## Introduction

In the decade since they were first introduced1, paper-based microfluidic devices (also called microfluidic paper-based analytical devices; µPAD) have emerged as an analytical platform that is capable of enabling the development of a wide range of analytical and bioanalytical tests2,3,4,5. In particular, paper-based microfluidic devices have the advantage of low cost and portability, which make them attractive for creating point-of-care diagnostic assays6. As new applications for paper-based microfluidic devices are developed, new device architectures often accompany them. While many assays can be performed using devices composed of a single layer of patterned paper7,8,9,10,11,12, there are numerous advantages to creating three-dimensional paper-based microfluidic devices. Three-dimensional devices, prepared from strategies that employ multiple layers of paper13 or a single folded layer of paper14, facilitate spatial separation of stored reagents (a requirement for multiplexed assays)15,16, performing sequential multistep reactions (e.g., immunoassays, nucleic acid detection)17,18,19,20, and the incorporation of sophisticated readouts into devices (e.g., barcodes or timers)21.

Three-dimensional devices fabricated from multiple sheets of paper, each of which comprise multiple copies of a device layer, rely on procedures that are similar to additive manufacturing processes22. The hydrophilic regions that connect layers of paper (e.g., channels and test zones) must overlap exactly to ensure reproducible capillary flow across all devices. Slight errors in device design may propagate significant inconsistencies that ultimately affect the performance of assays. Misalignment of zones in the fluidic pathway of a device, which can result from manufacturing error22 or patterning error, can lead to device failure. Consequently, the process of creating, optimizing, and improving upon designs for paper-based microfluidic devices can be slow and tedious. Even a single change to a design may require a complete readjustment of the entire device layout by the user. Devices become substantially more error-prone and time-consuming to fabricate when additional layers or complex channel structures are required. These problems are exacerbated further by the need to produce and test many iterations of the device, which is often required for the development and optimization of an assay. Ultimately, these concerns may limit the evaluation or use of such designs. Diminishing the impact of these factors on prototyping and manufacturing processes would reduce the amount of time and costs invested in creating, modifying, and fabricating devices and provide improved quality assurance for paper-based microfluidic devices.

We recognized that many of these challenges are associated with adapting existing programs that were initially intended for other purposes (e.g., graphic design or slide presentation software). For example, Adobe Illustrator is a popular option for graphic design, but can cost US$20 per month per license-holder23. CorelDRAW, another popular design software, costs nearly US$500 per permanent license24, while licenses for more powerful and precise drawing platforms can be even more cost prohibitive (e.g., US$695 for SketchUp Pro or US$185 per month per license-holder for AutoCAD)25,26. Consequently, we sought to develop a software program that was created purposefully to design and rapidly prototype paper-based microfluidic devices. We named this program AutoPAD—automated design of paper analytical devices. In order to ensure accessibility for a broad and global user base, we developed AutoPAD to be a cross-platform, open-source resource. Similar to other open-source scientific software (e.g., ImageJ27 and Jmol28), we make the source code for AutoPAD free to use and modify. This approach supports our goals to (i) reduce costs associated with starting and maintaining research laboratories in the field of paper-based diagnostics and (ii) enable new opportunities for researchers for whom costs of other design software may be prohibitive (e.g., high school students and laboratories in developing economies). Further, if device designs are commonly accessible and sharable, which is possible using AutoPAD text file outputs, a significant hurdle to information sharing and collaboration between international laboratories may be overcome29. Although AutoPAD does not have as many features and tools as commercial software options, all of its components were designed with developers of paper-based microfluidics in mind. AutoPAD’s streamlined features make it easy to use, and ensuring that the software remains open-source will allow for collaborative enhancement of the program’s capabilities by those who work within the field.

To develop AutoPAD, we adopted a relativistic design methodology whereby a device design is reduced to an interconnected network of objects, which do not have absolute positions but are instead located according to their inter-network connections. Through this organization, changing one object within the design causes all other objects to readjust their positions automatically, which is in stark contrast to conventional methods where each object within a design would need to be readjusted manually by the user. As a result, devices designed using AutoPAD scale well, as the amount of time required to make an adjustment does not change significantly between designs of varying size and complexity. In this manuscript, we demonstrate the development and use of AutoPAD with a number of case studies, each of which were chosen to highlight different capabilities of the software, and which, taken as a whole, show that our system has the capacity to produce almost any desired design.

## Experimental Design

### Design of Software Interface

The basis of the AutoPAD software is script files representing designs. A device could be created simply by writing a script into a text file and passing it through our program. However, we wanted to create a more user-friendly and visual interface that did not require the user to master the underlying scripting language to design devices. In AutoPAD, devices are represented as trees of interconnected shapes, where each node on the tree has child branches and its own parent branch from which it is sourced (Fig. 1). The same structure is used in computer file systems, where files are stored in folders that can be then nested inside of other folders. Recognizing this, we designed a “Tree Interface”, similar to a file browser, where each node is a folder containing its properties, such as a description of its shape and size, as well as all of the nodes attached to it. In this manner, when the properties of one node are modified (e.g., spatial location), then the corresponding properties of all attached nodes will be automatically reconfigured (Fig. 2). Additionally, we constructed the interface to automatically provide previews of the device while it is under construction, and the user may select parts of the device simply by selecting those parts on the preview image. The interface is a self-contained design solution, as it can be used to create, save, load, and compile script files.

### Identification of Desired Features

In order to identify the desired features of AutoPAD, we surveyed the literature for paper-based microfluidic devices and identified commonalities among their designs. Generally, paper-based microfluidic devices consist of paper channels that are either cut out (e.g., two-dimensional paper networks; 2DPN)30,31,32 or patterned with hydrophobic barriers (e.g., using photoresist or wax)33,34,35. We determined that a vast selection of differing designs could be generalized as a series of connected channels, circular zones, and rectangular pathways. As a result, we designed a system that describes devices as lists of these connected shapes, and drew inspiration from the literature to determine which geometric shapes should be supported most prominently with added support for generalized polygons for increased design flexibility. We also identified that text was commonly used in device designs. As such, we added support for rendering text in any size, font style, or rotation. We then looked at the various means of constructing devices, especially multi-layered devices; some device designs are cut out of paper or membranes32 or masked using stencils36, while many others are constructed using layers of patterned paper and adhesive37,38,39. To address all cases, we developed a system that would automatically generate cutting pattern outlines of device designs, which could be vectorized to work with any cutting plotter robot or laser cutter (Supplementary Information). In order to provide support for wax printing methods, we developed a buffering system whereby a device could be rendered either as white channels upon a black background (i.e., no wax deposited upon printing) or as white channels surrounded by shape-fitting black borders, which use less wax when printing. We also built tools for the construction of origami devices16,40 by developing features that allow layers to be arranged into grids, which could then be printed and folded. We recognized that many origami devices consist of the same layer designs rotated and reflected in various arrangements, so we additionally allowed users to reuse existing layers with different transformations rather than needing to redesign a layer several times for each rotation.

Device designs have specific desired dimensions, as it is important to control factors related to experimental performance (e.g., total channel volume), economy (e.g., minimizing material waste), and operability (e.g., holding or manipulating a device). However, an image by itself carries no dimensional data, because it can be printed at varying pixels-per-inch resolutions. Therefore, AutoPAD will compile images into PDF files of set dimensions based on the user’s specifications. Depending on the desired page size and the dimensions of each device, intact layer designs will be tiled across the page as many times as space allows. When these sheets are printed, the device will have the desired dimensions and will be ready for mass assembly. Files of PDF format are recognized by every operating system, every printer, and most graphics design software, which ensures the compatibility of our program and its outputs with other systems currently in use.

## Results and Discussion

### Tutorials and User Documentation

We have prepared extensive documentation in support of training new users of AutoPAD. We provide the following four documents in the Supplementary Information that accompanies this manuscript: (i) Materials and Methods. We describe technical details related to programming AutoPAD. Additionally, we include images of case studies for using AutoPAD to recreate several literature examples of paper-based microfluidic devices. (ii) Glossary. We provide a list of commands and a description of their uses. (iii) Introduction to the Tree Interface. This document instructs the user how to navigate the front-end interface of AutoPAD and highlights important components, panels, and principles required to create a design. (iv) Getting Started with AutoPAD. This document guides users through the process of designing paper-based microfluidic devices using AutoPAD—from conceptualization, through the use of commands, and ultimately to generating printable files. In essence, all previous documents support this guide. In addition, we have prepared five narrated tutorial videos in order to help guide new users through learning key functions of AutoPAD and begin to design their own paper-based microfluidic devices. These videos are also available online on the Mace Lab YouTube channel41.

### Case Studies

In order to demonstrate the broad and enabling capabilities of AutoPAD, we chose several case studies that highlight functions that are necessary to design, manufacture, and ultimately use paper-based microfluidic devices. Each case study is accompanied by: (i) a figure depicting the literature reference rendered as an image provided by the AutoPAD Interpreter, images of modified paper-based microfluidic devices designed in AutoPAD, and images of devices manufactured using designs created in AutoPAD; and (ii) examples of code lines that may be easily modified to make complex adjustments to design geometry. In these case studies, the appearance (e.g., feature size, border sharpness, or color) of paper-based devices made from designs produced in AutoPAD may differ slightly from their parent software designs. These differences can be attributed to the way that melted wax permeates chromatography paper34. We also provide text files (and accompanying PNG and PDF outputs) for all six case studies in a compressed zip file in a GitHub repository42, which can serve as the basis for the design of various types of paper-based microfluidic devices.

### Case Study 1. Single Layer Sample Splitting

The case of splitting one sample into three test zones on a single layer of paper represents the design used in the first report of patterned paper for bioassays by Whitesides and coworkers1. We highlight this design here because it also demonstrates many of the basic features of AutoPAD, such as the placement of geometric shapes and angles (Fig. 3, Case Study 1). Additionally, this design showcases the “snap-to-shape” buffer system that allows for clean outlining of continuous hydrophilic channels prepared from multiple discrete objects. Modifications to the design of these devices (e.g., the length of channels or shape of test zones) require changing only two or three numbers using the Tree Interface or within the script (Fig. 3). AutoPAD will automatically adjust the positions of relevant objects and generate new image files with minimal user input, which greatly facilitates the manufacture and testing of paper-based microfluidic devices.

### Case Study 2. Paper Microzone Plates

Patterned paper microzone plates were developed as low-cost alternatives to standard 96- and 384-well plastic microwell plates43. Because these microzone plates can be printed on-demand and offer users the flexibility to make design modifications (e.g., zone geometry or introduction of channels), numerous applications including paper-based ELISA44,45,46, array-based screening47, and development of tissue engineering scaffolds48,49 are made possible. We recreated a 96-well paper microzone plate in order to demonstrate how AutoPAD can be used to create arrays of geometric shapes and incorporate text labels (Supplementary Fig. S1, Case Study 2). This latter capability is additionally useful for providing users with clear identifiers for test zones of multiplexed assays (e.g., for unique markers or to differentiate tests from controls)50,51.

### Case Study 3. Two-Dimensional Paper Networks (2DPN)

Two-dimensional paper networks (2DPN) demonstrate both a basic application of layers and also the use of cut layers (Supplementary Fig. S2, Case Study 3)52. Unlike the previous case studies, this design requires the porous media (e.g., nitrocellulose membranes) to be laser cut instead of patterned with hydrophobic barriers. The AutoPAD interpreter is able to generate outlines using the $CUT property, which creates linewidths that can be traced into paths that are compatible with laser cutter software. Designs generated using the$CUT property can also be turned into paths compatible with robotic knife plotters, and used to cut layers of double-sided adhesive that facilitate the manufacture of three-dimensional paper-based microfluidic devices22.

### Case Study 5. Three-Dimensional Sample Splitting

We recreated a three-dimensional sample splitter13 in order to demonstrate AutoPAD’s ability to produce complex patterns that require precise alignment between multiple layers of paper. In total, this device design comprises 364 individual elements distributed across 9 layers of patterned materials: 4 objects on layer 1 (paper), 4 objects on layer 2 (adhesive), 20 objects on layer 3 (paper), 8 objects on layer 4 (adhesive), 40 objects on layer 5 (paper), 16 objects on layer 6 (adhesive), 144 objects on layer 7 (paper), 64 objects on layer 8 (adhesive), and 64 objects on layer 9 (paper). An automated approach to preparing such a design would limit operational errors (e.g., leaks or failures) caused by any misalignment between layers that could result from a manual design process. We present the layers of the three-dimensional sample splitter design in Fig. 4 (Case Study 5).

By using this referencing approach, splitting device designs with altered dimensions (Fig. 4) require only minor alterations to the script rather than a complete restructuring and repositioning of all device design elements across multiple layers. The simple modifications that we made to Layer 2 of this device caused numerous automatic changes in subsequent layers of the device. For example, changing the length of 1 rectangular zone in Layer 2 of this device causes the positions of 43 additional design elements—32 from paper layers and 11 from adhesive layers—to also change. When we subsequently modify the length of all 8 rectangular zones in Layer 2, which takes just 2–3 minutes in AutoPAD, 344 design elements are automatically updated to yield a properly aligned, functioning device design. In a commercial software like Adobe Illustrator, this type of design modification would require a significant time investment (ca. hours), as it requires repositioning of 95% of all device elements.

### Case Study 6. Origami Devices

An additional function of AutoPAD is the ability to construct origami-styled devices14,16,40. Because they are manufactured using a single layer of folded paper, components of origami designs require very careful spatial organization and alignment in order to ensure that proper contact is made between all zones to complete the desired, intact network of hydrophilic channels. Origami devices demonstrate the use of Combined-Layer commands, which is a functionality of the AutoPAD Interpreter that allows for multiple layer designs to be tiled into a sheet. The arrangement of individual tiles can be controlled through rotation, reflection, and positioning commands to create the desired array of zones. To demonstrate these concepts, we recreated the first demonstration of the use of origami to assemble three-dimensional paper-based microfluidic devices (Supplementary Fig. S4, Case Study 6)14. This design required the use of only three layers (seen in the first row), which were then rotated in order to complete the grid. Similar to the previous case studies, the origami design is also easy to refactor with changes to only a single parameter due to AutoPAD’s ability to perform simple arithmetic. Note that it is possible that some altered designs may not fold in a manner that creates functional fluidic connections; future versions of AutoPAD that are driven by user need could offer more sophisticated folding-alignment prediction capabilities.