability, and a third set of transformation functions configured to transform the modulated biophysical signatures based on a combination of cell size and cell deformability.
[0024] In one embodiment of the method, the transformation functions comprise one or more mathematical and / or statistical operations comprising statistical measures, histogram bin evaluations, gated population calculations, distribution profile comparisons, or combinations thereof.
[0025] In one embodiment of the method, the trained determination unit comprises a machine learning model based on one or more of linear models, tree-based models, neural networks, support vector machines, Bayesian models, and instance-based models.In one embodiment of the method, the one or more CTM features comprise size-related features, deformability-related features, or combined size and deformability-related features.
[0026] In one embodiment of the method, the cells comprise chimeric antigen receptor (CAR) T cells, and wherein the one or more properties of the plurality of cells comprise immune cell activation state, cell phenotype, and killing potency.
[0027] In one embodiment of the method, the sample comprises blood, urine, saliva, milk, or cultured cells; the sample is substantially free of externally added tags or labels; the sample is undiluted or untreated; the sample has a concentration falling in the range of from 0.5 x 106cells / mL to 80 x 106cells / mL; the sample has a volume from 10 pL to 100 pL; and / or the sample is profiled within a target time period of no more than 15 minutes.
[0028] In one aspect, there is provided a method of classifying cells, the method comprising, passing a volume of a sample comprising a plurality of cells through a microfluidic device comprising a plurality of pillars configured to modulate flow trajectories such that the plurality of cells are sorted into a plurality of location bins; capturing image data of the plurality of location bins using an imaging apparatus; processing the image data of the plurality of location bins to extract image data of the cells sorted into the plurality of location bins; and inputting the image data of the cells into a trained machine learning model to classify the cells into one or more categories based on one or more morphological characteristics.
[0029] In one embodiment of the method, the sample is substantially free of externally added tags of labels, and / or introduced into the microfluidic device without dilution or pre-treatment.
[0030] In one embodiment of the method, the method further comprises obtaining a distribution profile of the plurality of cells across the plurality of location bins, wherein the distribution profile comprises a number of cells sorted into eachlocation bin, and wherein the number of cells sorted into each location bin is segmented according to the one or more categories based on the one or more morphological characteristics.
[0031] In one embodiment of the method, processing the image data comprises applying blob detection to identify and analyze regions within the image data that represent cells.
[0032] In one embodiment of the method, the one or more categories comprise size and shape, wherein the size category comprises small, medium and large cells, and wherein the shape category comprises doublets, blobs and irregular shapes.
[0033] In one embodiment of the method, the cells are immune cells; and wherein immune cells having a diameter of less than 7 pm are classified as small, immune cells having a diameter of greater than 12 pm are classified as large, and immune cells having a diameter between 7 pm and 12 pm are classified as medium.
[0034] In one embodiment of the method, a doublet comprises two cells adhered together, a blob comprises more than two cells grouped together, and an irregular morphology comprises a single cell having a non-spherical shape.
[0035] In one embodiment of the method, the machine learning model is trained using augmented image data generated by horizontal flipping, vertical flipping, rotating, or circular shifting image data captured by the imaging apparatus.
[0036] In one embodiment of the method, the machine learning model comprises at least one of a support vector machine, random forest classifier, k-nearest neighbor model, logistic regression model, artificial neural network, convolutional neural network, or gradient boosting model.In one embodiment of the method, the machine learning model comprises a support vector machine configured to classify the cells by determining an optimal hyperplane separating different morphological classes.
[0037] DEFINITIONS
[0038] The term “substrate” as used herein is to be interpreted broadly to refer to any supporting structure.
[0039] The term “micro" as used herein is to be interpreted broadly to include dimensions from about 1 micron to about 1000 microns. Accordingly, the term “microstructure’’ , such as pillars, posts, or obstacles and the like as used herein may include a structure having at least one dimension that is less than about 1000 pm, less than about 900 pm, less than about 800 pm, less than about 700 pm, less than about 600 pm, less than about 500 pm, less than about 400 pm, less than about 300 pm, less than about 200 pm, less than about 100 pm, less than about 90 pm, less than about 80 pm, less than about 70 pm, less than about 60 pm, or less than about 50 pm.
[0040] The term “nano” as used herein is to be interpreted broadly to include dimensions less than about 1000 nm. Exemplary sub-ranges that fall within the term include but are not limited to the ranges of less than about 900 nm, less than about 800 nm, less than about 700 nm, less than about 600 nm, less than about 500 nm, less than about 400 nm, less than about 300 nm, less than about 200 nm, less than about 100 nm, less than about 50 nm, less than about 20 nm, or less than about 10 nm.
[0041] The term “particle” as used herein broadly refers to a discrete entity or a discrete body. The particle described herein can include an organic, an inorganic or a biological particle. The particle used described herein may also be a macroparticle that is formed by an aggregate of a plurality of sub-particles or a fragmentof a small object. The particle of the present disclosure may be spherical, substantially spherical, or non-spherical, such as irregularly shaped particles or ellipsoidally shaped particles. The term “size” when used to refer to the particle broadly refers to the largest dimension of the particle. For example, when the particle is substantially spherical, the term “size” can refer to the diameter of the particle; or when the particle is substantially non-spherical, the term “size” can refer to the largest length of the particle.
[0042] The terms “coupled” or “connected” as used in this description are intended to cover both directly connected or connected through one or more intermediate means, unless otherwise stated.
[0043] The term “associated with”, used herein when referring to two elements refers to a broad relationship between the two elements. The relationship includes, but is not limited to a physical, a chemical or a biological relationship. For example, when element A is associated with element B, elements A and B may be directly or indirectly attached to each other or element A may contain element B or vice versa.
[0044] The term “adjacent” used herein when referring to two elements refers to one element being in close proximity to another element and may be but is not limited to the elements contacting each other or may further include the elements being separated by one or more further elements disposed therebetween.
[0045] The term “and / or”, e.g., “X and / or Y” is understood to mean either “X and Y” or “X or Y” and should be taken to provide explicit support for both meanings or for either meaning.
[0046] Further, in the description herein, the word “substantially” whenever used is understood to include, but not restricted to, “entirely” or “completely” and the like. In addition, terms such as “comprising”, “comprise”, and the like whenever used, are intended to be non-restricting descriptive language in that they broadlyinclude elements / components recited after such terms, in addition to other components not explicitly recited. For example, when “comprising” is used, reference to a “one” feature is also intended to be a reference to “at least one” of that feature. Terms such as “consisting”, “consist”, and the like, may in the appropriate context, be considered as a subset of terms such as “comprising”, “comprise”, and the like. Therefore, in embodiments disclosed herein using the terms such as “comprising”, “comprise”, and the like, it will be appreciated that these embodiments provide teaching for corresponding embodiments using terms such as “consisting”, “consist”, and the like. Further, terms such as “about”, “approximately” and the like whenever used, typically means a reasonable variation, for example a variation of + / - 5% of the disclosed value, or a variance of 4% of the disclosed value, or a variance of 3% of the disclosed value, a variance of 2% of the disclosed value or a variance of 1 % of the disclosed value.
[0047] Furthermore, in the description herein, certain values may be disclosed in a range. The values showing the end points of a range are intended to illustrate a preferred range. Whenever a range has been described, it is intended that the range covers and teaches all possible sub-ranges as well as individual numerical values within that range. That is, the end points of a range should not be interpreted as inflexible limitations. For example, a description of a range of 1% to 5% is intended to have specifically disclosed sub-ranges 1% to 2%, 1% to 3%, 1% to 4%, 2% to 3% etc., as well as individually, values within that range such as 1%, 2%, 3%, 4% and 5%. It is to be appreciated that the individual numerical values within the range also include integers, fractions and decimals. Furthermore, whenever a range has been described, it is also intended that the range covers and teaches values of up to 2 additional decimal places or significant figures (where appropriate) from the shown numerical end points. For example, a description of a range of 1% to 5% is intended to have specifically disclosed the ranges 1.00% to 5.00% and also 1.0% to 5.0% and all their intermediate values (such as 1.01%, 1.02% ... 4.98%, 4.99%, 5.00% and 1.1%, 1.2% ... 4.8%, 4.9%, 5.0% etc.,) spanning the ranges. The intention of the above specific disclosure is applicable to any depth / breadth of a range.Additionally, when describing some embodiments, the disclosure may have disclosed a method and / or process as a particular sequence of steps. However, unless otherwise required, it will be appreciated that the method or process should not be limited to the particular sequence of steps disclosed. Other sequences of steps may be possible. The particular order of the steps disclosed herein should not be construed as undue limitations. Unless otherwise required, a method and / or process disclosed herein should not be limited to the steps being carried out in the order written. The sequence of steps may be varied and still remain within the scope of the disclosure.
[0048] Furthermore, it will be appreciated that while the present disclosure provides embodiments having one or more of the features / characteristics discussed herein, one or more of these features / characteristics may also be disclaimed in other alternative embodiments and the present disclosure provides support for such disclaimers and these associated alternative embodiments.
[0049] The description herein may be, in certain portions, explicitly or implicitly described as algorithms and / or functional operations that operate on data within a computer memory or an electronic circuit. These algorithmic descriptions and / or functional operations are usually used by those skilled in the information / data processing arts for efficient description. An algorithm is generally relating to a self-consistent sequence of steps leading to a desired result. The algorithmic steps can include physical manipulations of physical quantities, such as electrical, magnetic or optical signals capable of being stored, transmitted, transferred, combined, compared, and otherwise manipulated.
[0050] Further, unless specifically stated otherwise, and would ordinarily be apparent from the following, a person skilled in the art will appreciate that throughout the present specification, discussions utilizing terms such as “scanning”, “calculating”, “determining”, “replacing”, “generating”, “initializing”, “outputting”, and the like, refer to action and processes of an instructingprocessor / computer system, or similar electronic circuit / device / component, that manipulates / processes and transforms data represented as physical quantities within the described system into other data similarly represented as physical quantities within the system or other information storage, transmission or display devices etc.
[0051] The description also discloses relevant device / apparatus for performing the steps of the described methods. Such apparatus may be specifically constructed for the purposes of the methods, or may comprise a general purpose computer / processor or other device selectively activated or reconfigured by a computer program stored in a storage member. The algorithms and displays described herein are not inherently related to any particular computer or other apparatus. It is understood that general purpose devices / machines may be used in accordance with the teachings herein. Alternatively, the construction of a specialized device / apparatus to perform the method steps may be desired.
[0052] In addition, it is submitted that the description also implicitly covers a computer program, in that it would be clear that the steps of the methods described herein may be put into effect by computer code. It will be appreciated that a large variety of programming languages and coding can be used to implement the teachings of the description herein. Moreover, the computer program if applicable is not limited to any particular control flow and can use different control flows without departing from the scope of the invention.
[0053] Furthermore, one or more of the steps of the computer program if applicable may be performed in parallel and / or sequentially. Such a computer program if applicable may be stored on any computer readable medium. The computer readable medium may include storage devices such as magnetic or optical disks, memory chips, or other storage devices suitable for interfacing with a suitable reader / general purpose computer. In such instances, the computer readable storage medium is non-transitory. Such storage medium also covers all computer-readable media e.g. medium that stores data only for short periods oftime and / or only in the presence of power, such as register memory, processor cache and Random Access Memory (RAM) and the like. The computer readable medium may even include a wired medium such as exemplified in the Internet system, or wireless medium such as exemplified in Bluetooth technology. The computer program when loaded and executed on a suitable reader effectively results in an apparatus that can implement the steps of the described methods.
[0054] The example embodiments may also be implemented as hardware modules. A module is a functional hardware unit designed for use with other components or modules. For example, a module may be implemented using digital or discrete electronic components, or it can form a portion of an entire electronic circuit such as an Application Specific Integrated Circuit (ASIC). A person skilled in the art will understand that the example embodiments can also be implemented as a combination of hardware and software modules.
[0055] Additionally, when describing some embodiments, the disclosure may have disclosed a method and / or process as a particular sequence of steps. However, unless otherwise required, it will be appreciated the method or process should not be limited to the particular sequence of steps disclosed. Other sequences of steps may be possible. The particular order of the steps disclosed herein should not be construed as undue limitations. Unless otherwise required, a method and / or process disclosed herein should not be limited to the steps being carried out in the order written. The sequence of steps may be varied and still remain within the scope of the disclosure.
[0056] DESCRIPTION OF EMBODIMENTS
[0057] Exemplary, non-limiting embodiments of a microfluidic device and a method of performing biophysical profiling of a sample, and a method of classifying cells are disclosed hereinafter.1. Microfluidic device for biophysical profiling of a sample
[0058] In various embodiments, there is provided a microfluidic device / chip for biophysical profiling of a sample, the microfluidic device comprising, a first microfluidic channel and a second microfluidic channel configured to allow the sample comprising a plurality of cells to flow through; a first cell trajectory modulation region disposed within the first microfluidic channel, the first cell trajectory modulation region comprising a first array of microstructures, e g., pillars configured to sort the cells and provide one or more first modulated biophysical signatures comprising one or more first distribution profiles of the sorted cells; and a second cell trajectory modulation region disposed within the second microfluidic channel, the second cell trajectory modulation region comprising a second array of microstructures, e.g., pillars configured to sort the cells and provide one or more second modulated biophysical signatures comprising one or more second distribution profiles of the sorted cells; wherein the first modulated biophysical signatures are substantially different from the second modulated biophysical signatures of the sorted cells.
[0059] In various embodiments, the microfluidic device may be referred to as a cell trajectory modulation (CTM) device / chip. In various embodiments, the microfluidic device may be configured to perform a CTM assay for biophysical profiling of a sample comprising a plurality of cells. In various embodiments, the microfluidic device may be used in a method for performing biophysical profiling of a sample as disclosed herein. In various embodiments, the CTM assay may utilize controlled modulation of cell trajectories within the microfluidic device to generate measurable modulated biophysical signatures. In various embodiments, these modulated biophysical signatures may then be used for downstream analysis, including cell classification, correlation with cellular phenotype, or assessment of functional properties such as activation state or potency.In various embodiments, the microfluidic device may be configured to perform biophysical profiling of a sample based on one or more biophysical properties / characteristics of cells in the sample as the cells flow through the microfluidic device. In various embodiments, the biophysical profiling may be achieved through controlled interactions between the cells and microstructures within the microfluidic device under defined flow conditions, wherein the trajectories, positions and / or distributions of the cells are modulated in a manner dependent on the biophysical properties of the cells. In other words, the trajectories of the cells may be a function of the intrinsic biophysical properties of the cells. In various embodiments, the one or more biophysical properties may include but are not limited to, cell size, cell shape, cell clusters, and cell deformability. As used herein, cell size broadly refers to a diameter or a volume of a cell. As used herein, cell deformability broadly refers to the ability of a cell to deform under applied hydrodynamic forces, e g., shear stress, within the microfluidic device.
[0060] In various embodiments therefore, the microfluidic device may be configured such that variations in these biophysical properties result in distinguishable differences in cell trajectories, displacement, or spatial distribution within the device. For example, cells of different sizes and / or deformability may follow different flow paths, exhibit different lateral displacements, or occupy different output locations within the microfluidic device. These differences may be captured as modulated biophysical signatures including distribution profiles, position-based signatures, or histogram-based representations corresponding to the biophysical properties of the cells.
[0061] In various embodiments, the microfluidic device may be configured to simultaneously or sequentially probe multiple biophysical properties by subjecting the cells to different flow conditions and / or different microstructures, e g., pillar configurations. Such configurations may enable separate or combined assessment of biophysical properties such as size and deformability. In various embodiments, the resulting modulated biophysical signatures may be used toderive quantitative biophysical features for downstream analysis, classification, or correlation with cellular function or phenotype.
[0062] In various embodiments, the microfluidic device may be a microfluidic deterministic lateral displacement (DLD) device adapted / modified to perform a CTM assay. While CTM and conventional DLD techniques may have overlapping features, there are important differences between these two techniques. In various embodiments, the CTM technique relies on flow modulation of the cells / particles in addition to utilization of a unique microstructure, e.g., pillar structure, for cell interaction. For example, the CTM technique involves varying the flow rates to achieve a plurality of distribution profiles providing useful information on the cell property. On the other hand, conventional DLD is designed to operate within a specific flow range for efficient cell sorting.
[0063] In other words, the CTM technique differs from conventional DLD techniques in that the primary objective is not to achieve sorting of cells, but rather to probe and characterize the cells based on their modulated trajectories under varying conditions. While conventional DLD devices are generally designed to operate within a specific flow range to ensure predictable and efficient separation, the CTM microfluidic device may deliberately exploit deviations from such regimes, including variations in flow rate and deformation-induced behavior, to extract richer information about cell properties.
[0064] Accordingly, in various embodiments, CTM may be understood as a technique that utilizes principles of DLD in combination with controlled flow modulation and microstructure design to generate modulated biophysical signatures of cells. These modulated biophysical signatures may then be used for downstream analysis, including cell classification, correlation with cellular phenotype, or assessment of functional properties such as activation state or potency.In various embodiments, the microfluidic device may be fabricated from or comprise a polymer material, such as a synthetic polymer or elastomer. In one example, the microfluidic device comprises polydimethylsiloxane (PDMS). In other embodiments, the microfluidic device may comprise silicone-based polymers, thermoplastics, or other suitable materials. In various embodiments, the microfluidic device may be fabricated using a variety of techniques, including injection molding, imprint lithography, soft lithography, 3D printing, or computer numerical control (CNC) machining. In various embodiments, other materials such as plastics (biodegradable or non-biodegradable), glass (e.g., silica or quartz), or composite materials may be used depending on desired properties such as optical transparency, mechanical robustness, or chemical compatibility.
[0065] In various embodiments, the microfluidic device may be comprised in a microfluidic test kit, i.e., provided as part of a microfluidic test kit. In various embodiments, the microfluidic test kit may be individually packaged and may comprise the microfluidic device, one or more tubing connectors, and one or more pre-loaded buffer reservoirs. In various embodiments, the microfluidic device may be disposable. In some embodiments, the microfluidic device may be configured for single use or one-time use, thereby reducing the risk of contamination and facilitating use in clinical environments.
[0066] In various embodiments, the microfluidic device may have a compact footprint suitable for integration with a corresponding system. For example, the microfluidic device may have a footprint of approximately 7.5 cm by 10 cm. In various embodiments, the dimensions of the microfluidic device may be defined by a length X and a width Y, where X and Y may fall within a range with start and end points selected from the following group of numbers: 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 8.5, 9, 9.5, 10, 10.5, 11, 11.5, 12, 12.5, 13, 13.5, 14, 14.5, and 15 cm.
[0067] In various embodiments, the microfluidic device may be configured to achieve a sorting resolution of about 200 nm. In various embodiments, the sortingresolution may fall within a range with start and end points selected from the following group of numbers: 50, 100, 150, 200, 250, 300, 350, 400, 450, and 500 nm.
[0068] 1.1. Microfluidic channels
[0069] In various embodiments, the first microfluidic channel and the second microfluidic channel are configured to receive and convey a sample comprising a plurality of cells. In various embodiments, the first microfluidic channel and the second microfluidic channel may each define a respective flow path (e g., a first flow path associated with the first microfluidic channel and a second flow path associated with the second microfluidic channel) through which the sample may be introduced, transported and processed within the microfluidic device.
[0070] In various embodiments, the first microfluidic channel and the second microfluidic channel may be disposed on a common substrate or integrated within a single microfluidic device. In various embodiments, the microfluidic channels may be formed using microfabrication techniques, including but not limited to soft lithography, molding, etching, or machining.
[0071] In various embodiments, the first microfluidic channel and the second microfluidic channel may be fluidically independent of each other such that flow through one of the microfluidic channels does not pass through the other microfluidic channel. In this configuration, each microfluidic channel may be configured to process the same sample or different portions of the same sample independently. In some embodiments, the microfluidic channels may be arranged in a non-serial configuration, for example in a parallel, branched, or otherwise spatially separated arrangement. In some embodiments, the first microfluidic channel and the second microfluidic channel may be arranged substantially parallel to each other, or parallel to each other along at least a portion of their respective lengths.In various embodiments, each of the first and second microfluidic channels may further comprise one or more segments / regions configured to perform respective functions along the flow path. In various embodiments, the different segments within each microfluidic channel are flu idically coupled to one another. In various embodiments, the segments / regions may include, but are not limited to, sample introduction regions, pre-focusing regions, sheath flow regions, cell trajectory modulation regions, and detection regions. In various embodiments, the segments within the microfluidic channel may be arranged sequentially along the flow path such that the sample undergoes successive processing steps as it traverses the microfluidic channel. In some embodiments, the segments within the microfluidic channel may be sequentially arranged. In some embodiments, the segments within the microfluidic channel may be sequentially but not colinearly arranged. In some embodiments, the segments within the microfluidic channel may be sequentially and co-linearly arranged. In some embodiments, the co-linear arrangement of segments may advantageously improve sensitivity of results.
[0072] 1.2. Cell trajectory modulation regions
[0073] In various embodiments, the first cell trajectory modulation region and the second cell trajectory modulation region may collectively be referred to as a cell trajectory modulation (CTM) region / segment of the microfluidic device. In various embodiments, the CTM region is configured to modulate trajectories of cells as they flow therethrough to facilitate generation of modulated biophysical signatures.
[0074] In various embodiments, the first array of pillars may comprise a first pillar configuration and the second array of pillars may comprise a second pillar configuration. In various embodiments, the first and second pillar configurations may differ from each other in at least one of pillar dimension, pillar shape, pillar structure, pillar arrangement or pillar orientation, with respect to a direction of flow of the cells. In various embodiments, such differences in pillar configurations mayresult in distinct cell-microstructure interactions in the respective CTM regions and generate different modulated biophysical signatures for the same cell population.
[0075] In various embodiments, the CTM region may be configured to modulate flow trajectories of cells passing therethrough. In various embodiments, the flow trajectory taken by a cell passing through the CTM region may be affected by one or more of the following parameters: cell size, cell shape, cell deformability, flow conditions e.g., flow rate of the sample, and structural configuration of the array of pillars. As a result, cells having different biophysical properties may follow different trajectories, exhibit different displacement behaviors, or occupy different spatial positions within the microfluidic device.
[0076] In various embodiments, the microfluidic device is configured to provide multiple CTM conditions within a single device / chip. For example, the microfluidic device may advantageously integrate two distinct pillar configurations, e.g., a substantially L-shaped pillar configuration in the first CTM region and a substantially inverse L-shaped pillar configuration in the second CTM region. These configurations may be used to generate complementary or distinct flow trajectories, thereby enhancing sensitivity to different cell properties, such as size and deformability.
[0077] In various embodiments, the pillars in the first array and the second array may have shapes selected from a variety of geometries, including but not limited to a substantially L shape (L), a substantially inverse L shape (L-1), mirror reflections thereof or combinations thereof. For example, the pillars in the first array may take the form of substantially L-shaped (L) pillars and the pillars in the second array may take the form of substantially inverse L-shaped (L-1) pillars. However, it will be appreciated that the pillars in the first and second arrays are not limited to these shapes, and other suitable geometric shapes may be employed to achieve desired trajectory modulation effects. In variousembodiments, the pillars may be selected from one or more of the shapes (e.g., cross-sectional shape) shown in Table 1.
[0078] Table 1 : Examples of suitable geometries for the pillar
[0079]
[0080] In various embodiments, the pillars may be symmetric or asymmetric in shape. In some embodiments, symmetrical pillars may have no more than 1 line of symmetry, no more than 2 lines of symmetry, no more than 3 lines of symmetry or no more than 4 lines of symmetry. In some embodiments, the pillars in the first array and the pillars in the second array may be asymmetric in shape.
[0081] In various embodiments, the pillars in the first array may have the same or substantially similar dimension, shape and structure but may have a different orientation from the pillars within the second array (e g. with respect to the direction of cell flow). In some embodiments, the difference in orientation may be due to a rotation of the pillars (i.e. rotationally different) at an angle of from about 1° to about 359°, from about 10° to about 350°, from about 20° to about 340°, from about 30° to about 330°, from about 40° to about 320°, from about 50° to about 310°, from about 60° to about 300°, from about 70° to about 290°, from about 80° to about 280°, or from about 90° to about 270° (for e.g., about 180°).In various embodiments, the pillars in the first array and the pillars in the second array may be mirror images of each other.
[0082] In various embodiments, the pillars in the first array and / or the pillars in the second array may have two longitudinal sections / segments abutting each other (for e.g. an L shape or a T shape). In various embodiments, the pillars in the first array and / or the pillars in the second array may have at least one curved surface. The curved surface may be one that extends from one end of a first longitudinal section / segment to another end of a second longitudinal section / segment (e.g. see shapes number 5 and 6 of T able 1 ). It should be appreciated that while curved surfaces may offer certain advantages, the absence of a curved surface may also work. Therefore, in some embodiments, the pillars may be devoid of curved surfaces and comprise only comers and / or flat surfaces. In some embodiments, the pillars in the first array and / or the pillars in the second array may have at least one pillar protrusion. The pillars in the first array and / or the pillars in the second array may have at least one groove. In one example, the at least one groove has a shape of a quadrant (e.g. see shapes number 5 and 6 of Table 1 ).
[0083] In various embodiments, the pillars are microstructures e.g., micropillars. Thus, in various embodiments, the dimensions of the pillars are in the pm range for example, the dimensions of the pillars may be less than about 1000 pm, less than about 900 pm, less than about 800 pm, less than about 700 pm, less than about 600 pm, less than about 500 pm, less than about 400 pm, less than about 300 pm, less than about 200 pm, less than about 100 pm, less than about 90 pm, less than about 80 pm, less than about 70 pm, less than about 60 pm, less than about 50 pm, less than about 40 pm, less than about 30 pm, less than about 20 pm, or less than about 15 pm.
[0084] In various embodiments, the first and second arrays of pillars may be configured to resolve differences in cell trajectories for cells within a dynamic size range. For example, the microfluidic device may be configured to profile cells having effective sizes in a range from about 2 pm to about 25 pm. In variousembodiments, the dynamic range for measuring an approximated output size range may fall in the range with start and end points selected from the following group of numbers: 2, 2.5, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, and 25 pm. In various embodiments, the specific range may be selected based on the target application and the types of cells or particles being analyzed. In some embodiments, the dynamic range may be about 6 pm to about 12 pm with a single sample inlet and outlet. In some embodiments, the microfluidic device may be configured to profile smaller cells or particles, such as red blood cells, which may have characteristic sizes in a range from about 2 pm to about 4 pm.
[0085] 1.3. Detection regions
[0086] In various embodiments, the microfluidic device further comprises a first detection region disposed downstream of the first cell trajectory modulation region and a second detection region disposed downstream of the second cell trajectory modulation region. In various embodiments, the first and second detection regions may be configured to receive cells exiting the respective cell trajectory modulation regions and to enable measurement of spatial distributions of the cells corresponding to modulated biophysical signatures. In various embodiments, the first and second detection regions may collectively be referred to as a detection region / segment of the microfluidic device.
[0087] In various embodiments, the first detection region comprises a first set of location bins for allowing the cells that are sorted by the first array of pillars to flow into or flow through and the second detection region comprises a second set of location bins for allowing the cells that are sorted by the second array of pillars to flow into or flow through. In various embodiments, the location bins may correspond to discrete spatial regions or output positions within the microfluidic device into which cells are directed after passing through the respective arrays of pillars in the cell trajectory modulation regions. In various embodiments, the location bins may be defined by physical channel partitions. In operation, cellsexiting the cell trajectory modulation regions may be distributed across the location bins according to their respective trajectories.
[0088] In various embodiments, the location bins may represent the final trajectory positions / locations of the cells within the microfluidic chip. In various embodiments, the number of cells flowed into or through each location bin may be counted to obtain / generate a distribution profile of the sorted cells. In various embodiments, the distribution profile of the sorted cells may be represented as a raw histogram distribution of the sorted cells, e.g., a frequency distribution across a plurality of bins. In various embodiments, such histogram -based representations may correspond to raw modulated biophysical signatures associated with the sorted cell population.
[0089] In various embodiments, the distribution profile of the sorted cells may be further processed to extract biophysical markers / features, e.g., quantitative biophysical markers / features. In various embodiments, the biophysical markers / features may be referred to as CTM markers / features. In some embodiments, the biophysical markers / features may comprise descriptors such as statistical measures, histogram bin evaluations, gated population calculations, distribution profile comparisons, or combinations thereof. In some embodiments, such descriptors may be computed from the histogram data corresponding to the modulated biophysical signature. In some embodiments, transformation functions comprising mathematical and / or statistical operations may be applied to the histogram data to generate the biophysical markers / features. For example, statistical measures such as mean, variance, skewness, kurtosis, interquartile range, or other derived metrics may be computed from the histogram data. In some embodiments, multiple distribution profiles obtained under different cell trajectory modulation conditions, e.g., different flow conditions and / or different pillar configurations, may be combined or transformed to generate composite biophysical features for downstream analysis.In various embodiments, the microfluidic device may be used in conjunction with a detection setup configured to count / determine the number and / or position of cells within the detection region. In some embodiments, the detection setup may comprise an imaging apparatus and a processing unit configured to capture and analyze images corresponding to the cells. For example, the detection setup may include a high-speed camera, a smartphone camera, a machine vision camera, or an electrode system. The processing unit may be configured to perform image processing, object detection, tracking, and counting of cells within the respective location bins.
[0090] In various embodiments, when an imaging-based detection setup is used, e.g. a high-speed camera detection setup, the imaging apparatus may be configured to operate at a frame rate selected based on flow conditions, e.g., flow rates of the sample / cells, and field of view of the device. For example, the frame rate may range from about 15 frames per second (fps) to about 250 fps, including values such as 15, 30, 20 60, 90, 120, 150, 180, 210 and 240 fps. In some embodiments, the frame rate may be selected to ensure accurate detection and counting of cells under different flow velocities. For example, frame rates of about 15 fps, about 30 fps, about 60 fps and about 150 fps may be used for flow rates of about 2.5 mm / s, about 5.0 mm / s, about 10.0 mm / s and about 25.0 mm / s respectively. It will be appreciated that other frame rates and flow conditions may be employed depending on the application and device configuration.
[0091] In various embodiments, the detection region of the microfluidic device and associated detection setup may advantageously enable real-time or near real-time acquisition of cell distribution data, thereby facilitating rapid generation of modulated biophysical signatures for downstream analysis.
[0092] 1.4. Pre-focusing regions
[0093] In various embodiments, the microfluidic device may further comprise a first prefocusing region disposed upstream of the first cell trajectory modulationregion and / or a second prefocusing region disposed upstream of the second cell trajectory modulation region. In various embodiments, the prefocusing regions may be configured to pre-separate or partially separate cells within the sample prior to entering the cell trajectory modulation regions (i.e., the actual CTM assay). In various embodiments, the first and second prefocusing regions may collectively be referred to as a prefocusing region / segment of the microfluidic device.
[0094] In various embodiments, the first prefocusing region may comprise a first modified array of pillars and the second prefocusing region may comprise a second modified array of pillars. In various embodiments, the modified arrays of pillars may differ from the pillar arrays in the cell trajectory modulation regions in at least one of pillar spacing, pillar geometry and pillar arrangement. In some embodiments, the modified arrays of pillars may differ from the pillar arrays in the cell trajectory modulation regions in terms of spacing between the pillars. In some embodiments, the spacing between adjacent pillars in the prefocusing regions may be larger or smaller than the spacing between pillars in the cell trajectory modulation regions, depending on the desired prefocusing effect to be achieved.
[0095] In various embodiments, the prefocusing regions may be configured to perform a pre-separation of cells based on one or more biophysical properties, such as cell size or deformability. For example, the prefocusing regions may align cells into one or more streamlines, or separate cells into coarse groups prior to finer trajectory modulation in the downstream cell trajectory modulation regions. In various embodiments therefore, the prefocusing region may advantageously improve consistency of cell entry into the CTM regions and reduce variability in cell-microstructure interactions.
[0096] In various embodiments, the prefocusing regions may advantageously enable greater signal to noise ratio of detection with the first and / or second cell trajectory modulation regions and the first and / or second detection regions. In some embodiments, the prefocusing regions may be configured to focus andconcentrate target cells, such as immune cells. As a result, throughput (e.g., the number of cells processed per unit time) may be increased and detection efficiency may be enhanced. In some embodiments, the prefocusing regions may also selectively remove or reduce unwanted cells and particles. As a result, signal to noise ratio, where the signal corresponds to the number of target cells processed, may be further improved.
[0097] In various embodiments, the prefocusing regions may also advantageously reduce aggregation effects or irregular flow patterns that could otherwise interfere with accurate trajectory modulation and detection. In some embodiments, the prefocusing regions may comprise similar arrays configured to direct cells into desired flow paths. In some embodiments, the prefocusing regions may utilize different microfluidic techniques, including passive inertial techniques and / or active particle manipulation techniques, such as acoustic forces, dielectric forces, and / or magnetic forces, to augment focusing / concentration / preconditioning of the sample prior to downstream processing.
[0098] 1.5. Inlet(s)
[0099] In various embodiments, the microfluidic device further comprises one or more sample inlets fluidically coupled to the first microfluidic channel and the second microfluidic channel. In various embodiments, the one or more sample inlets may be disposed upstream of the cell trajectory modulation regions and the prefocusing regions, if present. In various embodiments, the one or more sample inlets may be configured to introduce a sample comprising a plurality of cells into the microfluidic device and to deliver one or more sample streams to the first and second microfluidic channels. In various embodiments, the one or more sample inlets may be configured to introduce a common sample stream to both the first and second microfluidic channels.In some embodiments, the microfluidic device may comprise a single sample inlet fluidically coupled to a flow junction, e.g., branching flow junction, configured to divide the sample stream into two or more portions which are then directed into the first and second microfluidic channels. This advantageously enables parallel processing of the same sample under different cell trajectory modulation conditions.
[0100] In some embodiments, the microfluidic device may comprise multiple sample inlets. For example, the microfluidic device may comprise a first sample inlet fluidically coupled to the first microfluidic channel and a second sample inlet fluidically coupled to the second microfluidic channel such that independent sample streams may be introduced into each microfluidic channel. This advantageously enables the microfluidic device to process different samples, or replicate the samples under similar or differing flow conditions, e.g., different flow conditions and / or different pillar configurations.
[0101] In various embodiments, the one or more sample inlets may be configured to receive the sample from an external source, such as a syringe, reservoir, pump system, or automated fluid handling system. In various embodiments, the flow of the sample through the sample inlets may be controlled using pressure-driven flow, syringe pumps, peristaltic pumps, or other suitable fluid actuation mechanisms.
[0102] 1.6. Sheath inlet(s)
[0103] In various embodiments, the microfluidic device further comprises one or more first sheath inlets and / or one or more second sheath inlets fluidically coupled to the first microfluidic channel and / or the second microfluidic channel. In various embodiments, the one or more first and / or second sheath inlets may be disposed upstream of the cell trajectory modulation regions and the prefocusing regions, if present. In various embodiments, the sheath inlets maybe configured to introduce one or more sheath fluids into the first and / or second microfluidic channels.
[0104] In various embodiments, the one or more first sheath inlets may be configured to introduce a first sheath stream, and the one or more second sheath inlets may be configured to introduce a second sheath stream into the first and / or second microfluidic channel. In various embodiments, the first and second sheath streams may be arranged relative to the sample stream such that the sheath streams hydrodynamically confine, direct or focus the sample stream within the microfluidic channel. In various embodiments, the sheath streams may be introduced from opposite sides of the sample stream such that the sample stream is positioned (i.e., sandwiched) between the first and second sheath streams, thereby forming a layered flow configuration.
[0105] In various embodiments, the first sheath stream, the sample stream, and the second sheath stream may be associated with respective hydraulic resistances R1, R2 and R3 within the microfluidic channel. In various embodiments, the hydraulic resistances R1 to R3 may be approximated as proportional to a ratio of channel length to channel width, i.e., R °c where L represents length of the channel, w represents width of the channel with an assumed fixed height. In various embodiments, the first microfluidic channel and / or the second microfluidic channel may be configured such that relative hydraulic resistance ratios between the sheath streams and the sample stream fall within predetermined ranges. For example, the microfluidic channel may be configured such that conditions (i) 0.33 < R3 / R2 < 49, and (ii) 2 < R1 / R2 < 94 are satisfied. In various embodiments, the resistance ratios may be selected to achieve the desired flow focusing.
[0106] In various embodiments, the microfluidic device may comprise different configurations of sheath inlets. In some embodiments, the microfluidic device may comprise one first sheath inlet and one second sheath inlet configured to provide the first and second sheath streams to both the first and secondmicrofluidic channels. In some embodiments, the microfluidic device may comprise two first sheath inlets and two second sheath inlet configured to provide the first and second sheath streams to the first and second microfluidic channels, respectively. It will be appreciated that other configurations of sheath inlets may be used.
[0107] Advantageously, the sheath inlets may ensure that cells enter downstream regions / segments, e.g., the cell trajectory modulation regions, in a controlled manner. As a result, this may enhance the consistency of cell-microstructure interactions and improve the quality of the resulting modulated biophysical signatures.
[0108] 1.7. Outlet(s)
[0109] In various embodiments, the microfluidic device further comprises one or more outlets fluidically coupled to the first microfluidic channel and the second microfluidic channel. In various embodiments, the one or more outlets may be disposed downstream of the detection region. In various embodiments, the one or more outlets may be configured to allow discharge of fluid from the microfluidic device after the sample has passed through the various regions / segments of the device, e.g., cell trajectory modulation regions and detection regions. In various embodiments, the one or more outlets may be configured to facilitate collection or removal of fluid exiting the microfluidic channels.
[0110] In some embodiments, the microfluidic device may comprise a single outlet fluidically coupled to the first and second microfluidic channels via a flow junction, e.g., a merging flow junction. In some embodiments, the microfluidic device may comprise two or more outlets, e.g., a first outlet associated with the first microfluidic channel and a second outlet associated with the second microfluidic channel.In various embodiments, the one or more outlets may be waste outlets for discharging waste fluid. In various embodiments, the waste fluid may comprise processed sample fluid and sheath fluid. In various embodiments, the one or more outlets may be fluidically coupled to a liquid handling unit having collection vessels for collecting waste fluid.
[0111] In various embodiments, the one or more outlets may be configured to be coupled to a pressure control unit. For example, the pressure control unit may be configured to drive fluid flow in a microfluidic device by applying negative pressure at an outlet of the microfluidic device and / or positive pressure at an inlet of the microfluidic device.
[0112] In various embodiments, the different segments of the microfluidic device may be sequentially arranged. For example, the different segments of the microfluidic chip may be sequentially arranged in the following order: input region, prefocusing region, cell trajectory modulation region, and detection region. In some embodiments, the different segments of the microfluidic device may be sequentially but not co-linearly arranged. In some embodiments, the different segments of the microfluidic device may be sequentially and co-linearly arranged.
[0113] 2. Method of performing biophysical profiling of a sample
[0114] In various embodiments, there is provided a method of performing biophysical profiling of a sample, the method comprising, passing a volume of the sample comprising a plurality of cells through a microfluidic device / chip comprising a plurality of microstructures, e g., pillars configured to modulate flow trajectories of the plurality of cells; obtaining one or more modulated biophysical signatures of the plurality of cells generated by the microfluidic device, the modulated biophysical signatures comprising distribution profiles of the sorted plurality of cells; processing the one or more modulated biophysical signatures using one or more transformation functions configured to map and integrate the one or more modulated biophysical signatures to obtain one or more biophysicalfeatures / CTM features; and inputting the one or more CTM features into a trained determination unit configured to correlate the one or more CTM features to one or more properties / attributes of the plurality of cells.
[0115] In various embodiments, the method of performing biophysical profiling may be referred to as a biophysical profiling assay or a CTM assay. In various embodiments, the modulated biophysical signatures may be used as a biomarker of the cellular properties / attributes such as cell phenotype or function. For example, activated T cells are more prone to be highly active to kill target cells such as cancer cells. Typically, a fully equipped laboratory with flow cytometry systems, confocal microscopes and spectrophotometry systems are required to probe the function of activated T cells. In various embodiments, the method may advantageously be used as a surrogate for conventional cell phenotype and functional assays, for rapidly profiling cells based on their cellular attributes such as cell phenotype and function.
[0116] 2.1. Passing the sample through a microfluidic device
[0117] In various embodiments, the sample may be introduced into and passed through a microfluidic device comprising one or more arrays of microstructures, e.g., pillars, configured to interact with the cells and modulate their trajectories as they traverse the microfluidic device. In some embodiments, the sample may be introduced into and passed through a microfluidic device as disclosed herein. In some embodiments, the microfluidic device may comprise a first array of pillars having a first pillar configuration and a second array of pillars having a second pillar configuration. In various embodiments, the first array of pillars may differ from the second array of pillars in at least one of pillar dimension, pillar shape, pillar structure, pillar arrangement, pillar spacing, or pillar orientation, with respect to a direction of flow of the cells. Such differences in pillar configurations may result in distinct cell-microstructure interactions and consequently different trajectory behaviors of the same cell population.In various embodiments, it will be appreciated that the method is not limited to use with the specific microfluidic device as disclosed herein. Alternative microfluidic devices capable of producing measurable modulated biophysical signatures may be used in conjunction with the method. For example, an alternative microfluidic device may comprise a single array of pillars, instead of two different arrays of pillars. Therefore, it will be appreciated that the microfluidic device may vary in architecture, provided that it is capable of generating modulated biophysical signatures corresponding to one or more biophysical properties of the cells.
[0118] In various embodiments, the flow trajectory taken by a cell passing through the microfluidic device may be affected by one or more of the following parameters: size of the cell, shape of the cell, deformability of the cell, flow rate of the sample, and configuration of the array of pillars. In various embodiments therefore, a unique modulated biophysical signature is generated when the sample is passed through the microfluidic device under a specific set of conditions. In various embodiments, the distribution profile of the sorted cells may comprise a histogram data of the number of cells sorted into a plurality of location bins in the microfluidic device.
[0119] In various embodiments, the step of passing a volume of the sample through the microfluidic device may be repeated for a plurality of times under different sets of conditions to obtain a plurality of modulated biophysical signatures. In various embodiments, the set of conditions may include, but are not limited to, pillar configuration, flow rate of the sample, duration of flowing the sample through the microfluidic device, and a combination thereof. In some embodiments, different sets of conditions may be achieved by varying one or more of these conditions across multiple, e.g., successive runs. In some embodiments, a separate microfluidic device may be used for each set of conditions, e.g., for single channel microfluidic devices, or for each pair of sets of conditions, e.g., for dual channel microfluidic devices.In various embodiments, the sample may be passed through the microfluidic device at different flow rates to induce different trajectory responses. In various embodiments, the sample may be passed through the microfluidic device at a first flow rate and a second flow rate, wherein the second flow rate is higher than the first flow rate. In various embodiments, the second flow rate may be higher than the first flow rate by about 2 times, about 3 times, about 4 times, about 5 times, about 6 times, about 7 times, about 8 times, about 9 times, about 10 times, about 11 times, about 12 times, about 13 times, about 14 times, about 15 times, about 16 times, about 17 times, about 18 times, about 19 times or about 20 times. In various embodiments, variation in flow rate may alter hydrodynamic forces acting on the cells and enable probing of different biophysical properties, e.g., cell deformability under higher flow conditions.
[0120] In various embodiments, the sample may be passed through the microfluidic device, e.g., the plurality of microstructures such as the array of pillars, at different flow rates or different flow velocities. In various embodiments, the flow velocities may fall within a range of from about 1 mm / s to about 1000 mm / s. In various embodiments, the flow velocities may fall within the range with start and end points selected from the following group of numbers: 1 , 1.5, 2, 2.5, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 100, 150, 200, 250, 300, 350, 400, 450, 500, 550, 600, 650, 700, 750, 800, 850, 900, 950, and 1000 mm / s. In various embodiments, the flow rate may fall within a range of from about 1.0 pL / min to about 100.0 pL / min. In various embodiments, the flow rate may fall within the range with start and end points selected from the following group of numbers: 1, 1.5, 2, 2.5, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, and 100 pL / min.
[0121] In various embodiments, the flow conditions e.g., flow rates and pillar configuration within the microfluidic device may cooperatively modulate trajectories of the cells based on one or more biophysical properties of the cells to generate a plurality of modulated biophysical signatures under different flow and pillar configurations. In various embodiments, a unique / distinct modulatedbiophysical signature may be generated for each set of conditions. Accordingly, by performing repeated runs under different conditions, a plurality of modulated biophysical signatures may be obtained for the same sample, thereby providing a multi-dimensional representation of the biophysical properties of the cells.
[0122] 2.2. Obtaining the one or more modulated biophysical signatures
[0123] In various embodiments, obtaining the one or more modulated biophysical signatures of the plurality of cells may comprise detecting the cells at a detection region of the microfluidic device, counting the cells that are sorted to a plurality of histogram bins at the detection region of the microfluidic device, and obtaining a set of histogram distribution of the cells.
[0124] In various embodiments, detection of the cells may be performed using a detection setup comprising an imaging apparatus and a processing unit configured to capture and analyze images corresponding to the cells within the detection region. For example, the detection setup may include a high-speed camera, a smartphone camera, a machine vision camera, or an electrode system. The processing unit may be configured to perform image processing, object detection, tracking, and counting of cells within the respective location bins.
[0125] In various embodiments, the detection region may comprise a plurality of bins / histogram bins corresponding to discrete spatial regions or output positions within the microfluidic device into which cells are directed after passing through the respective arrays of pillars. In various embodiments, counting the cells that are sorted to the plurality of histogram bins may be performed using image processing algorithms, signal processing techniques, or other detection methods configured to identify and count individual cells within each bin.
[0126] In various embodiments, the counts of cells associated with the plurality of histogram bins may be used to generate a histogram distribution representing the spatial distribution of the cells. In various embodiments, the histogram distributionmay comprise a set of values corresponding to the number or frequency of cells in each bin and may represent a modulated biophysical signature of the cell population under a given set of conditions. In various embodiments, the histogram distribution may be processed, e.g., normalized, scaled, to account for variations in total cell count, flow conditions, or other parameters. In some embodiments, multiple histogram distributions obtained under different conditions may be combined or compared to provide a comprehensive representation of the biophysical characteristics of the cells.
[0127] In various embodiments, the method may comprise obtaining a plurality e.g., four modulated biophysical signatures, each corresponding to a different set of conditions under which the sample is passed through the microfluidic device. In various embodiments, each modulated biophysical signature may represent a distribution profile of cells generated under a specific combination of device configuration and flow conditions, thereby providing complementary information relating to the biophysical properties of the cells.
[0128] In various embodiments, a first, second, third and fourth modulated biophysical signatures may be obtained by passing the sample through a microfluidic device under a corresponding first, second, third and fourth set of conditions, respectively. In various embodiments, the first set of conditions may comprise passing the sample through a microfluidic device having a first array of pillars at a first flow rate within a first time period. In various embodiments, the second set of conditions may comprise passing the sample through a microfluidic device having a second array of pillars at the first flow rate within the first time period. In various embodiments, the third set of conditions may comprise passing the sample through a microfluidic device having the first array of pillars at a second flow rate within a second time period. In various embodiments, the fourth set of conditions may comprise passing the sample through a microfluidic device having the second array of pillars at the second flow rate within the second time period.In other words, the first set of conditions may differ from the second set of conditions in the pillar configuration of the array of pillars in the microfluidic device. The third set of conditions may differ from the fourth set of conditions in the pillar configuration of the array of pillars in the microfluidic device. For example, the pillars in the first array may take the form of substantially L shape (L) pillars and the pillars in the second array may take the form of substantially inverse L shape (L-1) pillars. The third and fourth sets of conditions may differ from the first and second sets of conditions in the flow rates and the time periods at which the sample is passed through the microfluidic device. For example, the second flow rate may be 10 times higher than the first flow rate. For example, the first time period may be 4 minutes, and the second time period may be 1 minute.
[0129] In various embodiments, the combination of different pillar configurations and flow conditions may enable probing of different biophysical properties of the cells. For example, lower flow rates may emphasize size-dependent trajectory behavior, while higher flow rates may induce deformation-dependent effects and enable assessment of cell deformability. Therefore, the resulting set of modulated biophysical signatures may advantageously provide a multi-dimensional characterization of the cell population.
[0130] It will be appreciated that the different sets of conditions are not limited to the above combination of pillar configurations, flow rates, and time periods. Other suitable combinations of conditions may be used. It will be appreciated that the different arrays of pillars may be comprised in a single microfluidic device. For example, the different arrays of pillars may be comprised in respective microfluidic channels in a single microfluidic device. It will be appreciated that more than four sets of modulated biophysical signatures may be obtained, each corresponding to a different set of conditions under which the sample is passed through the microfluidic device.
[0131] 2.3. Obtaining the one or more biophysical features / CTM featuresIn various embodiments, one or more biophysical features / CTM features may be obtained by processing the one or more modulated biophysical signatures using one or more transformation functions configured to map and integrate the one or more modulated biophysical signatures.
[0132] In various embodiments, the one or more transformation functions may be configured to transform the one or more modulated biophysical signatures into one or more CTM features that are classified based on cell size, cell deformation, or a combination thereof. In various embodiments, the one or more CTM features comprise size-related features, deformability-related features, or combined size and deformability-related features. In various embodiments, multiple distribution profiles corresponding to multiple modulated biophysical signatures obtained under different cell trajectory modulation conditions may be combined or transformed to generate composite biophysical features for downstream analysis.
[0133] In various embodiments, the one or more transformation functions may comprise a first set of transformation functions configured to transform the modulated biophysical signatures based on cell size, a second set of transformation functions configured to transform the modulated biophysical signatures based on cell deformability, and a third set of transformation functions configured to transform the modulated biophysical signatures based on a combination of cell size and cell deformability.
[0134] In various embodiments, the first set of transformation functions may be used to process the modulated biophysical signatures that are obtained under the first and / or second sets of conditions. In various embodiments, the second set of transformation functions may be used to process the modulated biophysical signatures that are obtained under the third and / or fourth sets of conditions. In various embodiments, the third set of transformation functions may be used to process the modulated biophysical signatures that are obtained under the first, second, third and / or fourth sets of conditions.In various embodiments, the transformation functions comprise one or more mathematical and / or statistical operations comprising statistical measures, histogram bin evaluations, gated population calculations, distribution profile comparisons, or combinations thereof.
[0135] In various embodiments, the transformation functions may comprise: i. transformation functions for computing data values for one or more selected gated populations from one or more sets of histogram data obtained under one or more defined sets of conditions (e.g., one or more selected gated population values, summation, difference, product, and ratio between multiple selected gated population values),
[0136] ii. transformation functions for computing statistical values from one or more sets of histogram data obtained under one or more defined sets of conditions (e.g., weighted interquartile range, standard deviation, weighted median, mean, gated mean, average mean, skew, kurtosis, Kullback-Leibler (KL) divergence, fitted Gaussian distribution, summation, difference, product, and ratio between multiple statistical values).
[0137] Examples of transformation functions (non-exhaustive) for transforming the modulated biophysical signatures are provided in Tables 2 to 4.
[0138] Table 2: Examples of transformation functions (non-exhaustive) for transforming the modulated biophysical signatures based on cell size
[0139] CTM Feature Feature Description Function
[0140] S1 Selected gated population of H2 Z^ZB^,- S2 Selected gated population of H2 XiTjf /
[0141] S3 Weighted interquartile range H2 Weighted interquartile range of H2. , 0 < i < 35 S4 Weighted interquartile range H1 Weighted interquartile range of H,, , 0 < 1 < 35 S5 Standard deviation H1 Standard Deviation of H,, „
[0142] S6 Standard deviation H2 Standard Deviation of H3,
[0143] S7 Selected gated population of H2 X?=i fA,
[0144] S8 Selected gated population of H1 iXiWi,
[0145] S9 Kurtosis H1 Skewness Wlf
[0146] S10 Selected gated population of H2
[0147] S11 Mean value of H1 Hi, x i ) / II ,
[0148] x O / ZM Wi, + Q3=aH2Xi) / S12 Average means of H1 and H2 X?=o^2,] / 2
[0149]
[0150] S13 Weighted median of H1 Weighted median of 0 < i < 35, H2S14 Gated mean of H1 (Sf=7Wi,-x0 /
[0151] S15 Weighted median of H2 Weighted median of 0 < i < 35,
[0152] S16 Mean value of H2 (2^2,x0 / Sfio«2,
[0153] S17 Gated mean of H2 (E£7«2,xi ) / XLS7«2,
[0154] S18 Selected gated population of H2 Xt?2 W1]
[0155] S19 Selected gated population of H2 S’iswi,
[0156] S20 Difference of means for H1 and H2 Si=o^i,x‘x‘ ) / X / =o S21 Kurtosis of H2 Kurtosis H,
[0157] S22 Skew of H2 Skewness H2
[0158] S23 Selected gated population of H2 Zifiow2,
[0159] S24 Selected gated population of H2
[0160] S25 Skew of H1 + Kurtosis of H1 Skewness W4+ Kurtosis Hlf
[0161] S26 Product of Kurtosis of H1 and H2 Kurtosis Hlfx Kurtosis H2
[0162] S27 Skew of H1 Skewness W1(
[0163] S28 Selected gated population of H2
[0164] S29 Difference in Skew of H1 and H2 Skewness H1(- Skewness H2
[0165]
[0166] S30 Kullback-Leibler divergence of H1 and H2 Wijog (H2)
[0167] Table 3: Examples of transformation functions (non-exhaustive) for transforming the modulated biophysical signatures based on cell deformability
[0168] CTM Feature Feature Description Function
[0169] Sum of weighted [H2- g(H2.)] where i > 22 Sum of weighted difference of H2 values and its Where g(x) is the probability density function of a D1 fitted gaussian distribution Gaussian Distribution
[0170] D2 Difference of means between H3 and H4 0 / x0 / S=ow4, D3 Selected gated population of H3 #3,
[0171] D4 Standard Deviation H4 Standard Deviation of
[0172] D5 Weighted Interquartile range H4 Weighted interquartile range of H4, 0 < i < 35 D6 Standard Deviation H3 Standard Deviation of W3,
[0173] D7 Selected gated population of H4
[0174] D8 Selected gated population of H3
[0175] D9 Selected gated population of H4 S;ii6w4,
[0176] D10 Selected gated population of H3 Etu, H3,
[0177] D11 Weighted median of H3 Median of H3
[0178] D12 Gated mean of H3 QX57«3,xi ) / £1=- H3l
[0179] D13 Mean of H3 0 /
[0180] [(SF=OXi } / +(Zi=0^4jX* ) / D14 Average of means for H3 and H4 S^=o #4,1 / 2
[0181] D15 Selected gated population of H3 S;ii6H3,
[0182] D16 Selected gated population of H4 X-_’inW4,
[0183] D17 Mean of H4 <E£n«4,x' ) / X&, «4,
[0184] D18 Weighted median of H4 Weighted median of 0 < i < 35, H4,
[0185] D19 Gated mean of H4 <2£7«4,xi } / TSi- Ht
[0186] D20 Selected gated population of H4 2^2 W4,
[0187] D21 Weighted interquartile range H3 Weighted interquartile range of H3. , 0 < i < 35 D22 Skew of H4 Skewness H4
[0188] D23 Skew of H3 Skewness H3.
[0189] D24 Kurtosis of H4 Kurtosis H4
[0190] D25 Selected gated population of H4 ELi #4,
[0191] D26 Selected gated population of H3 %
[0192] D27 Difference between skew of H3 and H4 Skewness H3l- Skewness H,:
[0193]
[0194] D28 Kurtosis of H3 Kurtosis H3.Table 4: Examples of transformation functions (non-exhaustive) for transforming the modulated biophysical signatures based on a combination of cell size and cell deformability
[0195] CTM Feature Feature Description Formula
[0196] Sum of weighted [H1(—<<7(^1,)] where i > 22 Sum of weighted difference of H1 values and its Where g(x) is the probability density function of a C1 fitted gaussian distribution Gaussian Distribution
[0197] Ratio of Weighted interquartile range of H2: HtC2 Ratio of Interquartile range of H2 against H4 0 < i < 35
[0198] C3 Ratio of selected gated population H1 and H3 i
[0199] Product of Weighted interquartile range of C4 Product of interquartile range of H2 and H4 H2and 0 < i < 35
[0200] Ratio of Weighted interquartile range of : W3, C5 Ratio of interquartile range of H1 against H3 0 < i < 35
[0201] C6 Ratio of selected gated population H1 and H3 il-n, / SSi6#3,
[0202] Product of weighted interquartile range of H1 and Product of Weighted interquartile range of H1(: f / 3, C7 H3 0 < i < 35
[0203] Sum of weighted [ff2|-,o('H2()l where i < 22 Sum of weighted difference of H2 values and its Where g(x) is the probability density function of a C8 fitted gaussian distribution Gaussian Distribution
[0204] C9 Ratio of selected gated population of H2 and H4 / 2=16^4,
[0205] Ratio of means difference of H1 and H3 against [(SF=owi,x' ) / X; H3x i ) / H3(] 1 C10 means difference of H2 and H4 [CXr=o ^2,x0 / w2, - W*x0 / W4]
[0206] Sum of weighted [Hlf- s(Hlf)] where I < 22 Sum of weighted difference of H1 values and its Where g(x) is the probability density function of a C11 fitted gaussian distribution Gaussian Distribution
[0207] C12 Difference of skew of H2 and H4 Skewness H2l- Skewness H4,
[0208] C13 Difference of skew of H1 and H3 Skewness H1- Skewness H,
[0209] Kullback-Leibler divergence of H2and Kullback-Leibler divergence of H2 and fitted Where g(x) is the probability density function of a C14 gaussian distribution for H2 Gaussian Distribution
[0210] C15 Difference in means of H2 and H4 [CXr=ox' ) / w2, ~ < X ;=7, 77* x r ~j / Sto Weighted median of H2— Weighted median of C16 Difference in weighted median for H2 and H4 H4, 0 < i < 35
[0211] C17 Difference in means of H1 and H3x0 / _ xWeighted median of Hu- Weighted median of C18 Difference in weighted median for H1 and H3 H3j, 0 < i < 35
[0212] C19 Kullback-Leibler divergence of H2 and H4 )
[0213] y-35 u incr^H'|
[0214] C20 Kullback-Leibler divergence of H1 and H3Zj,=1*^3 /
[0215] Product of means difference for H1 and H3 withx‘ ) / ZFIowi, -x0 / C21 means difference for H2 and H4 ^2,x' ) / Z£ow2, -x
[0216] C22 Difference of means for H1 and H3 [(S=o^iixi) / S?=owii (Sf=0#3xi) / 2=ow3 Kullback-Leibler divergence of H, and Kullback-Leibler divergence of H1 and fitted Where g(x) is the probability density function of a C23 gaussian distribution for H1 Gaussian Distribution
[0217] C24 Kullback-Leibler divergence of H3 and H4 #3,l°g(h4)
[0218]
[0219] C25 Ratio of selected gated population for H1 and H3 S?=28^i,J[QX50W2, x i ) / v: . x i )! C26 Average of means of H2 and H4 X l=o / 2
[0220] [(S^=oxi ) / +(2 / =o ^3,xi ) / C27 Average of means of H1 and H3
[0221] Ratio of means difference for H1 and H2 against (2i=o ^ifxO / 2i=o^i, (E(=o^2,xO / Zt=o^2(C28 means difference for H3 and H4 QSH3(X‘) / - QX„ / i4, x i ) / C29 Means difference for H2 and H4 QXofXxi ) / - (S&>w4,-x' ) /
[0222] XO / Sp=o^2f+ X0 / 2
[0223] Difference in average means of H2 and H4 and [(S?=owiiX0 / +(E?lo^3xO /
[0224]
[0225] C30 average means of H1 and H3 ■>
[0226] In the “Feature Description” column, H1, H2, H3 and H4 represent histogram data of modulated biophysical signatures obtained by passing the sample through the microfluidic device under the first, second, third and fourth sets of conditions, respectively. In the “Function” column, Wconditionfcinrepresents the value of a particular location / histogram bin when the sample is passed through the microfluidic device under a particular set of conditions. For example, H2ZOrepresents the value of histogram bin number 20 under the sample is passed through the microfluidic device under the second set of conditions.
[0227] 2.4. Correlating the one or more CTM features to one or more properties / attributes of the plurality of cells
[0228] In various embodiments, the one or more CTM features may be input into a trained determination unit configured to correlate the one or more CTM features to one or more properties / attributes of the plurality of cells.
[0229] In various embodiments, the trained determination unit may comprise one or more computational models configured to analyze the CTM features and generate outputs indicative of cellular attributes. In various embodiments, the trained determination unit may comprise a machine learning model to perform one or more analytical task, including but not limited to classification, regression, clustering ora combination thereof. In some embodiments, the machine learning model may be configured for performing classification and regression tasks.In various embodiments, the trained machine learning model may comprise one or more models including, but not limited to, linear models (e.g., linear regression), tree-based models (e g., decision trees, random forest classifier), neural networks (e.g., convolutional neural networks, multilayer perceptron), support vector machines (SVMs), Bayesian models, instance-based models (e.g., k-nearest neighbors k-NN). In some embodiments, the trained determination unit may comprise a machine learning model based on one or more of linear models, tree-based models, neural networks, support vector machines, Bayesian models, and instance-based models. In some embodiments, the trained determination unit may be based on a support vector machine (SVM) algorithm designed for classification and / or regression tasks.
[0230] In various embodiments, the trained machine learning model may be configured to perform dimension reduction of the CTM features. For example, principal component analysis (PCA) or other dimensionality reduction techniques may be used to reduce the dimensionality of the CTM feature space while retaining relevant variance. In some embodiments, the CTM features may be reduced to a smaller number of principal components, e.g., two principal components, for visualization and / or clustering.
[0231] In various embodiments, the trained machine learning model may be configured to perform machine classification of the CTM features using existing modules available in libraries of programming languages known in the art, e.g., Python, with modules such as pandas, sci-kit learn and SciPy, for data normalization, statistical analysis, correlation analysis, PCA data and hierarchical clustering. In various embodiments, the trained machine learning model may be configured to perform statistical analyses. For example, sample means testing may be performed using a two-tailed Student’s t-test, while correlation analysis may be evaluated with Pearson’s correlation. In various embodiments, data visualization and plots may be performed using Matplotlib and Seaborn Python modules. Visualization tools may be used to generate plots, heatmaps, or other graphical representations of the CTM features and results.In various embodiments, the one or more properties / attributes of the cells determined by the trained determination unit may comprise cell function, cell phenotypes, and potency. In various embodiments, cell function-related attributes / functional attributes may include, but are not limited to, immune cell activation state (e.g., T cell activation), expression of markers such as CD127+, CD4+, CD8+, PD1+, LAG3+, TIM3+, CD57+, and TEMRA, or ratios such as CD4:CD8 and CD3:CD28 ratio. In various embodiments, functional phenotypes / cell phenotypes may include, but are not limited to, effector memory phenotype and central memory phenotype. In various embodiments, potency-related attributes may include, but are not limited to, killing potency, and cytokines such as CD19, IL-3, IL-5, IL-13, TNF, and GM-CSF.
[0232] In various embodiments, the cells comprise chimeric antigen receptor (CAR) T cells. In various embodiments, the trained determination unit may be configured to determine properties associated with cell therapy applications such as immune cell activation state, cell phenotype, and killing potency. In various embodiments, the determination of these properties may facilitate assessment of quality, efficacy or functionality of CAR T cell products.
[0233] In various embodiments, the method further comprises extracting image data of the cells and inputting the image data of the cells into a single cell machine classifier that is trained to classify the cells based on different morphologies of the cells, e g., cell size and cell shape. In various embodiments, the cells may be classified using a method of classifying cells as disclosed herein. In various embodiments, the output from the single cell machine classifier may be used independently or in combination with the CTM features to improve accuracy of classification or prediction of cell attributes.
[0234] In various embodiments, there is provided a non-transitory computer readable storage medium having stored thereon instructions for instructing a processing unit of a system to execute a method of performing biophysicalprofiling of a sample, the method comprising, passing a volume of the sample comprising a plurality of cells through a microfluidic device comprising a plurality of pillars configured to modulate flow trajectories of the plurality of cells; obtaining one or more modulated biophysical signatures of the plurality of cells generated by the microfluidic device, the modulated biophysical signatures comprising distribution profiles of the sorted plurality of cells; processing the one or more modulated biophysical signatures using one or more transformation functions configured to map and integrate the one or more modulated biophysical signatures to obtain one or more cell trajectory modulation (CTM) features; and inputting the one or more CTM features into a trained determination unit configured to correlate the one or more CTM features to one or more properties of the plurality of cells.
[0235] 3. Method of classifying cells
[0236] In various embodiments, there is provided a method of classifying cells, the method comprising, passing a volume of a sample comprising a plurality of cells through a microfluidic device comprising a plurality of pillars configured to modulate flow trajectories such that the plurality of cells are sorted into a plurality of bins / location bins / histogram bins; capturing image data of the plurality of location bins using an imaging apparatus; processing the image data of the plurality of location bins to extract image data of the cells sorted into the plurality of location bins; and inputting the image data of the cells into a trained machine learning model to classify the cells into one or more categories based on one or more morphological characteristics.
[0237] In various embodiments, the microfluidic device may comprise one or more microfluidic channels, each microfluidic channel comprising an array of microstructures, e.g., pillars configured to modulate flow trajectories of the cells such that, under continuous flow conditions, cells within the sample are directed along different paths based on one or more biophysical properties of the cells. In various embodiments, the one or more biophysical properties may include butare not limited to cell size, cell shape and cell deformability. In various embodiments, array of microstructures may be a structured array of pillars having predetermined geometric features, including but not limited to size, spacing, orientation, and arrangement. In various embodiments, the method may be performed using a microfluidic device as disclosed herein.
[0238] In various embodiments, the imaging apparatus may be configured to capture image data in real time, for example in the form of a continuous video stream or a sequence of image frames, as cells pass through the plurality of location bins. In various embodiments, the imaging apparatus may comprise a microscope system. In various embodiments, the imaging apparatus may be a low-cost imaging system comprising a camera and a fixed objective lens configured to acquire image data of cells flowing through the microfluidic device. In various embodiments, the low-cost imaging system may be suitable for in-field or clinical deployment, offering real-time sample classification for point-of-care diagnostics. In various embodiments, the low-cost imaging system may enable rapid and accessible acquisition of image data without requiring high-end microscopy equipment. In various embodiments, the combination of the microfluidic device and the imaging apparatus advantageously enables real time or near real time classification of cells based on their morphological and biophysical properties, facilitating applications such as point-of-care diagnostics, process monitoring in biomanufacturing, or on-site analysis of biological samples.
[0239] In various embodiments, processing the image data of the plurality of location bins to extract image data of cells sorted into the plurality of location bins comprises applying blob detection to identify and analyze regions within the image data that represent / correspond to cells. In various embodiments, the image data may be obtained as a sequence of image frames or video data captured by the imaging apparatus as the cells flow through the microfluidic device. In various embodiments, blob detection is applied to the image data to identify regions corresponding to cells flowing through the microfluidic device and to distinguish such regions from background features, including structures of themicrofluidic device. The identified regions are used to isolate individual cells from the image data. In various embodiments, processing the image data may further comprise extracting and cropping image regions corresponding to the identified cells to generate individual cell images. Each extracted cell image may be centered on the cell and represented as a fixed-size image. In various embodiments, the extracted cell images may comprise low-resolution images, for example having a size of approximately 20 x 20 pixels, corresponding to individual cells captured within the plurality of location bins.
[0240] In various embodiments, the method further comprises obtaining a distribution profile of the plurality of cells across the plurality of location bins. In various embodiments, the distribution profile comprises the number of cells sorted into each location bin. In various embodiments, the number of cells in each location bin may be determined by counting the cells corresponding to that location bin. In various embodiments, each extracted cell image may be associated with a corresponding location bin. In various embodiments, the number of cells sorted into each location bin may be segmented according to the one or more categories based on the one or more morphological characteristics. For example, the number of cells associated with each location bin may be grouped according to classifications based on morphological characteristics, such as size-based categories and shape-based categories. For example, the number of cells sorted into each location bin may be segmented in terms of the number of cells for each category of size (small, medium, large) and shape (doublets, blobs, irregular class for cells with abstract morphologies). In various embodiments, the segmented distribution profile advantageously provides a representation of the distribution of different cell morphologies across the plurality of location bins.
[0241] In various embodiments, the trained machine learning model may be configured to classify the cells based on size and shape. In various embodiments, the size category may comprise small, medium and large cells. In various embodiments, the shape category may comprise doublets, blobs and irregularshapes for cells with abstract morphologies. In various embodiments, the classification based on size may depend on the type of cell to be classified, as different types of cells have different ranges of sizes. In some embodiments, the cells may be immune cells. In some embodiments, images where the immune cells are less than about 7 pm may be considered small-sized cells, images where the immune cells are greater than 12 pm may be considered large cells and immune cells that fall between the range may be classified as medium -sized cells. In various embodiments, a region of interest in an image may be considered a doublet when two particles (e.g., cells and / or synthetic particles) are adhered to each other. In various embodiments, a region of interest in an image may be considered a blob when multiple (i.e., more than two) particles are grouped together. In various embodiments, single particles (e g., cells) that have non-spherical morphologies may be considered irregular.
[0242] In various embodiments, the method further comprising training the machine learning model to classify the cells based on different morphologies of the cells. In various embodiments, training the machine learning model may comprise preparing a set / dataset of pooled images for training, validating and testing purposes. In various embodiments, the dataset of pooled images may comprise image data of cells classified based on size (small, medium, large) and shape (doublets, blobs, irregular class for cells with abstract morphologies). In various embodiments, the dataset of pooled images may comprise original and augmented image data of cells, wherein the augmented image data of cells are derived from the original image data of cells by image augmentation.
[0243] In various embodiments, the method may further comprise performing image augmentation on the image data of the cells. In various embodiments, the dataset of pooled images may be enhanced with image augmentation techniques to compensate for imbalances in image data distribution, i.e., to compensate for classes having fewer image samples relative to other classes. In various embodiments, image augmentation may comprise one or more transformations applied to the image data captured by the imaging apparatus. In variousembodiments, the one or more transformations may include, but are not limited to horizontal flipping, vertical flipping, rotation (e.g., 180° rotation), and shifting (e.g., circular shifting) of the image data. In various embodiments, the augmented image data may be combined with the original image data to form the pooled dataset used to train the machine learning model.
[0244] In various embodiments, the trained machine learning model may comprise a single cell machine classifier that is trained to classify the cells based on different morphologies of the cells from image data. In various embodiments, the machine learning model is trained to distinguish between different cell morphologies, including categories based on size and shape, using image data of individual cells extracted from the plurality of location bins.
[0245] In various embodiments, the machine learning model may comprise at least one of a support vector machine, random forest classifier, k-nearest neighbor model, logistic regression model, artificial neural network, convolutional neural network, or gradient boosting model. In some embodiments, the machine learning model may comprise a single model. In some embodiments, the machine learning model may comprise a plurality of models used in combination.
[0246] In various embodiments, the machine learning model may comprise a support vector machine (SVM). Support vector machines are supervised machine learning algorithms that classify data by finding the optimal hyperplane that separates different classes while maximizing the margin between them. Support vector machines can perform complex classification boundaries on non-linear data. In various embodiments, the support vector machine may be configured to classify the cells by determining an optimal hyperplane separating different morphological classes.
[0247] In various embodiments, the trained machine learning model is configured to receive image data of individual cells as input and output a classification corresponding to one or more morphological categories. For example, each cellimage may be represented as a set of pixel values corresponding to a fixed-size image.
[0248] In various embodiments, the sample may be substantially free of externally added tags or labels (i.e. label free). The sample may also be undiluted / untreated. In various embodiments, there may be no need for additional laboratory equipment to pre-process the sample. Advantageously, in various embodiments, the method does not require additional and time-consuming steps to label and treat / process the sample prior to classifying. In various embodiments, the sample is introduced into the microfluidic device without dilution or pre-treatment.
[0249] In various embodiments, there is provided a non-transitory computer readable storage medium having stored thereon instructions for instructing a processing unit of a system to execute a method of classifying cells, the method comprising, passing a volume of a sample comprising a plurality of cells through a microfluidic device comprising a plurality of pillars configured to modulate flow trajectories such that the plurality of cells are sorted into a plurality of location bins; capturing image data of the plurality of location bins using an imaging apparatus; processing the image data of the plurality of location bins to extract image data of the cells sorted into the plurality of location bins; and inputting the image data of the cells into a trained machine learning model to classify the cells into one or more categories based on one or more morphological characteristics.
[0250] In various embodiments, the microfluidic device, system and method as disclosed herein may be configured to perform biophysical profiling of the sample within a target time period of no more than about 15 minutes. In various embodiments, the target time period may be no more than about 15 minutes, no more than about 10 minutes, or no more than about 5 minutes. In various embodiments, the time period may include one or more of sample loading, processing within the microfluidic device, image acquisition, data processing, and output of results. In various embodiments, the microfluidic device and method as disclosed herein may advantageously enable rapid biophysical profiling of thesample within a significantly shorter time period compared to conventional central laboratory testing methods, which typically require approximately 60 to 120 minutes to obtain results. This reduction in processing time facilitates near realtime assessment and may support timely clinical decision-making, particularly in point-of-care settings.
[0251] In various embodiments, the microfluidic device, system and method as disclosed herein may be configured to process a microvolume sample. In various embodiments, the sample may have a volume of from about 10 pL to about 100 pL. In some embodiments, the sample may have a volume of from about 20 pL to about 50 pL. In various embodiments, the volume of the sample may fall in the range with start and end points selected from the following group of numbers: 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 55, 60, 65, 70, 75, 80, 85, 90, 95, and 100 pL.
[0252] In various embodiments, the microfluidic device, system and method as disclosed herein are configured to perform biophysical profiling of a sample having a concentration of cells (e.g., analyte cells, cells of interest, cells to be measured) falling in the range of from about 0.5 x 106cells / mL to 100 x 106cells / mL. In various embodiments, the concentration of the sample may fall in the range having start and end points selected from the following group of numbers: 0.5 x 106cells / mL, 1 x 106cells / mL, 2 x 106cells / mL, 5 x 106cells / mL, 10 x 106cells / mL, 15 x 106cells / mL, 20 x 106cells / mL, 25 x 106cells / mL, 30 x 106cells / mL, 35 x 106cells / mL, 40 x 106cells / mL, 45 x 106cells / mL, 50 x 106cells / mL, 55 x 106cells / mL, 60 x 106cells / mL, 65 x 106cells / mL, 70 x 106cells / mL, 75 x 106cells / mL, 80 x 106cells / mL, 85 x 106cells / mL, 90 x 106cells / mL, 95 x 106cells / mL, and 100 x 108cells / mL. In some embodiments, the microfluidic device and associated method as disclosed herein are configured to perform biophysical profiling on a sample having a concentration falling in the range of from 0.5 x 106cells / mL to 80 x 106cells / mL. In some embodiments, thesample may further comprise additional non-analyte cells, and the above concentrations refer specifically to analyte cells to be measured.
[0253] In various embodiments, the sample may contain no more than 25,000 cells (e g., analyte cells, cells of interest, cells to be measured), no more than 24,000 cells, no more than 23,000 cells, no more than 22,000 cells, no more than 21 ,000 cells, no more than 20,000 cells, no more than 19,000 cells, no more than 18,000 cells, no more than 17,000 cells, no more than 16,000 cells, no more than 15,000 cells, no more than 14,000 cells, no more than 13,000 cells, no more than 12,000 cells, no more than 11 ,000 cells, no more than 10,000 cells, no more than 9,000 cells, no more than 8,000 cells, no more than 7,000 cells, no more than 6,000 cells, or no more than 5,000 cells. In some embodiments, the microfluidic device, system and method as disclosed herein are configured to perform biophysical profiling on a sample comprising no more than about 20,000 cells. In some embodiments, the sample may further comprise additional non-analyte cells and the above-defined cell numbers refer specifically to the analyte cells to be measured.
[0254] In various embodiments, the terms “analyte cells” “cells of interest” and “cells to be measured” refer to cells within a sample that are selected, targeted, or otherwise intended to be measured, analyzed, or profiled using the microfluidic device, systems, or methods as disclosed herein. In some embodiments, analyte cells may comprise a specific subpopulation of cells of interest within a heterogeneous sample. In some embodiments, analyte cells may include, but are not limited to, white blood cells, circulating tumor cells, stem cells, cultured cells, and the like. In some embodiments, analyte cells are present in a sample that further comprises additional non-analyte cells, such as red blood cells or other background cells, which are not the primary subject of measurement.
[0255] In various embodiments, the sample may be a biological sample. In various embodiments, the sample may be derived from a mammalian subject. In various embodiments, the biological sample may be a fluid sample. In variousembodiments, the sample may include, but is not limited to, bodily fluid (e g., blood, urine, saliva, breast milk), particulate solutions (e.g., emulsions and milk products), and cultured cells (e g., suspended cell culture, cells cultured in bioreactors, vessels, and flasks). In various embodiments, the fluid sample may comprise cells, e g., blood cells. In various embodiments, the cell type present in the sample that is used for biophysical profiling may be one of immune cells, leukocytes, red blood cells, stem cells, cancer cells, algae, yeast, or combinations thereof. In various embodiments, the cells may be mammalian cells. In one example, the biological sample is blood. In various embodiments, the sample may be substantially free of externally added tags or labels (i.e. label free). In various embodiments, the sample may also be undiluted / untreated. In various embodiments, there may be no need for additional laboratory equipment to pre-process the sample. Advantageously, in various embodiments, the microfluidic device, system and method as disclosed herein do not require additional and time-consuming steps to label and treat / process the sample prior to profiling or classifying.
[0256] In various embodiments, the sample may be obtained from a subject. In various embodiments, the subject may be a mammal. In various embodiments, the subject may be human. In various embodiments, the subject may be a premature human infant, a human infant, a human child, a human adult, or an elderly human. As used herein, the term “premature human infant” refers to a human infant bom at less than about 37 weeks, less than about 36 weeks, less than about 35 weeks, less than about 34 weeks, less than about 33 weeks, less than about 32 weeks of gestational age. As used herein, the term “human infant” refers to a newborn / neonate to about 1 year old human. As used herein, the term “human child” refers to a human that is about 1 year to about 18 years old. As used herein, the term “human adult” refers to a human that is about 18 years or older. As used herein, the term “elderly human” refers to a human about 65 years or older. In some embodiments, the subject is a human infant or a human child no more than 12 years old. In various embodiments, the microfluidic device, system and method as disclosed herein may be configured to perform biophysicalprofiling of a human child aged about 12 years and below, including a human infant.
[0257] In various embodiments, the methods disclosed herein may be implemented using the microfluidic device disclosed herein. In some embodiments, the methods may be implemented using alternative microfluidic devices or systems capable of modulating cell trajectories and generating measurable modulated biophysical signatures.
[0258] In various embodiments, the method of performing biophysical profiling of a sample and the method of classifying cells may be performed at one or more stages of a cell processing or manufacturing workflow. In various embodiments, the methods may be performed prior to cell expansion or genetic modification to assess baseline characteristics of starting cell populations. In various embodiments, the methods may be performed during processing, e g., during cell activation, cell expansion, to monitor changes in biophysical properties of cells. In various embodiments, the methods may be performed at or near the end of the manufacturing process, e.g., prior to infusion to evaluate quality attributes of the final cell product, including but not limited to phenotype, activation state and potency. In various embodiments, the methods may be performed post-cell processing or manufacturing. For example, the methods may be performed following cryopreservation, thawing, or transport to assess quality attributes of the final cell product. In various embodiments therefore, the methods may advantageously be used for monitoring, quality control, and decision making across multiple stages of cell therapy workflows, including but not limited to CAR T cell manufacturing processes.
[0259] In various embodiments, the method of performing biophysical profiling of a sample and the method of classifying cells may be implemented for the assessment of immune cell states in cell-based therapies, such as CAR T cell therapy. In various embodiments, the methods disclosed herein may leverage on the intrinsic / innate biophysical properties of immune cells to define / characterizean “immune state” of a patient. The biophysical immune state may provide quantitative information, offering alternative insights as compared to its biochemical counterpart assays. In some embodiments, the methods may be implemented in a point-of-care testing (POCT) system to enable rapid measurement of such biophysical properties in clinical settings. In various embodiments, the measured biophysical / CTM features may correlate with downstream immune phenotypes and clinical outcomes. In various embodiments, the measured biophysical / CTM features may be used to develop predictive models for treatment response and adverse events, such as cytokine release syndrome. Accordingly, the methods disclosed herein may be used for evaluating efficacy, monitoring safety, and supporting clinical decision-making across pre-infusion and post-infusion stages of CAR-T therapy.
[0260] In various embodiments, the method of performing biophysical profiling of a sample and the method of classifying cells may be a pre-treatment or posttreatment evaluation method. In various embodiments, the CTM features may be extracted directly from biological samples, such as whole blood from patients. In various embodiments, the CTM features may exhibit correlations with clinical outcomes, e.g., patient response to drug treatment. For example, immune cells present in a sample may exhibit modulated biophysical signatures that vary as a function of treatment response, such as in cancer therapies. In various embodiments, the CTM features may be used to develop predictive models capable of forecasting in vivo patient responses at later time points, e.g., at days 6, 10 and 14, based on measurements obtained at an earlier time point, such as day 0. Accordingly, the method may enable early prediction of treatment efficacy and therapy optimization.
[0261] In various embodiments, the method of performing biophysical profiling of a sample may be integrated with the method of classifying cells. In various embodiments, the integration of biophysical profiling and cell classification may provide enhanced analytical capability for applications including diagnostics, cell therapy manufacturing, quality control, and monitoring of treatment response. Invarious embodiments, the method of performing biophysical profiling of a sample (i.e., the CTM assay / technique) may be integrated with imaging cytometry to enable image-based recognition and classification of large-scale single-cell data. The combination of CTM and single-cell imaging cytometry may provide enhanced analytical capabilities for high-throughput cellular characterization. Such integration may further enable advanced applications in cell and gene therapy.
[0262] It will be appreciated that embodiments of the microfluidic device, system and methods as disclosed herein may be combined, integrated in various manners as appropriate. Likewise, the various example embodiments are not necessarily mutually exclusive as some may be combined with one or more embodiments to form new exemplary embodiments. Other modifications related to biological, chemical, structural, electrical and optical changes may be made without deviating from the scope of the invention. Various embodiments of the microfluidic device, system and methods disclosed herein may also utilize one or more features that are described in PCT / SG2021 / 050011, the contents of which are fully incorporated herein.
[0263] BRIEF DESCRIPTION OF FIGURES
[0264] FIG. 1 is a schematic block diagram illustrating a microfluidic device / chip for biophysical profiling of a sample in an example embodiment.
[0265] FIG. 2 is a schematic block diagram illustrating a system for performing biophysical profiling of a sample in an example embodiment.
[0266] FIG. 3 is a schematic drawing illustrating a microfluidic device / chip for biophysical profiling of a sample in an example embodiment.FIG. 4 is a schematic diagram of a microfluidic channel of a microfluidic device in an example embodiment.
[0267] FIGS. 5A to 5E are schematic drawings illustrating different configurations of sample inlet(s), sheath inlet(s) and outlet(s) of a CTM microfluidic device for performing a CTM assay in an example embodiment.
[0268] FIG. 6 is a schematic drawing of a computer system suitable for implementing an example embodiment.
[0269] FIG. 7 is a schematic drawing illustrating a CTM microfluidic chip in an example embodiment. The CTM microfluidic chip comprises a cell input, a first and second sheath buffer ports 1 and 2, a detection region, a waste outlet and integrates two CTM regions providing conditions of L- and inverse L-shaped structures. The final exit location of a cell can be observed in the detection region. The cell input, sheath buffer ports, detection region, waste outlet, and CTM regions are constructed and operate substantially in accordance with the various embodiments described herein. In particular, sheath buffer port 2 is configured to provide pre-focusing of the sample and to provide a clean background for cells transiting through the CTM regions. Sheath buffer port 2 may enhance detection performance by providing a stable and low-noise background signal for identifying cells that are in the flow of the sheath stream . Sheath buffer port 2 may be further configured to chemically probe biological cells during transit, e.g., by changing electrostatic charges, surface binding of markers and even addition of cell activators.
[0270] FIGS. 8A to 8D are schematic drawings illustrating microfluidic device segment specifications of a CTM microfluidic chip in an example embodiment. FIG. 8A illustrates the table specifications of each microfluidic segment. FIG. 8B illustrates the overall design and specifications for an L-shaped structure. FIG.
[0271] 8C illustrates the overall design and specifications for an inverse L-shaped structure. FIG. 8D illustrates four conditions that can be applied (Condition 1: L-shaped structure at 1x flow rate; Condition 2: inverse L-shaped structure at 1x flow rate; Condition 3: L-shaped structure at 10x flow rate; Condition 4: inverse L-shaped structure at 10x flow rate).
[0272] FIGS. 9A and 9B are schematic diagrams illustrating a CTM assay for label-free cellular biophysical profiling in an example embodiment. As illustrated in FIG. 9A, T cells are injected into a microfluidic device, and a CTM assay is performed by subjecting cells to varying flow parameters and microstructures (L and L-flipped) to modulate cell interactions and flow trajectories within the device. The modulated trajectory results in unique cell output positions and signatures / modulated biophysical signatures for each condition 1, 2, 3 and 4, corresponding to 4 different histograms H1 , H2, H3 and H4, respectively. FIG. 9B describes a downstream processing of the signatures using transform functions resulting in various biophysical features / CTM features based on the modulated biophysical signatures comprising heatmaps H1, H2, H3 and H4. These CTM features are grouped into size (S), deformation (D) and combined (C) categories. The CTM features are used in machine classifier models for correlations to T cell functions such as phenotyping and potency.
[0273] FIGS. 10A to 10D illustrate a CTM assay system with custom optics and microfluidic chip in an example embodiment. FIG. 10A is a photograph of the CTM assay system. FIG. 10B is a corresponding schematic drawing of the CTM assay system. The CTM assay system is a low-cost custom system developed to perform the CTM assay using a fixed objective and a mechanical stage. A flow system is controlled by a stepper motor, and a syringe setup is provided to facilitate withdrawal or compression of a syringe. Solenoid valves are used to contain the pressure in a tube pressure reservoir. FIG. 10C is a calibration plot of the number of steps corresponding to a driving pressure. FIG. 10D is an image of a custom software graphical user interface (GUI) for interfacing with the system and processing the sample in real-time for the processing of histograms data.FIGS. 11Aand 11 B illustrate single cell data point composition of CTM raw histogram data in an example embodiment. FIG. 11A is a graph showing H1 to H4 histogram plots and their corresponding heatmaps. FIG. 11 B is a single CTM histogram comprised of single cell positions within a microfluidic device detected in the highlighted dots.
[0274] FIG. 12 is a Pearson’s correlation heatmap showing correlation of quantitative phase imaging (QPI) features and CTM features in an example embodiment. Features with significant correlation (p value < 0.05) is denoted with a *.
[0275] FIGS. 13A to 13G illustrate CTM assay H1-4 biophysical signatures showing distinct regions of profiles for different T cell states of unstimulated and untreated control, CD3 / 28 activation, and cryopreservation in an example embodiment. FIGS. 13Ato 13C show the H1-4 CTM assay biophysical signatures for three T cell groups at 24 hours for unstimulated (control), CD3 / 28 activation state and exposure to DMSO for an additional 15 mins (mimicking cryopreservation), respectively. FIG. 13D shows unsupervised hierarchical clustering used to group the T cell states with respect to CTM features grouped using a second unsupervised hierarchical clustering. FIG. 13E depicts the grouped CTM features in a Pearson’s correlation heatmap. FIG. 13F are graphs of selected features from the different clusters, with distinct differences in CTM features between all three T cell conditions. FIG. 13G is a principal component analysis (PCA) plot showing clear clusters of the cell conditions. All statistical means analyses were performed using a two-tailed independent Student’s t-test with *, **, and **** denoting p-values < 0.05, 0.01 and 0.0001, respectively.
[0276] FIG. 14 is a set of CTM assay biophysical signatures and corresponding heatmaps illustrating temporal modulation of biophysical properties of T cells during activation in an example embodiment. T cells were stimulated with CD3 / CD28 beads for 0, 6, and 12 hours, and corresponding changes were measured using a CTM assay.FIGS. 15A to 15E illustrate longitudinal biophysical signatures of cultured CD3+ T cells in comparison with flow cytometry forward scatter measurements in an example embodiment. FIG. 15A illustrates an experimental setup in which T cells are profiled at alternate time points over a 14-day culture period following activation, using both flow cytometry and CTM assay. FIG. 15B are images of the cells at each time point taken using quantitative phase imaging (QPI) within culture wells, with a scale bar of 20 pm. FIG. 150 are frequency distributions and corresponding H1 to H4 histograms obtained from the CTM assay. FIG. 15D are graphs showing a comparison between interpolated forward scatter data from flow cytometry and CTM histograms, including Pearson’s correlation analysis, wherein the correlation coefficient is denoted by p and statistical significance is indicated by ***, **** and ns corresponding to p-values of <0.001, <0.0001, and not significant, respectively. FIG. 15E are flow cytometry forward scatter plots of the corresponding cells in culture.
[0277] FIGS. 16A to 16G illustrate correlations between CAR T cell biophysical signatures, cell phenotypes, and cytotoxicity from different bioreactors in an example embodiment. FIG. 16A is a schematic representation illustrating CAR T cell production using G-Rex and Breez bioreactors from peripheral blood mononuclear cell (PBMC)-isolated CD3+ T cells, resulting in endpoint CD19 CAR T cells after a 12-day manufacturing process. The resulting CAR T cell products are subjected to a functional cytotoxicity assay, flow cytometry profiling, and a CTM assay. FIG. 16B is a heatmap illustrating CTM features obtained from multiple runs (n = 3) across three donors, comparing CAR T cells produced using the two bioreactors. FIG. 16C is a scatter volcano plot showing significance of Pearson’s correlation and statistical fold differences between CTM and flow cytometry features for CAR T cells produced using the Breez and G-Rex bioreactors, with the fraction of significant features (p > 0.05) indicated in the legend. FIG. 16D is a heatmap illustrating expression levels of selected flow cytometry phenotypes with a comparative significance study shown between the two bioreactor groups. FIG. 16E are uniform manifold approximation andprojection (UMAP) plots for all the CTM and flow cytometry features for CAR T cells produced using the Breez and G-Rex bioreactors. FIG. 16F is a Pearson’s correlation heatmap comparing flow cytometry phenotype and CTM biophysical features with significant correlations denoted by a * FIG. 16G are representative scatter plots corresponding to selected correlations plots of CD127+, CD4:CD8 ratio, and CAR expression in FIG. 16F. All statistical means analyses were performed using two-tailed paired Student’s t-test with *, **, and *** denoting p-value < 0.05, 0.01, and 0.001, respectively.
[0278] FIG. 17 illustrates CAR T phenotype profiles obtained using flow cytometry markers in an example embodiment. CAR T cells produced using Breez and G-Rex bioreactors are profiled using a panel of markers corresponding to 152 phenotypic subtypes listed in the figure. Significance between the two bioreactor groups is measured using a Student’s two tailed t-test with *, **, ***, **** representing p <0.05, p < 0.01 , p< 0.001 and p < 0.0001 respectively.
[0279] FIG. 18 is a correlation heatmap illustrating a correlation analysis between flow cytometry phenotypes and CTM markers in an example embodiment. The correlation heatmap compares 152 flow cytometry phenotypes and 88 CTM markers. Statistically significant Pearson’s correlation of p < 0.05 is represented by *.
[0280] FIGS. 19A to 19E illustrate correlations between CTM feature(s), CAR T potency specific lysis and cytokine secretion profiles following co-culture with NALM6 tumor cell in an example embodiment. FIG. 19A are raw plots for donors 1 , 2 and 3 in a CAR T cell potency assay for CAR T cells manufactured using Breez and G-Rex bioreactors. FIG 19B is a correlation heatmap comparing CTM features with effectortarget (E:T) ratios. FIG. 19C are correlative plots of selected CTM features S5, D22 and C5 versus 1:5 E:T specific lysis %. As shown in the first plot of FIG. 19C, an E:T ratio of 1 :5 best describes the kinetics of the potency assay with close to 50% killing efficiency. paii represents the Pearson’s correlation of Breez and G-Rex measurements combined, while psreez and pcrex representthe Pearson’s correlation coefficient of Breez and G-Rex separately. FIG. 19D is a correlation heatmap of the CTM features and secreted cytokine. FIG. 19E are correlative plots of selected cytokine profiles with Pearson’s correlation profiles of pcytokine. All correlation heatmap profiles are normalized to -1 to 1 based on Pearson’s correlation analysis, and strongly correlated features with p-value < 0.05 are marked with * in the heatmap. For individual plots, Pearson’s correlative values are shown in p while the significances of the correlation are labeled as *, **, **** and ns, denoting p-value < 0.05, 0.01, 0.0001 and not significant, respectively.
[0281] FIG. 20 is a correlation heatmap between CTM features and secreted cytokines for samples obtained from three donors each measured in triplicates in an example embodiment. The figure provides an expanded cytokine panel correlation data corresponding to the correlation analysis presented in FIGS. 19A to 19E. All correlated features are marked in * where Pearson’s correlation coefficient is significant with p-value < 0.05. The cytokines are sorted based on hierarchical clustering.
[0282] FIGS. 21 A to 21 D illustrate aggregated CTM linear indexing showing distinct biophysical differences between CAR T cells manufactured using Breez and G-Rex bioreactors in an example embodiment. FIG. 21 A is a schematic flowchart illustrating dimensionality reduction of CTM profiled T cell biophysical features using linear principal component analysis (PCA) with the dominant principal component (PC1 ) used as an indexing measure, given its representation of the CTM biophysical features of maximal variance. FIG. 21 B illustrates biophysical indexing results for samples obtained from three healthy donors, each measured in triplicates (n = 3 donors, triplicates). FIG. 21 C illustrates biophysical indexing results for a single donor with multiple replicates (n = 6). FIG.
[0283] 21 D illustrates a comparison of biophysical indexing between CAR T cells derived from healthy donors and patient samples (n = 4) manufactured using the Breez bioreactor. The mean and standard deviation are denoted in the biophysical indexing plot. The two-tailed Student’s t-test is used to perform the statisticalanalysis with **, **** and ns denoting p-value < 0.01 , 0.0001, and not significant, respectively.
[0284] FIGS. 22A and 22B illustrate PCA analysis of CTM data obtained from multiple T cell and CAR T cell datasets in an example embodiment. The datasets include CTM data from tested T cells and CAR T cells from 24-well plate data as described in FIG. 13, CAR T cells from three donors measured in triplicates as described in FIGS. 16 and 19, and CAR T cells from FIG. 21 comprising single donor replicates, healthy donor and patient donor. FIG. 22A is a PCA plot illustrating the PCA data of all CTM tests clustered into two major clusters A and B separated primarily in Component 1. FIG. 22B is a PCA plot of cluster A comprising data from Breez and G-Rex-derived samples wherein component 1 (PC1 ) is used in the analysis of data in FIG. 21.
[0285] FIGS. 23A to 23C illustrate cross comparison of various experimental groups using biophysical indexing with Principal Component 1 (PC1) in an example embodiment. FIG. 23A illustrates a comparison between CAR T cell samples manufactured using Breez (n = 23) and G-Rex (n = 15) bioreactors. FIG.
[0286] 23B illustrates a comparison between cryopreserved CAR T cells (n = 14) and on-day measurement (n = 9) of CAR T cells manufactured using the Breez bioreactor. FIG. 23C illustrates a comparison between cryopreserved CAR T cells (n = 6) and on-day measurement (n = 9) for CAR T cells manufactured using the G-Rex bioreactor. All statistics were performed using Student’s independent two-tailed t-test with *** and **** denoting significance of p-value less than 0.001 and 0.0001 respectively.
[0287] FIGS. 24A and 24B illustrate aggregated PCA component 1 (PC1) indexing for normalized data from potency assay and flow cytometry for all three sets of CAR T cells from (a) 3 donor triplicates (n = 3), (b) single donor replicates (n = 6) and healthy versus patient CAR T cells (n = 4) in an example embodiment. All statistical significance were based on two tailed Student’s t-test with *, **, **** and ns representing p-value < 0.05, 0.01 , 0.0001 and not significant respectively.FIGS. 25A and 25B illustrate aggregated comparison of indexed CTM PC1 versus flow cytometry PC1 index and potency assay PC1 indexing in an example embodiment. Pearson’s correlation is performed on all the datasets and shown in FIG. 25A and FIG. 25B which describes the aggregated correlation analysis of all samples profiled for potency assay and flow cytometry respectively.
[0288] FIGS. 26A to 26C illustrate an overview of a workflow for observing and categorizing distinct cell morphologies in an example embodiment. FIG. 26A is a schematic flowchart illustrating an experimental protocol. A 20 pL sample of cultured cells was profiled using a combination of microscopy and a microfluidic chip. After profiling, the cells were categorized into distinct labels to train a machine learning model, which was later deployed to make predictions on new data. FIG. 26B illustrates a cross-sectional view of a CTM microfluidic device and a profiled cell distribution with decoupled frequencies of distinct cell morphologies. FIG. 26C illustrates multiple sets of 20 x 20-pixel images of observed cells with distinct morphological differences across classifications, enabling a detailed examination of CTM -profiled cell types. The unit length per pixel is 1.35 pm per pixel.
[0289] FIGS. 27A to 27C illustrate a custom Graphic User Interface (GUI) for manual labeling of single cell image data in an example embodiment. FIG. 27A illustrates the GUI with reference images for accurate labels. FIG. 27B illustrates a sizing guide comprising concentric circles for differentiating between cells of different sizes. FIG. 27C illustrates sample cell images of doublets and blobs for labeling.
[0290] FIGS. 28A to 28E illustrate a workflow for development of an image classifier model for prediction of identified cell morphologies in an example embodiment. FIG. 28A illustrates image augmentation methods applied on a dataset to balance class distribution through vertical flip, horizontal flip, 180° rotation, and circular shift. FIG. 28B is a bar chart illustrating a pooled datasetcomprising both original and augmented images, demonstrating a growth of the dataset resulting from augmentation in contrast with the original dataset size. FIG.
[0291] 28C illustrates splitting of the pooled dataset into training, validation and testing datasets before training a support vector machine, with the test dataset isolated from train-validate datasets to assess the final model performance. FIG. 28D illustrates cross-validation performance of the support vector machine on the original versus pooled datasets, emphasizing differences in individual class sensitivities after pooling. FIG. 28E illustrates a normalized confusion matrix for the 7-class classifier, showing the proportion of true versus predicted class for each prediction.
[0292] FIGS. 29A and 29B illustrates a comparison of performance across different machine learning models in an example embodiment. FIG. 29A is a schematic flowchart illustrating a workflow for model comparison. The workflow comprises labelling data, data augmentation, model training, model validation, and model comparison. FIG. 29B illustrates normalized confusion matrices for SVM, RFC, MLP and KNN models.
[0293] FIGS. 30A to 30C illustrate classification results generated by a SVM model in an example embodiment. FIG. 30A is a schematic flowchart illustrating an experimental workflow in which CAR T-cell and B-cell co-culture are processed to generate image data that is input into the SVM model for prediction on observed morphologies. FIG. 30B are box plots comparing the average model prediction for each sample type across the different classes. FIG. 30C are 8 x 8 grids of cell images for user adjudication. The SVM model prediction reveals distinct segmentation of morphologies for CAR T-cell and B-cell co-culture.
[0294] FIGS. 31 A to 31 C illustrate implementation of CTM based hardware gate to complement SVM model predictions for different cell morphologies in an example embodiment. FIG. 31 A illustrates a gated region projected onto the CTM region of interest representing sections indicating presence of highlighted morphology. FIG. 31 B are histograms illustrating actual predicted distributionsfrom SVM with trim lines indicating applied CTM gate. FIG. 31 C is a bar chart illustrating adjudicated model predictions for gated and ungated results benchmarked against original model test performance.
[0295] DETAILED DESCRIPTION OF FIGURES
[0296] Example embodiments of the disclosure will be better understood and readily apparent to one of ordinary skill in the art from the following discussions and if applicable, in conjunction with the figures. It should be appreciated that other modifications related to structural, material, electrical and optical changes may be made without deviating from the scope of the invention. Example embodiments are not necessarily mutually exclusive as some may be combined with one or more embodiments to form new exemplary embodiments. The example embodiments should not be construed as limiting the scope of the disclosure.
[0297] FIG. 1 is a schematic block diagram illustrating a microfluidic device / chip 100 for biophysical profiling of a sample in an example embodiment. The microfluidic device 100 comprises a first microfluidic channel 102 and a second microfluidic channel 104 configured to allow the sample comprising a plurality of cells to flow through; a first cell trajectory modulation region 106 disposed within the first microfluidic channel 102, the first cell trajectory modulation region 106 comprising a first array of microstructures, e g., pillars configured to sort the cells and provide one or more first modulated biophysical signatures comprising one or more first distribution profiles of the sorted cells; and a second cell trajectory modulation region 108 disposed within the second microfluidic channel 104, the second cell trajectory modulation region 108 comprising a second array of microstructures, e g., pillars configured to sort the cells and provide one or more second modulated biophysical signatures comprising one or more second distribution profiles of the sorted cells; wherein the first modulated biophysical signatures are substantially different from the second modulated biophysicalsignatures of the sorted cells. The direction of flow of fluid within the microfluidic channels 106, 108 is indicated by the block arrows. In the example embodiment, the microfluidic device 100 is a CTM device for performing a CTM assay as disclosed herein.
[0298] FIG. 2 is a schematic block diagram illustrating a system 200 for performing biophysical profiling of a sample in an example embodiment. The system 200 comprises a stage 202, e.g., mechanical stage, configured to hold a microfluidic chip / device, e.g., a microfluidic device as disclosed herein, containing the sample comprising a plurality of cells. The microfluidic device / chip is configured to modulate flow trajectories of the plurality of cells to obtain one or more modulated biophysical signatures comprising one or more distribution profiles of the plurality of cells. The system 200 further comprises an imaging apparatus 204 coupled to the stage 202, the imaging apparatus 204, e.g., a camera, configured to perform imaging of the plurality of cells in the sample when the microfluidic chip is held on the stage 202. The system 200 further comprises a liquid handling unit 206 configured to control fluid flow within the system 200. The liquid handling unit 206 may comprise one or more fluid delivery components such as a syringe, tubing, waste collection reservoir, and valves, for delivering a sample and one or more reagents, including buffer solutions or sheath fluids to the microfluidic device. The system 200 further comprises a pressure control unit 208 configured to regulate pressure and flow of liquid within the microfluidic device. The pressure control unit 208 is operatively coupled to the liquid handling unit 206 and is configured to drive fluid flow from the liquid handling unit 206 into and through the microfluidic chip under controlled conditions. For example, the pressure control unit 208 may comprise a stepper motor coupled to a syringe setup to facilitate withdrawal or compression of a syringe. The system 200 further comprises a processing unit 210 configured to: process the one or more modulated biophysical signatures using one or more transformation functions configured to map and integrate the one or more modulated biophysical signatures to obtain one or more biophysical features / CTM features; and input the one or more CTM features into a trained determination unit configured tocorrelate the one or more CTM features to one or more properties / attributes, e g., cell phenotype, activation state, killing potency, of the plurality of cells. The system 200 further comprises a user software interface 212 configured to enable user interaction and control of system operations. For example, a custom software graphical user interface (GUI) may be provided for interfacing with the system 200 and processing the sample in real-time for the processing of histograms data. In the example embodiment, the system 200 is a CTM assay system for performing a CTM assay as disclosed herein.
[0299] FIG. 3 is a schematic drawing illustrating a microfluidic device / chip 300 for biophysical profiling of a sample in an example embodiment. The microfluidic device 300 is constructed and functions substantially similarly to the microfluidic device 100 of FIG. 1. The microfluidic device 300 comprises a first microfluidic channel 302 and a second microfluidic channel 304 configured to allow the sample comprising a plurality of cells to flow through. The microfluidic device 300 further comprises a cell trajectory modulation region / segment 306 comprising a first cell trajectory modulation region disposed within the first microfluidic channel 302 and a second cell trajectory modulation region 308 disposed within the second microfluidic channel. The first cell trajectory modulation region comprises a first array of microstructures, e.g., pillars configured to sort the cells and provide one or more first modulated biophysical signatures comprising one or more first distribution profiles of the sorted cells. The second cell trajectory modulation region comprises a second array of microstructures, e.g., pillars configured to sort the cells and provide one or more second modulated biophysical signatures comprising one or more second distribution profiles of the sorted cells. The first modulated biophysical signatures are substantially different from the second modulated biophysical signatures of the sorted cells. Advantageously, the microfluidic device 300 is capable of generating two parallel processing of the same sample under different cell trajectory modulation conditions. Therefore, a plurality of modulated biophysical signatures may be obtained for the same sample, thereby providing a multi-dimensional representation of the biophysical properties of the cells.In the example embodiment, the microfluidic device 300 further comprises a detection region / segment 308 disposed downstream of the cell trajectory modulation region 306, the detection region 308 comprising a first detection region disposed downstream of the first cell trajectory modulation region and a second detection region disposed downstream of the second cell trajectory modulation region within the respective microfluidic channels. The detection region 308 is configured to receive cells exiting the cell trajectory modulation region 306 and to enable measurement of spatial distributions of the cells corresponding to modulated biophysical signatures.
[0300] In the example embodiment, the microfluidic device 300 further comprises a sample inlet 310 disposed upstream of the cell trajectory modulation region 306 within the respective microfluidic channels, the sample inlet 310 configured to introduce a sample comprising a plurality of cells into the microfluidic device 300. As illustrated in FIG. 3, a single sample inlet 310 is fluidically coupled to a branching flow junction configured to divide the sample stream into two portions which are then directed into the first and second microfluidic channels. This advantageously enables parallel processing of the same sample under different cell trajectory modulation conditions.
[0301] In the example embodiment, the microfluidic device 300 may optionally comprise a prefocusing region 312 disposed between the cell trajectory modulation region 306 and the sample inlet 310. The prefocusing region 312 comprises a first prefocusing region disposed upstream of the first cell trajectory modulation region and a second prefocusing region disposed upstream of the second cell trajectory modulation region within the respective microfluidic channels. The first prefocusing region comprises a first modified array of pillars and the second prefocusing region comprises a second modified array of pillars. The modified arrays of pillars in the prefocusing region 312 differ from the pillar arrays in the cell trajectory modulation region 306 in at least one of pillar spacing, pillar geometry and pillar arrangement. The prefocusing region 312 is configuredto perform a pre-separation of cells based on one or more biophysical properties, such as cell size or deformability. For example, the prefocusing region 312 may align cells into one or more streamlines, or separate cells into coarse groups prior to finer trajectory modulation in the downstream cell trajectory modulation region 306. Therefore, the prefocusing region 312 may advantageously improve consistency of cell entry into the CTM regions and reduce variability in cellmicrostructure interactions.
[0302] In the example embodiment, all the segments are fluidically arranged in a sequential manner and may optionally be fluidically arranged in a co-linear manner. The co-linear arrangement of segments may advantageously improve sensitivity of results.
[0303] FIG. 4 is a schematic diagram of a microfluidic channel of a microfluidic device in an example embodiment. Various embodiments of the microfluidic device may further comprise one or more sheath inlets fluidically coupled to the microfluidic channel and configured to introduce one or more sheath fluids into the microfluidic channels. As illustrated in FIG. 4, the microfluidic channel is configured to allow a sheath stream 1, sample stream 2, and sheath stream 3 to flow through. The sheath stream 1, sample stream 2, and sheath stream 3 respectively occupy a width Wi, W2 and W3 of the microfluidic channel having a width WT. The sheath streams 1 and 3 are introduced from opposite sides of the sample stream 2 such that the sample stream 2 is positioned (i.e. , sandwiched) between the sheath streams 1 and 3, thereby forming a layered flow configuration. The sheath streams 1 and 3 are configured to hydrodynamically confine, direct or focus the sample stream 2 within the microfluidic channel. The microfluidic channel further comprises a pre-focusing region / pre-CTM focusing region disposed upstream of the cell.
[0304] FIGS. 5A to 5E are schematic drawings illustrating different configurations of sample inlet(s), sheath inlet(s) and outlet(s) of a CTM microfluidic device for performing a CTM assay in an example embodiment.FIG. 5A illustrates a configuration for a single flow CTM assay with three inputs and one output. In other words, the microfluidic device comprises a single microfluidic channel. The single flow CTM assay of FIG. 5A comprises a sheath inlet 1, a sample inlet 2, and a sheath inlet 3 configured to introduce a sheath stream 1 , a sample stream 2, and a sheath stream 3 into the microfluidic channel of the microfluidic device. After passing through the microfluidic device, the sheath streams and sample stream are removed via a single outlet.
[0305] The sheath stream 1, sample stream 2, and sheath stream 2 are associated with respective hydraulic resistances R1 , R2 and R3 within the microfluidic channel. R1 to R3 may be approximated as proportional to a ratio of channel length to channel width, i.e., R ex where L represents length of the channel, w represents width of the channel with an assumed fixed height. The relative hydraulic resistance ratios between the sheath streams and the sample stream may be configured to fall within predetermined ranges. For example, where the resistance of the channel (R) satisfies the conditions (i) 0.33 < R3 / R2 < 49, and (ii) 2 < R1 / R2 < 94, these relationships result in a bounded parameter space as shown in Table 5 below.
[0306] Table 5: Ranges of stream widths based on 100% total width (WT) Stream width Lower Bound (%) Mid (%) Upper Bound (%) R1 / W1 60 49 94
[0307] R2 / W2 30 1 1
[0308] R3 / W3 10 49 5
[0309]
[0310] FIG. 5B illustrates a configuration for a double flow CTM assay with three inputs and one output. In other words, the microfluidic device comprises two microfluidic channels, i.e., a first microfluidic channel and a second microfluidic channel integrated on the same device. The double flow CTM assay of FIG. 5B comprises a sheath inlet 1, a sample inlet 2, and a sheath inlet 3. In the double flow CTM assay, the sheath inlet 1 is configured to introduce a sheath stream 1to the first and second microfluidic channels; the sample inlet 2 is configured to introduce a sample stream to the first and second microfluidic channels; and the sheath inlet 3 is configured to introduce a sheath stream 3 to the first and second microfluidic channels. After passing through the microfluidic device, the sheath streams and sample streams are removed via a single outlet. The equations and bounding conditions described with reference to FIG. 5A are also applicable for the double flow CTM assay of FIG. 5B.
[0311] FIG. 5C illustrates a configuration for a single flow CTM assay with three inputs and multiple outputs. The single flow CTM assay of FIG. 5C comprises a sheath inlet 1, a sample inlet 2, and a sheath inlet 3 configured to introduce a sheath stream 1, a sample stream 2, and a sheath stream 3 into the microfluidic channel of the microfluidic device. After passing through the microfluidic device, the sheath streams and sample stream are removed via the multiple outlets. The equations and bounding conditions described with reference to FIG. 5A are also applicable for the single flow CTM assay with multiple outlets of FIG. 5C.
[0312] FIG. 5D illustrates a configuration for a double flow CTM assay with three inputs and multiple outputs. The double flow CTM assay of FIG. 5D comprises a sheath inlet 1, a sample inlet 2, and a sheath inlet 3. In the double flow CTM assay, the sheath inlet 1 is configured to introduce a sheath stream 1 to the first and second microfluidic channels; the sample inlet 2 is configured to introduce a sample stream to the first and second microfluidic channels; and the sheath inlet 3 is configured to introduce a sheath stream 3 to the first and second microfluidic channels. After passing through the microfluidic device, the sheath streams and sample streams are removed via the multiple outlets. The equations and bounding conditions described with reference to FIG. 5A are also applicable for the double flow CTM assay with multiple outlets of FIG. 5D.
[0313] FIG. 5E illustrates a configuration for a double flow CTM assay with three inputs and multiple outputs. The configuration of FIG. 5E is substantially similar to the configuration of FIG. 5D and includes an added functionality of pre-focusingof samples within the CTM assay. The pre-focusing is implemented by a prefocusing region comprising a modified array of pillars (e.g., with modified spacing between pillars) for performing a pre-separation step prior to the actual CTM assay, which can improve signal measurements by enabling greater signal to noise ratio of detection within the cell trajectory modulation and detection regions.
[0314] FIG. 6 is a schematic drawing of a computer system 600 suitable for implementing an example embodiment. Different example embodiments can be implemented in the context of data structure, program modules, program and computer instructions executed in a computer implemented environment. A general-purpose computing environment is briefly disclosed herein. One or more example embodiments may be embodied in one or more computer systems, such as is schematically illustrated in FIG. 6.
[0315] One or more example embodiments may be implemented as software, such as a computer program being executed within a computer system 600, and instructing the computer system 600 to conduct a method of an example embodiment.
[0316] The computer system 600 comprises a computer unit 602, input modules such as a keyboard 604 and a pointing device 606 and a plurality of output devices such as a display 608, and printer 610. A user can interact with the computer unit 602 using the above devices. The pointing device can be implemented with a mouse, track ball, pen device or any similar device. One or more other input devices (not shown) such as a joystick, game pad, satellite dish, scanner, touch sensitive screen or the like can also be connected to the computer unit 602. The display 608 may include a cathode ray tube (CRT), liquid crystal display (LCD), field emission display (FED), plasma display or any other device that produces an image that is viewable by the user.
[0317] The computer unit 602 can be connected to a computer network 612 via a suitable transceiver device 614, to enable access to e.g. the Internet or othernetwork systems such as Local Area Network (LAN) or Wide Area Network (WAN) or a personal network. The network 612 can comprise a server, a router, a network personal computer, a peer device or other common network node, a wireless telephone or wireless personal digital assistant. Networking environments may be found in offices, enterprise-wide computer networks and home computer systems etc. The transceiver device 614 can be a modem / router unit located within or external to the computer unit 602, and may be any type of modem / router such as a cable modem or a satellite modem.
[0318] It will be appreciated that network connections shown are exemplary and other ways of establishing a communications link between computers can be used. The existence of any of various protocols, such as TCP / IP, Frame Relay, Ethernet, FTP, HTTP and the like, is presumed, and the computer unit 602 can be operated in a client-server configuration to permit a user to retrieve web pages from a web-based server. Furthermore, any of various web browsers can be used to display and manipulate data on web pages.
[0319] The computer unit 602 in the example comprises a processor 618, a Random Access Memory (RAM) 620 and a Read Only Memory (ROM) 622. The ROM 622 can be a system memory storing basic input / output system (BIOS) information. The RAM 620 can store one or more program modules such as operating systems, application programs and program data.
[0320] The computer unit 602 further comprises a number of Input / Output (I / O) interface units, for example I / O interface unit 624 to the display 608, and I / O interface unit 626 to the keyboard 604. The components of the computer unit 602 typically communicate and interface / couple connectedly via an interconnected system bus 628 and in a manner known to the person skilled in the relevant art. The bus 628 can be any of several types of bus structures including a memory bus or memory controller, a peripheral bus, and a local bus using any of a variety of bus architectures.It will be appreciated that other devices can also be connected to the system bus 628. For example, a universal serial bus (USB) interface can be used for coupling a video or digital camera to the system bus 628. An IEEE 1394 interface may be used to couple additional devices to the computer unit 602. Other manufacturer interfaces are also possible such as FireWire developed by Apple Computer and i. Link developed by Sony. Coupling of devices to the system bus 628 can also be via a parallel port, a game port, a PCI board or any other interface used to couple an input device to a computer. It will also be appreciated that, while the components are not shown in the figure, sound / audio can be recorded and reproduced with a microphone and a speaker. A sound card may be used to couple a microphone and a speaker to the system bus 628. It will be appreciated that several peripheral devices can be coupled to the system bus 628 via alternative interfaces simultaneously.
[0321] An application program can be supplied to the user of the computer system 600 being encoded / stored on a data storage medium such as a CD-ROM or flash memory carrier. The application program can be read using a corresponding data storage medium drive of a data storage device 630. The data storage medium is not limited to being portable and can include instances of being embedded in the computer unit 602. The data storage device 630 can comprise a hard disk interface unit and / or a removable memory interface unit (both not shown in detail) respectively coupling a hard disk drive and / or a removable memory drive to the system bus 628. This can enable reading / writing of data. Examples of removable memory drives include magnetic disk drives and optical disk drives. The drives and their associated computer-readable media, such as a floppy disk provide nonvolatile storage of computer readable instructions, data structures, program modules and other data for the computer unit 602. It will be appreciated that the computer unit 602 may include several of such drives. Furthermore, the computer unit 602 may include drives for interfacing with other types of computer readable media.The application program is read and controlled in its execution by the processor 618. Intermediate storage of program data may be accomplished using RAM 620. The method(s) of the example embodiments can be implemented as computer readable instructions, computer executable components, or software modules. One or more software modules may alternatively be used. These can include an executable program, a data link library, a configuration file, a database, a graphical image, a binary data file, a text data file, an object file, a source code file, or the like. When one or more computer processors execute one or more of the software modules, the software modules interact to cause one or more computer systems to perform according to the teachings herein.
[0322] The operation of the computer unit 602 can be controlled by a variety of different program modules. Examples of program modules are routines, programs, objects, components, data structures, libraries, etc. that perform particular tasks or implement particular abstract data types. The example embodiments may also be practiced with other computer system configurations, including handheld devices, multiprocessor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, personal digital assistants, mobile telephones and the like. Furthermore, the example embodiments may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a wireless or wired communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.
[0323] EXAMPLES
[0324] Study 1: Rapid Characterization of CAR T Cell Functional Microfluidic Cellular BiIntroduction:
[0325] Chimeric antigen receptor (CAR) T cell therapy has emerged as a breakthrough medicine for treating various hematological malignancies. It shows unprecedented success in clinical trials, leading to its rapid adoption and approval for clinical use. However, as CAR T cell therapy gains broader acceptance, challenges and gaps associated with consistent manufacturing have become apparent. Current autologous CAR T cells are extremely complex to produce, with each personalized batch of Food and Drug Administration (FDA) approved Axicaptagene Ciloleucel and Tisagenlecleucel products costing around $500,000. Unfortunately, 2 to 9% of these products are rejected due to manufacturing issues such as insufficient CAR T cells, low effector cell numbers or low cell viability. In some cases, a new batch of CAR T product is manufactured, resulting in an additional 3 to 4 weeks’ delay of the critical therapy.
[0326] The personalized nature of autologous CAR T cell manufacturing presents significant hurdles in managing scalability, cost-effectiveness, treatment efficacy and toxicity. Current CAR T manufacturing methods in bioreactors are akin to a “black box” where product functional phenotypes are not monitored until final product release tests. These single end-point measurements of cell quality attributes do not correlate well with clinical response. This inevitably results in CAR T cells that are manufactured with large variations in cell number, phenotypes, in vivo efficacy and potential side effects.
[0327] Critically, manufactured CAR T potency and functional tests are only performed before the addition of cryopreservation media and are not performed during the post-thaw, pre-product infusion. Thus, the CAR T cell attributes measured at the release test can be significantly altered at pre-infusion, potentially impacting in vivo efficacy. Both Tisagenlecleucel and Axicaptagene Ciloleucel products utilize IFN-y secretion as an indirect assay as part of potency assessment for product release. However, Novartis reported that IFN-y secretion varied greatly from batch to batch, suggesting that there may be significantuncontrolled variations in the safety, potency, and clinical effect of these products. These variations in measured quality are further confounded by downstream cellular changes during DMSO cryopreservation, shipping, and thawing, which potentially affect clinical efficacy. Coupled with inherent patient-to-patient heterogeneity, there is a strong need for new technologies with more specific and rapid cell functional phenotype measurements during, at the end of the manufacturing, and arguably at pre-infusion of CAR T cells, to effectively monitor product-specific variations that could affect the consistency and efficacy of cell therapy products.
[0328] In this study, a microfluidic T cell biophysical size and deformability profiling assay, called the cell trajectory modulation (CTM) assay, was developed to assess the quality of manufactured T cells, which shows strong correlations with immune phenotypes, activation, and CAR T potency. The CTM assay adopts specific micropillar structures engineered for precision cell interaction and measurement. The readout is based on the output of T cell trajectories corresponding to unique biophysical signatures. The CTM assay handles about 20 pL of unprocessed and unlabeled culture samples, using less than about 10,000 cells, and completes the profiling within less than about 10 mins, making it suitable for integration into the current CAR T manufacturing workflow, enabling rapid label-free measurements of the CAR T cells.
[0329] When benchmarked against flow cytometry cell profiling, the CTM assay revealed bioreactor-specific, donor-specific and cell processing (cryopreservation) variations of CAR T product. The nature of CAR T cells in manufacturing remains dynamic and heterogeneous, and insights into the temporal trends during manufacturing are currently not adopted as a critical quality attribute (CQA). The CTM assay offers a streamlined, label-free method for monitoring cell quality during CAR T cell production. By integrating this label-free assay into existing manufacturing workflows, the CTM assay provides timely insights into cell characteristics that may inform future process refinement and analytics. These findings highlight the potential of the CTM assay to deepenunderstanding of product variability and support efforts towards achieving greater consistency in cell therapy manufacturing.
[0330] Materials and Methods:
[0331] Microfluidic chip fabrication
[0332] The CTM microfluidic assay is performed using a single microfluidic chip (also referred to as a CTM chip / device) integrating two microfluidic devices with L and inverse-L micropillar structures (FIG. 7). In other words, the CTM chip comprises two microfluidic channels, each channel comprising a cell trajectory modulation region having an array of pillars (e g., L or inverse-L pillars) disposed therewithin. The CTM device is designed to have a dynamic range for measuring an approximated output size range from 6 to 12 pm with a single sample inlet and outlet. The specifications of the CTM device are further described in FIGS. 8A-8D and Table 6. The CTM device is composed of polydimethylsiloxane (PDMS) material (Sylgard 184, Dow Corning) bonded to a clear soda lime glass slide. The device is fabricated based on established soft photolithography means. Briefly, starting with computer-aided drawing, the designs are transferred to a glass chromed photomask (JD photodata), followed by using UV lithography means. An SU-8 coating of 20 microns is coated on the surface of a 4-inch wafer, baked and exposed to UV light using the chrome photomask. The final developed features on the wafer are coated with hydrophobic trichloro(1 , 1 ,2,2-perfluoroocytl)silane (Sigma, Singapore) using chemical vapor deposition. Finally, PDMS is mixed using a 1:10 curing ratio, degassed and poured onto the wafer to be baked at 80°C for 90 mins. The cured PDMS is peeled off the wafer and processed by cutting it into the correct device dimensions with reservoirs and tubing insert holes punched. The final device is made by bonding the fabricated PDMS on a clear glass slide using a plasma generator.
[0333] Table 6: Device segment specificationSegment Gradient -9
[0334] Segment N Length (Deg) (pm) 1 19.9 2.88 895 2 18.6 3.08 836 3 17.4 3.29 782 4 16.3 3.51 734 5 15.3 3.74 689 6 14.4 3.97 649 7 13.6 4.21 612 8 12.8 4.45 578 9 12.1 4.71 547 10 11.5 4.97 518 11 10.9 5.23 491 12 10.4 5.51 467 13 9.9 5.79 444 14 9.4 6.08 423 15 9.0 6.37 403 16 8.5 6.68 384 17 8.2 6.99 367 18 7.8 7.30 351 19 7.5 7.63 336 20 7.2 7.96 322 21 6.9 8.30 309 22 6.6 8.64 296 23 6.3 8.99 284 24 6.1 9.35 273 25 5.8 9.71 263 26 5.6 10.08 253 27 5.4 10.46 244 28 5.2 10.84 235 29 5.0 11.23 227 30 4.9 11.63 219
[0335]
[0336] 31 4.7 12.03 211
[0337] CTM assayThe CTM assay shown in FIG. 9A depicts a 4-condition process to modulate the trajectory of cell flow measured at the locating bins. At least 20 pL of T cell samples are directly loaded onto the sample loading reservoirs on the chip. The CTM assay can process undiluted cell concentrations at a minimum of 1 x 106cells / mL. Cell concentrations processed here range from 1 x 106cells / mL to 80 x 106cells / mL. Modulated biophysical signatures comprising distribution profiles / histogram distributions obtained under the four conditions are processed in accordance with the workflow illustrated in FIG. 9B. A custom CTM optical setup shown in FIGS. 10A and 10B uses a 4x Olympus infinite objective lens with a 2 mm field of view and is focused using a 60 mm lens to project the image to a machine vision camera (BFS-U3-16S2M, FLIR Blackfly2). The CTM assay is driven by negative pressure at the outlet with flow driving pressure for “1x” at -50 mBar while “10x” flow is driven at -500 mBar (see FIG. 10C). A custom software graphical user interface (GUI) is used to interface with the system and process the sample in real-time for the processing of Hi to HU histograms (FIG. 10D). The cell outlet trajectory positions are recorded on the camera to identify single cells (see FIGS. 10A and 10B) and capture the images at 15 FPS for Ixflow and 150 FPS for 10x flow. The time taken to record the locating bins at each outlet in FIG.
[0338] 9A for the 4 conditions is 4, 4, 1, and 1 minute to establish the modulated biophysical signatures in H1, H2, H3 and H4, respectively.
[0339] CTM data processing
[0340] To enable analysis to output the H1-4 biophysical signatures, video frames recorded from the camera are processed in buffered real-time data using Python scripts for background subtraction, blob detection and histogram bin tabulation for sequential conditions. The processed data is saved and post-processed to the CTM features using a secondary script to import the H1-4 biophysical signatures and process the transform functions resulting in 88 features / biophysical features. Machine classification of the features and phenotype data was performed using existing modules in Python, such as pandas, sci-kit learn and SciPy, for all data normalization, statistical analysis, correlation analysis, PCA data and hierarchicalclustering. All statistical analyses for sample mean testing were performed using a two-tailed Student’s t-test, while correlation analysis was evaluated with Pearson’s correlation. The data visualization and plots were performed using Matplotlib and Seaborn Python modules. All processed data were normalized on a logarithmic scale. The data of PCA models for CTM, flow cytometry and potency are tabulated in Table 7. Cytokine data is tabulated in Table 8.
[0341] Table 7: Principal component analysis for CTM, potency and flow cytometry data. The PCA component variance ratio percentage is shown at the top of the column for each principal component (PC).
[0342] CTM PCA Potency PCA Flow PCA component % component % component % 43.9% 12.4% 56.4% 22.9% 37.6% 20.2% Sample CTM CTM Potency Potency Flow Flow PC1 PC2 PC1 PC2 PC1 PC2 1 Breez - Donor 1 Run1 -1 .27 -1 32 -0 06 -0.41 -0.29 2 40 2 Breez - Donor 1 Run2 -1 .13 -0 97 -0 07 -0.41 -0.29 2 63 3 Breez - Donor 1 Run3 -0.86 -0 95 -0 04 -0.45 0 02 2 29 4 Breez - Donor 2 Run1 0 41 -0 35 -0 34 -0.1 1 0 42 -1 .58 5 Breez - Donor 2 Run2 -0.04 -0 46 -0 54 0.08 0 47 -1 .22 6 Breez - Donor 2 Run3 0 76 -0 48 -0 51 0.04 0 41 -0.60 7 Breez - Donor 3 Run1 0 36 -1 46 -0 21 -0.31 1 65 2 1 1 8 Breez - Donor 3 Run2 0 52 -0 98 -0 18 -0.37 1 26 2 17 9 Breez - Donor 3 Run3 0 51 -1 30 -0 20 -0.34 0 78 2 13 10 GRex - Donor 1 Run1 -2.17 0.36 0.74 -0.10 -2.42 1 25 11 GRex - Donor 1 Run2 -2.45 0.68 0.85 -0.12 -1.93 1 37 12 GRex - Donor 1 Run3 -2.37 0.30 0.91 -0.19 -1.82 1 87 13 GRex - Donor 2 Run1 -0.99 0.68 0.57 0.07 -1.34 -1 .35 14 GRex - Donor 2 Run2 -2.40 0.72 0.81 0.06 -1.15 -0.62 15 GRex - Donor 2 Run3 -1 .60 0.80 0.48 0.10 -1.15 -1 .40 16 GRex - Donor 3 Run1 -1 .77 0.14 0.46 -0.26 -1.02 1 43 17 GRex - Donor 3 Run2 -1 .69 -0 10 0.52 -0.22 -1.02 1 46 18 GRex - Donor 3 Run3 -1 .69 -0 07 0.47 -0.24 -0.80 1 75 19 Breez Replicate 1 1 85 0.01 0.00 0.02 2 54 -1 .20 20 Breez Replicate 2 1 39 0.46 0.05 -0.15 0 56 -0.93 21 Breez Replicate 3 2 86 -0 15 -0 59 0.19 0 08 -0.38 22 Breez Replicate 4 0 13 -0 66 -0 19 0.36 -0.75 -0.84 23 Breez Replicate 5 0 74 1.32 -0 05 0.37 -1.62 -1 .03 24 Breez Replicate 6 3 23 -0 25 -0 79 0.37 0 71 -0.68 25 GRex Replicate 1 -0.32 -0 14 0.21 -0.21 -3.41 -1 .16 26 GRex Replicate 2 -0.03 1.09 0.68 -0.05 -1.99 -1 .42 27 GRex Replicate 3 -0.39 0.08 0.42 0.55 -2.73 -1 .61 28 GRex Replicate 4 -1 .01 -0 60 0.03 0.30 -0.85 -1 .39 29 GRex Replicate 5 -0.74 0.01 0.75 1 .13 -1.95 -1 .11 30 GRex Replicate 6 -1 .24 -0 65 0.18 0.65 -2.84 -1 .58 31 Healthy D1 1 55 -0 43 -0 72 0.04 1 70 0 77 32 Healthy D2 2 13 0.13 -0 75 0.10 1 17 0 82 33 Healthy D3 0 57 0.66 -0 77 0.09 2 56 -0.03 34 Healthy D4 0 37 0.1 1 -0 06 -0.50 0 53 -0.28 35 Patient D1 2 37 -0 50 -0 81 0.13 4 08 -1 .12 36 Patient D2 1 84 1.40 -0 09 -0.13 5 52 -0.99
[0343]
[0344] 37 Patient D3 1 99 0.39 -0 82 0.16 2 68 -1 .39
[0345]
[0346] | 38 | Patient D4 | 0 60 | 2.48 | -0 35 | -0.23 | 2 22 | -0.53
[0347] Table 8: Average cytokine profiles in pg / mL for donor D1 , D2 and D3 for Breez and G-Rex
[0348] PD-L1
[0349] Class IFN-y GZB MIP-3b MIP-3a MIP-1a MIP-1 b CD40 GROa IP-10 GROb IL-8
[0350] B7-H1 Egf Breez D1 33585 4007.2 26 03 295.33 35.52 5070.4 1193 8 6635.4 57 44 42.91 106.79 1 1725 13.49 Breez D2 62748 5081.3 27 78 388.51 43.09 4636.8 24406 7708.7 61 39 78.02 328.45 23917 16.38 Breez D3 44583 8089.7 26 89 519.65 51.66 2141 .1 29261 7038.1 59 99 55.95 256.4 9086 6 14.95 GRex D1 82206 6866.1 27 74 299 70.05 3922.7 1839 2 6592.1 54 04 51.85 127.47 5140 9 15.37 GRex D2 112599 4040.6 31 49 550.66 98.31 5288 77864 7671.4 65 02 61.57 308.47 18919 16.14 GRex D3 265965 4834.9 35 49 11 15 8 123.16 4045.9 37388 8975.8 75 31 90.4 344.48 12105 19.69
[0351] GM- IL-12
[0352] Class Flt-3 G-CSF TNF-b IL-10 IL-13 IL-15 IL-17A IL-9 IL-1 b IL-1 ra IL-3
[0353] CSF p70
[0354] Breez D1 314.61 52.63 30329 662.26 14781 63 66 5912 8 9.66 5625 2220.3 7.15 124.13 12842 Breez D2 324.46 51.62 33856 472.85 26987 61 31 9013 9 10.57 1661 4 2150.7 9.48 117.75 10243 Breez D3 364.44 32.2 14678 180.9 48335 64 12 6104 1 9.64 4061 6 1939.5 8.87 477.47 4999.9 GRex D1 280.41 66.64 94743 989.13 22708 67 16 2363 9 8.95 873.28 1786.5 8.32 232.6 35068 GRex D2 340.83 83.37 117557 1108 3 29180 75.7 3603 2 10.86 640.12 2319.3 9.89 131.62 33426 GRex D3 316.84 83.64 107449 1263 7 101802 86 16 3597 8 12.82 6858 5 2483.6 10 54 131.37 44374
[0355] PDGF- RANTE
[0356] Class IL-22 IL-4 IL-5 IL-6 IL-7 TGF-a TNF-a TRAIL VEGF FGF AA S
[0357] Breez D1 50 86 721 .28 1480 444.98 8.16 156.49 1960 7 30.68 4026 8 62.15 20 29 42.92 Breez D2 54 17 2555.8 1910.6 288.57 8.79 157.98 10045 33.66 3241 6 68.8 35 53 45.87 Breez D3 50 53 303.97 1628.6 145.29 8.79 251.3 3398 4 31 .36 426.64 71.31 28 73 44.89 GRex D1 49 87 68.78 174 71 149.67 8.86 235.79 1768 1 33.39 9266 3 62.15 25 2 39.47 GRex D2 58 01 255.28 382 87 131.49 10.4 200.59 5224 9 38.87 13488 68.17 68 13 46.74 GRex D3 65 68 189.79 359 99 132.17 10.93 526.24 3692 5 40.86 14909 70.68 32 77 50.5
[0358]
[0359] Quantitative phase imaging (QPI)
[0360] Quantitative phase images were taken on a Leica microscope with incubator attachments for culture well plates to allow imaging of T cells in the culture wells for the longitudinal studies in FIG. 15. The QPI method and setup were implemented in accordance with previously established techniques. The T cell QPI images were acquired on DaysO, 2, 4, 6, 8 ,10, 12, and 14 to characterize the morphology of the cells. The raw QPI data shows strong correlations with CTM biophysical features (see FIG. 12).
[0361] T cell isolation from PBMCs
[0362] CD3+ T cells were isolated using EasySep Human T Cell Isolation Kit (STEMCELL Technologies). Activation using CD3 / CD28 Dynabeads (Thermo Fisher Scientific) were performed for 0 to 24 hours in 24-well plate cultures (Thermo Fisher Scientific) for naive control, CD3 / CD28 activation and 10% (v / v) DMSO experiments. CD3 / CD28 beads were removed prior to the CTM assay. T cell experiments in FIG. 15 were activated using ImmunoCult Human CD3 / CD28 T Cell Activator (STEMCELL Technologies). Cells were transferred into a 12-well glass-bottom plate for QPI imaging.
[0363] Cell lines
[0364] Lenti-X 293T Cell Line was purchased from Takara Bio Inc. and maintained in HyClone Dulbecco's Modified Eagle Medium (DMEM) with high glucose (Cytiva) supplemented with 10% Fetal Bovine Serum (Thermo Fisher Scientific) and 1% Penicillin-Streptomycin (Thermo Fisher Scientific). Jurkat, Clone E6-1 (TIB-152) and NALM6, clone G5 (CRL-3273) cell lines were purchased from ATCC and maintained in HyClone RPMI 1640 medium (Cytiva) supplemented with 10% Fetal Bovine Serum (Thermo Fisher Scientific) and 1% penicillin-streptomycin (Thermo Fisher Scientific).Plasmids
[0365] pMDLg / pRRE, pRSV-Rev, and pMD2.G packaging plasmids are Addgene plasmid # 12251, 12253, and 12259, respectively. A second-generation anti-CD19 CAR-IRESEGFP transfer plasmid, encoding a codon-optimized anti-CD19 CAR comprised of Mycepitope-tagged FMC63 single-chain variable fragments, lgG4 hinge, CD28 transmembrane domain, and human 4-1 BB and CD3zeta intracellular signaling domains, was generated in accordance with previously established techniques. Research-grade plasmid DNA was prepared by GenScript Biotech.
[0366] Lentiviral vector production and titration
[0367] Lentiviral vectors were generated by first transfecting the Lenti-X 293T Cell Line with the packaging plasmids and transfer plasmid complexed with Lipofectamine 3000 Transfection Reagent (Thermo Fisher Scientific). Medium was changed every 4-6 h after transfection, and lentiviral supernatants were collected 48-96 h after transfection. The supernatant was then centrifuged for 5 min at 300 g, 4°C, and filtered through a 0.45-pm low-protein-binding filter. The clarified supernatant was then ultracentrifuged for 1.5 h at 75,000g, 4°C. The lentiviral pellet was then resuspended in Opti-MEM Reduced Serum Medium (Thermo Fisher Scientific) at 4°C and stored in aliquots at -80°C. The concentrated lentiviral vectors were titrated on Jurkat cells. The percentages of GFP+ Jurkat cells were measured by flow cytometry two days post-transduction, and lentiviral titers were calculated and expressed in Tll / mL.
[0368] CAR manufacturing in bioreactors
[0369] Frozen human peripheral blood mononuclear cells (PBMCs) from 3 healthy donors were purchased from STEMCELL Technologies. For patient samples, PBMCs were obtained from discarded leukapheresis tubing sets (Spectra Optia Apheresis System, Terumo BCT) from four anonymous adultlymphoma patients participating in a CD19-targeted CAR T cell therapy clinical trial [SingHealth Centralized Institutional Review Board (CIRB) approval: #2018 / 2677 and #2022 / 2322], The PBMCs were isolated using Ficoll-Paque density gradient centrifugation (Cytiva) at 400g for 30 minutes, cryopreserved in CryoStor CS10 (STEMCELL Technologies), and later thawed for CAR T cell production. T cells were isolated using EasySep Human T Cell Isolation Kit (STEMCELL Technologies) and resuspended in AIM Medium (Thermo Fisher Scientific) supplemented with 2% human male AB serum (Merck Sigma-Aldrich) and 100 lU / mL recombinant human IL-2 (Miltenyi Biotec). The detailed protocols, preparation and operation of G-Rex and Breeze for CAR T manufacturing were implemented in accordance with previously established techniques. Briefly, for activation, purified T cells were mixed with DynaBeads Human T-Expander CD3 / CD28 (Thermo Fisher Scientific) at a 1:1 celkbead ratio and seeded into each gas-permeable well of a G-Rex 24-well plate (Wilson Wolf) and each microbioreactor cassette (Breez Biosystems, MilliporeSigma). One day after activation, lentiviral vectors were added to the wells or cassettes at a multiplicity of infection (MOI) of 5. T cells were expanded over 10 days for a total process duration of 12 days.
[0370] Potency assay
[0371] Cryopreserved end-of-production T cells were thawed, rested overnight (for experiments in FIGS. 16 and 19) or for 2 days, and co-cultured overnight with NALM6-Luciferase cells at various effector: target (E:T) ratios (CAR+ T cell: NALM6 cell) in RPM11640 medium (Cytiva) supplemented with 10% Fetal Bovine Serum (Thermo Fisher Scientific) and 1% Penicillin-Streptomycin (Thermo Fisher Scientific). For the luciferase luminescence-based cytotoxicity assay with NALM6-Luciferase cells, co-cultured cells were transferred to a white 96-well opaque-bottom plate. Bright-Glo Luciferase Assay reagent (Promega) was added to the wells, and luminescence signals were measured using an Infinite M200 Pro plate reader (Tecan).Cytokine profiling
[0372] Freshly produced T cells were co-cultured overnight with NALM6 cells at a 1:1 effector-to-target (E:T) ratio (CAR+ T cell:NALM6 cell) in AIM V Medium (Thermo Fisher Scientific), supplemented with 2% human male AB serum (Merck Sigma-Aldrich) and 100 lU / mL recombinant human IL-2 (Miltenyi Biotec). Cell culture supernatants were collected and stored at -80°C. Cytokine and chemokine secretion profiles were assessed using the Human XL Cytokine Luminex Performance Assay 46-plex Fixed Panel (R&D Systems) and analyzed on a Luminex FLEXMAP 3D system (Thermo Fisher Scientific). Due to high levels of Granzyme B and IFN-y exceeding the assay’s standard range, these cytokines were re-evaluated using the Human Granzyme B and IFN-y ProcartaPlex Kit (Thermo Fisher Scientific) on the same Luminex FLEXMAP 3D system.
[0373] Flow cytometry panels
[0374] For the experiments in FIGS. 16, 24 and 25, flow cytometry was performed using the following antibodies: LIVE / DEAD Fixable Violet (Thermo Fisher Scientific), BV510 anti-CD4 (clone OKT4), BV605 anti-CD45RA (clone HI100), BV650 anti-CD8a (clone RPA-T8), PerCP / Cy5.5 anti-LAG-3 (clone 11C3C65), PE anti-CCR7 (clone G043H7), PE / Dazzle 594 anti-CD127 (clone A019D5), PE / Cy7 anti-CD57 (clone HNK-1), AF700 anti-PD-1 (clone EH12.2H7), and APC / Fire 750 anti-TIM-3 (clone F38-2E2) (Biolegend). Cells were washed with PBS (Thermo Fisher Scientific) and stained by incubating with LIVE / DEAD Fixable Violet Stain for 20 minutes at room temperature. Following this, cells were washed with FACS Buffer [PBS containing 2% Fetal Bovine Serum (Thermo Fisher Scientific) and 0.1 % sodium azide (Merck Sigma-Aldrich)], then incubated with antibodies for 20 minutes at room temperature for staining. The cells were washed again with FACS Buffer before being analyzed on a CytoFLEX S flow cytometer (Beckman Coulter).
[0375] Results:CTM assay profiles T cell biophysical features
[0376] The CTM assay evolved from observations in cell sorting by deterministic lateral displacement (DLD) techniques, where for a given microfluidic pillar arrangement, sorting trajectories of biological particles can be resolved at 20 nm resolution. It was observed that cell sorting was significantly impacted at varying flow rates and with different DLD pillar structures. While these observations of variations are often deemed undesirable for sorting applications, this assay adapts these unique cell trajectories in differing conditions to characterize the T cell’s inherent biophysical properties, as shown in FIG. 9A. These CTM conditions result in unique trajectories of cells flowing within the device, and the interacting T cells would finally flow into the designated location bins within the microfluidic chip (FIG. 9A). The cumulative frequency histogram plots showing the modulated biophysical signatures result in the different histogram maps with Hi and H2 linked to size profiling at a slow flow rate, while H3 and H4 map the deformation profiles at higher flow rates. H1-4 denotes the 4 different conditions of the biophysical assay. The device used in this study for T cells has a designed cell trajectory specification for a size range from 6 to 16 pm. A high level of system integration and automation was achieved, as shown in FIGS. 7, 8A-8D 9A-9B and 10A-10D, a significant step in deploying and enabling direct sample profiling of cells from bioreactors.
[0377] The modulated biophysical signatures are processed in custom transformation functions (See Tables 2 to 4), with each function mapping and integrating the histogram array data into a single value corresponding to the T cell feature of size (S), deformability (D) or a combined (C) feature involving both size and deformability histograms. These biophysical assay features are finally fed into data classification models and statistical analysis to study the correlation to T cell activation, functional phenotype, and potency.
[0378] Distinct T cell biophysical states in CTM assayThe CTM assay was first experimentally evaluated to assess its ability to profile distinct physical morphology of T cells (n = 5) with known exposure to anti-CD3 / CD2824-hour activation or chemical triggers such as 10% (v / v) DMSO for an additional 15 mins post 24-hour culture (FIGS. 13A-13C). Exposure to DMSO is common in CAR T production due to the cryopreservation process, and it is known to cause mechanical changes to cells.
[0379] The unstimulated samples show sharper singular peaks for all H1-4 conditions while T cell size profiles (H1 and H2) increase distinctively during CD3 / CD28 bead activation with multiple peaks and an increase in histogram spread. Interestingly, exposure to DMSO for just 15 mins changes the biophysical properties of the cells with significantly reduced size and deformability differences across Hi to Fh. This suggests a decrease in cell deformability in high-flow conditions, which is expected upon exposure to DMSO. Changes in cell deformability are expected since these cryoprotectants interact with the cell membrane as well as intracellular structures. These DMSO concentrations are comparable to the concentrations spiked into CAR T products for storage and transportation. For example, Kymriah products add 7.5% (v / v) DMSO in the cryopreservation protocol and may be infused up to 30 mins post product thaw. Prolonged DMSO in in-vitro cell cultivation was shown to impact the properties of the T cell function, however in vivo studies have also reported to be superior to freshly produced CAR T cells.
[0380] The CTM features show very distinct profiles of these three groups in separate clusters via unsupervised hierarchical clustering, and the features correlation heatmap suggests cluster groups of similar features (FIGS. 13D and 13E). Selected CTM features also show significant differences among the three test groups (FIG. 13F). For example, CTM feature S16 shows low levels for unstimulated samples, while C18 shows an inverse relationship. The principal component analysis (PCA) in FIG. 13G shows clearly segregated clusters for allthree groups, which demonstrates proof of concept that biophysical properties can be used to sensitively probe changes in the T cell state.
[0381] Culturing T cells in bioreactors is a highly dynamic process with constantly changing media conditions, and autocrine and paracrine signaling, affecting the cell phenotype. A single end-point release test might fail to detect variance during CAR T manufacturing, which precludes any possibility of optimizing cell potency based on attributes measured during production. While the activated cells show clear differences at 24 hours using CD3 / CD28 beads, it is critical to detect cellular changes early in the activation cycle. These biophysical changes can be detected in CTM assay as early as 6 hours post-CD3 / CD28 activation (See FIG. 1 ).
[0382] Biophysical signatures in cell cultures comparable to flow cytometry forward scatter plots
[0383] T cell biophysical morphologies change dynamically over time. These changes can be tracked via flow cytometry or commercial microscopes. Here, the utility of CTM assay for rapid profiling of T cell temporal biophysical signatures under culture conditions was studied. CD3+ T cells were isolated from the peripheral blood mononuclear cells (PBMCs), activated, and cultured in a 12-well plate in AIM V media containing interleukin-2 (IL-2). The temporal cell dynamics were profiled every 2 days using flow cytometry, CTM biophysical assay, and quantitative phase imaging (QPI) for cell morphological characterization (FIGS.
[0384] 15A and 15B). It was shown that the cell biophysical features dynamically changed with multiple peaks across the CTM histogram in FIG. 15C. These dynamic cell changes are in line with known activation / expansion dynamics of CAR T production and were reflected in the CTM histograms.
[0385] Corresponding flow cytometry profiles were measured for T cells via forward scatter (FSC) plots, which are commonly used as a measure of cell size distribution. The FSC plots in FIG. 15E show similar histogram peaks and distributions compared to CTM features. The Pearson’s correlation comparingthe CTM H4 profile and FSC shows that H4 describes a deformed cell profile closely correlated to FSC cell profiles (FIG. 15D). The strong correlation to deformable H4 CTM histogram shows that CTM H4 measures a condition similar to high flow rates and shear experienced by immune cells in flow cytometry. These results demonstrate the use of biophysical size and deformability markers as a viable means to rapidly profile T cells in a label-free manner.
[0386] CAR T biophysical signatures distinguish CAR T from different bioreactors
[0387] Biophysical CTM profiling was studied to evaluate whether it can be used to detect any differences between CAR T cells manufactured using different bioreactors, particularly G-Rex gas permeable well plates and the Breez perfusion-based microbioreactor. G-Rex is one of the current FDA-approved culture platforms for CAR T cell manufacturing. The Breez perfusion-based microbioreactor was recently used for the manufacturing of CAR T cells, demonstrating comparable activation markers and CAR expressions to other methods of CAR-T cell manufacturing.
[0388] Briefly, the CAR T cell manufacturing protocol is performed via beadbased negative isolation of CD3+ cells from commercial frozen PBMC from 3 donors (FIG. 16A and Methods). Technical triplicates were performed for each donor and profiled using the biophysical assay. The T cells were transduced with lentivirus to produce anti-CD19 CAR T cells. In both bioreactors, the CAR T cells were expanded for 12 days, and the final products were profiled using CTM, flow cytometry and a potency cytotoxicity assay to measure the killing lysis of NALM6 CD19+ cells, which were later correlated with measured CTM biophysical attributes.
[0389] CAR T cells manufactured from the same donor samples manufactured using G-Rex and Breez yield distinctly different biophysical signatures, as shown in FIG. 16B. Unsupervised hierarchical clustering found distinct clusters in the CTM feature heatmap, showing clear regions of high feature expression for Breezbut low expression in G-Rex. This suggests that the manufacturing process may impact product attributes given the same or similar starting material, which is in line with other studies on the variation of CAR T in different bioreactors. The differences between CTM features of the two bioreactors were statistically meaningful, using a two-tailed independent t-test, with 21 out of 88 (23.9%) biophysical features showing significant differences (FIG. 16C).
[0390] Biophysical and phenotype distinction between bioreactors
[0391] A panel of T cell phenotypes was profiled (FIG. 17) with selected markers of flow cytometry shown in FIG. 16D. Donor-to-donor specific profiles were dominant in the flow cytometry markers, and distinct phenotypes were seen in the heatmap. 22.4% (34 / 152) of all flow cytometry markers showed significance between the CAR T cells of the two bioreactors (FIG. 16C). Particularly, the CAR % expression between both bioreactor modalities did not show a significant difference, while CD127+ and PD1+ expressions were significantly different between bioreactors.
[0392] A clear distinction can be seen in the unsupervised UMAP plots (FIG. 16E) showing two clusters’ groups for CTM profiles, separated by bioreactor types, with donor specific clusters. Similarly, flow cytometry UMAP plots showed donor specific clusters with phenotypic distinction between two bioreactors. The biophysical profiling distinguishes the products manufactured from two different bioreactors as well as donor specific CAR T profiles.
[0393] CAR T biophysical markers correlate with CAR T cell phenotypes
[0394] During CAR T cell manufacturing, T cells are cultured for up to 12 days, and often, the temporal dynamics of cell functional phenotypes (by flow cytometry) are not explicitly monitored. Here, an evaluation was conducted to determine if CTM markers were able to functionally correlate with manufactured CAR T cell phenotypes. Interestingly, there were strong correlations betweenCTM biophysical features and T cell phenotypic characteristics (FIG. 16F). Selected correlation plots critical in CAR T manufacturing, such as CD127+ %, CD4:CD8 ratios, and CAR %, were shown in FIG. 16G with strong Pearson’s correlation of -0.93, 0.85 and 0.79, respectively. This is the first instance of a CAR % correlation with a biophysical marker using the CTM assay.
[0395] CD127+ % and S12, which provide a measure of the average size of T cells, both showed significant differences and strongly correlated for CAR T cells manufactured in Breez and G-Rex. On the other hand, no difference in CAR expression was seen in both bioreactors, while correlations with CTM markers such as C1 , a combination of cell deformability and size feature, remain strong. A complete correlation heatmap showing 152 flow cytometry gatings and 88 CTM markers can be seen in FIG. 18. Biophysical profiles of CAR T cells show a correlation to potency.
[0396] CTM biophysical features correlate with CAR T potency and cytokine profiles
[0397] To functionally measure the CD19 targeting and killing potential of the CAR T cell product, the CAR T cells were challenged with target NALM6 cells to quantify the CAR T killing and lysis (FIG. 19A). Multiple CTM features were identified to correlate strongly with killing potency (FIG. 19B and Table 9). Three selected CTM features (S5, D22, and C5) were plotted against the 1:5 specific cell lysis %, showing strong linear correlations (FIG. 19C). 14 out of 88 CTM features (e.g., S5 H1 Std. Deviations) show strong correlations for both Breez and G-Rex CAR T cells (|paii| > 0.95) while CTM features such as (D22 and C5) show bioreactor specific correlations with different pBreez and pcrex suggesting functional biophysical correlations that are specific to the CAR T manufacturing platform (FIG. 19C). The CTM biophysical features show that the manufacturing protocols, in addition to donor-specificity, can impact the biophysical properties, cellular phenotypes, and corresponding functionality of CAR T cells.Following co-culture with target NALM6 cells, cytokine secretion of n=3 donors were also profiled for Breez (B1 ,2 and 3) and G-Rex (G1 ,2 and 3), and correlation analysis with CTM assay was performed (FIG. 19D and FIG. 20). Strong negative correlations (person value > 0) were seen for IL-3, TNF-|3, TNF-a and GM-CSF. The grouped correlation of IL-3, TNF and GM-CSF as known cofactors that regulate immune cell responses.
[0398] Strong positive correlations with CTM profiles were observed for IL-5 and IL-13. Both IL-5 and IL-13 have cellular pathways that are somewhat interrelated, as they both ultimately contribute to Th2 immune responses and have been shown to relate to certain pathophysiology, such as asthma. Minimal correlations were observed for IL-6, IFN-y, and GZB (granzyme b). The strongest correlation of GZB was observed in FIG. 19E with a correlation of 0.95 (p-value = 0.004). Collectively, CTM features show a strong correlation with CAR T cell function in potency and cytokine secretion profiling.
[0399] Table 9: Pearson’s correlation of CTM features with killing potency of CAR T cells manufactured from Breez and G-Rex
[0400] G-Rex
[0401] Label Pearson’s Breez Pearson’s
[0402] p_value p_value Pearson’s p_value D22 -0.5424 0.2662 1 0000 0 0042 0.9056 0.2788 C9 0 1929 0.7143 -0 9999 0 0090 -0 8997 0.2876 D24 -0.6039 0.2042 0 9999 0 0106 0.4287 0.7179 S3 0 5668 0.2408 -0 9998 0 0122 -0 9939 0.0701 C5 -0.2969 0.5677 0 9997 0 0143 0.9641 0.1711 D4 0 7568 0.0815 -0 9997 0 0158 -0 6041 0.5871 D16 0 3029 0.5595 -0 9995 0 0193 -0 7141 0.4937 D9 0 5734 0.2341 -0 9995 0 0203 -0 5745 0.6104 D25 -0.4527 0.3673 0 9995 0 0205 0.7051 0.5018 D17 0 5705 0.2371 -0 9961 0 0559 -0 6758 0.5276 S25 0 6497 0.1626 -0 9955 0 0604 -0 5452 0.6329 C7 -0.0892 0.8666 0 9950 0 0637 -0 0772 0.9508 S2 0 6422 0.1691 -0 9947 0 0658 -0 9977 0.0430 C17 0 7374 0.0944 0 9944 0 0675 0.6876 0.5173 C2 0 6044 0.2038 -0 9940 0 0697 -0 1 136 0.9275 C18 0 5737 0.2338 0 9929 0 0760 0.7252 0.4835 C1 0 9570 0.0027 0 9913 0 0839 0.7705 0.4400
[0403]
[0404] D5 0 5385 0.2703 -0 9896 0 0918 -0 6875 0.5174C22 0 8971 0.0153 0 9890 0 0945 0.4261 0.7198 S23 -0.1547 0.7698 0 9878 0 0997 0.8859 0.3071 C8 0 7966 0.0578 0 9873 0 1015 -0 9670 0.1640 S6 0 6191 0.1900 -0 9829 0 1 179 -0 9830 0.1176 D19 0 7032 0.1191 -0 9803 0 1267 -0 6936 0.5121 S9 0 6579 0.1555 -0 9799 0 1280 -0 5743 0.6105 S29 -0.5458 0.2626 0 9708 0 1542 0.7321 0.4770 D7 0 8304 0.0407 -0 9614 0 1774 -0 9013 0.2852 C19 0 0350 0.9476 0 9613 0 1777 0.9709 0.1539 D18 0 4357 0.3878 -0 9475 0 2073 -0 4732 0.6862 C25 -0.1242 0.8147 0 9390 0 2235 0.0009 0.9994 C20 0 5403 0.2684 0 9363 0 2284 0.9042 0.2810 C23 0 3484 0.4985 0 9332 0 2341 0.6078 0.5841 S27 0 5990 0.2090 -0 9221 0 2529 0 4538 0.7001 S20 0 7475 0.0876 0 9133 0 2670 0 9999 0.0064 S22 -0.4512 0.3691 0 9043 0 2808 0.9992 0.0247 D21 0 3861 0.4496 -0 8891 0 3027 0 7673 0.4432 S5 0 9598 0.0024 0 8825 0 3117 0.7896 0.4206 D14 0 4985 0.3142 0 8794 0 3159 0 8054 0.4039 D11 0 3346 0.5168 0 8684 0 3304 0 9055 0.2791 S21 -0.4042 0.4267 0 8631 0 3371 0.9706 0.1548 D1 0 6316 0.1786 0 8464 0 3575 0 9829 0.1180 C13 -0.4477 0.3733 0 8418 0 3629 0.8542 0.3481 D12 0 3713 0.4686 0 8366 0 3691 0 7663 0.4442 C6 0 2449 0.6400 0 8324 0 3739 0 9526 0.1969 C16 0 4949 0.3183 0 8130 0 3956 0.9571 0.1871 D15 0 4414 0.3809 0 7940 0 4160 0 9976 0.0437 D10 0 2013 0.7021 0 7931 0 4170 0 9995 0.0209 C21 0 8540 0.0304 0 7930 0 4170 0.8993 0.2882 D6 0 5281 0.2814 0 7841 0 4263 0 8645 0.3353 D26 -0.3586 0.4851 0 7737 0 4368 0.9993 0.0235 D13 0 4180 0.4095 0 7395 0 4701 0 9112 0.2704 C4 0 2918 0.5748 0 7184 0 4898 0 8274 0.3796 D23 -0.2953 0.5699 0 7059 0 5011 0.9930 0.0752 S7 -0.7191 0.1073 0 7057 0 5013 0 7340 0.4753 S19 0 9510 0.0035 0 7051 0 5018 0.8320 0.3744 D8 0 5701 0.2375 0 7048 0 5021 0 0605 0.9614 S30 0 4228 0.4035 0 6786 0 5252 0.9519 0.1984 S28 -0.2913 0.5754 0 6658 0 5362 0 9388 0.2238 D28 -0.3593 0.4842 0 6655 0 5364 0.9915 0.0828 C28 -0.0395 0.9408 0 6605 0 5407 0.2031 0.8698 S10 0 7583 0.0805 0 6234 0 5715 -0 3613 0.7647 D3 0 6022 0.2059 0 6200 0 5743 -0 6165 0.5771 C12 -0.5682 0.2394 0 5938 0 5953 -0 5838 0.6031 C26 0 8007 0.0556 0 5852 0 6020 0.2738 0.8234 S14 0 9060 0.0128 0 5339 0 6415 0.3732 0.7565 C3 0 2508 0.6316 0 5007 0 6662 -0 8556 0.3464 C14 0 0510 0.9236 0 4993 0 6672 0.9373 0.2266 S17 0 8169 0.0472 0 4981 0 6680 -0 0698 0.9555 C15 0 7144 0.1107 0 4939 0 6711 0.9640 0.1714
[0405]
[0406] S4 0 3645 0.4775 0 4348 0 7136 0.9920 0.0804S1 07088 0.1149 -04323 0 7154 -06045 0.5867 S13 08360 0.0382 0 4288 0 7178 0.6315 0.5649 S26 06507 0.1617 0 4155 0 7272 -0 5684 0.6151 D2 00783 0.8828 -0 3950 0 7415 -04285 0.7181 C10 -0.1979 0.7070 -0 3855 0 7481 1.0000 0.0008 C29 08127 0.0493 0 3703 0 7585 0.9976 0.0439 S11 08697 0.0244 0 3583 0 7667 -0 3103 0.7991 D27 -0.4829 0.3320 0 3388 0 7800 -0 9989 0.0303 S15 07555 0.0824 -0 3230 0 7906 0.5705 0.6135 C24 03945 0.4389 0 2997 0 8062 0.8367 0.3690 C27 07576 0.0810 -02691 0 8266 -06043 0.5869 C11 07946 0.0590 0 1985 0 8728 0.8710 0.3270 S12 08689 0.0246 0 1899 0 8784 0.3084 0.8004 D20 08275 0.0421 -0 0815 0 9480 -0 3094 0.7998 S8 -0.5485 0.2598 0 0729 0 9535 0.9947 0.0653 S24 06853 0.1330 0 0723 0 9539 -0 3668 0.7609 C30 -0.5906 0.2171 -0 0469 0 9701 0.9834 0.1162 S18 06147 0.1941 -0 0428 0 9728 -0 9297 0.2401
[0407]
[0408] S16 08408 0.0360 -0 0404 0 9743 0.7012 0.5053
[0409] Aggregated CAR T cell linear biophysical indexing
[0410] To enable comparison across all culture conditions and experiments, a biophysical index based on linear principal component analysis (PCA) was performed to reduce the 88 features into two principal components in PCA (FIGS.
[0411] 21A and 22A and 22B). The principal component 1 captures 44% of the aggregated CTM features variance and was used as the linear biophysical index (Table 7).
[0412] The CTM biophysical index showed a significant difference between mean ± SD for Breez (-0.083 ± 0.74) and G-Rex (-1.90 ± 0.46) manufactured CAR T cells, which was profiled on the final day of manufacture (FIG. 21 B). This data is in agreement with earlier analysis where individual CTM features showed statistically significant differences for CAR T cells in the n=3 donor each in a triplicate study shown in FIGS. 16B, 16D and 16F. To investigate the effects of donor CAR T heterogeneity during manufacturing, a single donor sample was used in n=6 independent batches of CAR T manufacturing runs for Breez and G-Rex each (FIG. 21 C). Each batch of the CAR T products was frozen in liquid nitrogen (LN2). Together, the samples were thawed, and the CTM biophysicalprofiling, flow cytometry and potency assay were performed for the donor replicates on the same day. The single donor replicates show that the variation between the same donor can be significantly large after 12 days of CAR T manufacturing for Breez (1.70 ± 1.10), while variations of G-Rex (-0.62 ± 0.42) remain comparable to FIG. 21 B CAR T cells from 3 donor triplicates (-1.90 ± 0.46). There is still a significant biophysical difference between cryopreserved CAR T products from Breez and G-Rex (FIG. 21 C).
[0413] CAR T cells from 4 cancer patients were manufactured in Breez and compared to 4 healthy donors (FIG. 21 D). Each CAR T batch was cryopreserved, and the CAR T cell characterizations were performed after resting for two days after thawing. The aggregated biophysical index showed that there is no significant difference between the patient (1.70 ± 0.66) and healthy donor (1.16 ± 0.72) CAR T cells manufactured in Breez. Interestingly, the patient CAR T cells have the same biophysical index as the same donor replicate Breez CAR T cells in FIG. 21C.
[0414] Collectively, in this study, all independent Breez (n = 23) and G-Rex (n = 15) CAR T samples showed significant biophysical profile differences of 0.91 ± 1.17 versus -1.39 ± 0.77, respectively. (FIG. 23). Principal component indexing performed on aggregated flow cytometry phenotype and potency assay show very similar indexing trends with the CTM biophysical index (FIG. 24 and Table 6). CTM index strongly correlates with CAR T potency and flow cytometry phenotype with Pearson’s correlations coefficient of -0.82 (p-value = 3.6 x 10-10) and 0.68 (p-value = 2.9 x 106), respectively (FIG. 25).
[0415] Discussion:
[0416] This study introduces a microfluidic assay (CTM assay) that offers a powerful means to rapidly profile the biophysical attributes of T cells, particularly CAR T cells, and correlate these attributes with functional phenotypes. The CTM assay addresses a critical need in CAR T cell therapy by enabling real-time, label-free monitoring of critical cellular characteristics during and at the end of the manufacturing process. While immune cell morphological features, such as size, deformability and shape, have shown to be strong indicators of cell activation, metabolic activity, cell cycle and disease pathology, these parameters are not yet adapted for direct assessment of autologous CAR T cells in manufacturing. This study represents the first-ever demonstration of using deformability characteristics of CAR T cells as correlative and potentially surrogate markers for critical quality attributes, including killing potency. Of note, the CTM assay detected distinct biophysical signatures between CAR T cells manufactured from two different culture platforms, Breez and G-Rex, despite using the same donors with the same transfection and activation modality. Using an aggregated biophysical index, comparison with single donor replicates and patient samples shows consistent biophysical profile characteristics for each bioreactor. This is consistent with studies that evaluated CAR T cells manufactured from different bioreactors and found significant phenotypic differences between CAR T cells manufactured using different bioreactors, albeit using potency functional assay and flow cytometry phenotyping. The results of the CTM assay suggest that the CAR T culture environment and manufacturing processes may affect product attributes, which in turn may impact CAR T in vivo efficacy.
[0417] Recent studies on morphological phenotypes of CD3 cells driving unique functional characteristics provide new evidence on potential biophysical correlation to cell states previously unknown. While discussions on cell deformability are lacking, it can be hypothesized with evidence from other groups on cell deformability that the immune cell architecture will influence the biophysical properties of size and deformability which determines function. In the case of CAR T cells for immunotherapy, biophysical changes due to variations in activation protocols may impact how T cells interact with the endothelium in different tissues, resulting in differences in in vivo cell distribution and trapping. One study showed that selected subpopulation activation showed different biophysical changes in naive T cells versus CD8+ T cells43. Another study investigated that a wide range of inflammatory conditions influence themechanical properties of immune cells, which can significantly impact their function independent of T cell surface maturation markers. Thus, the potency of CAR T cells to kill effectively depends on their ability to detect tumor chemokine signatures and on cell migration or extravasation to the tumor cells. The CTM assay specifically probes cell size and deformation of T cells, which may be indirectly related to the cell extravasation potential, Thus, a correlation between the CTM features and CAR T potency is surprising yet probable.
[0418] The mechanobiology of T cells is a nascent but growing field, where conventional techniques to probe the mechanical response of cells focus on static, single-cell responses in micro pipetting technique and atomic force microscopy. These techniques are manual and have low throughput; therefore, they are not adequate in cell manufacturing control where automation, speed, and consistency are required. Microfluidic technologies such as cell deformability cytometry and optical force cytometry offer high-content single-cell biophysical measurements albeit with the need for high magnification laser optical setup or costly high speed cameras; constriction pillar microfluidics are often easier to operate albeit with increased likelihood of clogging using gap sizes smaller than the target profiled cell; The use of viscoelastic fluids and impedance to probe cell deformability shows distinct in-vitro activated states of neutrophils. The CTM assay may involve using a sequential change of flow rates to induce cell deformation, and is capable of handling a large range of cell concentrations in native culture media and may use a 4x objective magnification optical setup with a standard machine vision camera.
[0419] In this study, the correlation between surface marker phenotyping of T cells was also investigated and it was found that certain biophysical features strongly correlated with CAR T cell phenotypes (e.g., CD4:CD8 ratio, p = 0.82 and CAR % expression, p = 0.82). T cell geometrical deformations directly induce nuclear morphological changes and alter gene expression through the nuclear factor of activated T cells and other signaling cascades. Thus, the biophysical and biochemical domains of T cells are complicated but clearly linked. In thecurrent manufacturing protocols of autologous CAR T cells, monitoring of cell phenotypes (CD4, CD8, Anti-CD19 CAR) are not performed daily, mainly because of the limited cell number produced, and the time and risk associated with the manual handling. Given the correlative potential, the label-free CTM assay only requires about 20 pL of sample and completes in about 10 mins, enabling potential use for perfusion and in-process monitoring of quality attributes of CAR T cells for the next generation of adaptive manufacturing.
[0420] Flow cytometry-based phenotyping between G-Rex and Breez-produced CAR-T cells showed significant, if subtle, differences (FIG. 17), which is believed to represent subtle potential functional and phenotypic differences. The CTM profiles are in agreement with this difference and the observed biophysical correlation and distinction of CAR T cells in both bioreactors would be a valuable area of future study. Insights from various ongoing research in biophysical deformability profiling measures the collective functionality and intracellular dynamics of cells can go beyond what flow cytometry surface markers can measure on their own.
[0421] Continuous testing during autologous CAR T manufacturing and wide testing at release is hampered by limited cell availability (e.g., Novartis T Charge with a total dose of >2.5 million CAR T cells). This necessitates the use of simple markers or surrogate assays for potency. Unfortunately, simple markers such as CD19 CAR expression levels and dose do not correlate well with clinical efficacy. There are no precise immunophenotypic profiles that are direct predictors of CAR T cell function. In vitro killing or surrogate assays are arguably the closest to validating in vivo CAR T potency. Existing killing assays typically require more than 10 hours of manual cell processing and handling, excluding sample preparation, making it not feasible for pre-infusion testing of autologous CAR T cell products. On the other hand, the biophysical label-free CTM assay, demonstrating potential correlation with function and in-vitro potency, emerges as a strong surrogate for in-process monitoring, requiring only a small volume (about 20 pL) with at least about 1000 T-cells for histogram distribution. Its rapidturnaround time is particularly advantageous for autologous products with li ited shelf life before patient administration, ensuring timely functional and potency testing. While there is a strong correlation (p > 0.82 shown in FIG. 19C and Table 9) between the CTM assay and in vitro killing assays, this does not suggest a correlation with in vivo efficacy, and downstream validations are in the works.
[0422] The CTM assay is a continuous flow and non -destructive cell profiling method and can be potentially integrated into the bioreactors for in-line Process Analytics Technology (PAT) or even as a closed-loop cell culture monitoring and control system. With a compact detection modality using commercially available machine vision cameras, low magnification bright field objectives and standard laptop processors, the CTM assay is adaptable, deployable and commercially viable when compared to current start-of-the-art cell deformability measurement tools utilizing ultra highspeed cameras and server specification computers for fast frame rate captures and image processing (FIG. 10).
[0423] This study introduces a microfluidic assay that offers a rapid and nondestructive means to profile the biophysical attributes of T cells, particularly CAR T cells, and to correlate these attributes with functional phenotypes. This assay provides a powerful, efficient means for predicting CAR T cell quality at critical stages of the manufacturing process, thereby potentially reducing batch failure rates and enhancing therapeutic consistency. Furthermore, this assay addresses a critical need in CAR T cell therapy by enabling real-time, label-free monitoring of critical cellular characteristics during and after the manufacturing process. The potential for such a system to rapidly measure functional phenotypes of CAR T cells when deployed closer to the bioreactors is invaluable for the growing field of CAR T bioprocessing, especially in adaptive manufacturing of next-generation CAR T products. Overall, this advancement supports the integration of real-time, biophysical profiling into adaptive CAR T cell production workflows, representing a significant improvement over existing phenotypic assessment methods and advancing efforts towards greater standardization of CAR T cell therapy manufacturing.Study 2: Label-Free Multichannel Cell Size and Shape Cytometry: Microfluidic Gating Approach for Enhanced Immune Cell Classification Introduction:
[0424] Cells exhibit remarkable complexity in function and structure, determined by a range of biophysical and biochemical phenomena. It is increasingly appreciated that biophysical features of cells, such as cell deformability, morphology, size and the like, are understudied yet potentially informative markers for cellular phenotypes and functions. Recent studies on T-cells suggest that the T-cell architecture is a probable marker for cellular response to antigens. Various groups have explored image-based techniques to profile cells through machine learning algorithms, model-based algorithms and computer vision. Many of these works implement micro / nano surface structures to observe how cells migrate and / or adhere to the surface, inducing changes to the cells through mechanical and spatial cues, which in turn affect cellular behaviors like migration. Cell biophysical properties, such as stiffness, deformability, and nuclear shape variability, have been shown to have strong correlations to metastatic potential, with changes in morphology (e.g., spherical vs. spread) influencing stem cell differentiation and cancer progression. Aging further alters these properties, offering insights into its impact on cellular function and health. These morphological shifts provide insights into the relationship between cell morphology, function and health. In recent years, various technologies have been developed that utilize image cytometry and label-free techniques to profile cells based on their morphology and biophysical properties. Microfluidic techniques, such as deformability cytometry (DC) and deterministic lateral displacement (DLD), are well-known label-free techniques to profile cell morphology and biophysical properties. Advances in high-throughput microfluidics have enabled vast data acquisition for biotechnology applications. Coupled with developments in deep learning, this integration has transformed and improved approaches to data analysis in cell profiling and the use of structured data.Deformability cytometry is a biophysical technique that measures the deformability of individual cells as they pass through a microfluidic channel, enabling high-throughput characterization of mechanical properties such as stiffness, elasticity, and viscosity. Various techniques are employed to manipulate and deform cells in specific regions of interest, including extrusion, hydrodynamic stretching, optical stretching, and acoustic stretching.
[0425] Deterministic lateral displacement (DLD) enables precise sorting of samples at the micro- and nanoscale. Under continuous flow, samples are directed through a structured array of geometric features and sorted based on size, shape, and deformability. Over the years, DLD has been modified and adapted to sort cells based on changes in cellular properties, including size, shape, and deformability.
[0426] State-of-the-art methods for analyzing cell morphology frequently integrate machine-learning techniques with imaging flow cytometry. These approaches often require high-specification cameras, meticulous sample preparation, and high-performance graphics processing units (GPUs). Machine learning enables the rapid analysis of large-scale, multi-parameter cellular images, extracting hundreds of morphological and spatial features per cell for classification and feature selection. For instance, time-stretch quantitative phase imaging (TS-QPI) combined with a microfluidic channel allows for rapid, label-free inference of cancer cells using raw waveform data. Unsupervised clustering models have been applied to efficiently process high-throughput cell images in imaging flow cytometry, while open-source platforms like CellProfiler streamline data processing for tasks such as label-free cell cycle analysis, reducing manual input and enhancing reproducibility. Automated cell classification on human white blood cells utilizing machine learning techniques and imaging flow cytometry was performed, showing the potential of label-free techniques in cell image classification tasks.However, these approaches face limitations regarding cost, scalability, and accessibility. Currently, no rapid, low-cost solution exists for cell morphology analysis using low-resolution imaging systems. Developing such a system would provide a low-cost and field-deployable alternative capable of capturing and analyzing data in real time. This would enable in-field or clinical deployment, offering real-time sample classification for point-of-care diagnostics. Additionally, it could be used as a process analytical technology, enabling real-time monitoring and control in closed-loop bioreactor systems.
[0427] In this study, the application of cellular morphology as a potential indicator of cellular function was explored by adopting a Cell Trajectory Modulation (CTM) technique. Using the characteristics of CTM microfluidic technology coupled with a custom low-cost microscope system, changes in the morphology of various cell types over time were captured and analyzed. This study proposed a classification model based on six distinct cell morphologies observed in immune cells, categorized by size (small, medium, large) and shape (doublets, blobs, and an irregular class for cells with abstract morphologies). It is believed that this system holds significant potential for enhancing diagnostic capabilities and improving understanding of cell behavior in health and disease.
[0428] Materials and Methods:
[0429] Microfluidic device fabrication
[0430] The Cell Trajectory Modulation (CTM) microfluidic device was fabricated using standard photolithography. Microfluidic channels were designed with computer-aided drawing and transferred to a glass chromed photomask (JD photodata). An SU-8 coating with an approximate thickness of 20 pm was coated on the surface of a 4-inch wafer, baked and exposed to UV light using the chrome photomask. The developed features on the wafer were coated with hydrophobic trichloro (1 ,1 ,2,2-perfluoroocytl) silane (Sigma, Singapore) using chemical vapor deposition. Polydimethylsiloxane (PDMS) (Dow Corning, Midland, Michigan) wasadded in a 1:10 ratio and poured onto the wafer and cured at 80°C for 1 hour for crosslinking. Finally, the PDMS was peeled out of the master mold and cut to the dimensions of the chip before using oxygen plasma surface activation and bonding to bond with a glass slide.
[0431] Imaging setup
[0432] Image data was collected using a low-cost imaging system with a camera (BFS-U3-16S2M, FLIR Blackfly2). A fixed objective lens, 2 x 0.15 NA (Edmund Optics), was attached to the camera and captured with the Spinnaker Software Development Kit (Teledyne Vision Solutions). The captured image frames were stored in “ mp4” file format using OpenCV 4.6.0. With the implementation of a custom Python code, single-cell images were extracted from the video file and stored in a data frame. FIGS. 26C and 26D show samples of the extracted singlecell images; each image is 20 x 20 pixels in resolution with a scale of 0.738 pm per pixel.
[0433] Experiment setup
[0434] FIG. 26A shows the full experimental process, cell culture, cell profiling and data collection, image classification, machine learning training, machine learning predictions and result analysis. T-cells were cultured within a bioreactor, and a small volume (20 pL) was sampled and channeled through a CTM microfluidic device. A frequency distribution may be obtained based on the profiled size and deformability of cells flowing through the assay with high precision. By deconvolving the distribution and taking single-cell captures of the cell population, further granularity can be achieved and the distribution can be broken down into 6 distinct categories, with each cell morphology contributing to a segment of the main distribution.
[0435] Blob detection was applied to the identified frames using OpenCV. Blob detection is an important step for identifying and analyzing regions within theimage that represent a cell flowing through the CTM. By leveraging OpenCV's algorithms, cell images were isolated from the pillar features and cropped into 20 x 20-pixel images, centering the cell within the pixel frame. The images were labeled by an experienced engineer into 6 distinct morphologies with the assistance of a custom Graphics User Interface (GUI) shown in FIG. 27A. The number of labeled images amounted to -22,000. Single-cell images with spherical morphology were categorized as “size” with 3 distinct sizes. Images where the cells are less than 7 pm were considered small-sized cells, images where the cells are greater than 12 pm were considered large cells and cells that fall between the range were classified as medium (FIG. 27B for size reference). Cells that appear to be grouped, adhered together or had non-spherical morphologies were categorized as “shapes” with 3 distinct shapes. FIG. 27C shows sample images interpolated with a scale bar for reference.
[0436] Dataset enhancement and image processing
[0437] The dataset was enhanced using data augmentation techniques due to imbalances in the data distribution, with certain morphologies having lower cell images (FIG. 28A). Typical dataset enhancement techniques for imaging include flipping, rotation, circular shift, scale transformation, noise application, and the like. The images captured by the camera were stored as greyscale images. The augmentation techniques implemented were limited to flipping, rotation, and circular shift to enhance the dataset while preserving image properties and features distinct from the specific class. With considerations to maintain the properties of cells and ensure that the augmented images do not deviate vastly from the original images, a modified version of circular shift was applied. This was achieved by splitting a part of the image (2 columns x 20 rows) of the leftmost portion of the image and shifting it towards the right. This ensures that the cell image is not cropped and allows for potential channel features to be replicated to the right of the image, creating a “new” image with a cell located on the left of a channel feature. The images were enhanced based on classes with low image counts, while classes with high image counts were not augmented. A secondaryconsideration for not augmenting the classes with high counts was the high abundance of images that were classified within the class. Horizontal flipping, vertical flipping, 180° rotation and circular shift (see FIG. 28A) were applied to large and blob classes to increase the number of images by a factor of 5, as shown in FIG. 28B. Doublets, irregular and small classes were horizontally flipped only to increase the number of images by a factor of 2. Once all images were augmented, under-sampling was performed to ensure that the number of images was balanced, resulting in 6050 images per class, as shown in FIG. 28B.
[0438] Machine learning model development
[0439] A support vector machine (SVM) algorithm was implemented for the classification of the six morphological classes. Support vector machines are supervised machine learning algorithms that classify data by finding the optimal hyperplane that separates different classes while maximizing the margin between them. SVMs can perform complex classification boundaries on non-linear data. An SVM model was chosen due to its ability to perform well with small datasets, fast training process time and low computational cost. SVM is a general classifier that relies on mathematical operations, allowing rapid operations with low computational costs for downstream applications. With the enhanced dataset, data preparation on the cell images was done by unraveling the capture images (20 x 20-pixels) into a 1 -dimensional array of 400 features. The array is trained within the SVM, with each value representing a different pixel feature of the actual cell image. The dataset, including processed images, was shuffled and split into three subsets using a 60-20-20 ratio. The largest subset was used for training, while the validation and test sets represented the remaining portions. The training and validation sets were employed for cross-validation and hyperparameter tuning. The test set was isolated and used only for a final, blind evaluation to assess the SVM model’s ability to generalize to new, unseen data. The Scikit-learn software package was employed for model training, optimization, and switching between different models. Its extensive library of pre-built model frameworks facilitated rapid experimentation under various conditions,streamlining the process of hyperparameter tuning and reducing optimization time.
[0440] Performance metrics
[0441] The performance metrics used to measure performance are broadly categorized, with the corresponding formulas provided. A confusion matrix (n x n array of values) captures the relationship between the predictions and ground truth. True positives represent correct predictions for a specific class, while true negatives are correct predictions for all other classes. False positives occur when a specific class is incorrectly predicted, and the true class is different. False negatives arise when the model fails to predict the correct class, predicting another class instead.
[0442] Macro-averaging was applied to precision and recall, calculating the arithmetic mean precision and recalls across the different classes and providing a macro view of model performance for comparative purposes.
[0443] General Performance Metrics:
[0444] Correct Predictions
[0445] Accuracy = — — — — - — : - Ail Predictions
[0446] True Positive
[0447] Precision =
[0448] True Positive + False Positive
[0449] True Positive
[0450] Recall - True Positive + False Negative
[0451] Precision x Recall
[0452] Fl — Score = 2 X
[0453] Precision + Recall
[0454] Macro-average Metrics:
[0455] PrecisionMacro-average PrecisionLabel A+ PrecisionLabel B+ ■■■ PrecisionLabel NN
[0456] Rccallbabel A "h ^^^-^lllLabel B "h ^^^^^Label N KecaiiMacro-average—
[0457]
[0458] N Where:N = Total Number of Samples Predicted
[0459] Data analysis
[0460] All post-processing of data in this manuscript was performed on Python 3.8.10 using NumPy 1.23.5, pandas 1.5.2, scikit-learn 1.3.2, OpenCV 4.6.0 and matplotlib 3.6.2. Statistical analyses were performed using a two-tailed student’s t-test, and plots were generated in matplotlib. The processed data images are provided in the source data.
[0461] Results:
[0462] Machine learning modelling and performance
[0463] With the single cell imaging and model development and creation of the dataset as discussed above, a Support Vector Machine (SVM) was implemented using Python 3.8 with scikit-learn 1.3.2 on a 64-bit Windows system, powered by an 11thGen Intel® Core™ i7-1165G7 processor, without Graphics Processing Units (GPU) acceleration. To evaluate the performance of the SVM, performance metrics such as recall, precision, accuracy, Matthew’s correlation coefficient, F1-score and a confusion matrix were employed. Training and prediction times were recorded during each run and implemented as an auxiliary metric.
[0464] To quantify the impact that augmenting the dataset had on the SVM, two models were developed: one using data augmentation and the other without. The sensitivity of each class was compared across the two models, showing significant improvements for the model trained using augmented data, particularly in the “Large” and “Doublets” categories, which saw improvements of +25.7% and +22.2%, respectively (See FIG. 28D). The augmented dataset provided greater diversity, enhancing the model’s generalization capabilities and making more accurate predictions. Although minor declines in performance were observed in the “Medium", “Irregular”, and “Blobs" categories, the overallimprovements were consistent with the study’s objectives. Additionally, comparing the average difference in sensitivity, an increase from 77.1% to 84.0% was observed, leading to a ~7% boost in prediction ability on the balanced augmented dataset.
[0465] Analyzing the Support Vector Machine (SVM) performance from the cross-validation steps, an average cross-validated accuracy of 84.4% was observed, with the macro-averaged precision and recall being at 84.2% and 84.2%, respectively. Training the model on 60% of dataset images (21,780 images) required only 24.99 seconds, while the average prediction time across cross-validation folds was 18.61 seconds for 20% of dataset images (7,260 images) per fold. All analyses were performed on an 11thGen Intel® Core™ i7-1165G7 processor, without Graphics Processing Units (GPU) acceleration.
[0466] The confusion matrix in FIG. 28E provides further insight into the performance of the SVM model, with the diagonal and off-diagonal values representing the normalized values for correct classifications and misclassifications. The model is shown to be able to predict each class accurately, with the percentage of correct predictions per class being at 87.0%, 80.0%, 86.0%, 84.0%, 83.0%, and 84.0% (small, medium, large, doublets, blobs, irregulars). The amount of misclassification values is also relatively small, with the highest values being 12.0% and 9.0%. The classifications also show consistent performance across the various classes with no bias in classification. From the confusion matrix, the model distinguishes between the small class and the large / doublets / blobs class while showing clear distinctions between the medium class and the doublets / blobs class with no misclassifications between those combinations of predictions.
[0467] Performance comparison of machine learning models
[0468] For comparative evaluation, 4 models Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest Classifier (RFC), and MultilayerPerceptron (MLP) were implemented. FIGS. 29A and 29B illustrate the performance across the different models, showing that SVM is superior in terms of performance when comparing precision and recall. Most notably, the training and prediction times of the SVM is considerably slower when placed against the other models. However, the higher macro-averaged precision and recall values prompted the decision to select the SVM as the model of choice. The low standard deviation across the different cross-validation steps further indicates that the SVM performance values are closely tied to the mean values reported.
[0469] Deployment of model for combinatorial assay imaging
[0470] To evaluate the trained SVM model, an experiment was performed from combinatorial assays involving CD19-positive Chimeric Antigen Receptor (CAR) T-cells and NALM6 cells. The experiment consists of 3 different sample types: CAR T-cells, NALM6 cells, and CAR T-Cells with NALM6 cells in a ratio of 1:2 cells, respectively. Each sample was processed through the CTM device, and single-cell images were extracted and pooled together for morphology prediction by the SVM model.
[0471] The generated predictions were consolidated, with the images from each predicted class pooled together and displayed for visual analysis. This approach allowed for a clearer assessment of the model's classification accuracy across the different morphological classes. FIGS. 30B and 30C illustrate the differences observed in predicting different morphologies for each sample type. From the plots, it was observed that the model prediction is closely related to the expected distribution from the various samples. With activated and inactivated T-cells, in general, being around 5 - 10 pm in diameter and NALM6 B-cells / cancer cells being around 8 - 10 pm, the distribution of cells would fall within the range of small / medium classifications for CAR T-cells. In contrast, NALM6 cells would fall within the range of medium / large classifications. With the co-culture of the cells, a reduction in the small I medium / large classifications was observed, which suggests a change in sample morphology because of cell-cell interactions. Tofurther analyze the model's performance, 3 different users were tasked to adjudicate 3 samples of predicted image data. FIGS. 30D and 30E shows a subsection of user-adjudicated data, with highlights around the image representing approved and rejected predictions within the class. The adjudication was done independently of each other, and an average of the 3 adjudicated datasets was taken for each class. Distinct morphological differences in sample types are observable, supporting the robustness of the model’s ability to differentiate between different cell populations and generate predictions based on cell morphology.
[0472] CTM-based dating enhancements to model performance
[0473] Following the results in the previous section, the profiling capabilities were utilized to complement the model predictions and improve classification performance. Gates were created based on the expected position of the cells within the different channels based on the classification type. By applying the gates after the prediction is made, the predictions were limited to the regions where a specific cell morphology is most concentrated, reducing the number of misclassified images. FIG. 31 A shows the gates for each of the 6 classes and the channels they fall within based on their CTM profiles. Decoupling the CTM distribution into the individual class distributions and applying the gates, it was observed that the peaks of the distribution are captured within the gates that were selected (see FIG. 31 B). The gates filter images by removing cell images from regions that do not correspond to the actual size of the cell. From the plots, it was observed that a large population of cells that were labeled as blobs but have apparent sizes of <12 pm. The size of blobs generally falls in the range of 11 pm - 13 pm, which suggests that there are misclassified images in the prediction of blobs. Data adjudication was performed again with the 3 users with a grid of images from the gated predictions. The average accuracy of the adjudicated predictions is visualized in FIG. 31 C. It was observed that the number of accurate predictions increased for most classes. A notable observation would be the high variance in the validation of blob images by the 3 users, which could be a resultof a different criterion for validating the images. Using the model test accuracy as a benchmark, a drop in performance for the average accuracy of the adjudicated gated and ungated predictions was observed.
[0474] Discussion:
[0475] In this study, a low-resolution image-based SVM model designed for rapid, label-free classification of cell morphologies was developed. Using 20 x 20-pixel resolution images, the model successfully categorized cells based on 6 morphological properties related to shape and size, achieving an accuracy of 84%. Through the inherent sensitivity of CTM test on cell size, the model was able to achieve a 75% adjudicated accuracy on an unseen dataset.
[0476] Through an exploration of existing image cytometry techniques and tools, this study demonstrates that the disclosed technique can achieve high prediction accuracy with a low sample volume and low-cost camera set up with a high throughput of cell detection. Most current products utilize high-speed cameras or lasers to perform imaging, resulting in either high cost and / or large form factor for their set up.
[0477] The variation within images, particularly in the 'Doublets' and 'Blobs' classes, presents a challenge against the generalization ability of the model. Unlike single cells, which are typically spherical and exhibit relatively low variability, doublets and blobs are morphologically diverse, complicating feature identification. The diversity is exacerbated by the scarcity of images from these classes, with fewer examples of 'Blobs' and 'Doublets', limiting effective learning and generalization by the model. From the image data collected, the ‘Blobs’ class often has aggregation of cells, these features could be a result of sample handling issues or culture conditions. Samples that are cultured provide a different profile of cell morphologies when compared to thawed samples. These differences tend to increase the amount of ‘Blobs’ prediction by the model and could be attributed to stresses on the cells during cryopreservation and thawing.A key challenge during the training step was the SVM model sensitivity to image brightness and contrast. This resulted in single cell images with a lower brightness and contrast to be misclassified as ‘Blobs’, resulting in reduced accuracy during initial blind tests. By augmenting the alpha and beta values, the images were processed to a similar median pixel value to match the original training data for prediction.
[0478] CTM data may be integrated with the training of machine learning models to optimize segmentation and feature extraction. Increasing the amount of image data to include a larger diversity of data points and exploring neural networks may result in better model performance and generalization. Inclusion of new sample types such as whole blood or stem cells may enable the development and training of a model with potential clinical applications.
[0479] Conclusion:
[0480] This study demonstrates that even with a low-cost, low-resolution setup paired with a CTM microfluidic device, high accuracy in cell prediction based on morphology is achievable. The potential application of such a model extends towards identifying cell viability or efficacy through the prediction of morphology and correlating these changes with biochemical results and the current state of art for biophysical measurements. With a low sample volume requirement, it brings potential for the technique to be integrated into existing bioprocessing pipelines for online and inline measurement of critical quality attributes during cell culture. The results show a rapid, low-cost deployment for image classification in a laboratory setting, providing real-time information on cell cultures and potentially clinical samples.
[0481] APPLICATIONSEmbodiments disclosed herein provide a microfluidic device and a method of performing biophysical profiling of a sample, and a method of classifying cells.
[0482] Advantageously, the microfluidic device and methods disclosed herein enable biophysical profiling of cells based on trajectory modulation within a microfluidic environment. This label-free approach offers an advantage over conventional techniques that rely on chemical labeling or complex preparation steps and require complex and expensive equipment such as flow cytometry systems, confocal microscopes and spectrophotometry systems, and other specialized platforms for mechanical characterization, including scanning electron microscopes, micropipette systems, optical force microscopes, and microfluidic deformability cytometry systems, which are typically costly (e.g., exceeding $200,000) and difficult to deploy in industry or clinical settings. Due to the limitations of such conventional systems, there is a lack of accessible tools and methodologies for rapidly quantifying, modeling, and utilizing biomechanical information of cells for profiling and decision-making. The microfluidic device, and methods disclosed herein address this gap by providing a rapid, scalable, and commercially viable approach for cell biophysical profiling, thereby facilitating broader adoption in clinical and biomanufacturing settings.
[0483] Advantageously, the microfluidic device and methods disclosed herein integrate multiple cell trajectory modulation conditions within a single device, e g., through different pillar configurations and / or flow conditions, thereby enabling acquisition of multiple complementary biophysical signatures from the same sample. Such biophysical profiling may serve as a surrogate indicator of cellular phenotype and / or function, enabling characterization that was previously difficult or not well established using conventional techniques.
[0484] Advantageously, the microfluidic device and methods disclosed herein utilize transformation functions and trained determination units comprising machine learning models to enable efficient processing and interpretation of the modulated biophysical signatures. Embodiments of the microfluidic device andmethods disclosed herein are capable of converting raw distribution profiles into quantitative biophysical / CTM features and correlating these features with cellular properties, thereby allowing rapid and automated analysis, which may be particularly advantageous in applications such as cell therapy manufacturing, quality control, and diagnostics.
[0485] Advantageously, the microfluidic device and methods disclosed herein provide a rapid, label-free approach for classifying cells based on morphological characteristics. The disclosed method enables efficient acquisition of structured image data without the need for complex sample preparation or labeling. The disclosed method may utilize a low-cost imaging apparatus in combination with automated image processing and classification for high-throughput analysis with reduced computational and hardware requirements. In particular, the disclosed embodiments enable rapid single-cell imaging within a compact and deployable setup, in contrast to conventional high-speed imaging cytometry systems that are typically large, complex, and not readily deployable.
[0486] It will be appreciated by a person skilled in the art that other variations and / or modifications may be made to the embodiments disclosed herein without departing from the spirit or scope of the disclosure as broadly described. For example, in the description herein, features of different exemplary embodiments may be mixed, combined, interchanged, incorporated, adopted, modified, included etc. or the like across different exemplary embodiments. The present embodiments are, therefore, to be considered in all respects to be illustrative and not restrictive.
Claims
1. CLAIMS1. A microfluidic device for biophysical profiling of a sample, the microfluidic device comprising,a first microfluidic channel and a second microfluidic channel configured to allow the sample comprising a plurality of cells to flow through;a first cell trajectory modulation region disposed within the first microfluidic channel, the first cell trajectory modulation region comprising a first array of pillars configured to sort the cells and provide one or more first modulated biophysical signatures comprising one or more first distribution profiles of the sorted cells; anda second cell trajectory modulation region disposed within the second microfluidic channel, the second cell trajectory modulation region comprising a second array of pillars configured to sort the cells and provide one or more second modulated biophysical signatures comprising one or more second distribution profiles of the sorted cells;wherein the first modulated biophysical signatures are substantially different from the second modulated biophysical signatures of the sorted cells.
2. The microfluidic device according to claim 1, wherein the first microfluidic channel and the second microfluidic channel are fluidically independent of each other such that flow through one of the microfluidic channels does not pass through the other microfluidic channel.
3. The microfluidic device according to claim 1 or 2, wherein the first array of pillars comprises a first pillar configuration; and the second array of pillars comprises a second pillar configuration; wherein the first and second pillar configurations differ in at least one of pillar dimension, pillar shape, pillar structure, pillar arrangement or pillar orientation, with respect to the direction of flow of cells.
4. The microfluidic device according to any one of claims 1 to 3, further comprising,a first detection region disposed downstream of the first cell trajectory modulation region, the first detection region comprising a first set of location bins for allowing the cells that are sorted by the first array of pillars to flow into or flow through; anda second detection region disposed downstream of the second cell trajectory modulation region, the second detection region comprising a second set of location bins for allowing the cells that are sorted by the second array of pillars to flow into or flow through.
5. The microfluidic device according to any one of claims 1 to 4, further comprising,a first prefocusing region disposed upstream of the first cell trajectory modulation region, the first prefocusing region comprising a first modified array of pillars; and / ora second prefocusing region disposed upstream of the second cell trajectory modulation region, the second prefocusing region comprising a second modified array of pillars;wherein the first and / or second prefocusing regions are configured to preseparate the cells in the sample prior to passing through the first and / or second cell trajectory modulation regions; andwherein the spacing between the first and / or second modified arrays of pillars in the first and / or second prefocusing regions are different from the spacing between the first and second arrays of pillars in the first and second cell trajectory modulation regions.
6. The microfluidic device according to any one of claims 1 to 5, further comprising one or more sample inlets fluidically coupled to the first and second microfluidic channels,wherein the one or more sample inlets are configured to deliver one or more sample streams to the first and second microfluidic channels.
7. The microfluidic device according to claim 6, further comprising one or more first sheath inlets and / or one or more second sheath inlets fluidically coupled to the first microfluidic channel and / or the second microfluidic channel;wherein the one or more first sheath inlets and / or the one or more second sheath inlets are configured to introduce a first sheath stream and / or a second sheath stream to the first microfluidic channel and / or the second microfluidic channel;wherein the first sheath stream and / or the second sheath stream are configured to focus the sample stream within the respective microfluidic channel.
8. The microfluidic device according to claim 7,wherein the first sheath stream, sample stream, and second sheath stream are associated with respective hydraulic resistances R1, R2 and R3; and wherein the first microfluidic channel and / or the second microfluidic channel are configured such that conditions (i) and (ii) are satisfied:(i) 0.33 < R3 / R2 < 49, and(ii) 2 < R1 / R2 < 94.
9. The microfluidic device according to any one of claims 1 to 8, further comprising one or more outlets fluidically coupled to the first and second microfluidic channels;wherein the one or more outlets are configured to allow discharge of fluid.
10. The microfluidic device according to any one of claims 1 to 9, wherein the first and second arrays of pillars are configured to resolve differences in cell trajectories for cells over a dynamic size range of from 2 pm to 25 pm.
11. A method of performing biophysical profiling of a sample, the method comprising,passing a volume of the sample comprising a plurality of cells through a microfluidic device comprising a plurality of pillars configured to modulate flow trajectories of the plurality of cells;obtaining one or more modulated biophysical signatures of the plurality of cells generated by the microfluidic device, the modulated biophysical signatures comprising distribution profiles of the sorted plurality of cells;processing the one or more modulated biophysical signatures using one or more transformation functions configured to map and integrate the one or more modulated biophysical signatures to obtain one or more cell trajectory modulation (CTM) features; andinputting the one or more CTM features into a trained determination unit configured to correlate the one or more CTM features to one or more properties of the plurality of cells.
12. The method according to claim 11 , wherein the microfluidic device comprises,a first array of pillars having a first pillar configuration; anda second array of pillars having a second pillar configuration; wherein the first array of pillars differs from the second array of pillars in at least one of pillar dimension, pillar shape, pillar structure, pillar arrangement or pillar orientation, with respect to the direction of flow of cells.
13. The method according to claim 11 or 12,wherein the step of passing a volume of the sample through a microfluidic device is repeated for a plurality of times under different flow rates; and wherein the flow rates and pillars within the microfluidic device cooperatively modulate trajectories of the cells based on one or more biophysical properties of the cells to generate a plurality of modulated biophysical signatures under different flow and pillar configurations.
14. The method according to any one of claims 11 to 13, wherein obtaining the one or more modulated biophysical signatures of the plurality of cells comprises,detecting the cells at a detection region of the microfluidic device; counting the cells that are sorted to a plurality of histogram bins at the detection region of the microfluidic device; andobtaining a set of histogram distribution of the cells.
15. The method according to any one of claims 11 to 14, wherein the one or more transformation functions comprise a first set of transformation functions configured to transform the modulated biophysical signatures based on cell size, a second set of transformation functions configured to transform the modulated biophysical signatures based on cell deformability, and a third set of transformation functions configured to transform the modulated biophysical signatures based on a combination of cell size and cell deformability.
16. The method according to any one of claims 11 to 14, wherein the transformation functions comprise one or more mathematical and / or statistical operations comprising statistical measures, histogram bin evaluations, gated population calculations, distribution profile comparisons, or combinations thereof.
17. The method according to any one of claims 11 to 16, wherein the trained determination unit comprises a machine learning model based on one or more of linear models, tree-based models, neural networks, support vector machines, Bayesian models, and instance-based models.
18. The method according to any one of claims 11 to 17, wherein the one or more CTM features comprise size-related features, deformability-related features, or combined size and deformability-related features.
19. The method according to any one of claims 11 to 18, wherein the cells comprise chimeric antigen receptor (CAR) T cells, and wherein the one ormore properties of the plurality of cells comprise immune cell activation state, cell phenotype, and killing potency.
20. The method according to any one of claims 11 to 19, wherein, the sample comprises blood, urine, saliva, milk, or cultured cells;the sample is substantially free of externally added tags or labels; the sample is undiluted or untreated;the sample has a concentration falling in the range of from 0.5 x 106cells / mL to 80 x 106cells / mL;the sample has a volume from 10 pL to 100 pL; and / orthe sample is profiled within a target time period of no more than 15 minutes.
21. A method of classifying cells, the method comprising, passing a volume of a sample comprising a plurality of cells through a microfluidic device comprising a plurality of pillars configured to modulate flow trajectories such that the plurality of cells are sorted into a plurality of location bins;capturing image data of the plurality of location bins using an imaging apparatus;processing the image data of the plurality of location bins to extract image data of the cells sorted into the plurality of location bins; andinputting the image data of the cells into a trained machine learning model to classify the cells into one or more categories based on one or more morphological characteristics.
22. The method according to claim 21, wherein the sample is substantially free of externally added tags of labels, and / or introduced into the microfluidic device without dilution or pre-treatment.
23. The method according to claim 21 or 22, further comprising obtaining a distribution profile of the plurality of cells across the plurality of location bins,wherein the distribution profile comprises a number of cells sorted into each location bin, andwherein the number of cells sorted into each location bin is segmented according to the one or more categories based on the one or more morphological characteristics.
24. The method according to any one of claims 21 to 23, wherein processing the image data comprises applying blob detection to identify and analyze regions within the image data that represent cells.
25. The method according to any one of claims 21 to 24, wherein the one or more categories comprise size and shape, wherein the size category comprises small, medium and large cells, and wherein the shape category comprises doublets, blobs and irregular shapes.
26. The method according to claim 25, wherein the cells are immune cells; and wherein immune cells having a diameter of less than 7 pm are classified as small, immune cells having a diameter of greater than 12 pm are classified as large, and immune cells having a diameter between 7 pm and 12 pm are classified as medium.
27. The method according to claim 25 or 26, wherein a doublet comprises two cells adhered together, a blob comprises more than two cells grouped together, and an irregular morphology comprises a single cell having a non-spherical shape.
28. The method according to any one of claims 21 to 27, wherein the machine learning model is trained using augmented image data generated byhorizontal flipping, vertical flipping, rotating, or circular shifting image data captured by the imaging apparatus.
29. The method according to any one of claims 21 to 28, wherein the machine learning model comprises at least one of a support vector machine, random forest classifier, k-nearest neighbor model, logistic regression model, artificial neural network, convolutional neural network, or gradient boosting model.
30. The method according to any one of claims 21 to 29, wherein the machine learning model comprises a support vector machine configured to classify the cells by determining an optimal hyperplane separating different morphological classes.
