Resistant starch, microbiome, and precision modulation

ABSTRACT Resistant starch, microbiome, and precision modulation. Mounting evidence has positioned the gut microbiome as a nexus of health. Modulating its phylogenetic composition and function has become an attractive therapeutic prospect. Resistant starches (granular amylase-resistant α-glycans) are available as physicochemically and morphologically distinguishable products. Attempts to leverage resistant starch as microbiome-modifying interventions in clinical studies have yielded remarkable inter-individual variation. Consequently, their utility as a potential therapy likely depends predominantly on the selected resistant starch and the subject’s baseline microbiome. The purpose of this review is to detail i) the heterogeneity of resistant starches, ii) how resistant starch is sequentially degraded and fermented by specialized gut microbes, and iii) how resistant starch interventions yield variable effects on the gut microbiome.


Introduction
Dietary fibers promote human health, in large part through the gut microbiome. A prebiotic is defined as "a substrate that is selectively utilized by host microorganisms conferring a health benefit". 1 Non-digestible carbohydrates are prebiotics that gut microbes ferment into beneficial metabolites like butyrate. 1 Butyrate is recognized as a key contributor to host health by maintaining immune homeostasis, gut barrier integrity, and metabolism (reviewed in ref. 2). 2 Consequently, non-digestible carbohydrates have attracted immense interest as potential adjuvant therapies for many health disorders.
Resistant starch (RS) is among the recent foci of non-digestible carbohydrate therapies. Boosting gut butyrate production has been the objective of several RS intervention studies associated with aging, 3 insulin resistance, 4 metabolic syndrome, 5 kidney disease, 6 and schizophrenia, 7 and may be especially relevant for illnesses characterized by dysregulated epithelial integrity and immune function, like inflammatory bowel disease. 8 Upon reaching the colon, RS acts as a communal resource that is degraded and fermented by a hierarchy of specialized gut microbes. RS selectively feeds upstream keystone species, which produce substrates that are cross-fed upon by butyrogenic bacteria (recently reviewed in ref. 9). 9 However, the variable effects of RS on the gut microbiome are striking, whereby RS supplementation may increase butyrate production in one person and lower it in another. 10,11 Such variability underscores the need for precision nutrition to confer health benefits. Predicting the most optimal RS for individual demands understanding the complex structures of RS granules, the enzymatic machinery required to degrade RS, and the individual baseline microbiome. Hence, the goal of this review is to: (1) Describe the physicochemical and morphological heterogeneity of starch, and explain how these properties confer resistance to host and bacterial hydrolysis; (2) Detail the hierarchy of gut bacteria known to degrade and ferment resistant starch; and (3) Highlight the importance of precision resistant starch interventions.

Section 1: Resistant starch
In humans, digestible starch is susceptible to hydrolysis by salivary and pancreatic α-amylases, which hydrolyze α-1,4-glycosidic bonds. 12 Starch that reaches the large intestine without being fully digested is termed resistant starch (RS). Resistance depends on several physicochemical features, including physical encasement in non-digestible material (type 1), native supramolecular structure and morphology (type 2), retrogradation via hydrothermal-cycling (type 3), chemical modifications (predominantly ester cross-linking) (type 4), and amylose-lipid complexes formed during cooking (type 5). Types 2, 3, and 4 RS are the primary RS types used in human studies and will be the focus of this review. Extrinsic factors, such as host amylase gene copy number, 13 oro-cecal transit time, 14 and amylase inhibition, 15 further complicate starch bioavailability. Hence, starch digestibility should be considered as a kinetic property (slower to faster) affected by host-specific factors, rather than as a binary trait (resistant or nonresistant). 12

Type 2 RS -Native starch
Starch is synthesized in the amyloplast and chloroplast organelles of plants, forming mixtures of amylose and amylopectin. These molecules both consist of chains of glucose subunits linked by α-1,4-and α-1,6-glycosidic bonds, but differ in their chain length (i.e. degree of polymerization; DP) and branching (α-1,6 bonds). Amylose possesses a DP below 6,300 glucose subunits, almost entirely (>99.3%) bonded by α-1,4-glycosidic linkages. 16 Conversely, amylopectin forms much larger molecules (DP up to 26,500) with dense networks of short chains (mean DP [15][16][17][18] branching from longer chains (mean DP 48 to 60). 16 The intraand intermolecular interactions of amylose and amylopectin impart starch granules with a complex hierarchical structure ( Figure 1). Starch's supramolecular structure begins with pairs of amylopectin branch chains and amylose chains intertwining to form crystalline double helices. At amylopectin junction points (i.e. α-1,6 bonds), amorphous regions are thermodynamically favored over helices. Helices can either be dense and orthorhombic (A-type) or open and hexagonal (B-type). 17 Alternating amorphous and crystalline lamellae are thought to form 20-500 nm intermediary structures called "blocklets" 18,19 with amorphous amylose forming an intermolecular matrix or "glue" between blocklets. 20 Recent in silico modeling proposes that the blocklet architecture follows phyllotaxic rules, whereby interlocking crystalline platelets form ellipsoid fractals. 21 Depending on the amylopectin branching density and chain lengths, blocklets might be more amorphous ("defective blocklets") or crystalline ("normal blocklets"). 18,22 Blocklets arrange into arrays of alternating crystalline and semicrystalline concentric rings, disrupted by amorphous channels and veins. 19,20,22,23 Foresti and colleagues elegantly demonstrated that amorphous layers are preferentially degraded by soluble α-amylase, leaving behind a crystalline skeleton. 24 Overall, blocklet type and arrangement are associated with surface smoothness, porosity, and resistance to hydrolysis.
Across botanical sources, native starch granules vary by size, degree, and type of crystallinity; surface porosity and texture; relative amylose and amylopectin content; and amylopectin branch chain length and density. As a result, starch digestion rates can vary remarkably. For instance, starch granules from tubers tend to be among the most hydrolysis-resistant native starches, possibly because they are larger, 23,25,26 enriched in B-type crystallites, 25 possess longer amylopectin branch chains, 25 and have a smoother surface texture with fewer pores. 23,[25][26][27] Huang and colleagues showed that smaller, densely packed blocklets form a resilient shell on the surface of potato starch granules, while the interior is composed of larger, loosely packed blocklets. 28 The surface porosity, crystallinity, and RS content of corn starch granules correlate with amylose content; high-amylose varieties exhibit less porous surfaces, higher proportions of B-type crystallites, longer amylopectin side chains, and higher resistance to hydrolysis than varieties with no amylose (i.e. "waxy" corn starch). 25,29 Intriguingly, corn starch resistance peaks with an amylose content of 68%, 30 suggesting that both amylose and amylopectin are required to confer resistance to hydrolysis.

Type 3 RS -Retrograded starch
Retrogradation occurs when starch undergoes gelatinization followed by a thermodynamically driven reconfiguration of amylose and amylopectin into a new ordered state. First, when starch is heated in the presence of water, 31 amorphous regions absorb water, the granule swells, and amylose leaches through surface pores. 19 The hydrogen bonds stabilizing the helices become disrupted, causing amylose and amylopectin to unwind and dissociate. 32 Upon cooling, new helices are formed, but the degree to which crystallinity increases depends on the amylose content, cooling temperature, and number of repeated cycles. Retrogradation occurs more rapidly for amylose (over the span of minutes to hours) than amylopectin (over days), due to amylopectin's complex branching networks. 33 Colder retrogradation conditions promote the formation of B-type crystallites over A-type, 34,35 and additional cycles of retrogradation are understood to further increase starch crystallinity and resistance to hydrolysis. 27,36 It is thought that the formation of smaller blocklets during retrogradation is favored in starches with higher amylose content and longer amylopectin chains. 37 It seems that low surface porosity, enriched B-type crystallinity, and small blocklets formed during retrogradation could confer resistance to hydrolysis analogous to the intrinsic features seen in potato starch.

Type 4 RS -Chemically modified starch
Chemical treatments that introduce cross-links strengthen starch structural networks and generally increase resistance to hydrolysis. Treatment with sodium trimetaphosphate (STMP) and sodium tripolyphosphate (STPP) creates phosphoester bridges between glucose residues in amylose and amylopectin. 38 Higher concentrations of STMP and STPP (10-12% w/w) increase the degree of crosslinking and resistance to hydrolysis. 39 However, lower concentrations (5%) may increase the digestibility of high-amylose corn starch (HACS), likely because the imparted cross-linking is not sufficient to compensate for the gelatinization that occurs during heat treatment. 40 Cross-linking can also be induced through specific heat and acid treatment, which forms esters between glucose hydroxyl groups. 41 Cross-linking is restricted to the granule surface, 42 roughening its texture, 39,40,43 but leaving the overall morphology intact. 38,40,44 The reported effect of cross-linking on the RS content depends on the starch's botanical source and amylose content. Shin and colleagues found that cross-linking potato, high-amylose corn, and wheat starches results in higher RS content than cross-linking regular corn and rice starches, but precise values depend on the RS measurement method used. 45 In agreement, cross-linking wheat and high-amylose corn starches has been independently shown to increase RS content more than cross-linked regular corn starch. 39,40,46 Other starch modifications include hydroxypropylation (reviewed in ref. 47), 47 acetylation (reviewed in ref. 48), 48 and octenyl succinylation. 49 While these modifications increase resistance to varying degrees, 50 they are generally intended to alter rheological properties, 42 or encapsulate drugs for delivery to the colon. 49 Furthermore, USA federal regulations require that esterification with these agents in food products do not exceed 0.1% to 2.5%, 51 indicating an upper limit to the resistance of chemically modified starches in commercial products.
Some starch modifications occurring during food processing or found in nature can affect its digestibility. Proteins can fortify food matrices, 52 cell-derived lipids can form complexes at the granule surface, 53 or glucose molecules can be phosphorylated during glycan synthesis. 54 In fact, around 0.5% of native potato starch glucose residues are phosphorylated, 54 up to 5.5 times that of rice starch. 55 These phosphate monoesters might inhibit exo-acting hydrolases, 46 or sterically hinder helix packing and reduce overall crystallinity, 56 particularly during retrogradation. 32

Summary:
The available lines of evidence converge to suggest that a starch's surface microstructure is the principal factor affecting its digestibility. Relative amylose content, amylopectin branch chain density, and crystallinity appear to influence the size, type, and packing density of blocklets, which then determine granule surface texture and porosity. Retrogradation and cross-linking modify starch surface crystallinity and intermolecular networks, respectively, thereby increasing resistance to hydrolysis. It remains unclear whether blocklets simply affect surface area and integrity, or constitute "discrete structures" 57 that complement amylase active sites. Moreover, these questions tie into whether different bacteria preferentially degrade certain starches more than others based on binding site availability or recognition of discrete microstructures.

Section 2: Resistant starch degradation by microbes
RS is degraded by the colon's complex ecosystem of microbes, triggering a cascading web of metabolic interactions. Primary degraders grow on RS in monoculture. They penetrate the outer surfaces of intact RS granules, exposing pores and deeper concentric matrices while liberating oligosaccharides and generating metabolites like lactate and acetate. 58,59 Secondary degraders grow on starch in monoculture, but degrade intact RS poorly or not at all. Instead, they may adhere to abrasions and pores on RS before participating in its degradation, and opportunistically utilize solubilized oligosaccharides produced by other RS degraders. Crossfeeders do not grow on starch in monoculture. They utilize the by-products generated by upstream degraders, helping to maintain stoichiometric equilibrium and thermodynamically favorable (i.e. unconstrained) fermentation. 60 Most of the metabolites generated are acidic, which may further stabilize the ecosystem. 58,61 Together, the subsystem of microbes involved in RS degradation and fermentation participates in a complex network of cross-feeding interactions. 62 In maintaining microbiome homeostasis, the RS nutrient web expands the scope of what could be considered a "beneficial" gut microbe to a cluster of metabolically interconnected microbes.
Caveats with this model's classification scheme can be raised. Baxter and colleagues consider secondary degraders and cross-feeders (described below) as butyrogenic "secondary fermenters". 63 Likewise, Cerquiera and colleagues define nonprimary degrader, starch-active bacteria as "secondary starch scavengers". 9 We advance that secondary degraders can be delineated by their ability to grow on starch in monoculture, while a tertiary category (cross-feeders) comprises starch-inactive microbes that play a unique and important role in transforming upstream metabolites. Lastly, some secondary degraders show a limited ability to degrade RS in vitro, 64,65 or toggle their starch activity in the presence of fitter degraders. 66 Therefore, the "secondary degrader" designation is a fluid concept, but nonetheless useful for the purpose of characterizing individualized responses to RS.

Unlocking starch: Molecular machinery
Microbial degradation of RS is initiated at the granule surface, which primary and secondary degraders must first recognize and adhere to. Generally, these bacteria possess polysaccharide utilization loci (PULs) encoding transport proteins and modular Carbohydrate-Active Enzymes (CAZymes). 67 CAZymes contain recognition and catalytic domains. 68 Endo-acting α-amylases, which hydrolyze internal α-1,4-glycosidic bonds within starch molecules, primarily belong to the family of 13 glycoside hydrolases (GH13). 69 Other GH families include exo-acting α-glucosidases and the starch debranching enzymes: limit dextrinase, pullulanase, and isoamylase (reviewed in ref. 70). 70 Alas, most microbial GHs have not been studied in the context of the human gut. 70,71 This has left a deficit in our understanding of precise GH ligand specificities in relation to RS microstructures.
Starch-active GHs possess one or more carbohydrate-binding modules (CBMs), which are contiguous protein domains that facilitate substratespecific enzyme docking and enhance catalytic efficiency (reviewed in refs. 69 and 72). 69 73 Early protein crystallization studies showed that CBMs 20, 25, 26, and 34 each recognize glucose residues exposed in starch helical structures. 73 CBM20 possesses two separate starch surface binding sites with higher specificity for parallel amylose helices than amorphous coils, and an ability to disrupt the structure of shorter helices. 69,74 Amylases possessing repeated homogenous (e.g. two CBM25 copies) 75,76 or heterogeneous (e.g. CBMs 25 and 26) 77 CBMs confer higher avidity and thus tighter adsorption to corn starch granules than those with single CBMs. Of note, most studies investigating CBM ligands use cyclodextrins to infer potential α-glucan helix binding, rather than RS.
Some CAZymes show preferences for starches of different botanical sources. [78][79][80] For instance, CBM74 enhances surface pore formation by GHs with higher binding affinity to potato starch granules than wheat and waxy corn. 78 MaAmyA, a GH13 produced by Microbacterium aurum B8. A, uses duplicate CBM25 domains to form pores on wheat, but not potato starch granules. 79 AmyP, a GH13 found in marine bacteria, exhibits 10-fold higher hydrolytic activity on raw rice compared to potato and wheat starches. 80 This remarkable difference may simply be due to rice starch's higher intrinsic digestibility, 26 but may also be explained by specificity imparted by its starch-binding domain, CBM69. 81 At present, there is no obvious pattern by which the copy number of GH13's or CBM family domains can predict whether a bacterial strain can degrade starch or RS (Table 2).

Primary degraders
By using complex enzymatic machinery to dock and digest starch granules, certain RS-degrading specialists find a niche in "unlocking" RS granules for other RS guild members to access and ferment. Ruminococcus bromii and Bifidobacterium adolescentis are the best characterized primary degraders thus far.

Figure 2. RS guild members show preferences for different RS in vivo.
Primary degraders, secondary degraders, and cross-feeders discussed in-text that have been reported to significantly increase in relative abundance across 16 clinical trials where the microbiome was monitored. Lines indicate the frequency of reported associations. *Species-level differences among B. adolescentis and E. rectale/ Roseburia could not be resolved in every study. Full data can be found in Supplemental Table 1.

Bifidobacteria adolescentis
Bifidobacteria possess a multi-modular carbohydrate-utilization system (reviewed in ref. 9) 9 enriched with GHs with specificity for a broad range of glycan substrates. 87 CBMs 25,26,and 74 are thought to work synergistically to enable Bifidobacteria to not only dock starch, 9 but also agglutinate starch granules into clusters. 85 This behavior is thought to sequester granules away from competing starch-degraders. 88 90,91 Fifteen strains of B. adolescentis have been isolated from humans. 64,[92][93][94] Of the strains whose activity on starch has been studied in vitro, two (P2P3 and L2-32) 65 93 and NCFB 2229). 95 Nonetheless, strains that can ferment RS generate lactate, acetate, and formate. 58 B. adolescentis uses more of its starch degradation by-products than R. bromii, including glucose, signaling a more competitive relationship with other RS guild members. 65 Intriguingly, Li and colleagues found that an unspecified strain of B. adolescentis produces butyrate, 96 in contrast with prior reports that the species is not butyrogenic. 58 B. adolescentis shows strain-specific preferences for different RS. Strain P2P3 utilizes HACS (Hi-Maize 958) to a significantly higher degree than cross-linked corn starch (Versafibe 2470), crosslinked potato starch (Versafibe 1490), retrograded corn starch (Novelose 330), and most intriguingly, a different HACS (Hi-Maize 260). 93 Strain VTT E-001561 exhibits preferential binding (i.e. adherent cell counts normalized per gram of starch) to HACS (Hylon VII), and lowest binding to potato starch. 94 These results require careful interpretation because potato starch granules have smaller surface area-to-volume ratios (and thus less binding area) than corn starch. Furthermore, strain IVS-1 shows preferential growth on HACS (Hi-Maize 260), lesser but considerable growth on native potato starch and cross-linked tapioca starch (Versafibe 3490), and little growth on cross-linked corn (Versafibe 2470) and potato (Versafibe 1490) starches. 64 The putatively butyrogenic B. adolescentis strain described earlier will adhere to and grow better on partially hydrolyzed than non-hydrolyzed retrograded tuber starch granules, likely because of its rougher yet more crystalline surface. 96 Together, these studies seem to indicate B. adolescentis has a nuanced preference for HACS over other types of RS in vitro; however, this finding has not yet been recapitulated by in vivo microbiome studies.
B  95 and M115 grows on retrograded corn starch (Novelose 330). 89 However, Novelose 330 is approximately half digestible starch by weight, 98 thereby detracting the likelihood that this strain is a true primary degrader.
Primary degraders are thought to be necessary for RS degradation in the human gut. Work by the Flint laboratory revealed that individuals consuming RS whose butyrate-production did not increase (considered to be "RS non-responders") harbored nearly undetectable levels of R. bromii as compared to RS responders. 5 In a follow-up in vitro study, Flint et al. cultured stools from RS non-responders with spiked-in R. bromii L2-63, which boosted RSfermentation to levels comparable to RS responders. 65 This work was seminal for developing a model whereby primary degraders (e.g. R. bromii) "unlock" RS for other community members to degrade and ferment.

Secondary degraders
Secondary degraders possess extracellular amylases to degrade regular starch, but their contribution to initiating RS degradation is negligible compared to that of primary degraders. Instead, they may require primary degraders to erode smooth RS granule surfaces before adhering to RS and/or scavenging for "substrate spillover" (i.e. excess oligosaccharides generated by primary degraders). 58 Conceivably, it is advantageous for secondary degraders to position themselves near primary degraders docked to RS. Among secondary degraders, preferences for certain starches over others have not been greatly elucidated by metaproteomic, co-culture, or clinical studies.

Eubacterium rectale
Eubacterium rectale is a prominent member of the butyrate-producing Clostridium cluster XIVa, which is implicated in maintaining gut homeostasis. 99 E. rectale L1-86 is equipped with 11 GH13s, of which two have been deeply characterized: Amy13K and Amy13B. 100 Amy13K contains duplicate CBM26 domains and additional CBM41, 82, and 83 domains, enabling E. rectale to bind to certain starch granules (discussed below). 101 In vitro, Amy13K hydrolyzes amylopectin with twice the activity than amylose, releasing maltotetraose and maltopentaose. 100 These solubilized substrates, liberated either by E. rectale or primary degraders, may then be further hydrolyzed by Amy13B, which does not bind larger substrates. 100 Finally, malto-oligosaccharides of DP less than 8 are transported into the cell via EUR_01830, EUR_31480, and EUR_01240 ABCtransporters. 100 Like primary degraders, E. rectale exhibits RS preferences. E. rectale DSM 17629 shows detectable growth on cross-linked corn starch (Versafibe 2470), but not cross-linked potato (Versafibe 1490) or tapioca (Versafibe 3490) starches. 64 Furthermore, it can bind and hydrolyze regular and high-amylose corn starches (Hylon VII, 100,102 Hi-Maize 260, 64,101 and Hi-Maize 958), 65 but binds potato and cross-linked wheat starch (Fibersym) granules less efficiently. 101 Human fecal inocula cultured in a batch-fermentation system revealed that E. rectale responds equally to A-type and B-type retrograded HACS. 35 In this study, E. rectale's growth lagged behind Bifidobacteria, potentially indicating a reliance on primary degraders to erode RS first. 35 In agreement with its behavior in vitro, in vivo microbiome studies show that the relative abundance of E. rectale is increased in response to regular (Hi-Maize 260), retrograded (Novelose 330), and cross-linked (Versafibe 2470) HACS ( Figure 2).
E. rectale enrichment is positively associated with increased butyrate following potato starch intervention. 11 It interacts synergistically with R. bromii in co-culture, 65 and these species

Bacteroides thetaiotaomicron
B. thetaiotaomicron is endowed with 88 different polysaccharide utilization loci (PULs) encoding 226 GHs, 109 enabling it to utilize at least 32 different types of glycans. 110 B. thetaiotaomicron also possesses an intricate starch-utilization system (Sus). The first PUL to be described, 111 the sus operon encodes independent intracellular (SusR), periplasmic (SusAB), transmembrane (SusC), and cell wallanchored (SusDEFG) proteins with coordinated action, briefly described in their order of action here (thoroughly reviewed in ref. 112). 112 SusD is a CBM-analog necessary for growth on starch granules, 113 and binds the endo-regions of αglucan helices. 114 SusG is a GH13 α-amylase that contains CBM58 and a starch surface-binding site, with a twofold higher activity on soluble potato starch than amylopectin and strict hydrolyticspecificity for α-1,4-glycosidic bonds. 115 SusG cooperates with SusE and F to bind soluble and granular starch. 113 Eliminating all SusEFG starchbinding sites slows growth on amylopectin derived from corn more than potato. 113 While not required for cell growth, 116 SusE assists in capturing DP 7-18 malto-oligosaccharides, 117 and both SusE and SusF bind wound and unwound helices. 116 SusC contributes to starch binding and imports malto-oligosaccharides into the cell, 118 which are degraded into maltose and glucose by periplasmic SusA and B. 111 Lastly, constitutively expressed SusR is activated by intracellular maltose to rapidly upregulate sus gene expression. 119 B. thetaiotaomicron generates lactate, acetate, and propionate, and while it does not produce butyrate, 120 its metabolites drive butyrate production by other bacteria via cross-feeding interactions. 121 Most of B. thetaiotaomicron's GHs localize to the periplasm or outer membrane, thereby contributing to the soluble glycan pool for community use. 122 B. thetaiotaomicron may also use SusE and SusF to sequester starch from its competitors. 116 Furthermore, when B. thetaiotaomicron and E. rectale co-colonize gnotobiotic mice, B. thetaiotaomicron upregulates its starch and host-glycan degradation pathways, while E. rectale downregulates its GH expression and increases expression of transport proteins, relying instead on butyrate-producing pathways to generate energy. 66 Interestingly, when provided both pectic galactan and amylopectin, B. thetaiotaomicron will downregulate Sus expression and upregulate expression of other PULs, indicating a substrate preference for non-starch polysaccharides. 123 Indeed, starch availability and interspecies competition may relegate secondary degraders to cross-feeders, or vice versa.
B. thetaiotaomicron can utilize less than 20% of raw (Hi-Maize 958) and retrograded (Novelose 330) HACS after a 72-hour incubation, but cannot utilize a different type of HACS (Hi-Maize 240). 65 However, B. thetaiotaomicron grows with these starches better if co-cultured with R. bromii or B. adolescentis, or if the starches are autoclaved or boiled. 65 Hence, while B. thetaiotaomicron prefers other glycans, it is equipped with a sophisticated system to access starch, and can also influence the activity of other RS guild members.

Bifidobacteria
Several Bifidobacteria beyond those mentioned above show strain-specific potential to be secondary degraders (Table 1). These species include  B. longum, B. bifidum, B. breve, B. dentium,  B. infantis, B. pseudolongum, B. thermophilum,  and B. angulatum. B. breve produces ApuB, a GH13 amylase with CBMs 25, 41, and 48 domains. 124 ApuB cleaves both α-1,4-and α-1,6-glycosidic bonds using different active sites, and is required by B. breve UCC2003 to grow on starch. 125 B. breve 20213 shows similar growth patterns as B. adolescentis L2-32 on waxy corn, highamylose corn, wheat, and rice starches, 65 and an unspecified strain of B. breve has been shown to bind HACS (Hylon VII). 102 B. cuniculi and B. magnum both grow better on starch when co-cultured than in monocultures. 87 In contrast, only B. thermacidophilum subsp. porcinum's growth on starch is improved when cocultured with B. longum subsp. suis. Indeed, B. longum subsp. suis growth is unchanged, suggesting porcinum outsources starch degradation to suis. 87 While these two pairs of Bifidobacteria were isolated from rabbit and pig feces, respectively, they serve to illustrate that relationships between secondary degraders range from mutualistic to commensal.
Other potential starch-degraders have been identified but studied less extensively than those listed above. Lactobacillus amylovorus uses AmyA, an αamylase containing five tandem CBM26 repeats, to bind to raw corn starch. 76,126 Parabacteroides distasonis ATCC 8503 grows on native and crosslinked corn (Versafibe 2470) and potato (Versafibe 1490) starches in culture without a distinct preference, but achieves stationary phase after 4 days rather than 12 to 24 hours as seen with B. adolescentis and R. bromii. 64 Two uncharacterized species belonging to Ruminococcaceae and Clostridiaceae have been inferred to be RS degraders that are markedly enriched in subjects consuming either HACS (Hi-Maize 260) or potato starch. 63 Lastly, a sequenced taxon most closely related to Ruminoclostridium [Eubacterium] siraeum increased in relative abundance in one subject fed potato starch whose R. bromii and B. adolescentis did not enrich. 11

Cross-feeders
RS cross-feeders utilize starch by-products or metabolites generated by upstream RS degraders, such as acetate, lactate, formate, and succinate. R. hominis A2-183 does not grow on starch in monoculture, but it does grow in co-culture with B. adolescentis L2-32, by utilizing substrate spillover and/or the acetate and lactate produced by B. adolescentis. 58 Furthermore, R. hominis A2-183 is unable to degrade amylopectin or amylose, 67 suggesting that it degrades smaller maltooligosaccharide fractions. The R. hominis genome encodes fewer GH13s (8) than R. intestinalis (12 to 13) and R. inulinivorans (10 to 12), and it does not encode GH13s possessing CBM26 or CBM41 domains. 67 Considering the high prevalence of these CBMs across known starch degraders (Table  2), their absence may account for R. hominis' limited capacity to utilize starch-derived molecules, and would therefore be better considered a crossfeeder than secondary-degrader. 67 Similarly, R. gnavus ATCC 29149 cannot grow on soluble or retrograded starch alone, but cross-feeds upon substrate spillover when grown in co-culture with R. bromii. 59 Describing all known gut bacteria capable of utilizing these substrates exceeds the scope of this review, but one other example is noteworthy. F. prausnitzii is a prominent butyrate-producing commensal, comprising 1.5% to 9.5% of fecal bacteria in European individuals. 127 F. prausnitzii utilizes maltose and acetate to generate butyrate. 86,128 Among 10 F. prausnitzii strains, growth on starch and amylopectin is negligible or undetectable in monoculture. 97,104,129 Other butyrate-producing bacteria found in the colon are reviewed in ref. 130. 130 Of note, F. prausnitzii (and primary and secondary degraders) are depleted in health conditions associated with low butyrate production, like inflammatory bowel disease. 131,132

Section 3: In vitro and In vivo microbiome studies
Our nascent understanding of how individual bacteria respond to different RS can be complemented by monitoring their activity in the full ecosystem of the gut microbiome. Two common approaches are fermentation (in vitro) and intervention (in vivo) studies. Batch fermentations involve inoculating fecal material into a pH-, temperature-, and anaerobically maintained system that attempts to replicate the gut microenvironment. 40,[133][134][135] By administering RS to these cultures, temporal changes occurring in microbial populations can be readily monitored. While used less frequently, continuous flow fermentations involve cycling gases and substrates in and out of the system to further maintain physiologically relevant conditions. 35,136 RS intervention studies generally involve doubleblinded clinical trials, whereby subjects ingest RS and/or non-RS placebo for several weeks. In parallel-armed trials, subjects are randomly divided into RS and control groups. However, inter-individual variation is better controlled with cross-over designs, whereby subjects alternate between RS and control groups, often punctuated by 2-week washout periods to restore the microbiota toward baseline. Fecal samples are collected before, during, and after the intervention to analyze bacterial, metabolite, and protein abundances. Although informative in their own right, animal models will not be discussed here because of uncertainties introduced by host factors (e.g. oro-cecal transit time, amylase activity, and hindgut fermentation) that have not been robustly shown to recapitulate human physiology.

In vitro microbiome studies
The advantages of in vitro microbiome studies include eliminating upstream host variables, predicting bacterial responses in different biogeographical regions of the colon by tweaking media pH, and monitoring changes in bacterial populations over time. For example, Lesmes and colleagues devised a three-compartment continuous-flow fermentation system with pH ranges corresponding to the proximal (pH 5.5), transverse (pH 6.2), and distal (pH 7.1) colon. 35 Using stools obtained from three healthy volunteers, they found that Bifidobacteria grew equally well with B-type crystallite-enriched HACS across all pH ranges, but grew most poorly with A-type HACS at pH 5.5. Growth curves indicated that total bacteria declined over the first 10 hours of culture, followed by a compensatory climb for 14 hours, and that E. rectale's growth lagged 5 hours behind that of Bifidobacteria when grown with HACS, consistent with their putative roles in the RS guild. Of note, B-type HACS induced a fivefold higher increase in butyrate compared to A-type after 24 hours of culture.
By incubating stool with HACS cross-linked under different concentrations of STPP/STMP, Wang and colleagues showed that R. bromii enrichment is inversely proportional to the degree of RS cross-linking. 40 Regular HACS significantly enriches R. bromii after 4 hours followed by a depletion in the species after 12 hours of incubation, whereas HACS cross-linked in a 12% w/w STMP/STPP solution yields a less pronounced but stable enrichment over 4-24 hours. This work also indicated that RS can potentially become too resistant to hydrolysis even for microbes specialized to degrade it. Furthermore, enrichment of R. faecis and an unclassified Clostridiales member lagged behind R. bromii, parallel to the relationship seen between Bifidobacteria and E. rectale in Lesmes and colleagues' work. 35 Li and colleagues introduced a novel highthroughput stool fermentation assay, called RapidAIM, 134 which they employed to assess type 2, 3, and 4 RS' in stool slurries collected from six healthy volunteers. 135 While no effect was seen in response to cross-linked wheat starch, regular HACS (Hi-Maize 260) and retrograded HACS (Novelose 330) enriched E. rectale, R. faecis, Roseburia, and Lachnospiraceae. The retrograded HACS also enriched Bifidobacterium, Subdoligranulum variabile, and Ruminococcaceae.

In vivo microbiome studies
One rationale behind clinical trials involving RS is to selectively enrich butyrate-producing bacteria and increase butyrate production. To our knowledge, 16 studies have reported changes in the fecal microbiome of individuals following RS interventions (Supplemental Table 1). While acknowledging that these studies' methodologies and experimental designs differ, patterns have emerged to show that RS guild members respond to RS in vivo according to RS type ( Figure 2). Together, these studies underscore the nuanced responses of previously identified primary and secondary degraders, and cross-feeders to RS ingested over several weeks. Beyond these observations, they also lend insight into bacterial co-associations, butyrate production, and inter-individual variability.

Bacteria co-associations
Among individuals supplemented with potato starch or HACS (Hi-Maize 260), Baxter and colleagues identified several co-associations between primary degraders and butyrogenic taxa. Hi-Maize 260 significantly enriched R. bromii, which was significantly associated with co-enrichments of E. rectale. Likewise, potato starch significantly enriched the B. faecale/adolescentis/stercoris group, which was significantly associated with coenrichments of Anaerostipes hadrus. 63 A. hadrus has previously been shown to grow on starch in coculture with R. bromii, but not in monoculture. 137 After 12 weeks of a cross-linked wheat starch (Fibersym) intervention, positive co-associations between 18 pairs of bacteria were observed, including co-enrichments of Bacteroides acidifaciens and Bacteroides ovatus, and Christensenella minuta and Ruminococcus lactaris. 138 Of note, F. prausnitzii did not correlate with any of these species, yet butyrate production was significantly increased following RS intervention, indicating potentially novel butyrogenic RS guild members.
Deehan and colleagues identified 55 taxa that were affected by dose-dependent cross-linked RS intake, which they further categorized into seven distinct co-abundance response groups (CARGs). 64 These co-abundance response groups (CARGs), containing 3 to 18 taxa each, were constructed by hierarchical clustering of Spearman's correlation distances. Then, each CARGs' constitutive taxa were summed together into a single entity for downstream statistical analyses. They found significant correlations among CARG1 (including B. adolescentis, P. distasonis, Eubacterium hallii, and others) and CARG7 (including several Eubacterium and Bacteroides species) with cross-linked tapioca starch (Versafibe 3490), and CARG3 (including Ruminococcus callidus and Bacteroides plebeius) with cross-linked corn starch (Versafibe 2470). 64 Employing CARG-based analyses in this way enables the detection of underlying ecological guilds, boosting otherwise weak signals of individual taxa. Guild-based approaches have been proposed by others. 139 Interestingly, Deehan et al. also reported a significant co-exclusion between R. bromii and an unannotated Ruminococcus taxon (OTU27), suggesting a potential novel primary degrader that competes for R. bromii's niche.

Butyrate production
Butyrate is a fermentation end-product known for its immunomodulatory bioactivities. 2 Among the 16 studies listed in Supplemental Table 1, seven reported the butyrate concentrations in fecal samples, and potato starch consistently leads to significant increases in butyrate concentrations. 3,11,63 Cross-linked wheat starch (Fibersym) was also found to significantly increase fecal butyrate. 138 In contrast, various HACS led to no significant changes between groups. 63,140,141 Cross-linked HACS (Versafibe 2470) led to significant increases in butyrate concentrations, plateauing at a dose of 35 g RS per day, while no significant changes were seen with any dose of cross-linked potato (Versafibe 1490) or tapioca (Versafibe 2490) starches. 64 In the cross-linked HACS (Versafibe 2470) group of this study, R. bromii was significantly negatively associated with the proportion of butyrate relative to acetate and propionate, yet there was a significant positive association in the cross-linked potato starch group. Lastly, retrograded HACS (Novelose 330) significantly decreased butyrate in fecal samples from British males. 142 Since it is estimated that 95% of shortchain fatty acids are absorbed by the host epithelium, fecal butyrate concentration has been implicated as a low-fidelity proxy for intestinal butyrate production. 143 Thus, the discrepant and counterintuitive changes in butyrate concentrations may be explained by methodological limitations, or nuanced RS guild dynamics and underlying effects seen at the individual, but not group level, described below.

Inter-individual variability
Several in vivo studies report bacterial responses for each subject, showcasing how R. bromii, B. adolescentis, and E. rectale, may increase, remain unchanged, or even decrease in response to different RS depending on the individual. 4,64,103,142,144 Based on the in vitro behavior of RS guild members, we would expect to see these taxa more consistently enriched following RS interventions. This discrepancy may be explained by biological factors and methodological limitations.
Biological factors that may explain interindividual variability include the presence/absence of keystone RS guild members, 65 functional subgroup variability, 145 metabolic flux sensitivities, 136 and baseline diet variability. 10 First, Ze and colleagues demonstrated that R. bromii can be necessary for RS degradation in vivo. 65 Second, if keystone RS degraders are present in an individual, they may lack downstream subgroups required for converting RS into butyrate. 145 For instance, possessing lactate-producing B. adolescentis, but lacking lactate-utilizing butyrogens (e.g. E. hallii) may result in a poor butyrate response to RS; likewise, if one possesses net acetate-producing R. bromii but lacks acetate-utilizing butyrogens (e.g. F. prausnitzii). Third, the dominating metabolites produced by the RS guild may exert feedback on the global microbiome structure. For instance, without sufficient microbial lactate utilization, lactate produced by B. adolescentis may accumulate, lower luminal pH, 59 stress pH-sensitive microbes, and destabilize the microbiome. 136 So far, these insights point to the potential utility of an RS guild member probiotic, synbiotic, or consortium which may enable RS non-responders to benefit from RS therapies. Lastly, McOrist and colleagues observed that individuals with the highest baseline butyrate production tended to decrease butyrate production following a dietary intervention high in RS. 10 They commented that these individuals' pretreatment diets and microbiomes were likely configured optimally to generate butyrate, and exchanging their dietary patterns for the intervention diet may have disrupted their butyrogenic system. These biological factors underscore that interventions ought to be personalized where butyrateproduction is a primary endpoint.
Methodological limitations that may contribute to inter-individual variability include poor taxonomic resolution, underpowered sample sizes, and compositionally insensitive statistical techniques. First, 16S amplicon sequencing and genus-level qPCR primers used in several of these studies were unable to resolve and differentiate RS-and starch-degrading species and strains (e.g. R. bromii and B. adolescentis) from their starch-inactive phylogenetic relatives (Table  1). 5,7,15 Second, many studies may not have been sufficiently powered to draw robust conclusions about species-level changes, with sample sizes ranging between 8 and 46 subjects per treatment group. With larger sample sizes, subgroup analyses like those employed by Venkataraman et al. may boost the detection of otherwise weak signals at the study population level. For instance, k-means clustering of participants based on changes in fecal butyrate concentration revealed that B. adolescentis and R. bromii significantly enrich in RS responders, and E. rectale significantly enriches in individuals with high butyrate production at baseline. 11 Lastly, microbiome sequencing data are compositional, meaning that gene amplicon read counts do not necessarily reflect bacterial absolute abundances. 146 Instead, read counts are typically normalized to sum to 100%. For this reason, the relative abundances of smaller keystone communities (e.g. primary degraders) may increase, but appear to decrease simply because cross-feeders increase in relative abundance to a greater extent. 146 While compositionally sensitive techniques like the centered log ratio transformation can help mitigate this technical artifact, 146 microbiome responses can also be evaluated by monitoring beta-diversity (i.e. overall microbial community composition). Plotting fecal microbiota compositions in two-dimensional space reveals multidirectional changes across individuals before and after RS interventions, signaling that changes are driven by different taxa communities. 7,144 Together, these limitations illustrate the necessity of sufficiently powering RS interventions where microbiome composition is the primary endpoint, collecting critical baseline data and employing appropriate statistical techniques.

Conclusion
RS-based interventions have been proposed as a safe and economical approach to restoring gut microbiome homeostasis through the selective enrichment of butyrate-producing bacterial communities. However, there are several commercial RS products available, each with different physicochemical properties affecting their hydrolysis by endogenous and bacterial amylases. Within the microbiome, we see that microbes show preferences for different types of RS, which is made clearer by categorizing primary degraders, secondary degraders, and cross-feeders based on their ability to grow on RS and regular starch. We advance that the involvement of starch-inactive crossfeeders is critical to the net metabolic outcome of RS interventions. Indeed, how the microbiome composition and function changes is dependent on microbe-, host-, and RS-specific factors.
At present, there are several knowledge deficits surrounding these factors, which once elucidated, may enable data-driven RS-selection based on individual microbiome features: • What determines whether RS guild members are co-exclusive or co-abundant? Multiple species likely compete for the same role in the RS guild, but generate different metabolites that affect the composition of downstream members. For instance, the co-exclusive primary degraders R. bromii and B. adolescentis predominately produce acetate and lactate, respectively, which likely favors guild membership to acetate-or lactateconsuming secondary degraders and crossfeeders. • What determines which type of RS bacteria will bind and degrade? Do bacterial CAZymes recognize discrete microstructures on starch granules (thus conferring substrate selectivity), are certain combinations of CAZymes necessary for effective penetration of RS, or does metabolic feedback select for certain RS guild members over others? • To what degree do host factors (e.g. amylase gene copy number) influence RS digestion prior to entry into the colon? Incorporating these data may improve RS-selection for individuals. Habitual dietary RS intake may also contribute to variable responses, 147 but dietary records are rarely collected. • How do other kingdoms (e.g. fungi) contribute to RS degradation in the gut? • Lastly, butyrate production is not the sole indicator of microbiome functionality. Can RS restore gut dysbiosis in other ways, such as by regulating microbial bile acid or tryptophan metabolism? 148