Among the biological sources of variability in responses to FF consumption, biological sex is one of the most studied, with 23 publications identified from the literature search (Table 1). The majority of the articles were from observational studies, with only two articles focusing on interventions for MetS. Diverse outcomes have been studied, ranging from MetS, associated parameters and metabolic pathologies (n = 9, BMI, type 2 diabetes (T2D) blood lipids, blood pressure) (21–29), to psychological disorders (n = 6) (30–35), cancer (n = 4, colorectal, gastric, bladder) (36–39), gut microbiota (n = 1) (40), or aging (n = 2) (41, 42). These studies were conducted in populations from different origins, with mostly one article per country (North of Europe: Norway, UK, Netherlands, Denmark, Poland; South of Europe: Spain; Asia: Japan, Korea; South America: Uruguay; Oceania: Australia). There were two articles implemented in Finland, Sweden, and France, and four in the USA. The sample sizes of the populations studied in these publications also varied, ranging from tens of volunteers in interventional studies to cohorts of nearly 100,000 participants. In terms of FF studied, the majority of them were dairy products, with yogurt being the most studied FF.

We noted that biological sex is rarely, if ever, studied in isolation, and variability of response attributed to biological sex is generally not considered a primary outcome. It's generally used as an adjustment parameter, together with BMI, for instance (29). It serves as a means to highlight the significant effects of FF on targeted outcomes when variability within cohorts is too great to allow for the measurement of a significant effect of FF at the overall population level. The consequence of this is that several biases can occur: comparison between males and females is conducted between cohorts that can be different also based on other parameters [e.g., the Health Professionals Follow-up Study (HPFS) cohort includes males, while the Nurses' Health Study (NHS) cohorts include females (21)]; inclusion of populations from distant places with varied nutritional habits and lifestyles, for example, Norway and Australia (25). Additionally, baseline health or physiological characteristics (e.g., BMI) and lifestyle habits (e.g., alcohol consumption and smoking) often differ between males and females, with females generally exhibiting higher baseline inflammation levels. This may explain some differences that are not strictly sex-related sensu stricto [e.g., see (24, 41)]. Therefore, it remains challenging to establish a general pattern for the role of sex in the differential responses observed.

However, a few sex-related patterns did emerge from our literature review results (Table 1). For metabolic outcomes, in females, yogurt consumption was positively associated with a reduced risk of MetS, a decreased BMI (21), lower inflammation (29), and weight loss (24) in normal-weight individuals. In males, it was associated with decreased weight in overweight people (24). Cheese consumption was associated with improvements in the lipid and glucose profiles, as well as a decrease in cardiovascular risk, based on various biomarkers across different studies (22, 23), although this association was not consistently observed [e.g., see (24)]. However, butter consumption was associated with lower blood glucose levels only in males (22). Bread consumption had a small effect on the serum uric acid (SUA) levels in females, while it decreased lipids and SUA in males in Norway (no effect on SUA in Australia) and had a general positive effect on type 2 diabetes (T2D) in males (25).

For psychological disorders, in females, a contrasting effect of yogurt according to fat content was observed, with a negative effect of low-fat yogurt consumption on depressive symptoms (32), whereas whole-fat yogurt was beneficial. In the same study, no effect of yogurt, regardless of fat content, was observed in males. However, in another study, yogurt was demonstrated to be associated with decreased depression and anxiety in males (with a mild effect on anxiety in females) (33). Butter and fermented milk beverage consumption patterns were also modified in females with mild cognitive impairment compared to females with normal cognitive functions (35). This was not the case for males. Finally, natto consumption was associated with decreased dementia in females only (34).

In the field of cancer research, yogurt was associated with a decreased colorectal cancer development in males only (36), whereas butter was linked to an increased risk of bladder cancer in females only (37). In females, bread consumption was associated with an increased risk of colorectal cancer (38), and miso soup with a decreased gastric cancer risk (39).

Finally, in musculoskeletal health studies, yogurt intake was associated with an increase in muscle mass in males (41). Conversely, cheese consumption was associated with a decrease in handgrip strength in females, and bread consumption was associated with a decrease in handgrip strength in males (42).

Based on these data, variability in responses between males and females has been demonstrated across various studies and for various outcomes. However, it is also clear that the variability is rarely consistently confirmed. One suggestion, considering the studies highlighting a sex differential response to FF (observed frequently in dairy foods in observational studies), is that additional studies with a higher number of participants and increased precision in FF intake assessment, as well as identification and adjustment for potential sex-specific confounders, are needed. Indeed, even in studies examining MetS parameters (the most extensively studied outcome), the number of publications is too low to draw a definitive conclusion. This is further explained in the corresponding articles: when studied, a significant sex-specific response is not always consistently observed; when such a response exists, it may fail to reach significance due to an inadequate control for potential confounding factors. This was also observed in the present work, even if the studies were discarded from the search because they were non-significant, for example, see (43) on MetS or (44) on all causes of mortality.

Although a few underlying mechanisms of action may contribute to the observed sex differences, the majority of explanations found in the literature relate to potential confounding factors. These include incomplete evaluation of alcohol consumption [that can be higher in males but not necessarily estimated (40, 41)], BMI at baseline (27), differences of intakes (in nature and quantity) between sexes, general inadequate evaluation of the FF intake (and impossibility to evaluate differences between males and females on this parameter), and undiagnosed pathologies or metabolic shifts at baseline, including microbiota dysbiosis (32). Only a few studies have mentioned that some sex differences in metabolism could explain the differential response to FF on various health outcomes, including higher susceptibility of females to inflammation-related tumors (36); differential calcium homeostasis in females due to sex hormones (37); and differences in rectal mucosa between males and females (38). However, even in these latter studies, the potential mechanisms of action remain hypothetical and have not been tested.

Three studies (two observational and one interventional) demonstrated a differential response to FF in populations from various geographical origins, as well as between black and white populations, or Caucasian/non-Caucasian populations (Table 2). The observational studies included a cross-sectional study by Zykova et al. (25), conducted in Australia and Norway, and a case-cohort study by Rosen et al. (47) in the USA, both of which focused on pregnant females with singleton pregnancies. The interventional study by Wolever et al. (45) conducted in Canada compared Caucasian/non-Caucasian populations. Finally, a meta-analysis compared the impact of yogurt consumption on colorectal cancer incidence, using data from studies conducted in various countries worldwide (including America, Asia, Africa, and Europe) (46). Here, in addition to ethnicity, variability between obese and non-obese subjects, as well as between males and females subjects, was also examined simultaneously. The studies investigated yogurt, cheese, and/or bread, assessed either through dietary assessments or direct interventions. Key outcomes varied widely between these studies, ranging from SUA levels, glycemic response, and vaginal microbiota composition (and associated risk of pregnancy outcome) to colorectal cancer.

Zykova et al. (25) highlighted distinct responses between Norwegian and Australian populations to yogurt. A more favorable effect observed in the Australian cohort was most likely attributed to a more extensive consumption of live probiotic bacteria in yogurt preparations compared to Norway at baseline, suggesting that the difference in response between the two cohorts was more linked to variation in nutritional habits than to an impact of ethnicity or location of the study per se. Differential responses were also observed following the consumption of high-fiber bread (Table 2). Similar to other studies, dietary habits were suggested as a possible factor in differential responses, with lower fat intake reported in the Norwegian cohort, which may influence intestinal lipid absorption mediated by dietary fiber. In the same line, the fact that yogurt consumption was associated with a lower risk of cancer in Europe and Africa but not in Asia and North America was not due to ethnicity or location per se but to the imbalance in the number of studies carried out in the various continents (more studies in Europe) and variable nutritional habits (and the quantity of yogurt ingested) between countries (46).

The Rosen et al. (47) study showed that yogurt consumption was linked to a higher likelihood of harboring a beneficial Lactobacillus crispatus vagitype rather than Lactobacillus iners, resulting in a lower risk of preterm birth. While slightly stronger associations were found among Black females, the overall effects were not significantly different between white and black females. When Wolever et al. (45) examined the blood glucose area under the curve and the glycemic index in response to white bread intake, the effects were significantly different across the ethnic groups. More specifically, non-Caucasians exhibited a greater glycemic response and a higher glycemic index compared to Caucasians, suggesting possible genetic differences in salivary amylase gene copy number. This gene variation, associated with increased salivary amylase activity, may impact starch digestion in the mouth and potentially gastrointestinal (GI) function.

In our literature search, we did not find studies that showed a significant differential health status response to FF in females populations during pre- and post-menopause. The outcomes generally studied in this field are hormone-related cancers or osteoporosis. For these outcomes, some studies failed to show variable responses following higher consumption of FF [e.g., fermented soy paste associated with reduced risk of breast cancer in both pre- and post-menopausal females (48)]. In other cases, data were too scarce (publications available only as an abstract) to evaluate whether risk factors for developing osteoporosis varied between pre- and post-menopausal females following the intake of FF (namely, hard cheese and yogurt) (49). However, some studies have shown a differential response depending on menopausal status following coffee intake (50) (described in part 3.3.1).

Similar to biological sex, age is commonly examined as a clustering variable, often in conjunction with BMI, dietary patterns, and health conditions. Among the studies identified in our search and evaluating age as a potential source of variable response to FF, only three reported a significant differential health response to FF depending on age (Table 3). Two observational studies investigated the impact of fermented dairy products (yogurt and cheese) and legumes (natto and miso) on psychological disorders, respectively (30, 34). Sun et al. (30) have shown a negative association between yogurt intake (< 131 g/day) and depressive symptoms in populations 60 years old and above, which was not the case for younger populations. Murai et al. (34) showed a sex/age combined effect, reporting an inverse relationship (significant or trending) between the risk of developing dementia and the intake of natto and miso in females under 60 years of age. This was not the case in males of all ages and females above 60 years old.

The proposed mechanisms of action by the authors include the presence of macro- and micronutrients in fermented dairy products and nattokinase in natto [e.g., see (52)] as possible explanations for the beneficial effects of the two categories of FF on psychological disorders. However, no explanation is provided or hypothesized to explain why females may be more responsive to these foods than males. Finally, a Cochrane meta-analysis examining the impact of chocolate and chocolate products on blood pressure found a slightly lower beneficial effect of cocoa on blood pressure reduction in populations aged 50–73 years compared to those aged 18–49 years (51). These data come from short-term interventional studies only. The lower response to cocoa with age was biologically explained by an age-related increase in arterial stiffening. In the short-term studies included in the review, an age-related decrease in vascular reactivity could limit the capacity to respond to the physiological stimulus (chocolate and its associated flavanols).

We noticed that age and stage of life are among the most studied sources of variability factors and are often considered potential confounding factors (data generally available in cohorts) that can be responsible for differential responses to food intake or metabolic challenges (5). This can be explained by age-related decreases in metabolic flexibility in response to the metabolic challenge that meal intake represents (53). In females, menopause can also overlap with (and be partially the cause of) this loss of flexibility. Surprisingly, very few studies have demonstrated significant differential responses to FF depending on the age of individuals, possibly because of the narrow age range of subjects enrolled in interventional studies. This can make sense if one considers the fact that, beyond age, the health status of each individual is more responsible for the degree of response of physiology and metabolism to a nutritional challenge. The variability of response in a category of age (e.g., above 60 years old) with variable health status may obscure any potential age-related differential response to nutritional challenges, including FF. However, differential age-dependent regulation of the circulatory levels of metabolites (some of which serve as markers or sentinels of our health status) was observed following the ingestion of a challenge meal (yogurt) (54). A more detailed study would be worthwhile since age has one of the highest potentials for further development of more targeted nutrition recommendations.

Several interventional studies reported variability associated with BMI and obesity. Pei et al. assessed the effects of low-fat yogurt consumption over 9 weeks in obese and non-obese individuals, examining biomarkers of chronic inflammation and gut permeability in both the fasted state (57) and fed state (58). Across these companion studies, lower tumor necrosis factor alpha (TNF-α) levels and TNF-α/sTNF-RII (TNF receptor II) ratios were observed in both obese and non-obese individuals following yogurt intake, indicating a reduced inflammatory status (57). This was accompanied by increased plasma IgM EndoCAb (immunoglobulin M endotoxin core antibodies) and decreased LBP/sCD14 (lipopolysaccharide-binding protein/soluble cluster of differentiation 14) levels, regardless of obesity status, suggesting a protective effect of this FF against chronic endotoxemia (57). These findings were further supported in the fed state by a lower incremental area under the curve (iAUC) for the LBP-to-sCD14 ratio and IL-6 concentration (58). However, differential responses to yogurt were also observed depending on the obesity status of volunteers. In the fasted state, yoghurt fed obese individuals exhibited lower diastolic pressure and higher plasma levels of 2-arachidonoylglycerol, which are markers indicative of improved intestinal barrier function, compared with obese controls. However, no similar effects were observed in normal-weight volunteers. In the fed state, pre-meal yogurt consumption resulted in a 72% higher net iAUC of sCD14 in the obese group (compared with consumption of a control snack), while non-obese individuals showed no significant response difference (treatment × obesity: P = 0.03). These results, although modest compared to the overall effects of yogurt on the measured parameters, suggest a more pronounced effect of yogurt in obese populations. Another interventional study investigated the effects of Streptococcus thermophilus-fermented milk in healthy or mildly hypercholesterolemic individuals with elevated LDL cholesterol level who consumed the product once daily for 12 weeks (60). Significant reductions in malondialdehyde-modified LDL (MDA-LDL), the MDA-LDL/LDL-cholesterol ratio, systolic blood pressure, and diastolic blood pressure were observed in participants receiving the S. thermophilus-fermented milk compared to controls. Notably, these reductions were statistically significant only among individuals with baseline MDA-LDL levels above the median, suggesting that yogurt may be more efficient in populations with already elevated MDA-LDL levels. The authors proposed that long-term suppression of oxidative stress by S. thermophilus-fermented milk could contribute to blood pressure reduction in individuals at a higher risk of cardiovascular disease, particularly those with elevated MDA-LDL levels. A similar differential beneficial effect of dairy was observed in an observational-longitudinal 6-year follow-up study, where overweight males with higher yogurt intake (highest quartile: 1.1–4.5 servings/day) showed lower weight gain and smaller increases in waist circumference compared to those consuming less yogurt (lowest quartile: 0–0.2 servings/day) (24). This association was not observed in normal-weight populations. Notably, similar results were also observed for milk consumption, suggesting that the fermentation of milk may not be the primary cause of the observed effect. Conversely, a different pattern emerged in a 7–10 year longitudinal study of perimenopausal females, where cheese consumption above one serving per day was associated with reduced weight gain among females with a BMI of ≥25 (56). Among females with a BMI of < 25, consumption of cheese, milk, or sour milk yielded the same effects.

Regarding the different responses of populations in lipid profiles and glycemic parameters, other FF were studied, particularly apple cider vinegar. Hadi et al. (65) conducted a systematic review and meta-analysis to assess the effects of apple cider vinegar on lipid profiles and glycemic parameters. The analysis revealed significant reductions in total cholesterol (TC), fasting plasma glucose, and HbA1c (glycated hemoglobin) concentrations. However, no significant effects were observed on low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), fasting insulin, or the homeostatic model assessment for insulin resistance (HOMA-IR). Notably, as previously shown above with dairy products, the lipid-lowering effects were more pronounced in individuals with “non-healthy” or “metabolically disturbed” conditions, such as those with type 2 diabetes, as well as those undergoing interventions longer than 8 weeks. The effect of consumption of fermented soy was also evaluated in a 5-year follow-up observational longitudinal study with 7,552 volunteers on serum lipids. It showed that fermented soy and natto were capable of limiting the increase in cholesterol, non-HDL cholesterol, and LDL cholesterol in populations with a BMI of ≥25. In contrast, no significant effect was observed in populations with a BMI of < 25 (55).

Finally, the effects of cocoa, chocolate, and chocolate product consumption on blood pressure have been well studied in the literature, with three meta-analyses identified through our search (51, 66, 67). All three meta-analyses concluded an overall beneficial effect of cocoa on the reduction of blood pressure (with high heterogeneity between studies in the effect on systolic, diastolic blood pressure, or both). As summarized by Ried et al. (51), cocoa has a significant beneficial effect on reducing systolic blood pressure in the hypertensive population, with trends toward a similar but non-significant effect on prehypertensive and normotensive groups.

Our interpretation is that, as shown in nearly all the studies mentioned so far in this field, the beneficial effect of the FF studied (fermented dairy, cocoa, and apple cider vinegar) appears more pronounced in populations already presenting metabolic alterations or metabolic pathologies. Since the studies discussed above were interventional, one can argue that it may be more feasible to observe an improvement in a metabolic parameter or a biomarker that is out of its normal physiological range than to improve a parameter that is already at equilibrium and not supposed to be improved by any nutritional intervention.

Variable responses to carbohydrate-enriched foods have been reported in the literature, particularly following the ingestion of a frequently consumed fermented product, such as bread (see Table 4). Moazzami et al. (61) showed, in an interventional study, that whatever the bread tested (refined wheat vs. whole wheat bread), postprandial insulin response to the bread intake was higher in females, presenting higher blood concentrations of leucine and isoleucine and lower blood plasma sphingomyelins and phosphatidylcholines concentrations in the fasted state. These markers are associated with the early stages of insulin resistance development. This means that, in addition to the bread itself (which can contain more or less digestible carbohydrates), the metabolic profile of the study participants affects the insulin response. This was confirmed by Korem et al. (59) Kovatcheva-Datchary et al. (62), who demonstrated variations in blood glucose and insulin responses to white and/or whole breads that were not due to the different compositions of the breads. Indeed, Korem et al. (59) demonstrated that after 1 week of bread consumption (three meals of 50 g of available carbohydrate), high interpersonal variability was observed in the postprandial glucose response, which is attributed to the individual's fecal microbiome characteristics. This was also the case in the interventional study by Kovatcheva-Datchary et al. (62), who showed that microbial diversity in the gut could be modified by the kernel-based bread in some “responder” populations, whereas it was not or less the case in the “non-responder” population, with a Prevotella/Bacteroides ratio higher in “responders” than in “non-responders” after kernel-based bread intake. Another study by Li et al. (63) confirmed this concept in participants supplemented with fermented rye bran, where “responders” and “non-responders” groups were established based on their response in terms of a decrease in fasting blood glucose following 12 weeks of consuming rye bran bread. Participants presenting a decrease in fasting blood glucose of more than 10% were considered “responders.” The group of “responders” also presented a specific microbiota (higher baseline Bacteroides, increased Faecalibacterium and Erysipelotrichaceae_UCG.003) and altered secondary bile acids relative to “non-responders,” suggesting again a role of the individual gut microbiota (and its metabolism) in the variation of response of host glucose response to the fermented rye bran (and possibly the dietary fibers it contains). Variability of response to fiber-enriched breads (multi-fiber bread) has also been highlighted in another recent study (64), on metabolic flexibility assessed by respiratory quotient variation (as a proxy of the ability of metabolism to switch between fat and glucose oxidation) following a mixed-meal tolerance test. The variations in respiratory quotients before and after supplementation with the breads were not due to the nature of the bread itself but due to higher baseline fasting LDL-C and greater post-mixed-meal tolerance test triglyceride excursion (MMTT ΔTG). This suggests that the intervention was more beneficial to subjects presenting initial metabolic dysregulations. The analysis of the data does not allow us to evaluate the potential role of gut microbiota in the adaptations mentioned above, even if a companion study has shown an alteration of gut microbiota composition (and function) by the same multifiber bread (69).

In conclusion, bread appears to have been more extensively studied due to its potential to produce products with variable contents of starches and carbohydrates, presenting varying degrees of digestibility (i.e., more or less digestible fibers). However, the glycemic response to these breads depends more on the health status (and potential susceptibility to developing insulin resistance) and the microbiota diversity and composition of the host than on the fact that the food used to test glycemic response is fermented.

Rintamäki et al. (31) investigated, in an observational cross-sectional study, whether cheese, tea, and fermented milk product intake could differ in populations suffering from depression, anxiety, or alcohol use. As individuals with anxiety disorders can also present increased or decreased BMI, the interaction between fermented dairy products, BMI, and the development of these psychological disorders should be studied in more detail. In the field of cancer, Kawakita et al. (68) found an inverse association between higher yogurt intake and a decreased risk of erodigestive tract cancer in an observational case-control study. This was not observed for butter and milk, suggesting a specific effect of yogurt. As for other outcomes presented earlier, the decreased risk was observed in populations with a BMI of >20, suggesting that the beneficial effect of yogurt to decrease the risk of digestive tract cancer is present in normal weight but not in lower weight populations (a population that may, due to its corpulence, consume less food, and yogurt).

For tea and coffee, the final number of selected studies is 22 (19 meta-analyses, two systematic literature reviews, and one analysis combining three large cohorts). The majority of the meta-analyses and literature reviews selected in our search focused on the health effects of coffee but also addressed the question of variability of response (Table 5). The majority of the studies concerned cancer outcomes, including digestive, biliary tract, esophageal, renal, breast, hepatic, prostatic, ovarian, and lung cancers (50, 70–80). Other studies examined metabolic pathologies (MetS, Metabolic Dysfunction Associated Steatotic Liver Disease or MASLD, and hypertension) (85–89), urinary disorders (81, 82), hip fracture risk (91), and psychological disorders (90). A few studies have focused on the impact of coffee on functions such as inflammatory responses and pathologies, as well as C-reactive protein (CRP) levels specifically (83, 84).

Based on the results of these meta-analyses, it is challenging to draw unequivocal conclusions about the significant beneficial or detrimental effects of coffee intake (or increased intake) on the outcomes listed above. This is due either to the high variability between studies or to the fact that different meta-analyses examining similar outcomes arrived at inconsistent conclusions. For instance, in the case of cancer, as summarized by Pauwels and Volterrani (92), if coffee intake is associated with a lower risk of all-cause mortality [as also described by Poole et al. (93)], the picture becomes less clear when it comes to specific types of cancer.

The authors highlight important uncertainties both between and within studies that can explain the difficulty in drawing definitive conclusions. Consequently, because the high variability in responses within studies could mask potentially significant correlations, the authors conducted additional a posteriori analyses to further explore possible mechanisms of action. These analyses considered several sources of variability by (i) analyzing certain studies separately (e.g., based on study design or location of the population studied), (ii) stratifying data within studies according to specific population characteristics (e.g., biological sex and smoking status), and (iii) including these sources of variability as covariates in statistical models. This stratified analysis revealed that, for some cancer risks, the location of the study (Europe, America, or Asia), as well as biological sex and hormonal status in hormone-dependent cancers, may lead to different outcomes. For instance, while increased coffee intake is not clearly correlated with a higher/lower risk of breast cancer, a negative correlation has been observed in estrogen receptor-negative populations [though not confirmed by Lafranconi et al. (50)], European and American (but not Asian) populations [see (74), also not confirmed by Lafranconi et al. (50)], and in post-menopausal females (50). In conclusion, stratification can help reveal significant correlations; however, this approach reduces the number of studies included in the meta-analysis and, consequently, the power and reliability of the conclusions (as shown in the example cited above, where inconsistent results were obtained after stratification in meta-analyses conducted on breast cancer). This limitation is clearly highlighted in the study by Wong et al. (86), where the conclusion regarding the effect of coffee on MetS differed depending on analytical approach: no effect was observed when data were adjusted for biological sex, whereas significant beneficial effects of coffee were found for both males and females when stratification was applied. Therefore, the analytical method, the selection of studies, and the relative importance of some studies with large sample sizes can skew (or significantly impact) the significance and/or variability of the final results.

In the concluding remarks of the meta-analyses on the health effects of coffee intake, the authors emphasize the difficulty of accounting for all confounding factors. For cancer, even if some constituents of coffee are suspected to carry some antiproliferative, antioxidant, and anticancer properties [e.g., caffeine, polyphenols, diterpenes (cafestol and kahweol), and chlorogenic acids] (92), too many confounding factors are not taken into account, sometimes intertwined and not detailed in observational studies. These sources of variability are population characteristics (biological sex, age, physical activity, education level, body weight, abdominal obesity, genotype [e.g., CYP1A2 variants, estrogen receptors + or -, hormonal status (pre- or post-menopause)], location of studies (Europe, USA, Asia; more or less associated with ethnicity), study design included in the analysis (case-control and/or cohort and/or interventional study), sample size, utilization of food frequency questionnaires (FFQ) for evaluation of consumption only [quantity—generally the notion of cup is used, habits of coffee consumption (e.g., with or without milk/sugar)], lack of precision in coffee-making process [type of coffee used, brewing process, temperature of coffee ingested (and impact on esophageal cancer for instance)], life habits that could be correlated with coffee intake (e.g., diet, smoking, alcohol consumption), the type of disease (e.g., for a same organ, type of cancer), design of studies. If some of the variables can be included in models as confounding factors (e.g., biological sex and body weight), others are not detailed in the protocols (e.g., quantity of coffee, coffee-making process, and details on diet). All these parameters impact the quantity of available molecules present in coffee capable of driving the healthy/non-healthy effect. The description of the coffee intake is also closely intertwined with the geographical location of studies, as the composition and quantity of a cup of coffee vary between Europe, America, and Asia due to population-specific consumption habits. It should be noted that data are lacking for other populations with coffee consumption (Africa, South America). This issue concerning confounding factors applies to all FF studied above but is particularly evident in the case of coffee, possibly due to the importance of the factors involved or because of the larger number of studies that allowed for a more extensive analysis of these factors.

Either way, all authors agreed on the necessity of increasing the number of studies and data to minimize biases and facilitate a proper evaluation of the risks and benefits associated with coffee in various health outcomes mentioned above. However, when looking at the wide sources of variability, particularly when it comes to coffee intake, and because the majority of the evaluations are made using data from observational studies, a detailed evaluation/standardization of the notion of a “cup of coffee,” combined with information on the sources of beans and the processes of grinding, fermentation, and brewing, are needed.

Finally, we noticed that one of the fields of research where interventional studies have been carried out extensively on the impact of coffee on health and variability of response evaluation is the genetic capacity to metabolize caffeine, which is closely associated with some clearly identified genetic variants. As caffeine is hypothesized to have an impact on health on both an acute and long-term basis, variants in populations have been more thoroughly investigated. To explain important variabilities of responses observed in populations in terms of mortality risk, metabolic pathologies (T2D and cardiovascular), cancer, or psychological disorders to coffee intake, genome-wide association studies have been carried out, and genetic variants were found (near Cholesterol 7α-hydroxylase—CYP1A1/2 genes, Aryl Hydrocarbon Receptor—AHR genes, and Adenosine 2A Receptor [ADORA2A] gene). Specific alleles of these genes were associated with higher or lower coffee consumption, reflecting the differences in metabolic capacity or susceptibility to caffeine. This is caused by the capacity of these populations to metabolize coffee, as these genes (and others) are involved in caffeine metabolism and/or sensitivity/susceptibility to coffee/caffeine [for details, see (82, 87, 94)]. Other genetic variants have also been discovered in the field, but they have been studied or are being studied more recently.

Among the three observational studies identified in our search, health outcomes associated with yogurt, fermented milk, and bread differed according to genetic variants. Karami et al. (95) reported a genotype-specific association between yogurt consumption and renal cell carcinoma (RCC), observing an increased risk of RCC with higher yogurt intake among individuals with the wild-type CC genotype at single-nucleotide polymorphisms (SNPs) rs3118538 and rs10776909 in the Retinoid X Receptor Alpha (RXRA) gene. Zhang et al. (96) demonstrated variation in cardiovascular (CVD) outcomes based on lactase persistence genotype (rs4988235), wherein participants with high fermented milk intake, CVD, and CVD mortality risk were significantly higher in lactase-persistent individuals (CT/TT). Additionally, Westerman et al. (97) demonstrated variability in glycemic control response (HbA1c) to bread consumption modulated by the Transient Receptor Potential Cation Channel, Subfamily M, Member 2 (TRPM2) rs62218803 polymorphism, where alternate allele carriers experienced a significantly reduced effect compared to reference homozygotes.

Three intervention studies similarly highlighted genetic variability in responses to FF, including doenjang (Korean fermented soy paste) and cheese. Lee et al. (98) showed that 12-week supplementation with doenjang improved visceral fat catabolism, specifically in individuals carrying the mutant G allele of uncoupling protein-1 (UCP-1), which was potentially mediated by increases in free fatty acids and insulin. Similarly, Cha et al. (99) found that doenjang supplementation led to a significant reduction in visceral fat area, more pronounced among individuals with the mutant T allele of the peroxisome proliferator-activated receptor (PPAR-γ2) gene compared to wild-type carriers. Antioxidative responses (measured by oxygen radical absorbance capacity and catalase activity) also differed by genotype and were increased in wild-type C allele carriers. Rajendiran et al. (100) identified variability in triglyceride levels following cheese intake, linked to SNPs (NPC1L1-rs2073547, PPAR-rs6008259) and apolipoprotein E (APOE) isoforms, which affect both the magnitude and direction of the lipid response.

Collectively, these findings highlight genetic polymorphisms in genes such as PPAR-γ2, APOE, TRPM2, and RXRA, as well as those related to lactase persistence, as important modulators of cardiovascular, metabolic, and cancer-related outcomes or parameters linked positively or negatively to FF intake. Future studies should investigate the clinical relevance of these genetic variations and their impact on health responses to inform personalized dietary recommendations more effectively. An ongoing consolidation of data on genetic variants, specific FF types, intake patterns, and health outcomes is also necessary. Public health strategies could eventually integrate knowledge of the prevalence of genetic variants of interest, the clinical significance of responses, and the strength of evidence to prioritize dietary guidelines based on genetic predisposition.

The gut microbiota is a dynamic, spatiotemporal interface between diet and host health, facilitating the metabolism or transformation of dietary components into end-products of fermentation, such as short-chain fatty acids (SCFAs) (101). It plays a critical role in health and has been identified as altered in cardiometabolic diseases, cancer, and psychological disorders (102). There are significant interindividual differences in gut microbiota composition and function, with the subject accounting for the major source of variation, up to 70%, in composition (103). Host and environmental factors, such as sex, genetics, age, or geographical origins, contribute to the variable part (104). Recent studies have highlighted the contribution of gut physiology and environment (e.g., transit time, pH) to variations in gut microbiota (105). Diet is among the most extensively studied environmental factors influencing the gut microbiota, with the entire diet contributing the most to variation (106). All these variables may affect the gut microbiota, which, in turn, could differentially interact with the host and/or confound results in the context of comparative analysis.

Among diet components, FF items may contribute to overall (even if small) variation (107). To date, few studies have specifically investigated the association of FF, including yogurt and coffee, with the gut microbiome and health outcomes in large-scale studies, primarily in Western populations (108–110). However, no studies have examined factors associated with variation in this association. In addition, these studies differed in covariate adjustment, with some adjusting for overall diet quality to specifically disentangle the contribution of specific FF to microbiota variation beyond overall healthier dietary habits (108). However, studies on how variation in gut microbiota can explain differential responses to FF consumption are limited, partly due to a lack of power to stratify both the microbiome and the response to FF in intervention studies. More specifically, in this section, we will describe how FF consumption can affect the gut microbiota composition and function, leading to differential effects on health outcomes, and how variable responses to FF consumption can depend on the initial microbiota function and composition.

Overall, the consumption of fermented products containing live microbes consistently results in a transient increase of fermented-food taxa in the gut (108, 109, 111), with a dose-response effect. This has been mostly studied in the field of fermented dairy products (112) and rarely in plant-based fermented foods (113). The viability of microbial composition in these foods depends on factors such as pH, water activity, and nutrient composition, which influence their potential to colonize the gut upon ingestion and interact with resident communities (114). Recently, a large observational study (NHANES 2001–2018) reported a positive association between dietary intakes of live microbes and a variety of health outcomes, including lower systolic blood pressure, lower plasma triglycerides, and a lower BMI (115). Since many FF contain high concentrations of lactic acid bacteria (10⁵-10⁹ UFC/ml), they are an important source of live microbes, which could then promote a better health status (1). Besides live microbes, compounds produced in the product via fermentation processes may have a direct or indirect (through action on the host gut microbiome) effect on host parameters.

The gut microbiome of individuals plays a role in responses to fermented foods, leading to the concept of clinical “responders” and “non-responders.” For example, the consumption of a 5-strain fermented milk product for 28 days led to an improvement in gas-related symptoms in some people (“responders”) and not in others (“non-responders”) (116). In “responders,” fermented milk intake induced an increased abundance of Faecalibacterium prausnitzii, changes in bacterial motility and lysine degradation pathways, as well as enhanced methanogenesis, reducing intestinal gas volume. These findings suggest that F. prausnitzii may reduce gas by shifting microbial activities, promoting methane production, and stimulating the growth of beneficial bacteria. F. prausnitzii also alleviated symptoms induced by a flatulogenic diet, with bacterial species activating CAZymes for carbohydrate breakdown, likely metabolizing an excess of carbohydrates. The response difference between groups was attributed to baseline microbiota composition, with “responders” having a different initial microbiota profile (116).

One example of the effect of FF consumption on the gut microbiota and health differential response was indirectly shown by the measurement of plasma soluble CD14 as a biomarker of gut barrier function in a cross-sectional study from the two US cohorts, the Nurses' Health Study (NHS) and the Health Professionals Follow-up Study (HPFS) (40). Higher yogurt consumption (at least two cups per week) was inversely associated with plasma concentrations of soluble CD14 in males, but not in females; however, the mechanism is unknown. The absence of effect in females was attributed to the fact that females's fecal microbiota had a higher fecal pH, as well as a higher abundance of total bacteria, Bifidobacterium, and the Lactobacillus gasseri subgroup, possibly limiting the effect of ingested bacteria from yogurts due to a lower available metabolic niche.

While the majority of studies have focused on the fecal microbiome, a growing number of studies examine the effect of probiotic and fermented milk products on the small intestine microbiota, an intestinal location with poorly characterized host-microbiota interactions. Especially, lactic acid bacteria are metabolically well-suited for the small intestine environment, with the ability to metabolize simple carbohydrates (117). One study explored the activity and effect of two fermented milk products (yogurt containing starter culture bacteria Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus) and one containing a strain of L. rhamnosus (~10¹¹ CFU/day) on the ileal microbiome in patients undergoing ileostomy. The most abundant strains of both products (L. rhamnosus and Streptococcus thermophilus) were transiently detected (8.5 and 5.2% of the total microbiota, respectively, with relative abundances of up to 90%), exhibiting high intra- and inter-subject variability. Microbial taxa such as Peptostreptococcaceae were positively associated with a low abundance of ingested bacteria, possibly due to competition in the small intestine (118). Other microbiota located outside the gut can also be influenced by FF intake, as shown in an observational study involving 600 females, which reported a positive correlation between dairy consumption and vaginal Lactobacillus profile, potentially related to the risk of preterm birth (47). Thus, analyzing microbial compartments other than feces could lead to the identification of additional between-subject variation, which may be targeted with personalized nutrition approaches.

In addition to a direct effect of live microbe ingestion, the presence of nutrients and compounds within FF could be responsible for the overall FF effect, as well as the potential role of these molecules in the variability of the host's response to FF. There are very few studies on the potential specific role of the fermentation process (or the presence of specific molecules or microbial components in the product) on inter-subject microbiota variations. As explained above, individual variation in the host microbiota may also be responsible for a lack of effect of the nutritional intervention. Indeed, dietary acculturation to the Western diet among migrants is accompanied by a loss in gut microbiome diversity and function, leading to an inability to metabolize fibers, thus predisposing individuals to metabolic diseases (119). Consequently, gut-microbiota-targeted diets are suggested to compensate for this loss by enriching microorganisms with fiber-degrading activity, along with dietary fibers in the diet (120). Currently, it is unknown if fiber-rich FF containing live microbes capable of degrading dietary fibers could enrich and complement gut microbiota functionality.

The pioneering work by Zeevi et al. (121) aimed to assess the influence of food intake on gut microbiota and the impact of gut microbiota on glycemic responses. The authors demonstrated in a cohort of 800 individuals that the postprandial glucose response varied significantly among individuals and that gut microbiome composition was a significant explanatory factor. They developed an algorithm capable of identifying multiple metabolic parameters, dietary habits, physical activity, and bacterial taxa as either beneficial or non-beneficial. They were able to predict the postprandial glucose response to personalized dietary intervention. This protocol is among the first steps toward individualized nutritional recommendations, but it was not designed to address the specific effect of FF supplementation efficiency, even if the diets contained some FF (121). In another study, glycemic response to bread could be predicted from the baseline microbiome (59). The importance of microbial enterotype composition and function in the differential response to whole grain products with fermented rye bran, compared with refined grain, was also demonstrated in a study of Chinese adults (79 participants, 12 weeks of consumption). Suggested mechanisms included alterations in SCFA and bile acid metabolism as potential mediators of the observed beneficial effect of whole grain bread on glucose metabolism (63).

Some data are also available on dietary patterns rather than the specific effects of FF on different health outcomes. A recent study examined whether dietary patterns related to MetS differ according to gut microbial enterotypes among 348 Korean adults aged 18–60 years, recruited between 2018 and 2021, in a cross-sectional study. The authors suggested that dietary factors associated with MetS risk may differ based on the gut microbiomes of Korean adults. More precisely, the Bacteroides enterotype, mainly found in people eating more refined rice-based diets, and the Prevotella enterotype, more frequent in people with a low fermented food-based diet, were associated with an increased risk of MetS (122).

Links between a pathology, gut microbiota composition and function, and diet, including FF, can be highlighted in observational studies; however, causality links are not always possible to assess due to the study design. Indeed, in the study from Baragetti et al. (123), the authors showed associations between diet, changes in taxonomic gut microbiome, and subclinical carotid atherosclerosis (SCA), with subjects without SCA reported to consume a higher amount of cereals (P = 0.009), starchy vegetables (P = 0.027), dairy products and beverages (P = 0.004 and P = 0.016, respectively), yogurts (P = 0.047), and bakery products (P < 0.001) as compared to those with SCA. These dietary intakes were more closely correlated with bacterial genera in subjects with SCA compared to those without SCA. Data supported the interaction between dietary exposure and changes in gut microbiota at the early stages of SCA. A reduced contribution of pathways (such as starch degradation and sulfur oxidation, and the biosynthetic routes of purine and pyrimidines) encoded by F. prausnitzii was found in metagenomes of +Intima-Media Thickness (IMT)/+SCA subjects. Notably, F. prausnitzii has been previously reported to be actively involved in gut permeability through the production of the anti-inflammatory compound butyrate. To address causality, an increasing number of interventional studies specifically dedicated to FF consumption were recently implemented. Wastyk et al. (124) conducted an FF-enriched nutritional intervention (36 participants, 27 weeks in duration) where the quantity and diversity of FF were increased based on participants' preferences, including yogurt, kefir, fermented cottage cheese, kombucha, vegetable brine drinks, and fermented vegetables such as kimchi. A positive correlation was observed between the total number of servings and microbial diversity. Yogurts and vegetable brine drinks were the most commonly consumed FF, and a stronger correlation was observed with these types of FF. The authors suggest that FF consumption was not primarily due to the consumption of microbes but rather indirectly affected microbiota diversity by increasing the representation of strains present but below the detection level. FF intake was also associated with a decrease in markers of host inflammation. Despite significant advances and the fact that each individual served as their own control (interventional study), the study lacked a control arm (on a usual diet) or comparative data using non-fermented, corresponding products.

The distinction between the effect of fermentation and the ingestion of microorganisms together with the FF was illustrated recently in a randomized controlled trial comparing pasteurized vs. fresh sauerkraut using a crossover design (87 participants, 4-week consumption) (113). Despite the global absence of effects on the microbial profile in the two intervention periods, some specific outcomes were related to interpersonal variations. For example, significant variations in the relative abundance of species from the Lachnospiraceae family were observed among overweight or older participants. The changes in microbial diversity were also consistently smaller for participants with a higher baseline diversity. As expected, the main bacterium found in sauerkraut (Lacticaseibacillus paracasei) was found in fecal samples after consumption of fresh sauerkraut. Future studies comparing FF to different controls (non-fermented, fermented, and pasteurized) are needed to precisely assess the effects of fermentation (bioactive compound production) and possibly disentangle the effect of live microbes from dead strains.

The concept of “responders” and “non-responders” has also been demonstrated for parameters other than gut microbiota. In a randomized controlled study, the glycemic index of eight experimental food products, including two types of bread, a cake, a cookie, and a fruit drink, was assessed in 10 volunteers. Variability in the glycemic index was substantial, with iAUC for bread ranging from 18 to 182, indicating considerable variability in glycemic responses both within individuals and between individuals (125). The variability of responses to bread and associated issues is detailed in Part 3.2.2.

Genetic and epigenetic factors also contribute to individual variability in responses to FF, particularly genetic variations in enzymes involved in carbohydrate and fat metabolism. In an interventional study, the effects of dairy fat consumption on lipid metabolism were assessed (126). Participants consumed dairy fat daily from full-fat cheese, reduced-fat cheese, butter, butter with calcium caseinate, and a control. They were categorized into “responders” and “non-responders” based on changes in total cholesterol (TC), low-density lipoprotein cholesterol (LDL-c), and high-density lipoprotein cholesterol (HDL-c). “Responders” showed significant reductions in TC and LDL-c, with increases in HDL-c, while “non-responders” had increases in TC and LDL-c and decreases in HDL-c. Baseline lipid levels influenced response: “responders” had higher baseline TC, HDL-c, and lower triglycerides (TAGs), “non-responders”. The study suggests that lipid metabolism responses to dairy fat may depend on baseline characteristics, with higher cholesterol levels being associated with more significant reductions. The food matrix, particularly dairy fat in its natural form, also contributes to cardiovascular health benefits. These findings highlight the complexity of individual variation in dietary responses. Another study evaluated the validity of a global assessment tool for gastrointestinal (GI) wellbeing (127). Participants consumed either a probiotic fermented dairy product (containing Bifidobacterium lactis, Streptococcus thermophilus, Lactobacillus delbrueckii subsp. bulgaricus, and Lactococcus lactis) or a non-fermented dairy product over 6 weeks. Subjects were categorized into “responders” and “non-responders” based on improvements in GI wellbeing (reporting improvements for at least 2 weeks during the 4-week intervention). The study found that “responders” reported significant improvements in digestive symptoms compared to “non-responders.” Sensitivity analysis showed a decrease in digestive symptoms by 41.8% in “responders” and 10.9% in “non-responders.” Stool frequency improvements were no longer significant when a stricter “responder” definition was used. These findings highlight individual differences in response to probiotic interventions and support the validity of the global GI wellbeing assessment tool.

Food characterization is critical for identifying possible factors of the FF effects, with an increasing number of studies complementing traditional nutrient profiling by analyzing both microbial strains and metabolites. Among the studies identified in our systematic search (Figure 1), cereal products, primarily bread, were the most studied (n = 7), followed by dairy products, mainly yogurt (n = 21). Ingredients (macronutrients) of FF were defined in the majority of studies (n = 10 in total), while in five studies, the “basic” chemical composition of FF was defined. These studies included low-fat yogurt (53, 54), white and sourdough bread (59), fermented milk (60), and refined rye bread, whole-meal rye bread, and refined wheat bread (61). On the other hand, the technological process by which the FF was produced was described in only three publications, including the production process for rye and wheat bread (28), white and sourdough bread (59), and barley kernel-based bread (62).

In three studies, genera, species, or specific strains were identified. Lactobacillus bulgaricus and Streptococcus thermophilus were used as starters for milk fermentation (57, 58, 60). Two studies described bread fermentation by yeast and lactobacilli (61) and by yeast alone (62). Metabolites formed during fermentation that may contribute to the effect of FF were described in six studies. Carbohydrates and dietary fiber were identified as active components in three studies (28, 45, 62). Regarding protein components, various amino acids, including leucine and isoleucine (61), were suggested as bioactive metabolites. Ito et al. (60) concluded that components in S. thermophilus cells exert antioxidant activity. Finally, SCFAs and bile acids were analyzed as active metabolites in the study by Li et al. (63).

Safety issues were not mentioned in any study, likely because the microorganisms used for fermentation had a Generally Recognized as Safe (GRAS) status. Live microorganisms were present in the final product in approximately half of the studies, including fermented milk (57, 58, 60) and cheese (29).

Based on our literature search, 27 observational studies were identified. In some studies, only food categories were listed, such as soy products, dairy products, and cereals, while in the majority of the selected studies, specific FF were mentioned, including yogurt, rye bread, cheese, and others. In some studies, variability was only considered at the level of macronutrients. It is important to note that the majority of (large) observational studies used food frequency questionnaires (FFQ), which are more indicative of long-term dietary habits. Other, more precise diet assessments, such as 24-h dietary recalls or food records (which are short-term but more precise in terms of the nature of food eaten and quantities), were less commonly used in large cross-sectional studies. Additionally, the majority of the articles in observational studies grouped these foods into food groups or mentioned, for instance, dairy products without distinguishing whether they were fermented or not or did not report the frequency of consumption of specific foods. Additionally, no composition of raw materials was specified in the description of FF intake, no production process was defined, and there were no reports on the microorganisms (or their viability) in the products. Overall, it is frequently difficult to determine from observational studies whether the food that study participants refer to has been fermented and contains live microorganisms. Within and between countries, there are also variations in the same type of product (e.g., fermented or not, pasteurized, such as butter). The ambiguity becomes even greater when we consider that the study participants may not recognize certain foods as fermented (128).