Introduction
A wide range of diets has been developed over the past decades to reduce weight and/or to improve health(Reference Jayedi, Ge and Johnston1–Reference Forouhi, Misra and Mohan7). Reducing the amount of any of the macronutrients, fat, carbohydrate or protein, is often used as a dietary strategy(Reference Jayedi, Ge and Johnston1,Reference Ge, Sadeghirad and Ball2,Reference Johnston, Kanters and Bandayrel4,Reference Freire8) . Such dietary alterations have been applied for the treatment of several diseases, including type 2 diabetes (T2D)(Reference Nield, Moore and Hooper9–Reference Bhatt, Choudhari and Mahajan11), chronic kidney disease(Reference Palmer, Maggo and Campbell12–Reference Kovesdy, Kopple and Kalantar-Zadeh15), epilepsy(Reference Martin-McGill, Bresnahan and Levy16–Reference Haridas and Kossoff18) and inflammatory bowel disease(Reference Limketkai, Iheozor-Ejiofor and Gjuladin-Hellon19–Reference Lewis and Abreu21). It has been suggested that an important effect of diet on health is mediated via the gut microbiome(Reference Power, O’Toole and Stanton22), and evidence is emerging that microbial metabolites may affect health by acting as signalling molecules(Reference Morrison and Preston23). The gut microbiome, also referred to as the forgotten organ(Reference O’Hara and Shanahan24), is an essential component of the human body. The human digestive tract harbours a diverse community of primarily anaerobic microorganisms. The conditions, as well as the numbers of bacteria differ considerably in the various sections of the gastrointestinal tract, which hosts up to 103 colony-forming units (cfu) per millilitre (cfu ml−1) in the stomach and duodenum, while the numbers increase in jejunum and ileum (104–108 cfu ml−1) and rise to even higher levels in the colon (109–1012 cfu ml−1)(Reference Blaut and Clavel25). Hundreds of different bacterial species can be present in a single individual, of which particular species are present in most individuals. Approximately 94% of all species in healthy adults belong to the phyla Bacteroidetes (new nomenclature; Bacteroidota), Firmicutes (Bacillota), Actinobacteria (Actinomycetota) or Proteobacteria (Pseudomonadota)(Reference Scott, Gratz and Sheridan26,Reference Thursby and Juge27) .
Faecal samples can be collected after bolus transit through the gastrointestinal tract to characterise the gut microbiome. Interindividual variability and plasticity of the gut microbiota composition make identifying a ‘healthy’ microbiome profile challenging, which remains a heavily debated topic(Reference Gentile and Weir28). However, richness and diversity generally provide the gut ecosystem with stability and resilience and are therefore associated with health(Reference Li, Zhou and Liang29,Reference McCann30) . Richness can be quantified as the total number of bacterial species in a sample; alpha diversity further incorporates relative abundance profiles (microbiota diversity within an individual sample), whilst beta diversity reflects the diversity between samples (inter-variability)(Reference Claesson, Clooney and O’Toole31). Healthy individuals generally have higher richness and diversity than people with metabolic dysfunction or chronic diseases(Reference Gentile and Weir28). Reduced gut microbiome diversity and richness are associated with a myriad of diseases, including T2D, rheumatoid arthritis, inflammatory bowel disease and several types of cancer(Reference Hajela, Ramakrishna and Nair32).
Not only diversity but also the relative abundance (distribution of individual bacterial taxa within a sample) of individual bacterial taxa in the gut may be associated with health or disease(Reference Hajela, Ramakrishna and Nair32). Some bacteria are assumed to be primarily health-promoting, such as Lactobacillus and Bifidobacterium, which are known to produce microbial compounds important for healthy gut function(Reference Marco33). Other bacteria may confer pathogenic effects since their abundance is related to adverse health outcomes(Reference Ducarmon, Zwittink and Hornung34). Several diseases are associated with an alteration in the abundance of specific bacteria. For example, people with T2D have lower faecal numbers of at least one of the genera Bacteroides, Bifidobacterium, Roseburia, Faecalibacterium and Akkermansia as compared with healthy controls(Reference Gurung, Li and You35), whereas colorectal cancer has been associated with an increase in the relative abundance of a core set of twenty-nine bacterial species(Reference Wirbel, Pyl and Kartal36).
The complex bacterial ecosystem in the human digestive tract has a myriad of functions, including vitamin synthesis(Reference LeBlanc, Milani and de Giori37), provision of colonisation resistance against incoming pathogens(Reference Ducarmon, Zwittink and Hornung34), mediation of immune responses, and digestion of macronutrients into metabolites by the production of a great array of enzymes(Reference Thursby and Juge27). The processing of macronutrients starts in the upper gastrointestinal tract. Carbohydrates are partly digested by salivary amylase, pancreatic enzymes and enzymes on the surface of small intestinal cells and subsequently absorbed by the small intestine wall(Reference Whitney, Rolfes and Crowe38). Some carbohydrates are easily digested in the small intestine, while others are more difficult to digest(Reference Whitney, Rolfes and Crowe38,Reference Rastall, Diez-Municio and Forssten39) . The non-digestible carbohydrates (NDC) thus largely pass through the small intestine into the colon, where they are fermented by the intestinal microbiota. Some NDC are associated with health benefits, such as laxation or lowering of blood cholesterol or glucose levels(Reference Rastall, Diez-Municio and Forssten39,Reference Holscher, Chumpitazi and Dahl40) . They are primarily metabolised by the gut microbiome into short-chain fatty acids (SCFA), including acetate, propionate and butyrate(Reference Scott, Gratz and Sheridan26). SCFA are partly consumed by the colonic mucosa and absorbed by intestinal cells, where they confer local effects. Some of the SCFA are partly transported through the basolateral membrane towards the bloodstream and can act on receptors at different body sites. The rest of the SCFA are excreted in the faeces. SCFA appear to regulate hepatic lipid and glucose homeostasis by decreasing glucose output, lipogenesis and free fatty acid accumulation. Also, associations with adipocyte lipolysis and adipogenesis have been reported(Reference Hong, Nishimura and Hishikawa41,Reference Byrne, Chambers and Morrison42) . Moreover, they affect appetite regulation by increasing anorexigenic signalling in appetite centres and affect energy homeostasis through several metabolic pathways activated in parallel(Reference Byrne, Chambers and Morrison42–Reference Bastings, Venema and Blaak44). Fat can be digested and absorbed in the small intestine after it is partially emulsified by bile acids and broken down into smaller fragments by pancreatic and intestinal lipases(Reference Whitney, Rolfes and Crowe38). A small part of ingested fat is not absorbed in the small intestine and can be metabolised by gut microbiota or excreted(Reference Scott, Gratz and Sheridan26,Reference Hullar and Fu45) . The gut microbiome can convert bile acids into secondary bile acids, which are suggested to play a role in epithelial cell integrity, host immune response and gut bacterial composition(Reference Blacher, Levy and Tatirovsky46). Proteins are broken down by gastric, pancreatic and intestinal proteases into smaller protein fragments, tripeptides, dipeptides and individual amino acids, which are partly absorbed by the small intestine(Reference Whitney, Rolfes and Crowe38). In the colon, protein fermentation produces diverse metabolites, including SCFA, ammonia, tryptophan metabolites and the branched-chain fatty acids (BCFA) isobutyrate, 2-methylbutyrate and isovalerate(Reference Scott, Gratz and Sheridan26,Reference LaBouyer, Holtrop and Horgan47,Reference Agus, Planchais and Sokol48) . Tryptophan is a precursor for crucial compounds, including serotonin and kynurenine, which are important for neurobiological functions, gut–brain signalling, gut motility, platelet functions and immune homeostasis(Reference Agus, Planchais and Sokol48). Macronutrient processing thus leads mostly to the absorption of metabolites by the gut, and only a minority of metabolites is excreted in the faeces. These metabolites can be used as an approximate indication for carbohydrate, fat and protein metabolisation by the microbiota(Reference Scott, Gratz and Sheridan26,Reference Blaak, Canfora and Theis49) .
Several interventions that would potentially be capable of altering the gut microbiome composition and/or its products to improve health status include the following: (1) supplements of dietary substrates that are selectively utilised by host microorganisms conferring a health benefit (prebiotics); or (2) live microorganisms that, when administered in adequate amounts, confer a health benefit on the host (probiotics); or (3) a mixture comprising live microorganisms and substrate(s) selectively utilised by host microorganisms (synbiotics); or (4) inanimate microorganisms and/or their components (postbiotics); or (5) faecal microbiota transplantations(Reference Marco33,Reference Salminen, Collado and Endo50) . However, diet is the most natural daily modulator of the gut microbiome and health(Reference Dahl, Rivero Mendoza and Lambert51). An elaborate modification of the diet may represent an excellent strategy to alter the microbial community composition and function for improved health. However, little is known about the effects of restriction of macronutrient levels on the gut microbiome. Therefore, this review aims to give an overview of the effects of diets restricted in carbohydrates, fat or protein on the bacterial composition of the human gut microbiome and on faecal metabolites.
Methods
Eligibility criteria
The study characteristics were defined as human studies with an intervention described as a low-fat diet (LFD), low-carbohydrate diet (LCD) or low-protein diet (LPD) with gut microbiome as an outcome measure. Studies had to be published in English or with an available English translation. Exclusion criteria included animal studies, paediatric studies, studies with no relevant extractable data or studies with no full text available. The following study designs were included: RCT, non-randomised trials, cohort studies and observational studies. Reviews and case reports were excluded.
Information sources and search strategy
The search strategy (supplementary material) was used to search PubMed, Embase, Web of Science and the Cochrane Library. It was adapted for each dietary intervention (low carbohydrate, low fat and low protein). Articles were selected for screening on 3 June 2021.
Selection and data collection process
Covidence (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia) was used for the screening process. After removing duplicates, the title and abstract screening and subsequent full-text screening were performed with the pre-defined inclusion and exclusion criteria by two independent reviewers (M.S. and N.J. or P.P.). A third review member (H.P.) was available for discussion in case of inconsistencies. Since there is no worldwide-accepted definition of low-fat, low-carbohydrate and low-fat diets, the following definitions were adopted: low-carb, <30 energy% intake of carbohydrates; low-fat, <30 energy% intake of fat; low-protein, <10 energy% intake of protein(Reference Bhandari and Sapra52–Reference Westerterp-Plantenga54).
Reported data
Outcome domains include changes in alpha diversity, relative bacterial abundance and/or metabolites between baseline and after intervention. Outcome data are reported as either increased or decreased only when a significant difference from baseline was observed. Data are included in the results section only when reported in at least three independent studies to enhance the reliability of our conclusions. Tables with all outcome data are included in the supplementary file. Furthermore, the macronutrient composition of the dietary intervention, participant characteristics (including age, BMI, gender and eventual disease), number of participants, intervention time, wash-out period in case of cross-over, and time of analyses (directly after intervention or at a later moment) were extracted and are reported in tables 1–6.
Quality assessment
The Cochrane Collaboration risk of bias tool was used to assess the methodological quality of the included studies on outcome level. The ROBINS-I tool(55) was used for non-randomised studies, and the RoB 2·0 tool was used for randomised studies. The risk of bias was independently reviewed by two reviewers (M.P. and N.J. or P.P.) and discussed until a consensus was reached. A third reviewer (H.P.) was available for consultation when consensus was not reached.
Results
Study selection
The literature search resulted in 1178 articles (Supplementary Figure 1). After removing 100 duplicates, 1078 articles were screened on title and abstract. A total of 938 articles were deemed irrelevant to the research question and were excluded, for example, due to the inclusion of animals or lack of gut microbiome outcomes. Four reports could not be retrieved. Full-text screening on eligibility was conducted on the remaining 136 articles, of which 100 were excluded, resulting in the inclusion of thirty-six articles. Excluded articles often only reported change between intervention groups and did not describe the effect from baseline per individual group. Of the thirty-six included articles, nineteen conducted LCD interventions, twenty conducted LFD interventions and five conducted LPD interventions. Six studies had LCD as well as LFD intervention groups(Reference Fragiadakis, Wastyk and Robinson56–Reference O’Keefe, Li and Lahti61), and two had both LCD and LPD intervention groups(Reference Ferraris, Meroni and Casiraghi17,Reference Tagliabue, Ferraris and Uggeri62) .
Study characteristics
LCD, LFD and LPD study features describe the year of execution, design and patient- and intervention characteristics (Tables 1–3). The LCD studies were published between 2006 and 2021, the LFD studies between 1978 and 2021, and the LPD between 2016 and 2021. Study designs included randomised prospective, randomised cross-over, non-randomised cross-over and non-randomised trials. Some studies used healthy subjects; however, more often, participants with overweight/obesity or specific diseases were included. In the low-carbohydrate studies, participants were often obese. In the low-fat studies, obese persons and persons with multiple sclerosis (MS) were often included. Low-protein studies often examined persons with chronic kidney disease. Study group size differed from six to 246, with most studies including fewer than thirty participants. Not all studies reported the number of subjects in the specific diet groups(Reference Haro, Garcia-Carpintero and Alcala-Diaz63–Reference Rocchetti, Di Iorio and Vacca66). Some studies use several study groups, which all undergo either LCD, LFD or LPD, where interventions differ in the source of the nutrition or additional supplements(Reference Scott, Gratz and Sheridan26,Reference Murtaza, Burke and Vlahovich59,Reference Basciani, Camajani and Contini67–Reference Lai, Molfino and Testorio70) . The average age varied between 23·3 and 70·5 years, although the average age was often not reported. Most studies included males and females, while some included only males(Reference Murtaza, Burke and Vlahovich59,Reference Haro, Garcia-Carpintero and Alcala-Diaz63,Reference Cummings, Wiggins and Jenkins71–Reference Ang, Alexander and Newman74) . The male/female numbers were not always reported. In papers reporting the average BMI, it varied between 21·7 and 35·9 kg/m2; however, most papers reported an average BMI of >25 kg/m2 (overweight), and the average BMI was >30 kg/m2 (obese) in the majority of studies evaluating the effects of LCD. Intervention time varied substantially between studies, with the shortest intervention time of 2 weeks and the longest of 3 years, while most studies had an intervention time of less than 6 months. In cross-over studies, wash-out time (if reported) varied from zero days to 3 months. In the majority of studies, data collected directly after intervention were used for analysis, except in the studies of Pataky, Russell and Gutierrez-Repiso, where the outcome was measured 3 weeks(Reference Pataky, Genton and Spahr75), 5 weeks(Reference Russell, Gratz and Duncan73) or 2 months(Reference Gutiérrez-Repiso, Hernández-García and García-Almeida57) after the end of the intervention.
Study characteristics of included low-carbohydrate intervention studies
BMI, body mass index; DRE, drug-resistant epilepsy; GLUT 1DS; glucose transporter 1 deficiency syndrome; MCI, mild cognitive impairment; MS, multiple sclerosis; NAFLD, non-alcoholic fatty liver disease; NASH, non-alcoholic steatohepatitis; NI, not indicated; NRC, non-randomised cross-over; RC, randomised cross-over; RP, randomised parallel.
Study characteristics of included low-fat intervention studies
BMI, body mass index; DRE, drug-resistant epilepsy; GLUT 1DS, glucose transporter 1 deficiency syndrome; HC, high carbohydrate; HGI, high glycaemic index; MCI, mild cognitive impairment; MS, multiple sclerosis; NAFLD, non-alcoholic fatty liver disease; NASH, non-alcoholic steatohepatitis; NRC, non-randomised cross-over; LFD, low-fat diet; LFD per, low-fat diet periodised; LGI, low glycaemic index; RC, randomised cross-over; RP, randomised parallel; T2D, type 2 diabetes.
Study characteristics of included low-protein intervention studies
BMI, body mass index; CKD, chronic kidney disease; DRE, drug-resistant epilepsy; GLUT 1DS, glucose transporter 1 deficiency syndrome; NI, not indicated; RC, randomised cross-over; RP, randomised parallel.
Macronutrient composition
The macronutrient composition of the diet was very heterogeneous among the included studies (Tables 4–6). Not all percentages add up to 100%, often without an explanation from the authors. Macronutrient content was sometimes reported in grams, so for this review, the percentage was calculated using the formula ‘grams × energy per gram × 100)/consumed kcal/d’. The energy per gram of carbohydrate and protein is 4 kcal (16.7 kJ) and, per gram of fat, 9 kcal (37.7 kJ). The macronutrient composition of the LCD, LFD, and LPD diets will be described.
Macro-nutrient composition of low-carbohydrate diets
Overview of macro-nutrient composition demonstrating the percentage of carbohydrate, fat, and protein of every intervention. When percentages were lacking, we calculated the percentage from the number of grams per macro-nutrient with the formula “amount of grams*energy per gram*100)/consumed kcal/day”.
DHA, docosahexaenoic acid. LGI, low glycemic index. NI, not indicated. PUFA, polyunsaturated fatty acids. SFA: saturated fatty acids.
Macro-nutrient composition of low-fat diets
Overview of macro-nutrient composition demonstrating the percentage of carbohydrate, fat, and protein of every intervention. When percentages were lacking, we calculated the percentage from the number of grams per macro-nutrient with the formula “amount of grams*energy per gram*100)/consumed kcal/day”.
DHA, docosahexaenoic acid. LGI, low glycemic index. NI, not indicated.
Macro-nutrient composition of low-protein diets
Overview of macro-nutrient composition demonstrating the percentage of carbohydrate, fat, and protein of every intervention. When percentages were lacking, we calculated the percentage from the number of grams per macro-nutrient with the formula “amount of grams*energy per gram*100)/consumed kcal/day”.
NI, not indicated.
In the LCD interventions (Table 4), the carbohydrate content varied between 4% and 25% of total calories, fat content between 14% and 87% of total calories, and protein between 9% and 68% of total calories. The calorie content varied between 600 and 2526 kcal/d; however, only nine out of nineteen papers reported calorie content. In the paper by Gutierrez-Repiso(Reference Gutiérrez-Repiso, Hernández-García and García-Almeida57), the number of grams of carbohydrates, fat and protein was only reported for the first 2 months of the intervention, and the number of calories derived from additional vegetables was not reported. The following 2 months of intervention were not specified, although calorie intake was higher than the first 2 months (800–1500 kcal/d). The studies of Basciani(Reference Basciani, Camajani and Contini67) and Lundsgaard(Reference Lundsgaard, Holm and Sjøberg68) used several study groups; in the trial of Basciani, the protein source differed between groups, and Lundsgaard supplemented either polyunsaturated fatty acids (PUFA) or saturated fatty acids (SFA). Overall, LCD studies were very diverse in regard to dietary composition.
In the LFD interventions (Table 5), the percentage of fat of total calories varied between 8% and 28%. In two studies, the exact fat percentage was not reported, only that it was below 30%(Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Santos-Marcos, Haro and Vega-Rojas76) . Carbohydrate content varied between 13% and 78%. One study examined two study groups consuming the same energy% of carbohydrates, differing in glycaemic index (relative rise in the blood glucose level 2 h after consuming that food)(Reference Fava, Gitau and Griffin69). Protein content varied between 14% and 68% of total calorie intake in studies where the content was indicated. Eight of twenty papers reported the total calorie intake varying between 600 and 2684 kcal/d. Again, LFD interventions were very heterogeneous in macronutrient composition.
The percentage of protein in LPD interventions varied between 3% and 9% of total calories (Table 6). In two out of six papers, the carbohydrate and fat content are not reported(Reference Lai, Molfino and Testorio70,Reference Di Iorio, Rocchetti and De Angelis77) . The carbohydrate percentage of total calories varied from 4% to 62%, and fat percentages ranged from 32% to 87% of total calories. Furthermore, in two studies, supplementation of keto-analogues was used(Reference Rocchetti, Di Iorio and Vacca66,Reference Di Iorio, Rocchetti and De Angelis77) ; in another, inulin was supplemented(Reference Lai, Molfino and Testorio70).
Risk of bias
The risk of bias was assessed for randomised (Supplementary Figure 1) and non-randomised (Supplementary Figure 2) studies. Six out of twenty-four randomised trials were judged to be at high risk, thirteen at moderate risk and five at low risk of bias. Studies were classified as being at high risk of bias for different reasons, including not reporting potential cross-over effects in a cross-over trial, deviations from the intended intervention, and missing outcome data. Of the twelve non-randomised trials, four were judged as at high risk, four as at moderate risk and four as at low risk of bias. Most risks of bias were judged as moderate or high due to a lack of reported study procedures, by not mentioning any possible confounders or how confounding factors were controlled for. Blinding of dietary interventions is often not feasible, especially when participants must prepare their food. Therefore, the risk of bias arising from the randomisation process was often judged as moderate.
Outcomes
Change in alpha diversity of bacterial gut microbiota
Alpha diversity was reported in seven papers documenting the effects of LCD interventions (Table 7). No difference in alpha diversity was found after the intervention compared with baseline in all but one study, which examined just a small group of nine participants(Reference Gutiérrez-Repiso, Hernández-García and García-Almeida57), where a higher alpha diversity was measured after 2 months of an LCD. Alpha diversity was documented in eleven LFD intervention groups. Five studies reported increased bacterial diversity(Reference Gutiérrez-Repiso, Hernández-García and García-Almeida57,Reference Cuevas-Sierra, Romo-Hualde and Aranaz78–Reference Wan, Wang and Yuan81) , whereas the other six groups measured no difference in bacterial diversity between baseline and post-intervention. In the study of Cuevas-Sierra(Reference Cuevas-Sierra, Romo-Hualde and Aranaz78), only men displayed an increase in diversity in response to LFD, whereas there was no change in women. Only one paper reported alpha diversity in response to an LPD intervention(Reference Rocchetti, Di Iorio and Vacca66) and found no difference between baseline and post-intervention. Overall, there is not much evidence that LCD or LPD interventions change alpha diversity, while an increased alpha diversity was measured in response to an LFD in several studies.
Alpha diversity change after dietary intervention compared with baseline
↑ significantly higher diversity post-intervention.
= non-significant difference in diversity post-intervention.
Change in the relative abundance of gut bacteria
The relative abundance of various bacterial taxonomic groups changed from baseline to post-intervention in response to the various diets (Supplementary Tables 1–3). However, changes in the abundance of a specific taxonomic group were often reported in just one paper. To provide a more accurate picture of the influence of diet on the relative abundance of bacterial groups as reliably as currently possible, only the taxa that were reported in at least three intervention groups will be discussed.
Eleven bacterial taxa were reported in three or more different LCD intervention groups (Table 8). These groups are part of five phyla: Actinobacteria, Bacteroidetes, Firmicutes and Verrucomicrobia.
Change in relative abundance of gut bacteria after a low-carbohydrate diet compared with baseline
Most studies documented a lower relative abundance of the phylum Actinobacteria in response to an LCD. Bifidobacterium was reported in nine study groups, of which seven had a relatively lower abundance in response to an LCD(Reference Murtaza, Burke and Vlahovich59,Reference Nagpal, Neth and Wang60,Reference Lundsgaard, Holm and Sjøberg68,Reference Duncan, Lobley and Holtrop72,Reference Brinkworth, Noakes and Clifton82–Reference Swidsinski, Dörffel and Loening-Baucke84) , while it did not significantly change in the other two(Reference Tagliabue, Ferraris and Uggeri62,Reference Duncan, Belenguer and Holtrop85) . Bacteria belonging to the phylum Bacteroidetes were often more abundant after an LCD(Reference Fragiadakis, Wastyk and Robinson56,Reference Gutiérrez-Repiso, Molina-Vega and Bernal-López58,Reference Basciani, Camajani and Contini67,Reference Lundsgaard, Holm and Sjøberg68,Reference Ang, Alexander and Newman74,Reference Pataky, Genton and Spahr75,Reference Ley, Turnbaugh and Klein86) . A minority of studies documented a decrease in the relative abundance of the genera Bacteroides (Reference Russell, Gratz and Duncan73,Reference Pataky, Genton and Spahr75) . Bacteria belonging to the Firmicutes phylum were generally reported to decrease after an LCD(Reference Gutiérrez-Repiso, Molina-Vega and Bernal-López58,Reference Murtaza, Burke and Vlahovich59,Reference Basciani, Camajani and Contini67,Reference Duncan, Lobley and Holtrop72–Reference Pataky, Genton and Spahr75,Reference Mardinoglu, Wu and Bjornson83,Reference Duncan, Belenguer and Holtrop85,Reference Ley, Turnbaugh and Klein86) . Just the taxonomic sublevels Lachnospira (Reference Fragiadakis, Wastyk and Robinson56,Reference Lundsgaard, Holm and Sjøberg68) and Streptococcus (Reference Mardinoglu, Wu and Bjornson83) were reported to increase in some studies. The phylum Proteobacteria and its taxonomic subgroup Enterobacteriaceae were measured in response to LCD in six studies, showing no change in relative abundance except for two studies showing an increase(Reference Murtaza, Burke and Vlahovich59,Reference Ang, Alexander and Newman74) . The genus Akkermansia from phylum Verrucomicrobia was reported in three studies, with one reporting an increase(Reference Murtaza, Burke and Vlahovich59) and the others measuring no difference(Reference Nagpal, Neth and Wang60,Reference Swidsinski, Dörffel and Loening-Baucke84) by use of an LCD.
In summary, the currently available evidence suggests that an LCD impacts the relative bacterial abundance in our gut, inducing an overall decrease of Actinobacteria, an increase of Bacteroidetes and a lower or stable abundance of Firmicutes, while it generally does not appear to affect the relative abundance of Proteobacteria or Verrucomicrobia.
The relative abundance of twenty-three bacterial taxonomic groups was reported in three or more study groups at baseline and after an LFD (Table 9). These twenty-three groups originate from five phyla: Actinobacteria, Bacteroidetes, Firmicutes, Proteobacteria and Verrucomicrobia.
Change in relative abundance of gut bacteria after a low-fat diet compared with baseline
The effect of LFD on Actinobacteria and its subtypes varied, as three papers reported no difference in relative abundance(Reference Murtaza, Burke and Vlahovich59,Reference Nagpal, Neth and Wang60,Reference Kahleova, Rembert and Alwarith87) , one paper with two study groups reported an increase(Reference Fava, Gitau and Griffin69) and two papers reported a decrease(Reference Fritsch, Garces and Quintero20,Reference Fragiadakis, Wastyk and Robinson56) . Bacteroidetes, documented at the phylum level, increased in four study groups(Reference Fritsch, Garces and Quintero20,Reference Fragiadakis, Wastyk and Robinson56,Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Ley, Turnbaugh and Klein86) and were not different in another four groups(Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Ren, Zhang and Qi80,Reference Kahleova, Rembert and Alwarith87) after the use of an LFD. Fourteen papers reported change within its taxonomic subgroups in response to LFD, of which four (Bacteroides, Prevotella, Parabacteroides and P. distasonis) were reported by a minimum of three papers. The majority reported an increase(Reference Fritsch, Garces and Quintero20,Reference Fragiadakis, Wastyk and Robinson56,Reference Gutiérrez-Repiso, Molina-Vega and Bernal-López58,Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Fava, Gitau and Griffin69) or no change(Reference Nagpal, Neth and Wang60,Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Haro, Montes-Borrego and Rangel-Zúñiga65,Reference Fava, Gitau and Griffin69,Reference Cummings, Wiggins and Jenkins71,Reference Ren, Zhang and Qi80,Reference Wan, Wang and Yuan81,Reference Kahleova, Rembert and Alwarith87) in relative abundance; none reported a decrease. Changes in abundance of the phylum Firmicutes and its taxonomic members in response to LFD differed widely. Members of the family of Oscillospiraceae, Faecalibacterium and F. prausnitzii, showed an overall increase(Reference Fritsch, Garces and Quintero20,Reference Gutiérrez-Repiso, Hernández-García and García-Almeida57,Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Haro, Montes-Borrego and Rangel-Zúñiga65,Reference Fava, Gitau and Griffin69,Reference Wan, Wang and Yuan81,Reference Kahleova, Rembert and Alwarith87) or no difference(Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Haro, Montes-Borrego and Rangel-Zúñiga65,Reference Fava, Gitau and Griffin69) , while its member Ruminococcus decreased(Reference Fragiadakis, Wastyk and Robinson56,Reference Ren, Zhang and Qi80) or showed no difference(Reference Haro, García-Carpintero and Rangel-Zúñiga64,Reference Haro, Montes-Borrego and Rangel-Zúñiga65) . Many genera (Lactobacillus, Streptococcus, Lachnospiraceae and its taxonomic members Roseburia and Ruminococcus) decreased or remained unchanged in response to LFD, while an increase, decrease or no difference in abundance was reported for others, including Clostridium and Dorea (see Table 9 for references). Likewise, studies documenting the phylum Proteobacteria and its taxonomic unit Enterobacteriaceae yielded a decrease(Reference Gutiérrez-Repiso, Hernández-García and García-Almeida57,Reference Kahleova, Rembert and Alwarith87) or no difference in abundance(Reference Nagpal, Neth and Wang60,Reference Cummings, Wiggins and Jenkins71,Reference Kahleova, Rembert and Alwarith87) in response to an LFD compared with baseline. The abundance of the phylum Verrucomicrobia after an LFD was reported in three study groups and did not change in any of them(Reference Murtaza, Burke and Vlahovich59,Reference Nagpal, Neth and Wang60,Reference Kahleova, Rembert and Alwarith87) .
To conclude, current evidence paints a diverse picture of gut bacterial abundance in response to an LFD. Thus, conclusions regarding the impact of an LFD on the gut microbiome are difficult to draw at present, although some trends were observed, including the increase in several Bacteroidetes and its subgroups, a decrease in several Firmicutes subgroups (except for the family Oscillospiraceae and its taxonomic members Faecalibacterium and F. prausnitzii, which tended to increase), and a tendency of Proteobacteria and subgroups to decrease.
The change in relative gut bacterial abundance in response to an LPD was measured in only two studies (Table 10). One study had two arms using an LPD(Reference Lai, Molfino and Testorio70). Its impact on just two bacteria was reported in at least three study groups. Lactobacillaceae from the phylum Firmicutes decreased in response to an LPD in three study groups(Reference Lai, Molfino and Testorio70,Reference Di Iorio, Rocchetti and De Angelis77) , and Enterobacteriaceae from the phylum Proteobacteria decreased in two out of three study groups(Reference Lai, Molfino and Testorio70,Reference Di Iorio, Rocchetti and De Angelis77) . Thus, the scarcity of data documenting the gut bacterial response to LPD precludes any conclusion as to the effect of such a diet on the gut microbiome.
Change in relative abundance of gut bacteria after a low-protein diet compared with baseline
Note: Pre–post-intervention change in bacterial taxonomic levels that were reported by three or more studies are included in this table.
↓ significantly lower abundance post-intervention.
↑ significantly higher abundance post-intervention.
= no significant difference in abundance post-intervention.
Change in faecal metabolites
Many metabolites were reported in the included papers (Supplementary Tables 4–6). As with the relative abundance of species, we will only report the metabolites that were documented by at least three trials (or trial groups), which specifically concerned SCFA and lactate. Unfortunately, bile acids and tryptophan/indoles were not reported in three or more papers.
Faecal metabolite concentrations in response to an LCD were documented by seven papers (Table 11). The total SCFA concentration was reported in five of them, all showing a decrease after an LCD compared with baseline(Reference Ferraris, Meroni and Casiraghi17,Reference Russell, Gratz and Duncan73,Reference Brinkworth, Noakes and Clifton82,Reference Mardinoglu, Wu and Bjornson83,Reference Duncan, Belenguer and Holtrop85) . Six papers reported faecal acetate, propionate and butyrate concentrations, which consistently decreased in response to an LCD(Reference Ferraris, Meroni and Casiraghi17,Reference O’Keefe, Li and Lahti61,Reference Russell, Gratz and Duncan73,Reference Brinkworth, Noakes and Clifton82,Reference Duncan, Belenguer and Holtrop85) . Valerate concentration decreased in two studies(Reference O’Keefe, Li and Lahti61,Reference Duncan, Belenguer and Holtrop85) , while it increased or did not change in one other(Reference Russell, Gratz and Duncan73), depending on the measured concentration or proportion of SCFA. Isobutyrate was measured in four studies examining the effects of an LCD. Two studies did not find an effect(Reference Ang, Alexander and Newman74,Reference Duncan, Belenguer and Holtrop85) , one study demonstrated an increase in both tested study groups(Reference Russell, Gratz and Duncan73) and one showed a decrease in concentration(Reference Ferraris, Meroni and Casiraghi17) after the intervention. Faecal isovalerate concentration increased in one study in both study groups(Reference Russell, Gratz and Duncan73), decreased in one other study(Reference Duncan, Belenguer and Holtrop85) and did not change in yet another study(Reference Ferraris, Meroni and Casiraghi17). Lactate decreased in one trial(Reference O’Keefe, Li and Lahti61), with no difference in the two other trials(Reference Russell, Gratz and Duncan73,Reference Duncan, Belenguer and Holtrop85) .
Change in faecal metabolites after a low-carbohydrate diet compared with baseline
SCFA, short-chain fatty acids; BCFA, branched-chain fatty acids.
↓ significantly lower post-intervention.
↑ significantly higher post-intervention.
= no significant difference post-intervention.
Faecal metabolites were measured in three studies evaluating the effects of an LFD (Table 12). Acetate increased after an LFD compared with the baseline in two studies(Reference Fritsch, Garces and Quintero20,Reference O’Keefe, Li and Lahti61) and did not change in one other(Reference Fava, Gitau and Griffin69). The quantity of propionate and butyrate increased in one study(Reference O’Keefe, Li and Lahti61) and did not change compared with the baseline in the two others(Reference Fritsch, Garces and Quintero20,Reference Fava, Gitau and Griffin69) .
Change in faecal metabolites after a low-fat diet compared to baseline
SCFA, short-chain fatty acids.
↓ significantly lower post-intervention.
↑ significantly higher post-intervention.
= no significant difference post-intervention.
Just one study(Reference Ferraris, Meroni and Casiraghi17) measured metabolites in response to an LPD, so no conclusions can be made concerning the effect of LPD on metabolite concentration.
In concert, the available evidence suggests that faecal SCFA concentrations decline in response to an LCD, while it remains unclear if faecal BCFA concentrations are affected by LCD. An LFD may increase faecal acetate levels. Just one study examined faecal metabolite concentrations in response to LPD, which precludes meaningful conclusions regarding the effects of this dietary intervention.
Discussion
This systematic review summarises current data documenting the impact of dietary macronutrient composition on human gut microbiota. Gut bacteria play a pivotal role in host health through the biosynthesis of vital nutrients such as vitamins, essential amino acids and SCFA(Reference Bäckhed, Ley and Sonnenburg88). Dietary intake can reproducibly change the human gut microbiome(Reference David, Maurice and Carmody89), and knowledge of the impact of dietary interventions on gut microbiota composition and metabolic activity is important for understanding their health effects and safety. We summarise available data on the effects of carbohydrate, fat or protein restriction on alpha diversity, the relative abundance of taxonomic units of the major phyla, and faecal metabolites.
Alpha diversity
There is inconclusive evidence to support the notion that the alpha diversity of human gut microbiota is significantly altered by LCD or LPD. In contrast, diets low in fat increased alpha diversity in five out of twelve study groups, while there was no change in response to LFD in the other seven. Low-fat diets are necessarily (relatively) high in carbohydrate and/or protein content, and indigestible carbohydrates (fibres), in particular, are well known to impact gut microbiota(Reference Makki, Deehan and Walter90). However, the low-fat diets in the studies demonstrating a higher alpha diversity varied widely in macronutrient content, comprising both high or low carbohydrate or protein energy percentage. Therefore, the effect of LFD on alpha diversity cannot (exclusively) be explained by high contents of either (indigestible) carbohydrates or protein, which is in line with a previous review documenting the effects of dietary fibre on the abundance of Bifidobacterium and Lactobacillus spp. without significant impact on alpha diversity(Reference So, Whelan and Rossi91). Relatively low microbial alpha diversities have been linked to several acute and chronic disorders(Reference Gentile and Weir28,Reference Li, Zhou and Liang29,Reference Manor, Dai and Kornilov92) . Thus, the increase in alpha diversity that is generally observed in response to LFD interventions, particularly in people with metabolic disease, may confer health benefits. Notably, four out of five studies demonstrating an increase of alpha diversity in response to an LFD examined overweight or obese participants with or without type 2 diabetes, while only three out of seven showing no effect studied overweight or obese people. Obesity and metabolic disease are well known to be associated with low alpha diversity of the gut microbiome, and low baseline values provide more room for improvement. Thus, the currently available data on the impact of LFD on alpha diversity may well have been confounded by sampling bias.
Relative abundance of taxonomic units
The relative abundance of specific taxonomic units of gut bacteria varies widely between individuals, primarily driven by multiple environmental and lifestyle conditions, and alteration of relative abundance is not necessarily related to health outcomes(Reference Arumugam, Raes and Pelletier93). However, the relative abundance (or absence) of specific bacterial taxonomic units has been observed to relate to human health. Here, we will discuss our findings concerning the potentially relevant changes in relative microbial abundance in response to dietary intervention for taxonomic units per phylum.
Actinobacteria (Actinomycetota)
Actinobacteria are one of the four major phyla of the gut microbiota and, even though they represent only a small percentage, are pivotal in maintaining gut homeostasis(Reference Binda, Lopetuso and Rizzatti94). Bifidobacterium is a genus that, in healthy breastfed infants, dominates the intestine and has much lower but relatively stable levels in adulthood. The different species of Bifidobacterium that are present change with age, from childhood to old age(Reference Arboleya, Watkins and Stanton95). Bifidobacterium fulfils important functions in the human gut. Bifidobacterial genera are involved in the protection of the gut mucosal barrier, in the bioavailability of B vitamins, antioxidants, polyphenols and conjugated linoleic acids, and in the production of several short-chain fatty acids(Reference Rivière, Selak and Lantin96). Decreased numbers of Bifidobacterium have been associated with a variety of disorders(Reference Rivière, Selak and Lantin96,Reference Schwiertz, Taras and Schäfer97) , although one study also found high numbers of Bifidobacterium in an elderly nursing home population(Reference Ducarmon, Terveer and Nooij98). In seven out of nine included studies examining Bifidobacterium abundance, it declined in response to an LCD, which possibly could have unfavourable effects that could counteract the health benefits of carbohydrate restriction. The studies examining the impact of LFD on Bifidobacterium abundance produced highly variable results, while there is a lack of data on the effects of LPD on Bifidobacterium, precluding any conclusion as to the effects of either LFD or LPD in this context.
Bacteroidetes (Bacteroidota)
Bacteroides spp., which form ∼30% of human gastrointestinal microbiota(Reference Arumugam, Raes and Pelletier93), are acknowledged to play a critical role in gut bacterial colonisation and (host) health through their capabilities to metabolise (host) glycans, their role in protein metabolism, deconjugation of bile acids, modulation of immune responsiveness to infections and protection against various auto-immune disorders(Reference Eribo, du Plessis and Chegou99–Reference Wexler105). Because of their broad metabolic potential, the role of the Bacteroidetes in the gastrointestinal microbiota is complex. Reduced abundance of Bacteroidetes and its taxonomic subunit Bacteroides have been associated with obesity(Reference Ley, Turnbaugh and Klein86), inflammatory bowel disease(Reference Singh, Chang and Yan106,Reference Liu, Zhao and Lan107) and asthma(Reference Wang, Li and Liang108), while increased abundance is associated with type 1 and 2 diabetes(Reference Larsen, Vogensen and van den Berg109). The phylum Bacteroidetes and its taxonomic members were typically reported to increase in response to both LCD and LFD interventions included in this review. It has been speculated that the loss of body weight, which usually accompanies both carbohydrate and fat-restricted dietary interventions, could be responsible for the increase of Bacteroides spp. abundance in response to both LCD and LFD(Reference Fragiadakis, Wastyk and Robinson56), but several studies contradict this argument(Reference Gutiérrez-Repiso, Molina-Vega and Bernal-López58,Reference Cuevas-Sierra, Romo-Hualde and Aranaz78) . Recent genomic and proteomic advances have greatly facilitated our understanding of the uniquely adaptive nature of Bacteroides species(Reference Mills, Dulai and Vázquez-Baeza110,Reference Aggarwal, Kitano and Puah111) . Nevertheless, given the previously mentioned diverse biological features of this phylum, conclusions on health effects from the intervention studies presented here are hampered due to a lack of information.
Firmicutes (Bacillota)
A substantial part (∼40%) of the human gut microbiome comprises Firmicutes spp.(Reference Arumugam, Raes and Pelletier93). Members of this phylum generally contribute to host health by being involved in gut permeability, inflammation, glucose metabolism, fatty acid oxidation, synthesis and energy expenditure, partly through the production of butyrate and anti-inflammatory metabolites(Reference Sun, Zhang and Nie112). Indeed, the relative abundance of Firmicutes taxonomic units is decreased in people with several diseases. Faecalibacterium was, for example, decreased in non-alcoholic fatty liver disease, hypertension and gestational diabetes mellitus, and F. prausnitzii was decreased in type 2 diabetes, colorectal cancer, coeliac disease, inflammatory bowel disease and several other auto-immune disorders compared to healthy controls(Reference Sun, Zhang and Nie112,Reference Parsaei, Sarafraz and Moaddab113) . The relative abundance of most taxonomic members of the Firmicutes phylum seems to decrease in response to an LCD. The effects of LFD on the relative abundance of Firmicutes vary among taxonomic units of this phylum, with, for example, a decline of Roseburia and an increase of Faecalibacterium and its species F. prausnitzii. As LFD interventions appear to exert mixed effects on the abundance of distinct Firmicutes taxonomic units, their potential impact on (gut) health remains unclear.
Metabolites
SCFA produced by gut bacteria play a pivotal role in the gut as well as systemic health(Reference Nogal, Valdes and Menni114,Reference Koh, De Vadder and Kovatcheva-Datchary115) . Distinct SCFA can be fuel for intestinal epithelial cells, strengthen the gut barrier function, have immunomodulatory functions and improve glucose homeostasis, and may play protective roles against cancer and colitis(Reference Rivière, Selak and Lantin96,Reference Parada Venegas, De la Fuente and Landskron116) . SCFA are primarily produced by colonic bacteria through anaerobic fermentation of complex carbohydrates that escape digestion and absorption in the small intestine(Reference van der Hee and Wells117). Most of the studies reported a reduction of acetate, propionate and butyrate concentrations in faeces in response to an LCD, which is in concordance with literature describing an increase in SCFA by high-carbohydrate interventions(Reference Brinkworth, Noakes and Clifton82,Reference Mueller, Zhang and Juraschek118) . However, it should be noted that only SCFA not absorbed by the (healthy) host can be measured in faeces(Reference Blaak, Canfora and Theis49), and these results do not represent all SCFA produced in vivo.
LFD are often (relatively) carbohydrate-rich and, therefore, often (but not always) provide plenty of substrates for SCFA production. SCFA levels were indeed increased or stable in the majority of the included studies documenting the impact of LFD on faecal metabolite content. This is in accordance with the decrease in SCFA in high-fat interventions(Reference Brinkworth, Noakes and Clifton82,Reference Mueller, Zhang and Juraschek118) . Thus, the fact that SCFA tend to decline in response to LCD calls for careful consideration of the potential dangers of long-term LCD intervention. In particular, it seems prudent to make sure that the diet provides sufficient fibre (i.e. 25–30 g/d according to the Dietary Guidelines for Americans (https://www.dietaryguidelines.gov/) and many other international guidelines) if carbohydrates are restricted for longer periods to sustain adequate SCFA production.
Limitations
A major difficulty in interpreting the results of studies evaluating the effects of an isolated class of macronutrients in our diet concerns the fact that such a component is never consumed alone. Moreover, the considerable variability of compounds within the macronutrient categories can lead to variable effects, even when macronutrient levels are similar. Within the carbohydrate category, literature has demonstrated differential effects on the microbiome when comparing simple and complex carbohydrates(Reference So, Whelan and Rossi91,Reference Martínez, Kim and Duffy119,Reference Moszak, Szulińska and Bogdański120) . There is increasing but still limited knowledge of the relationship between the physiochemical structure characteristics and functional properties of non-digestible carbohydrates in the gut microbiome(Reference Rastall, Diez-Municio and Forssten39). Both increases and decreases in fibre content seem to alter the gut microbiota(Reference So, Whelan and Rossi91,Reference Moszak, Szulińska and Bogdański120,Reference Siracusa, Schaltenberg and Kumar121) , and various types of dietary fibre have exhibited functional distinctions in their impact on the composition of human faecal microbiota(Reference Lai, Molfino and Testorio70,Reference Martínez, Kim and Duffy119) . In the protein category, the source of protein, whether animal or plant based, has also been shown to exert varying effects on the gut microbiome(Reference Scott, Gratz and Sheridan26). Additionally, distinctions emerge when considering the fat content, in which unsaturated versus saturated fats demonstrate differential effects on the gut microbiome(Reference Singh, Chang and Yan106). Moreover, specific types of polyunsaturated fatty acid or saturated fatty acid(Reference Gutiérrez-Repiso, Hernández-García and García-Almeida57,Reference Lundsgaard, Holm and Sjøberg68,Reference Fu, Wang and Gao122) can have divergent impacts. Furthermore, dietary availability or supplementation of specific compounds in the diet, such as polyphenols(Reference Cardona, Andrés-Lacueva and Tulipani123,Reference Edwards, Havlik and Cong124) and keto-analogues(Reference Rocchetti, Di Iorio and Vacca66,Reference Di Iorio, Rocchetti and De Angelis77) , can affect the composition and function of the gut microbiome. Polyphenols are thought to influence carbohydrate metabolism at many levels, including inhibition of carbohydrate digestion(Reference Hanhineva, Törrönen and Bondia-Pons125), influence fat metabolism via the interaction with bile acids(Reference Han, Haraguchi and Iwanaga126) and affect protein metabolism through the phenolic compounds binding influence to protease activity and protein substrate accessibility(Reference Cirkovic Velickovic and Stanic-Vucinic127). Caloric content varied across studies, influencing the quantity of consumed macronutrients, and very low caloric content could affect the gut microbiome independent of macronutrients(Reference Ducarmon, Grundler and Le Maho128). Thus, the type and amount of (other) nutrients and availability of other compounds in each of the specific dietary interventions that were examined in the studies included in this review may have influenced the results. Moreover, the included studies turned out to be very heterogeneous in terms of participant features (healthy or sick, normal weight or obese), age, duration of the interventions and outcome data (highly variable taxa). In this context, it is also pivotal to acknowledge the gut microbial community the macronutrients are introduced into and the microbiota’s metabolic potential to utilise such substrates, as the maturity and metabolic potential of the gut microbiome varies throughout life and with health status(Reference Ottman, Smidt and de Vos129,Reference Turnbaugh, Hamady and Yatsunenko130) . There is also accumulating evidence that gut transit time is a key factor in shaping gut microbiota composition and activity, which are linked to human health(Reference Procházková, Falony and Dragsted131). These factors may have affected the included outcomes and, therefore, complicate drawing uniform conclusions. This review did not differentiate between the methodologies used in relation to collection, fixation, storage, shipping, extraction, library preparation, sequencing and bioinformatic processing. As none of these steps is standardised, the variability created among different studies for each of these steps may cause bias(Reference Ducarmon, Hornung and Geelen132,Reference Costea, Zeller and Sunagawa133) , which made risk-of-bias assessment of sample collection and processing of samples challenging. Since there is a lack of access to samples from different sites of the intestine and only faecal samples are available, it is not possible to fully unravel the influence of an intervention on the complete gut microbiome(Reference Whitney, Rolfes and Crowe38). Finally, papers often reported only taxonomic units that changed in response to a particular intervention, excluding critical evaluations of unaffected species at the endpoint. Thus, although our review appears to unveil the effects of the restriction of distinct dietary macronutrients despite all these caveats, its results need to be judged in the context of these (partly unavoidable) limitations.
Recommendations for future research
To create a complete overview of the effects of dietary restriction of specific macronutrients on the gut microbiome and its metabolites, it is important to provide a comprehensive and integrated analysis of the microbiome and metabolite changes induced by dietary interventions, where not only taxa and metabolites exhibiting significant change are reported. It is also important to provide detailed information on the diet, including caloric content, the quantity of all macronutrients and the availability of specific compounds such as polyphenols. To enhance the adequacy of the interpretation of data from studies examining the effects of macronutrient restriction, it is important to recognise the potential influence of fibre and caloric content on the gut microbiome. Therefore, researchers could strive to maintain fibre and calorie intake close to what is consumed at baseline, thereby minimising the risk of bias by these dietary characteristics. Participant features should also be described in detail, including health status, and preferably, more extended information should be shared, such as individual transit time. To reduce bias created by the variability in the methodology of sample processing, it could be interesting to obtain raw sequence data for all the studies and then uniformly process them bioinformatically so that at least variation in that step would be removed. Furthermore, 16S rRNA gene amplicon sequencing, which was the most often used technique for microbiota profiling in nutritional studies, is somewhat limited, and the implementation of metagenomics for gut microbial community analysis will allow the generation of in-depth knowledge on the microbial community dynamics as well as the metabolic potential of specific microbial communities(Reference Livingstone, Ramos-Lopez and Pérusse134,Reference Dao, Sokolovska and Brazeilles135) . Thus, future studies should use integrated advanced metagenomics and metabolomics analyses to foster our understanding of the impact of manipulating dietary macronutrients on gut microbiota and its metabolites.
Conclusions
We have reviewed available studies evaluating the impact of the restriction of distinct dietary macronutrient components on gut microbiota composition. The results, which must be assessed in light of certain limitations, suggest that carbohydrate restriction reduces the abundance of several health-promoting bacterial species as well as the faecal concentration of SCFA. In contrast, low-fat diets appear to have opposite effects on SCFA production and relative abundance of health-promoting bacteria, which is in line with current knowledge on the effect of the fibre content of the diet on the gut microbiome. As to the impact of protein restriction on gut microbiome composition and metabolite production, there are not enough data to draw any conclusions to date.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0954422424000131
Acknowledgements
We thank J.W. Schoones for his contribution and guidance in formulating the systematic review search strategy. We thank Nienke A. Jansen for her contribution to title and abstract screening, full-text screening and quality assessment.
Financial support
There was no specific funding for this review, but the salaries of E.B. and M.S. (PhD students) were funded in the context of the FIT (Fasting in diabetes Treatment) trial. The FIT trial was co-funded by Health∼Holland, Top Sector Life Sciences & Health and the Dutch Diabetes Foundation. The study funders had no role in the design, analysis or writing of this manuscript.
Competing interests
None.
Authorship
The authors’ responsibilities were as follows: M.S., P.P., H.P. and M.N. developed the overall research plan; M.S. devised and executed the literature search; M.S. and N.J. or P.P. performed the title/abstract as well as the full-text screening; M.S. and N.J. or P.P. performed the risk of bias; M.S. was responsible for the construction of data tables and summary and reporting of results; P.P., E.B., H.P., M.N., Q.D. and M.W. reviewed and edited the manuscript; M.S. had primary responsibility for final content; all authors read and approved the final manuscript.
Data sharing
Data described in the manuscript will be made available upon request pending. All proposals requesting data access will need to specify how the data will be used, and all proposals will need the approval of the trial co-investigator team before the data will be released.
Open access
